5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen
Monday 13 May 2013 - Wednesday 15 May 2013 Technical University of Denmark
Book of abstracts
5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Tuesday 24 June 2014 Book of abstracts 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
Page 2 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
Book of abstracts 5th International Seminar on Railway Operations Modelling and Analysis RailCopenhagen2013 13-15 May, 2013
Page 3 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
2 - Stability of railway dispatching solutions under a stochastic and dynamic environment Presenter: Dr. QUAGLIETTA, Egidio (Delft, University of Technology); Dr. GOVERDE, Rob M.P. (Delft, University of Technology) Railway traffic is strongly influenced by random disturbances during operations which cause deviations from the original schedule and thereby reducing performances. In order to cope with small perturbations, the design of a robust timetable can be an effective solution; but if larger disturbances or service disruptions are observed, it is necessary to adopt real-time dispatching measures to reschedule (reorder, retime or reroute) train services into new updated conflict-free train path plans. The rescheduling process involves a first phase in which the traffic state is predicted over a pre-set time horizon (called prediction horizon) on the basis of current traffic information (e.g. train positions, speeds) that is communicated at irregular time intervals to the Traffic Management System. If conflicting train paths are detected, a second phase is activated to compute suitable solutions addressed to restore planned operations as quick as possible or to minimize impacts of conflicts on service availability. The effectiveness of a rescheduling decision not only depends on the operational strategy identified to solve detected conflicts, but also on the actual traffic state that holds when such solution is implemented. During the time lag between the last field measurement and the practical application of the rescheduling plan (the so called control delay), the traffic state could evolve differently from the prediction due to its stochastic and dynamic nature. The control delay depends on different factors such as the type of monitoring system installed (e.g. train describers, GSM-R) and the frequency of train data updates to dispatching centres, the time to predict conflicts and elaborate a new suitable scheduling strategy, as well as the time needed by dispatchers to evaluate and communicate such solution to the interlocking system and possibly to the train. Intuitively, the larger such delay the higher is the probability that a certain rescheduling measure will be ineffective and even counterproductive for mitigating effects of detected conflicts. In this case, the solution should be recomputed and adjusted accordingly to current conditions of the network. During real operations, control measures cannot be updated and implemented continuously, since from one side there are constraints on technical times to obtain and transfer new plans to the system, while from the other side there is a practical unfeasibility for dispatchers to manage different alternatives in short time periods. In this context, particular attention must be paid to the stability of rescheduling plans. A plan can be defined as stable when its (initial) structure is invariant to random perturbations occurring on the network within a given time period Δt. In other words, the first part of a stable control strategy is the same for successive computations after each Δt with respect to updated traffic information and can thus be confidently implemented. In literature only little work [1] deal with stability of dispatching measures under a stochastic and dynamic environment. Instead, numerous approaches have been proposed so far [2, 3, 4] for efficiently generating optimal schedules to minimize train delays, through an open-loop optimization process which primarily involves assumptions of certain and deterministic conditions. In practice, since no advanced decision support tool is generally available, traffic controllers take real-time decisions according to their own experience or rules-of-thumb, without taking into account potential new conflicts due to the stochastic evolution of traffic states [5]. This paper addresses the stability of real-time schedules within a stochastic and dynamic system. To this aim an innovative framework is developed which integrates the Alternative-Graph based tool ROMA [4] for computing optimal rescheduling plans with a stochastic microscopic model for simulating railway traffic. Optimal plans returned by ROMA for a given perturbed scenario are transferred to and implemented in the simulator as if this latter is the real field. An application to a real case study is performed to evaluate open-loop optimization solutions while considering random and dynamic changes in traffic conditions during the control delay. In particular an investigation is conducted to understand how sensitive schedule stability is with respect to the control delay and the prediction horizon. This also gives insight in the maximal time before occurrence of the expected conflicts that a stable solution can be pursued. Then optimal rescheduling plans are recomputed at regular time intervals with respect to current traffic states, assuming continuous updates of train information and the minimum technically possible control delay to put such plans into operation. These successive solutions are compared in order to understand how the optimal control strategy changes over time due to stochastic disturbances and dynamic evolution of traffic behaviour. Understanding the stability of successive dispatching solutions is a first step into closed-loop railway traffic control and is preparatory to an effective management of real-time perturbations. In a rolling horizon approach the stable part of a dispatching plan can be implemented at some point with confidence that this will be the optimal choice for any control decisions at a later stage.
References [1] Meng L., Zhou X., Robust Single-Track Train Dispatching Model Under A Dynamic And Stochastic Environment: A Scenario-Based Rolling Horizon Solution Approach, Transportation Research Part B, Vol. 45, pp.1080-1102, 2011. [2] Dorfman M.J., Medanic J., Scheduling Trains On A Railway Network Using A Discrete Event Model Of A Railway Traffic, Transportation Research Part B, Vol. 38, pp. 81-98, 2004. [3] Tornquist J., Persson J., N-Tracked Railway Traffic Re-Scheduling During Disturbances, Transportation Research Part B, Vol. 41, pp.342-362, 2007. [4] D’Ariano A., Improving Real-Time Train Dispatching: Models, Algorithms And Applications, PhD Thesis, Delft University of Technology, the Netherlands, 2008. [5] Sahin I., Railway Traffic And Train Scheduling Based On Inter-Train Conflict Management, Transportation Research Part B, Vol.33, pp. 511-534, 1999.
Page 4 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
Page 5 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
3 - Railway line capacity consumption of different railway signalling systems under scheduled and disturbed conditions Presenter: Dr. GOVERDE, Rob (Delft University of Technology) The characteristics of a signalling and automatic train protection (ATP) system have a significant impact on the capacity and stability of a railway line. The capacity consumption of a railway line can be calculated using the UIC timetable compression method for given infrastructure characteristics, rolling stock characteristics, and timetable pattern (UIC, 2004). This compression method is based on a deterministic microscopic calculation of conflict-free train paths with minimum headway times using blocking time theory (Hansen and Pachl, 2008). In addition, the UIC capacity method gives empirically derived guidelines on the total required buffer time in a timetable pattern to be stable for delays. However, this method does not consider the actual capacity consumption under disturbed conditions.
In the presence of delays, train path conflicts occur depending on the scheduled buffer time between train paths. In practice this means that a train has to brake in response to the signalling system and possibly wait in rear of a stop signal. This leads to changed train trajectories with increased blocking times so that following trains may also be affected. The actual response to the signals depends on the specific signalling system and constraints of the ATP system, and can be quite different. In a standard three-aspect signalling system, trains have to brake from two signals in advance of the occupied block to a restricted speed and continue with this restricted speed until a final brake before a red signal. In contrast, a cab signalling system as ETCS allows a train to proceed until an approach indication point that is determined by a dynamically computed braking curve onboard. This braking curve depends on the actual train speed and braking characteristics as well as on the infrastructure description until the end of the movement authority (such as slopes). Also the time of re-acceleration after an improved signal aspect depends on the (intermittent or continuous) ATP system.
This paper evaluates the capacity consumption on a railway line both under scheduled and disturbed conditions. Disturbed conditions are modelled as distributions for the delay of each train at the beginning of the line. For the scheduled condition the standard UIC compression method is used, while the computation of capacity consumption under disturbed conditions requires multiple simulation runs. The average capacity consumption of all replications is used as a measure of the capacity consumption for disturbed situations. Note that the disturbed capacity consumption depends on the (time allowances in a) given timetable, the assumed delay distributions, and the train and traffic control applied.
For the analysis we used the rescheduling tool ROMA (D’Ariano and Pranzo, 2009). ROMA is based on blocking time theory and is thus applicable to any signalling/ATP system. For this study ROMA was extended with different signalling/ATP systems so that the braking behaviour of hindered trains is correctly simulated in the different configurations. Also the sectional release route locking principle was taken into account for accurate blocking time calculations in the station areas (D’Ariano et al., 2009). ROMA was also used to compute the compressed timetable with conflict-free train paths without rescheduling. For the disturbed scenarios ROMA is used in a Monte Carlo simulation set up.
We consider a case study of the Dutch Utrecht-Den Bosch railway line with different signalling/ATP systems. Utrecht-Den Bosch is a mixed traffic line of about 40 km with intercity, local and freight trains, and six intermediate stops for the local trains. Halfway the line is an overtaking station where the intercity trains overtake the local trains and some of the freight trains also stop to be overtaken. Four signalling/ATP systems are considered: the current Dutch NS’54 speed signalling system with ATB train protection, NS’54/ATB with short blocks, ETCS Level 2 with existing block lengths, and ETCS Level 2 with short blocks. The disturbances are modelled as stochastic entrance delays for all trains with train-type dependent distributions given by three-parameter Weibull distributions fitted by empirical data. Two traffic control scenarios are considered for the disturbed scenarios: a basic traffic control with simple FCFS rules and the advanced rescheduling algorithms in ROMA that minimize the maximum consecutive delay (D’Ariano et al., 2007ab, 2008).
The results show that the scheduled capacity consumption decreases from NS’54/ATB to ETCS Level 2 with short blocks. The capacity consumption increases considerably for NS’54/ATB when delays are considered. This can be explained by the increasing blocking times when trains have to brake and run at lower speeds. For NS’54/ATB with short blocks the capacity consumption doesn’t change much; the smaller blocks cause quicker releases of the blocks by which the headway times do not increase much. For both ETCS cases the capacity consumption improves when delays are considered: the braking distances decrease when trains run at lower speeds, which has a stabilizing effect on headway times and the throughput.
References
Corman, F., Goverde, R.M.P., D'Ariano, A. (2009). Rescheduling dense train traffic over complex station interlocking areas. In: R.K. Ahuja, R.H. Moehring, C.D. Zaroliagis (Eds.), Robust and Online Large-Scale Optimization: Models and Techniques for Transportation Systems, LNCS 5868, Springer, Berlin, pp. 369-386.
D’Ariano, A., Corman, F., Pacciarelli, D., Pranzo, M. (2008). Reordering and local rerouting strategies to manage train traffic in real-time. Transportation Science, 42(4), 405–419.
Page 6 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
D’Ariano, A., Pacciarelli, D., Pranzo, M. (2007a). A branch and bound algorithm for scheduling trains in a railway network. European Journal of Operational Research, 183(2), 643–657.
D’Ariano, A., Pranzo, M. (2009). An advanced real-time train dispatching system for minimizing the propagation of delays in a dispatching area under severe disturbances. Networks and Spatial Economics, 9(1) 63–84.
D'Ariano, A., Pranzo, M., Hansen, I.A. (2007b). Conflict resolution and train speed coordination for solving real-time timetable perturbations. IEEE Transactions on Intelligent Transportation Systems, 8(2), 208-222.
Hansen, I.A., Pachl, J. (Eds.) (2008). Railway Timetable & Traffic: Analysis, Modelling, Simulation. Eurailpress, Hamburg.
UIC (2004). UIC Code 406: Capacity. International Union of Railways, Paris, 1 september 2004.
Page 7 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
7 - An online railway traffic prediction model Presenter: KECMAN, Pavle (Delft University of Technology) Real-time prediction of train positions in time and space is a basic requirement to effective route setting, traffic control, rescheduling, and passenger information. However, in practice only the last measured train delays are known in the traffic control centres and dispatchers must predict the arrival times of trains using experience only, without adequate computer support. This often results in a simple extrapolation of the current delays as the expected arrival delays. Some railways use a linear shift of the timetable to extrapolate the current delays to the future. This method neglects the fact that some trains may (partially) recover from delay using running time supplements, while others may get (more) delayed due to route conflicts. Better predictions could be obtained by microscopic simulation models but excessive computation times still prohibit online application of microscopic simulation to densely operated large-scale networks.
This paper presents a real-time model for continuous online prediction of train traffic using process mining, a method of analysing and extracting information about processes from event data logs using the process model. The process model is an acyclic timed event graph representing the precedence relations of the events including train runs and stops, connections, and minimum headways between events, as well as the scheduled event times. The arc weights are estimations of the process times based on historical event data of track usage and train paths together with actual train positions and routes from the actual route process plan. The level of detail can be selected from macroscopic with only train events at stations to microscopic with all signal passages and section releases. The timed event graph is updated regularly when new information is available on train positions or traffic control decisions, including changes in the route process plan. The predictions of the actual event times are computed by a critical path algorithm over the acyclic timed event graph, and incorporate the usage of running time supplements and buffer times, as well as time loss due to route conflicts based on a conflict detection scheme within the critical path algorithm (Goverde, 2010).
The historical event data is obtained from TROTS train describer records pre-processed by another (off-line) process mining tool (Kecman et al., 2011a). In this pre-processing step train numbers are coupled to occupation and release messages from track sections, and moreover, route conflicts are identified based on signalling logic and blocking time theory as described in Daamen et al. (2009). The resulting event data is suitable for fast one-pass searches by an online algorithm for rapid model building and updating. From the event data only the relevant events are selected corresponding to the actual routes of the running train numbers, which are processed further using statistical methods. Robust estimators for the minimum process times are used for process time updates, like a small percentile for conflict-free running times, conditional on time-of-day where relevant. Moreover, running times may depend on the current delay in which case a robust regression fit is used to determine the best estimate conditional on the current delay. The best estimators are selected and validated off-line by splitting the event data in two parts: one for the estimation of process times and the other for evaluation of the result by comparison of the event time estimations with the realized event times (Hansen et al., 2010). Note that the arc weights relate to the minimum process times, while time loss due to timetable and infrastructure constraints are provided by the critical path algorithm over the timed event graph. A route conflict is detected when a critical path traverses a minimum headway arc, implying that a train had to wait for another train and proceeds only after a minimum headway time.
The predictive traffic model supports route setting and traffic control decisions and it could be interactively used by signallers and traffic controllers. First, the model predicts route conflicts for a given actual route plan and train positions. This could be used by the signaller to proactively resolve the conflict by e.g. changing routes or the order of trains at points. The impact of any control decision can be checked by an update of the predictive model leading to new conflict and arrival time predictions. If a control decision leads to satisfying results it can be implemented in the actual process plan. If on the other hand a route conflict cannot be avoided, then the signaller could give speed advice (or new target passage times) to the relevant train drivers so that the impact of the route conflict is minimal and energy can be saved by preventing unnecessary braking and reacceleration. Second, the arrival time predictions could be used to check connections in the case of arrival delays. When a connection conflict is detected the signaller may decide to secure or cancel a connection in advance. This way up-to-date passenger information can be provided, both at stations and in the delayed trains.
The predictive model provides effective decision support to signallers and traffic control and contributes to a better utilisation of railway infrastructure, improved reliability of train services, and more reliable and dynamic passenger information. The developed model will be embedded in a closed-loop model-predictive railway traffic control framework where online optimization algorithms will automatically resolve detected conflicts and propose control decisions to traffic controllers together with the predicted conflicts (Kecman et al., 2011b). This way an intelligent railway traffic management system will be obtained that proactively monitors the railway traffic and supports traffic controllers with decisions that optimize the traffic on a network level, beyond the traditional local control areas.
References
[1] Daamen, W., Goverde, R.M.P., Hansen, I.A. (2009). Non-Discriminatory Automatic Registration of Knock-On Train Delays. Networks and Spatial Economics, 9(1), pp. 47-61. [2] Goverde, R.M.P. (2010). A Delay Propagation Algorithm for Large-Scale Railway Traffic Networks. Transportation Research Part C, 18(3), pp. 269-287.
Page 8 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
[3] Hansen, I.A., Goverde, R.M.P., Van der Meer, D.J. (2010). Online train delay recognition and running time prediction. Proceedings of the 13th International IEEE Conference on Intelligent Transportation Systems (IEEE ITS), Funchal, September 19-22, 2010, pp. 1783-1788. [4] Kecman, P., Goverde, R.M.P., Hansen, I.A. (2011a). Train describer records as a source of information for infrastructure monitoring, performance analysis and traffic management. Proceedings of the 5th IET conference on Railway Condition Monitoring and Non-Destructive Testing (RCM), Derby, UK, November, 2011. [5] Kecman, P., Goverde, R.M.P., Van den Boom, T.J.J. (2011b). A model-predictive control framework for railway traffic management. Proceedings of the 4th International Seminar on Railway Operations Modelling and Analysis (RailRome 2011), Rome, Italy, February 16-18, 2011
Page 9 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
8 - On Implicit versus Explicit Max-Plus Modeling for Rescheduling of Trains on a Railway Network Presenter: Mr. KERSBERGEN, Bart (Delft Center for Systems and Control, Delft University of Technology) Introduction
In many countries railway traffic already covers a large part of the public transportation needs, while the number of passengers using trains is steadily increasing. Because of the increase of passengers the railway network becomes more heavily utilized, resulting in timetables with less buffer times to compensate for delays. As a result, a small delay of a single train may cause numerous secondary delays and propagate through a large part of the network. Current practice of many railway operators is to divide the network into several dispatching areas, each with their own dispatcher. The dispatcher tries to reduce the number of secondary delays by taking dispatching actions such as rescheduling or rerouting trains, canceling trains, and breaking connections in his dispatching area. These actions are based on predetermined sets of rules and the experience of the dispatchers. As a result these actions may be optimal for the dispatching area, but may cause unnecessary and unforeseen delays in other parts of the network. Many researchers have been researching this re-scheduling problem and have developed systems to support the dispatcher in making his decisions [2, 3, 4]. This paper will continue the work of [4], where a re-scheduling method is introduced that uses an implicit switching max-plus linear (SMPL) model and determines the optimal control actions for the entire network in case of delays. We will continue this work by introducing a new description of the implicit SMPL model, rewriting it into its explicit form, and comparing the computational performance of the rescheduling problem for the implicit and explicit model descriptions.
Switching Max-Plus as Max-Plus Affine Model
In general an SMPL model can model different kinds of behavior of the same system, also called modes, by switching between system matrices describing these different behaviors. The switching is achieved using a switching mechanism that may depend on several system properties as well as the control inputs. The matrices of the different modes may have very different structures; however in the case of modeling railway traffic, while limiting ourselves to the rescheduling of trains, the matrices of the different modes are very similar and only the headway times between the trains change when the trains are rescheduled. As a result it is possible to describe the matrices of the different modes of the implicit SMPL model as max-plus affine functions (i.e. affine in the max-plus algebra) of the control variables. The total model can then be combined into a single implicit max-plus affine function of the control variables.
Implicit to Explicit
An implicit max-plus linear equation can be rewritten in its explicit form using the Kleene star operator on the matrix representing the implicit relation between the events [1]. However, the Kleene star only exists if the communication graph of the ,atrix has no circuits or if all circuit weights are less or equal to zero. In general it is not possible to calculate the Kleene star of an implicit SMPL equation, since for some of the modes the Kleene star may not exist. In order to calculate the Kleene star for an implicit SMPL equation we need to show that these modes are infeasible. This can be shown by using the fact that all process times are positive, and therefore any circuit in the graph of the matrix will have a positive weight. This means that if a circuit exists the mode is infeasible, since a circuit with positive weight results in a relation of the occurred time x to itself of the form x = x+a, with a > 0, which has no solution in RU{-infinity}. Therefore, no circuits should exist, which in turn means the Kleene star will exist. So if we can determine the combinations of control actions that result in circuits in the graph of the matrix and remove them, we can compute the Kleene star for the implicit SMPL model. A circuit in the graph of the matrix exists if and only if for some of the max-plus powers of that matrix the diagonal elements are different from -infinity (i.e. the max-plus zero element). If a diagonal element is found that is not equal to -infinity, it will be a combination of process times and control inputs that corresponds to an infeasible mode and that should therefore be removed. This is done by adding a constraint to the calculation that states that this combination of control inputs is equal to -infinity. By doing this for every combination of control inputs found on one of the diagonal elements of one of the powers of the matrix all infeasible modes can be determined and removed and the Kleene star can be computed, yielding the explicit SMPL model.
Performance Comparison
The last part of this paper will consist of rewriting the rescheduling problem using the implicit and explicit SMPL models into mixed integer linear programming (MILP) problems, solving these MILP problems, and comparing the the computation time needed to solve them. This will show that computing the MILP solution using the explicit model description requires less time.
Page 10 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
Acknowledgements
Research funded by the STW project ‘Model-Predictive Railway Traffic Management’ (11025).
References [1] F. Baccelli, G. Cohen, G. Oslder, and J. Quadrat. Synchronization and Linearity: An Algebra for Discrete Event Systems. Wiley, 1992. [2] A. D’Ariano and M. Pranzo. An advanced real-time train dispatching system for minimizing the propagation of delays in a dispatching area under severe disturbances. Networks and Spatial Economics, 9(1):63–84, 2009. [3] J. Törnquist Krasemann. Design of an effective algorithm for fast response to the re-scheduling of railway traffic during disturbances. Transportation Research Part C: Emerging Technologies, 20(1):62 – 78, 2012. [4] T. J. J. van den Boom, B. Kersbergen, and B. De Schutter. Structured modeling, analysis, and control of complex railway operations. In Proceedings of the 2012 51th IEEE Conference on Decision and Control (CDC), Maui, Hawaii, Dec. 2012.
9 - Methods to improve railway capacity by integration of automatic train operation with centralized traffic management Presenter: Ms. RAO, Xiaolu (Swiss Federal Institute of Technology Zurich; systransis Ltd.) The demand for railway capacity increase has probably never been greater than today with the change brought about by growing urbanization and increasing expense in private transport. How to increase capacity on limited infrastructure becomes an unavoidable topic in railway optimization. In this regards, infrastructure planning and train operation are therefore fields in tackling the railway capacity challenge. From planning side, it focuses on planning capacity by analyzing infrastructure occupation and capacity utilization balance. From operation side, the objective is to reduce deviation between operation and planning to a minimum capacity lost. In this paper, a specific attention will be given to the railway operational capacity improvement. In the railway operation, disruption leads to capacity lost, which might happen at any time and any place. Some are caused by temporary disturbances such as maintenance and some are derived from inappropriate driver behaviors. At present, centralized traffic management and automatic train operation are regarded as the common solutions to resolve operational disruptions. Real time rescheduling increases disruption-resolving efficiency in the centralized traffic management, while automatic train operation is an on-board accurate control to minimize the loss of efficiency caused by manual operation. However, efforts for each method were of limited success. This is because rescheduling takes train behaviours as given and it has no direct impact on the accuracy of traffic plan execution, while automatic train operation concept is centred on train level optimization resulting lack of knowledge about other trains or disturbances in the railway network. Until now, there is a lot of research about rescheduling and automatic train operation as separate subjects, but rather limited scientific attention is on the closely coordinating the optimization strategies in both sides. Therefore, this paper aims at combining the capacity improvement knowledge from both the infrastructure side and the train side. To this aim, this paper is expected to deliver a description of an integrated algorithm and a model used for capacity evaluation as well. If this integration strategy has a positive response, achieving in improved stability and accurate implementation ability, then it might imply less capacity constraints for the infrastructure planning.
References [1] UIC, Capacity (UIC code 406), International Union of Railways (UIC), 2004. [2] Landex, A., Methods to estimate railway capacity and passenger delays, PhD thesis, Department of Transport, Technical University of Denmark, 2008. [3] Corman, F., Ariano, D. & Hansen, I. A., Disruption handling in large railway networks, Computers in Railways XII, Vol114, 2010. [4] Metha, F., Rößiger, C. & Montigel, M., Latent energy savings due to the innovative use of advisory speeds to avid occupation conflicts, Computers in Railways XII, Vol 114, 2010. [5] Lüthi, M., Improving the efficiency of heavily used railway networks through integrated real-time rescheduling, Diss.ETH No.18615, 2009. [6] Caimi, G. C., Algorithmic decision support for train scheduling in a large and highly utilized railway network, Diss.ETH No.18581, 2009. [7] Poré, J., ATO for suburban and main lines, IRSE-ITC, 2010. [8] Rao, X., Montigel, M., Weidmann, U., Holistic optimization of train traffic by integration of automatic train operation with centralized train management, Computers in Railway XIII, 2012. [9] Rao, X., The research and simulation of automatic train operation algorithm, Southwest Jiaotong University Master Degree Thesis, 2007. [10] Yasui, Y., Automatic train operation system for the high speed Shinkansen train, Computers in Railways X, Vol 88, 2006. [11] Yasunobu, S., Miyamoto, S. & Ihara, H., A fuzzy control for train automatic stop control, Trans. Of the Society of Instrument and Control Engineers, VolE2, 2009.
Page 11 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
Page 12 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
12 - Comparative Analysis of Algorithms for Running-Time-Calculation Presenter: Mrs. JAEKEL, Birgit (TU Dresden) Comparative Analysis of Algorithms for Running-Time-Calculation Birgit Jaekel, Frank Thonig, Thomas Albrecht August 23, 2012 1 motivation Since the early 20th century, train running simulation has been a field of research in particular for timetabling [BD11]. Today, train running simulation is also used for other related tasks, e.g. in . dynamic traffic management algorithms to determine the consequences of dispatching decisions (e.g. delays of trains). This task has very high requirements on computation time, but also on precision, as imprecise predictions of running time might lead to non-optimal dispatching decisions. . driver advisory systems for energy-efficient driving (DAS), which have high requirements on the accuracy of simulation results in order to actually achieve the desired effects of energy-saving and to obtain high user acceptance. In recent years, coupling centralized traffic management systems with on-board driver advisory systems is regarded as promising option to further increase capacity. If both kinds of systems shall work together, a compatible running time simulation module is needed which fulfills both requirements of computation speed and accuracy. The paper analyses different methods for running time computation with the focus on quantitative differences between the different models and on computation time. Furthermore, the impact of the height profile on the computations is examined. For applications of energy-efficient driving in DAS, the height profile is an important input. It has so far mainly been ignored in re-scheduling applications. Today, height profiles are often given in geographic information systems (GIS) with a high level of detail, which complicates all approaches of train running simulation and increases computation times. The authors have developed a method to simplify height pro les without loss of important information. The method and the quantitative influence of this simplification on the simulation results as well as its influence on computation time are examined in this paper.
2 methodology In this paper, three different approaches to running simulation are compared: . discrete time step as proposed in [Wen03], . the analytical-deq-solving-method proposed in [Sch11], . solving the deq with gauÿ-quadrature as proposed in [BD11] In the first part these approaches are used under a mass point train model and as an addition it will be evolved whether the algorithms are suitable to fit a mass band model in the latter part of the paper. With the three algorithms applied on tracks with different slope pro les we are going to analyse the influence of the discretisation pitches on both computation errors and times. As these depend on the train behavior especially during acceleration an braking phases we have used for each train model several train parameters.
3 example tracks To evaluate the applicability of these algorithms in an environment of practical relevance it is irremissible to use them on real-world tracks. In this paper, we have decided to use some tracks of German regional trains of about 20 and 60 km length and minimal running times up to 2000s with their speed-restriction-pro le and their gradient profile. To gain slope pro les of these tracks the height pro les were measured with GPS and transformed into an RailML-infrastructure format with the use of an algorithm depending on minimal pitches p and minimum differences d in gradient between two consecutive values (described in [TAK11]). The lower the values of p or d the more gradient changes contains the output pro le. Considering different values of p and d it is possible to observe the influence of the level of detailedness of the slope pro le on the calculated velocities and running-times and the calculation times.
4 results 4.1 computation times The computation time over a track depends on several factors, which are for both of the algorithms . the amount of changes in gradient, . the amount of changes in speed restriction and . the amount of changes in acceleration-spline.
These have less effect on the Euler's method than on the analytical solver because of the design of the first as a step-method. The number of steps of a time-step-method depends on the time which is needed to transit a track, so the discrete algorithm is additionally affected by . the track length, . the trains velocity and . the time pitch used for discretisation. The benefit of the analytical algorithm towards the Euler's algorithm lowers with the rising amount of gradient steps and increasing
Page 13 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
time pitch.
4.2 influence of the used methods on the exactness in position, time and velocity Position errors sum with increasing running time, while errors in velocity don't, because of being tared at cruising phases. The grade of exactness of the Euler's method depends on the . the time pitch used for discretisation, . the trains acceleration/braking behavior The deviation in braking phases depends on the way of computing forward with an overestimation of braking times or an underestimation if calculated backwards with a decreasing braking-force-function. 4.3 influence of the gradient discretisation pitches on the exactness in position, time and velocity The underestimation of the running times on this track tends to be enlarged on this track with increasing gradient pitches. The deviation in velocity remain negligible except for the constant track. Computing the track the other way round leads to an overestimation of the running time by the numerical algorithm.
References [BD11] O. Brünger and E.: Dahlhaus. Running time estimation. In Pachl J. Hansen, I.A., editor, Railway Timetable & Tra c, pages 58 82. eurailpress, 2011. [Sch11] Thomas Schank. A fast algorithm for computing the running-time of trains by infinitesimal calculus. 2011. RailRome. [TAK11] Frank Thonig, Thomas Albrecht, and Jürgen Krimmling. Nutzung einer Sensorkombination aus GPS und Barometer zur Bestimmung des Neigungsprofils von Bahnstrecken. In DGON-Symposium Positionierung und Navigation für Intelligente Verkehrssysteme, POSNAV ITS 2011, 2011. [Wen03] Dietrich Wende. Fahrdynamik des Schienenverkehrs. Teubner Verlag, 2003.
Figure 1: deviations in trajectory of fastest run over track Goslar-Halberstadt over time-pitches Figure 2: deviations in time of fastest run over track Goslar-Halberstadt over gradient pitches
Page 14 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
13 - Assessing the absolute traffic carrying capacity: a train timetabling based approach Presenter: Dr. MENG, Lingyun (School of Traffic and Transportation, Beijing Jiaotong University) Capacity assessment is a classical and difficult topic of railway infrastructure planning and keeps attracting research interests. In the past decades, a great number of studies have been devoted to the issue of capacity assessment (e.g. UIC 406 norm (2004), Landex (2009)). However, few studies have been focused on calculating the absolute traffic carrying capacity (ATCC), although it is obviously very important to know the maximum trains that can be operated on a railway line. Recently, Burdett and Kozan (2006) proposed an approach for determining ATCC on railway lines or networks. In their study, the mix of different train types is treated as a key step for capacity determination. In this paper, absolute traffic carrying capacity (ATCC) is defined as the maximum number of trains that can be accommodated across the entire railway line or certain critical sections in a given time window with given conditions (infrastructure data, train information, and level of service). We propose a train timetabling based approach to calculate ATCC. The approach consists of five key steps as outlined below. Note that this approach is discussed in the context of classical railway lines in China on which both passenger trains and freight trains are operated. On such lines, the passenger trains are predetermined and fixed (i.e. fixed framework of passenger trains), and then freight trains are scheduled with much more flexibility. Step 1 Initialization. Input data: railway network layout, train types with route, stops, minimum running times, train operation time standards (e.g. additional acceleration/deceleration times), dwell times, minimum headway times, proportion, schedule period (e.g. a day), passenger trains running framework (e.g. start and arrival times of passenger trains at stations), and maintenance time window. Set a small amount of freight trains to be scheduled (e.g. 10 trains). Step 2 Build the train paths in each direction respectively using given running and dwell times including running time supplements, based on the given framework of passenger trains. Step 3 If all freight train paths could be scheduled and result is feasible within a fixed time threshold, move to step 4, otherwise move to step 5. Step 4 Increase the amount of freight trains to be scheduled by one in each direction, and move back to step 2. Step 5 Output the current train amount N. Note that this train amount N is the maximum number of trains that could be scheduled, i.e., the ATCC.
In order to implement the step 2 of the approach, we build a train timetabling model by means of integer programming technique. This model is solved by the commercial solver Lingo optimizer, rather than a heuristic algorithm proposed by Meng and Goverde (2011). Furthermore, considering the statement by Burdett and Kozan (2006) that the train mix is a key step for capacity determination, this paper takes the proportional mix of trains into account and investigates the impact of train mix on ATCC. Finally, we take the Wuhan-Guangzhou Railway classical Line as an example to test and verify the above approach. Results demonstrate better adaptabilities compared to existing ATCC calculation methods.
Keywords absolute traffic carrying capacity; integer linear programming, train timetabling
Key References [1] Burdett, R.L., Kozan, E. Techniques for absolute capacity determination in railways, Transportation Research Part B 2006; 40(8): 616-632. [2] Landex, A. Evaluation of railway networks with single track operation using the UIC 406 capacity method. Networks and Spatial Economics 2009; 9(1): 7-23. [3] Meng, L., Goverde, Rob M.P. Advanced monitoring and management information of railway operations[C]. Proceedings of the 4th International Seminar on Railway Operations Modelling and Analysis (RailRome 2011), February 16-18, Università degli Studi di Roma La Sapienza, Rome, Italy. [4] UIC, Leaflet 406 – Capacity, 2004.
Page 15 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
15 - Adaptive, Rule-Based Infrastructure Modeling Presenter: Dr. SEYBOLD, Bernhard (trafIT solutions gmbh); Dr. KUCKELBERG, Alexander (VIA Consulting & Development GmbH) Different railway operation tools and applications use quite specific and more or less individual infrastructure models. Keywords like microscopic, mesoscopic or macroscopic infrastructure models are usually mentioned in this context with a commonly accepted idea about the characteristics of these granularity levels and how they are distinguished [2]. The infrastructure model for a specific tool is usually chosen due to functional needs, legacy conditions or available specifications and standards. Microscopic simulation, timetable construction and analytic tools like LUKS® [3] require detailed knowledge about topology, available train protection systems, gradients or stopping positions to compute travel and occupation times to the second. In contrast, macroscopic tools require more abstract data like stations and minimum headway times between them.
Generally, current infrastructure modeling approaches follow a topology-based approach, which transforms the topology of infrastructure into comparable data and object structures. This leads to a graph model, which normally represents the topology one-to-one. In a microscopic world, switches or crossings build the nodes and connecting tracks the edges, subdivided by signals, gradients or stopping positions as inner nodes if needed. Macroscopic views may consider stations, marshaling yards or control points as nodes and routes as connecting edged. Additional operational information can be assigned to nodes or edges. Especially for microscopic railway operation tools specifications like railML® [4] or XML-ISS and XML-KSS [5] are developed and available. They define the structure of XML documents in a quite rigid manner, concentrating on data syntax instead of data semantics. These models are not really flexible or easily adaptable to extended application fields. Moreover, these specifications require (and imply) high data consistency, since small changes usually have strong influence on computational results. This makes it difficult and time-consuming to provide consistent and complete (base) data. And it is often a barrier when starting new projects. The step to feed systems with a sufficient minimal set of data often requires a lot of work that should and cannot remain unconsidered.
With our tool OnTime, an alternative, rule-based approach was chosen to support modeling and data representation functionalities, which can be extended stepwise and provides a good and sufficient base for the OnTime algorithms even with a simple and small set of rules.
The complexity of applications and data models following the topology-based approaches still rises when versioning and temporal validity aspects are introduced. On the other hand, managing versions or temporal modifications of infrastructure availability decreases processing performance and enlarges data sizes. The interrelationship between validity, topology modification, operational and technical content of these modeling approaches complicate the derivation of data between granularity levels. The derivation of data of one granularity level from data of another granularity level is complex – especially when granularity rises – but achievable [6]. With these traditional, topology-oriented modeling approaches, situation dependent adaptive infrastructure modeling becomes hard and ambitious. In this context, adaptive modeling means, that the granularity of data can be changed due to required functionality or availability of information as well as information can be added when it is available without data consistency problems.
The alternative, a rule-based approach for adaptive infrastructure modeling presented in this paper, primarily requires rules for infrastructure characteristics as far as available. With this rule system, an infrastructure versioning becomes easy, data redundancy is decreased and data can be provided up to the granularity basically required for serious and reliable results.
Even if there is still a relationship between the granularity of data and the reliability of results as it is true for topology-oriented modeling, in our approach input data with a low granularity doesn’t lead to inconsistent data, but can be processed. Nevertheless, it is still a matter of granularity, which result quality can be achieved.
The rule-based modeling approach is easily extensible. It follows top-down reasoning. The most abstract information provided by modeling rules is evaluated first and become more and more precise and detailed with more concrete rules.
On the other hand, the rule-based approach does not support known features of explicit models like fixed views on the underlying data respectively rules. Some “dumping” functionality has to be realized to provide discrete views on the basic data and to get “dynamic snapshots” to provide data for tools using topology data.
Infrastructure modeling with rules requires a good application performance and efficiency. The number of rules may be very large when granularity rises. Nevertheless, the management of rules, structured access and evaluation is not implied by the infrastructure topology any more. Therefore, storing, retrieving and evaluating rules can be optimized using proven mathematical concepts, which is promising for oncoming larger projects and optimization requirements. Memory usage can be optimized and reduced as soon as rules and objects considered by rules can be merged.
[1] OnTime (http://www.ontime-rail.com) [2] Pachl, J. (eds.), Railway Timetable & Traffic. Analysis - Modelling - Simulation. Eurailpress 2008, 228 ., ISBN 978-3-7771-0371-6
Page 16 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
[3] Janecek, D.; Weymann, F.; Schaer, T.: LUKS – integriertes Werkzeug zur Leistungsuntersuchung von Eisenbahnknoten und -strecken. ETR 59 (2010) 1+2, [4] railML ( http://www.railml.org/) [5] Brünger, O. ; Gröger, T.: Fahrplantrassen managen und Fahrplanerstellung simulieren. Proceedings der 19. Verkehrswissenschaftlichen Tage (VWT), Dresden [6] Radtke, A; Sewcyk, B.; Wilfinger, G.: Combining Microscopic and Macroscopic Infrastructure Planning Models. International Seminar on Railway Operations Modelling and Analysis, IAROR 2007.
16 - Optimizing the network in German wagonload traffic Presenter: SENDER, Julia (Institute of Transport Logistics, TU Dortmund) In wagonload traffic, flows of single wagons with different origins and destinations are consolidated on their routes through the railway network in formation yards. The formation yards build a specific hub-and-spoke network. Depending on the size and function, a formation can handle only a certain number of wagons. On the one hand, the consolidation of wagons can decrease the transportation costs, but on the other hand, it generates new costs due to establishing and operating hub facilities. Hence, the network design has a great influence on the future production costs in wagonload traffic. The strategic location and network planning problem for wagonload traffic in railway logistics belongs to the class of hub location problems. Hub location problems consider the location of hubs (formation yards) and the allocation of origin-destination and hub nodes in order to route each origin-destination demand through the network. In order to (re-)optimize the current network structure, we develop a specific hub location problem for the strategic network design of wagonload traffic worked out together with our partner Deutsche Bahn AG. The model covers the main characteristics of wagonload traffic. We consider a multiple allocation hub location problem with capacities on arcs as well as in hubs. We involve the decision of dimensioning hub facilities by introducing multiple hub capacity levels. So far, this decision has hardly been considered in literature. The number of hubs needed is not predisposed but defined via the optimization. Due to practical requirements, we limit the number of outgoing connections from hubs. The objective is to route the wagons through the network at minimum transport and hub costs with regard to arc and hub capacities. In order to reduce the complexity of the problem and to decrease the solution-time, we modify the formulation in several ways. We test the computational effect of (dis-)aggregation constraints. We also prove sets of constraints to be redundant. We solve the problem with CPLEX on test data sets obtained from Deutsche Bahn. Computational results are presented and compared. Due to the difficulty to solve real-sized instances to (near-) optimality, we develop heuristic solution approaches. We summarize our preliminary (heuristic) solution approaches.
Page 17 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
17 - OnTime - Networkwide Analysis of Timetable Stability Presenter: Mr. FRANKE, Burkhard (trafIT solutions gmbh) Predicting the operating quality of a timetable or comparing the quality of timetable scenarios is subject to/of several scientific papers (for instance: [1]). Based on research presented at the railZurich in 2009 [2] the tool OnTime made it into effective use at railways such as SBB CFF FFS and Infrabel and in consulting projects for the networkwide analysis of timetable stability. Using distribution functions for the mapping of delays and the analytical calculation of delay propagation leads to short calculation times. Even comprehensive and complex railway networks like Switzerland or Germany can be simulated in well below one hour considering all traffic and its interaction, thus providing the possibility for an iterative optimization of the timetable.
Up to now assessments of the operating quality of timetables are performed – if at all – with the aid of simulations ([3], [4]). Most simulation tools need a microscopic infrastructure and a large number of simulations (“Monte-Carlo-Simulation”) to produce reliable results. The high effort needed restricts the use of those simulations predominantly to infrastructure planning and the determination of an optimal design of rail equipment. The abilities of such simulation tools to assess infrastructure and train movement on a microscopic level obstruct on the other hand the possibility to assess the effects of train interaction in large, complex networks as these microscopic simulation tools can hardly cope with the large amount of data of nationwide timetables. Another option for a quality assessment of timetables is the planner`s expertise. Based on experience or by analyzing operational data, an expert`s assessment is a good guideline when improving a given timetable. But the complex interactions of major changes in timetable structure, operation or major infrastructure upgrades are probably too hard to be quantified even by experts.
This leaves especially strategic decisions in early planning phases without any objective prognosis of timetable stability. Variant decisions on timetable concepts, operating-process alterations or infrastructure development are, therefore, usually taken without considering their impact on operating quality. This is quite unsatisfactory as these strategic decisions determine future operating quality to a great extent.
This need triggered research work conducted at RWTH Aachen University which already led to a usable program for the evaluation of timetable punctuality [5], [6] with the following key features: - Reproducible and scientifically sound results using operational key figures to forecast the stability and punctuality of operation - Short calculation time for fast feedback on timetable quality - Coping with varying levels of knowledge about infrastructure in short term and long term planning As such a tool was seen as an amendment to a planning tool, it does not need complex own data management but uses only timetable data, selected infrastructure properties and a smart use of delay calculation. The critical success factors proved to model and manipulate the delay as random variables. The mapping of delay as distribution functions fits the stochastic nature of delays and the analytical calculation of delay propagation lead to a short calculation time.
Major improvements in the modeling of railway operation [7] led to the development of the tool OnTime, ordered and supported by SBB CFF FSS and Infrabel. It proved possible to implement this approach into an application within one year. OnTime is available since summer 2011 and is currently licensed to SBB and Infrabel. Both infrastructure managers use OnTime to assess the operating quality of timetables and future timetable concepts as well as to analyze the impacts of particular changes in the timetable or dedicated incidents, e.g. temporary speed restrictions. Though some studies are restricted to just a network sector, the main scope is the analysis of the network wide effects in the complex national timetable, which is feasible with OnTime for the first time. In this paper, we will point out the problems of assessing the timetable stability throughout different planning stages and realizing a software tool for the usage not only in academic but in a railway business environment. A short introduction in the basic concept of delay modeling and the calculation of delay propagation is followed by describing the implementation of railway`s requirements into an application. We demonstrate the consequences of calculating with distribution functions with reference to software design resulting in further performance gains by parallel computing. The short calculation times allow for new options when designing timetables. We give an outlook on new planning processes with short cycle times and iterative improvement for better timetables.
Experience gathered in projects in Switzerland, Belgium and Germany in different planning contexts ranging from short-term to strategic planning are presented and the respective solutions for adapting to customer-specific data formats, planning tools and planning faults. Further research is conducted on universally valid key figures describing the stability of timetables and on enhancements for the assessment of a “timetable quality”.
List of cited literature: [1] Goverde, R.: “Railway timetable stability analysis using max-plus system theory”, Transportation Research Part B: Methodological 41, 2007. [2] Büker, Th.; Wendler, E.: “Efficient modelling of delay distribution functions”, Proceedings RailZurich2009, 2009. [3] Huerlimann, D.; Longo, G.; Medeossi, G.: Stochastic micro-simulation as a timetable robustness estimation tool, Proceedings RailZurich2009, 2009. [4] Siefer, T.; Radtke, A.: Railway-Simulation: Key for Better Operation and Optimal Use of Infrastructure, Proceedings RailDelft 2005, 2005.
Page 18 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
[5] Waas, K.; Büker, Th.: „Effiziente Analyse von Fahrplankonzepten“, Eisenbahntechnische Rundschau, Issue 11, 2008. [6] Akermann, H.; Büker, Th.: „Ziel Pünktlichkeit: Fahrplanstabilitätsuntersuchung für die SBB”, Eisenbahntechnische Rundschau (ETR), Issue 11, 2008. [7] Büker, Th.: „Ausgewählte Aspekte der Verspätungsfortpflanzung in Netzen“, PhD thesis submitted at RWTH Aachen University, 2010.
Page 19 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
18 - Train Speed Profile Estimation: Model Calibration and Validation Presenter: BESINOVIC, Nikola (Delft University of Technology) Train timetables are designed to allow conflict-free train traffic. However, conflicts still take place due to random disturbances which may force some trains to make unplanned stops causing delays throughout the railway network. Therefore, railway infrastructure managers strongly need reliable estimations of actual train running times as well as their variations, in order to effectively design railway timetables which are robust with respect to such stochastic perturbations. To this purpose real-time train information collected at dispatching centres by means of train describers or GSM-R systems can be used to reliably predict actual speed profiles and recognize their statistical pattern.
So far, the most common type of train information available to infrastructure managers is constituted by track occupation data, namely discrete time-distance data reported for each train after having occupied or released a certain detection section (e.g. track circuits, axle counters). This information can be used to understand the variability in train running times, but does not allow to easily comprehend speed profiles and driver behaviours, whose variations are relevant for both robust timetabling and planning of energy-efficient operations.
In the past years some effort has been carried out (Albrecht et al., 2006; Albrecht et al., 2010) that used track occupation information to reconstruct speed profiles using kinematic motion equations and calibrating speed and acceleration against observed data. Such approaches can approximate quite accurately the speed trajectory and the energy consumption relative to a specific train composition with given physical-mechanical features as well as environmental conditions. However, they cannot reproduce speed trajectories of trains when changing respective characteristics and compositions.
To design a robust timetable and evaluate its performance, variations in microscopic details of trains must be considered. To this aim it is necessary to generate a statistically significant number of different train trajectories by randomly varying the train running-related features such as the number of wagons, the type of traction unit, the service braking rate, driving behaviour and weather conditions. This can be done only if: i) dynamic motion equations are used to reproduce train trajectories on the basis of its physical and mechanical characteristics, ii) probability distribution functions of train parameters are known. Furthermore, features affecting the train behaviour can significantly alter train resistances and the tractive effort used (Lukaszevicz, 2001). According to this, the approximated parameters given by manufacturers or infrastructure managers (Radosavljevic, 2006) cannot be adopted, but have to be computed for each train run.
This paper presents an approach which implements dynamic motion equations to derive most probable speed profiles of train runs from corresponding track occupation. A microscopic train running model based on the Newton’s motion formula is used and expressed as an ordinary differential equation solved by means of the Runge-Kutta method with an adaptive step size (Butcher, 2003). In particular, parameters of the tractive effort-speed curve, the resistance equation, and the braking effort are estimated for each train against its measured space-time data. The estimation process is formulated as a simulation-based optimization problem which aims at minimizing the difference between simulated and observed train passage times at the given locations. Track occupation information collected for different trains at a dispatching centre in the Netherlands (Kecman and Goverde, 2012) are considered. Realized speed profiles are further adjusted for better fitting the observed train path by tuning acceleration and deceleration curves as well as by introducing coasting phases into the train profile. Estimating such equations for a significant set of trains, consents to calibrating a probability distribution for each one of the considered parameters. A validation procedure is successively implemented to compare obtained train speed profiles with corresponding train GPS data collected for the same corridor.
The model presented in this paper provides the most probable speed profile and the most probable train driver behaviour for a certain train run. Such aspects can give insights on different driving regimes adopted and are useful to understand if some relationship exists between certain driving regimes and the delay suffered by trains.
References
Albrecht T., Goverde R.M.P., Weeda V.A., Van Luipen J. (2006). Reconstruction of train trajectories from track occupation data to determine the effects of a Driver Information System, In: Allan J., Brebbia C.A., Rumsey A.F., Sciutto G., Sone S., (eds.), Computers in Railways X, WIT Press, Southampton.
Albrecht T., Gassel C., Knijff J., van Luipen, J. (2010). Analysis of Energy Consumption and Traffic Flow by Means of Track Occupation Data, Proceedings of the 4th International Conference on Railway Traction Systems (RTS 2010), Birmingham.
Lukaszevicz P. (2001). Energy Consumption and Running Time for Trains, PhD Thesis, Royal Institute of Technology, Stockholm.
Radosavljevic A. (2006). Measurement of train traction characteristics, Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit, Sage, London.
Page 20 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
Butcher J.C. (2003). Numerical Methods for Ordinary Differential Equations, Wiley, London.
Kecman P., Goverde R.M.P. (2012). Process mining of train describer event data and automatic conflict identification, In: Brebbia C.A., Tomii N., Mera J.M., (eds.), Computers in Railways XIII, WIT Press, Southampton.
19 - Waiting and loss probabilities for route nodes Presenter: Dr. NIEßEN, Nils (VIA Consulting & Development GmbH) One major aim of railway operations research is to identify bottlenecks of the infrastructure. In the past, lines were been analyzed in detail but junctions had often been neglected because of their complexity. The consequences become evident at the “ends” of various European high-speed projects. For the aim of capacity assessment junctions can be split into station track groups and route nodes. Route nodes contain the switch zones in the front end of stations, linking the adjacent lines and/or the station tracks. Thereby train runs can be operated simultaneously, if they do not fully or partly exclude each other. Besides simulation approaches analytical methods based on queuing theory are applied for capacity assessment of railway infrastructure. These analytical methods create a relation between the workload of the infrastructure and resulting performance indicators, such as waiting times or waiting probabilities. The analytical methods can even be applied without knowledge of a specific timetable. Instead, already the number of the different train types is sufficient; the train mixture is calculated stochastically. Input to the calculations are the minimum headway times between train runs. Analytical models are used for long term or strategic network planning. For such long planning horizons usually there is no detailed timetable available but just some general information about the intended operating programme. Another advantage of the analytical approaches is a fast computing time. For lines and station tracks analytical methods for capacity calculation are well known and are implemented in a number of software tools. However for the assessment of route nodes only approximate solutions exit until yet. This paper describes an algorithm to calculate the waiting probability and the loss probability for a route node. The approach uses a multiresource queue to model a route node. Because of the multi-channel feature of a route node two or even more train runs can be operated simultaneously. In a first step the general settings of the system are introduced. Afterwards a formula to calculate the exact loss probability of the system is derived. An approximation to deduce the waiting probabilities is drawn. Comparing the calculated waiting probabilities with an acceptable “level of service” the capacity of the system is derived. The complexity of the system, which depends on the dimension of the route node and the number of different train types, increases enormously. This paper shows a method to reduce the complexity of large system to a manageable size.
Page 21 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
21 - Optimization Models for Integrated Timetable Based Design of Railway Infrastructure Presenter: Prof. RAIDL, Günther (Vienna University of Technology); Prof. SCHöBEL, Andreas (Vienna University of Technology) The design of new railway infrastructure is nowadays strongly guided by pre-specified integrated timetables that have been derived from expected traffic to be served [1]. Integrated timetables synchronize the traffic in major nodes (e.g., main railway stations in major cities) at regular time intervals, ensure connectivity between different lines with minimum waiting times, and allow passengers to easily remember the regular departure and arrival times. In many European countries, integrated timetables have been successfully introduced in the last years and could prove their substantial advantages.
Implementing the concept of integrated timetables, however, imposes major challenges and constraints. In fact, the almost simultaneous arrival of the most relevant trains at a station and the strongly regulated travel times between stations, which must be multiples of a basic cycle interval, frequently demand extensions of existing railway infrastructure. Substantial financial investments are typically necessary for building further tracks, platforms, and other elements in order to be able to realize an integrated timetable and benefit from its long-term efficiency and higher flexibility.
Considering an integrated timetable as a central dogma when building new or extending existing railway infrastructure has a major impact on the design process. Unfortunately till today decisions like where and how to strengthen an existing single route frequently are done from a predominantly local perspective considering the specific route's properties and demands almost only, and timetables are adapted thereafter. Nowadays with an integrated timetable, dependencies between routes of different trains are much stronger, and impacts of certain design decisions have a more global influence. A more systematic optimization approach is thus required in the design in order to achieve a cost-effective solution that guarantees the constraints imposed by the integrated timetable.
Today's state of the art in developing the infrastructure layout is given by graphical procedures upon the required arrival and departure times [2] and validation by microscopic simulation of railway operation [3]. In this paper we present a concrete graph-theoretic approach for modeling the basic problem. It considers existing railway infrastructure as well as various extension possibilities on a fine-grained track-segment based way, speed limits on segments in dependence of trains and chosen routes (entry situation in stations), and various kinds of costs for installing new elements of infrastructure (tracks and switches) and realizing the connections as specified by the integrated timetable. The model is flexible in the sense that it can be relatively easily adapted to further, more special or alternative requirements.
From a complexity theoretical point of view, the problem is shown to be NP-hard. Most promising approaches to exactly solve instances of small to medium size currently appear to be constraint programming [4] and column generation based mixed integer programming techniques [5]. For larger instances of more realistic size, metaheuristics including variable neighborhood search and tabu search and in particular hybrid techniques [6] seem to be well suited to obtain excellent approximate solutions. Concrete implementations and experimental evaluations using artificial test instances derived from real-world scenarios are work in progress.
Bibliography
[1] M. Lichtenegger: Der Integrierte Taktfahrplan: Abbildung und Konstruktion mit Hilfe der Graphentheorie, Eisenbahntechnische Rundschau (ETR) 1991.
[2] A. Schöbel, G. Besau: Timetable based design of railway infrastructure. Proceedings of EURO-ZEL 2012.
[3] D. Hürlimann: Objektorientierte Modellierung von Infrastrukturelementen und Betriebsvorgängen im Eisenbahnwesen, Dissertation ETH Zürich, 2001.
[4] F. Rossi, P. Van Beek, T. Walsh: Handbook of Constraint Programming, Elsevier, 2006.
[5] G. Desaulniers, J. Desrosiers, and M. M. Solomon (Eds.): Column Generation, Springer, 2005.
[6] G. R. Raidl, J. Puchinger, and C. Blum: Metaheuristic Hybrids, in M. Gendreau and J. Y. Potvin: Handbook of Metaheuristics, 2nd edition, Int. Series in Operations Research & Management Science, volume 146, Springer, pp. 469-496, 2010.
Page 22 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
22 - Benchmark Analysis of Railway Networks and Undertakings Presenter: Prof. HANSEN, Ingo (Delft University of Technology) Research on performance assessment of railway networks and companies has been stimulated by the European policy of deregulation of transport markets, the opening of national railway networks and markets to new entrants and separation of infrastructure and train operation. Recent international benchmarking studies of railways including Lan & Lin (2006), Nash & Smith (2007), Yu (2008), Growitch & Wetzel (2009), Cantos et al. (2010), and Smith (2012) show a considerable variation of scope concerning the input and period of data analysis, while the selected output performance measures are mostly limited to costs per track-km and per train-km respectively.
These benchmark studies are based on statistical data from different periods and sets of railway infrastructure managers and train operating companies compiled by UIC. The input data concerning network length and number of railcars used in the different studies may be biased by different interpretations of network length and track length respectively reported by the railway undertakings as well as the big variation of the number of passenger and freight cars and the train composition with regard to length, weight and capacity. Furthermore, the number of staff reported by vertically integrated companies is not always clearly split into infrastructure management, passenger train operation and freight train operation, while the amount of subcontracting of maintenance and repair works is not well reported. The impact of network characteristics, as density, share of single tracks, share of electrified tracks, and distribution of personnel between passenger and freight traffic on effectiveness and efficiency is not reflected so far. Simple efficiency indicators like costs per track-km, per train-km, per passenger-km and per freight tonne-km, as well as the use of index and panel data from different periods for Data Envelopment Analysis restrict the dependability and comparability of some of the results especially in relation to the effects of deregulation, separation, and privatization of enterprises and services.
In this paper, a comprehensive approach for benchmark analysis is proposed for the determination of technical and economic key performance criteria of railway networks, infrastructure management and train operations. These consist of the relevant criteria and a number of indicators for assessing the transport and traffic output, effectiveness, productivity, quality of service, and efficiency performance of infrastructure management and train operations respectively. Interesting results of a recent benchmark analysis (Hansen et al., 2011, Tweede Kamer der Staten Generaal, 2012)) for a number of mid-size European railway networks and undertakings compared to one of the mayor Japanese railway undertakings are discussed with regard to selected key productivity and efficiency indicators for the use of infrastructure, rolling stock, personnel, expenses and revenues.
Page 23 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
23 - A Lagrangian heuristic for a real-life integrated planning problem of railway transportation resources Presenter: Dr. GUYON, Olivier (SNCF - Direction Innovation & Recherche (SNCF - DIR)) Railway planning requires three scarce and heterogeneous resources: Train paths (infrastructure), rolling stock and train drivers. In the current industrial approach at SNCF (French National Railway Company), these resources are essentially planned through a sequential approach which typically starts from (1) train paths and goes further on to (2) rolling stock and finally (3) train drivers. SNCF has already developed optimization tools for Steps (2) and (3).
Motivated by previous research applied to the airline industry and public transportation, where cost reductions ranging from 5 to 10% are reported, we developed a Lagrangian approach to solve a model that integrates the planning of rolling stock and train drivers.
In IAROR-RAILROME 2011, we presented a mixed integer linear programming model with coupling constraints for a simplified integrated problem of railway production resources. We also proposed a Lagrangian relaxation heuristic. In this approach, sub-problems were solved thanks to a standard mathematical programming solver. First numerical experiments were conducted on a reduced data set, extracted from an actual instance from a French region (Bretagne). The results obtained were promising but showed that the resolution with a standard solver was too costly in terms of computational times for real-world instances and that the model had to be improved for implementation in a Lagrangian relaxation framework.
Since 2011, the mathematical model has been improved and numerous operational constraints have been integrated in order to tackle real-life integrated planning problems at SNCF. The Lagrangian relaxation heuristic has been updated consequently. As already mentioned, SNCF has already developed two independent optimization tools for planning rolling stock and train drivers. The Lagrangian approach has also been adapted so that the resulting sub-problems of this mathematical decomposition method can now be solved with the two dedicated tools.
We thus can now address real-life instances and solve each sub-problem of the specific Lagrangian heuristic with proprietary software.
Preliminary computational results show the interest of our method. Compared to a sequential approach, the Lagrangian heuristic leads to substantial cost reductions and generates good solutions in a reasonable CPU time. This is thus an interesting tool for human planners who want to experiment and quantitatively evaluate different scenarios (e.g. other train-path distribution, specific rolling stock, train drivers with other capabilities…).
If this abstract is accepted for the conference, we will submit a draft paper with the following outline: 1) Introduction of the problem with its industrial context and a brief state-of-the-art, 2) A mixed integer linear programming model for the integrated planning problem of railway resources we address, 3) A description of the Lagrangian Relaxation approach we developed, 4) Some details of implementation of the decomposition method, especially the adjustments that had to be made to be able to use the two optimization tools already developed by SNCF, 5) Computational results on real-life instances from a French region (Bretagne).
In our opinion, a first interest of presenting our work in IAROR-RAILCOPENHAGEN 2013 is that we address a real-life planning problem that has not yet been really studied in a railway context. Also, we propose an interesting mathematical decomposition model which has the advantage to be based on two sub-problems that have already been tackled at SNCF. Finally, preliminary computational results show the interest of our heuristic.
Page 24 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
24 - On the Dynamic of Primary and Secondary Delay Presenter: Dr. LABERMEIER, Helga (Swiss Federal Railway (SBB)) Delay analysis is part of the daily job of a timetable planner at the Swiss Federal Railway (SBB). Therefore real-time arrival and departure delay of every train is given for about 1000 stations. With this data, all delays can be analysed in the dimensions train, time and location. But it’s not possible to say if a delay was caused due to technical or human influences (so called primary delays) or if the train was influenced by another trains due to conflicts on its route or connections it had to wait (secondary delays).
The paper will present a method how SBB was able to differentiate and quantify primary and secondary delays on a time scale of seconds out of real-time data. This was done with the very detailed data provided by the new dispatching system “Rail Control System” (RCS), which not only knows the arrival and departure times of all trains in the stations but also the time when the signals turn to green and when the trains pass the signals.
Also some results will be presented which allow to get a deeper insight into the dynamic of delay expansions at different punctuality levels. The analysis was performed over 192 working days (13.6 million stops and 32 million sections). For example, the results show impressively, that secondary delays occurring at stops play the main role at days with a low punctuality value. This is amongst others attributed to the high importance of connections in the swiss timetable.
Finally the paper will give an overview on the additional benefit we gained from being able to separate primary and secondary delays. For example, we were able to calibrate the new simulation software OnTime, which needs primary delays as input data, almost automatically in a very high quality. Furthermore it will help to support the definition of the dispatching principles – how long a train has to wait to ensure a connection – to achieve the optimum for the entirety of the passengers. Combined with researches on the capacity to recover delay on the whole net, these results will also help the timetable planners to optimize the time margins.
26 - A suboptimal control scheme of multiple trains based on mode vector constraints Presenter: Mrs. WANG, Yihui (Delft University of Technology) I send the PDF files via email. Please check it. Thank you!
Page 25 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
27 - Stepwise capacity increase of single-track railway lines Presenter: Dr. LINDFELDT, Olov (Vectura Consulting) Extended abstract
Background and introduction Most railway lines in Sweden are single-track. Several of the lines are of great importance for the industry and in times of prosperity the capacity utilization gets so high that the demand for freight transport is difficult to meet. Some lines are used also for fast passenger services (200 km/h) which gives a complicated combination of bidirectional and mixed traffic. There are long term plans for upgrade of these lines into double-track. Such upgrades will most likely be performed as a successive and stepwise upgrade where a new line track is constructed along the existing one for one line section at a time. In a shorter perspective, limited capacity increases can be achieved through construction of additional crossing loops located at strategic points along the lines. A major goal is to find upgrade strategies that make the available capacity follow the development of capacity demand. This includes finding the relationship between capacity and infrastructure design, where the infrastructure successively goes from single to double-track. This kind of result can then be used in Cost Benefit Analyses to value upgrade measures on different parts of the railway line.
Objectives The main objective of the work described in the proposed paper is to evaluate a lot of alternative sequences of upgrades from single to double-track. This means that: • it is possible to describe how capacity rises as more and more line sections are upgraded from single- to double-track, • it is possible to find line designs with a capacity level in-between single and double track capacity. The capacity of mixed lines, i.e. with a combination of double-tracked and single-tracked sections, is evaluated and described. These capacity values can then be combined with construction costs to evaluate the societal profitability of different extension strategies, • an optimal order of extension of different sections can be found. This includes interaction effects between partial double-tracks located apart from each other with single-track sections between them. This kind of relationship can be used to increase profitability and efficiency of infrastructure investments, • the effect of complementing (new) crossing loops located on sections that remain single-track will be examined. The evaluation will be made for a freight line with sparse passenger traffic. However, the proposed method can also be used on other lines where the traffic mix is more complicated. Model and method Two complementing methods will be combined: • Factorial design where upgraded sections are systematically shifted. • Capacity evaluation through timetable analysis (automated, computer generated timetables). Factorial design will be applied to choose feasible combinations of upgraded sections and additional (new) crossing loops. Several samples of 1,000 – 15,000 designs will be generated. The plan is to evaluate four different levels with one sample at each level: 1. Only crossing loops (to find the potential of minor investments), 2. 10% double track, 3. 25% double track, 4. 50% double track. The combinatorial, timetable generating model, TVEM, presented at RailZürich and RailRome, will be used to evaluate the capacity of each design within each sample.TVEM, Timetable Variant Evaluation Model, uses asynchronous scheduling. The traffic is divided into train patterns according to a presumed regular timetable (freight traffic is scheduled non-periodically) and the patterns are then systematically scheduled in different time locations. The strength of TVEM is that both infrastructure and timetable factors may be systematically changed and analyzed. Several alternative timetables will be constructed for each infrastructure design which gives not only the capacity for each design but also a measure of timetable sensitivity, i.e. how much the capacity depends on the timetable. The strategy with four levels makes it possible to apply a simplified form of Branch and Bound technique to eliminate combinations that are found to be unprofitable (low capacity compared to investment).
Application The proposed methods, combining factorial design, combinatorial scheduling (TVEM) and Branch and Bound techniques will be exemplified on an existing line in middle Sweden. The aim is to find a capacity function where the capacity effect of additional double-track sections is plotted as a function of expected construction costs (length of upgraded sections). Different investment strategies, i.e. orders of upgrade for the sections, will give rise to different capacity functions and the effect of alternative strategies will be seen.
Outline of the paper The proposed paper will give a brief introduction to the operation of existing single-track lines in Sweden. Infrastructure measures considered for capacity increase will be explained. The need to find efficient strategies to increase capacity stepwise from single to double-track will be commented. In the following section the TVEM-model will be described and the modelling of special single-track features will be emphasized.
Page 26 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
The handling of the two major factors of interest, double-track sections and crossing loops, will also be a part of this section. A major part of the paper will focus on how the TVEM-model can be combined with factorial design for evaluation of a great number of infrastructure designs to find out the effect of upgraded sections (double-track) on different locations along a single-track line. This will end up in a relationship that shows how capacity grows when additional line sections are upgraded and how this growth depends on the order (strategy) in which the line sections are upgraded.
Page 27 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
30 - A mixed-integer linear program for the real-time railway traffic management problem modeling track-circuits Presenter: Dr. PELLEGRINI, Paola (IFSTTAR - Univ. Lille Nord de France) At peak hours, railway timetables extensively exploit the infrastructure for accommodating traffic. This extensive exploitation often translates into many trains traveling through critical junctions within short time horizons, where junctions are physical areas in which multiple lines cross. In these areas, unexpected events, even of apparently negligible entity, may cause a relevant deviation with respect to the scheduled timetable. In fact, according to the timetable, trains may be scheduled to traverse the same track segment at a very short time distance. If one of them is delayed due to an unexpected event conflicts may emerge, i. e., multiple trains may claim the same track segment concurrently: in this case trains may have to stop or slow-down for ensuring safety. As a consequence, conflicts may generate a severe delay propagation.
In the practice, conflicts are often solved in a first-come-first-served manner: the first train claiming a track segment is allowed to use it first. This is of course a sub-optimal scheduling strategy, but dispatchers in charge of managing traffic in a control area typically do not dispose of automatic tools for selecting a better one. As an additional degree of freedom for dispatchers, typically multiple routes can be used for traversing a junction, going from an origin to a destination location. In principle, the exploitation of these multiple routes may increase the efficiency of the system, but, as in the case of train scheduling, few automatic tools are available for selecting an effective rerouting. As a consequence, the decrease of delay propagation by concurrently rerouting multiple trains is a hardly achievable goal for dispatchers.
The selection of the train routing and scheduling for minimizing delay propagation has been formalized as the real-time Railway Traffic Management Problem (rtRTMP) [5]. In this study, we propose a fixed-speed mixed-integer linear programming formulation for optimally solving the rtRTMP, which we shortly introduced in [5]. We model the infrastructure in terms of track-circuits, which are the track segments on which a train presence can be automatically detected. Thanks to this detection, the signaling system imposes the suitable headway distance between consecutive trains. This headway must always be at least equal to the train braking distance: it is achieved by combining sequences of track-circuits into block sections, and protecting them through signals. Signals indicate the behavior that the driver must hold: if no other train is detected on a track-circuit within the arriving train braking distance, then the signal assumes the green aspect for indicating the authorization of running at the scheduled speed. Otherwise, the signal imposes to decrease the train speed, possibly up to the complete stop, as a function of the distance of the preceding train on the route. While directing trains along track-circuits, our formulation assesses all possible alternatives for train rerouting in the infrastructure and all rescheduling alternatives for trains along these routes. To the best of our knowledge, we present and assess the first formulation that solves this problem to optimality. Most of the models proposed in the literature limit the representation to block sections [2,3,4,7,8], imposing an artificial restriction to junction capacity. The others either consider heuristic solution approaches [6] or restrict a-priori the possibilities to be taken into account for rerouting or rescheduling trains [1].
In a thorough experimental analysis, we test our formulation on perturbations of a real instance representing traffic in the control area including the main station of Lille in the North of France, i. e., the Lille-Flandres station. In particular, we considered a one-day timetable including 589 trains. All rolling stocks are used for both an arriving and a departing train, but for what concerns the first trains departing in the morning (which arrived the day before to the platform) and the last ones arriving at night (which will leave the platform the day after). Besides 259 turn-arounds, the timetable contains 8 joins and 10 splits. Starting from the original timetable, we impose a delay to 20% of trains that do not represent shunting movements: we randomly select the trains to be delayed and we randomly draw their delay in the interval between 5 and 15 minutes. Moreover, we consider different scenarios for what concerns the infrastructure, forbidding the use of different number of track-circuits, and hence of different numbers of routes.
This analysis allows the quantification of the improvement, in terms of reduction of delay propagation, that can be achieved by passing from modeling block sections to modeling track-circuits. Our results show that the consideration of track-circuits may allow a reduction of delay propagation that is both non-negligible and statistically significant. Furthermore, the performance of our formulation are very promising in terms of computation time requested for finding the optimal solution to the instances.
[1] G. Caimi, M. Fuchsberger, M. Laumanns, and M. Lüthi. A model predictive control approach for descrete-time rescheduling in complex central railway station approach. Computers & Operations Research, 39:2578–2593, 2012. [2] F. Corman, A. D’Ariano, D. Pacciarelli, and M. Pranzo. A tabu search algorithm for rerouting trains during rail operations. Transportation Research Part B, 44:175–192, 2010. [3] A. D’Ariano, D. Pacciarelli, and M. Pranzo. A branch and bound algorithm for scheduling trains in a railway network. European Journal of Operational Research, 183:643–657, 2007. [4] M. Mazzarello and E. Ottaviani. A traffic management system for real-time traffic optimisation in railways. Transportation Research Part B, 41:246–274, 2007. [5] P. Pellegrini, G. Marlère and J. Rodriguez, Real time railway traffic management modeling track-circuits, ATMOS 2012, 2012. [6] J. Rodriguez. A constraint programming model for real-time train scheduling at junctions. Transportation Research Part B, 41:231–245, 2007.
Page 28 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
[7] J. Törnquist and J.A. Persson. N-tracked railway traffic re-scheduling during disturbances. Transportation Research Part B, 41:342–362, 2007. [8] J. Törnquist Krasemann. Design of an effective algorithm for fast response to rescheduling of railway traffic during disturbances. Transportation Research Part C, 20:62–78, 2012.
31 - Computing multiple running times for railway timetabling: a speed-level based model for constructing alternative speed profiles Presenter: Dr. CHEVRIER, Rémy (IFSTTAR); Dr. PELLEGRINI, Paola (IFSTTAR) In this paper, we tackle the problem of estimation of the appropriate running time for the timetabling process. Although this problem is crucial in the timetable design, in practice, the running times are based on the fastest journey multiplied by an arbitrary factor slightly greater than one. These obtained time supplements allow the train driver to adapt speed to the traffic by possibly, in case of perturbation, running slower than scheduled. Indeed, the time supplement could also be used to save energy.
We propose an original approach to compute train running times by concurrently minimizing both energy consumption and running time. Usually, the latter is estimated for an entire trip and the time supplement is spread over it. However, before considering this fact, we consider here the estimation of running time between two stations. Since a trip includes a succession of journeys from one station to another, the approach proposed performs a running time estimation of one journey.
The estimation of running time between two stations can be done by building the speed profile that the train driver must follow. The speed profile indicates the speed at a certain position and it is provided for the train driver in his roadmap. The speed profile is built according to a set of specific rules that we propose to determine the order of driving regimes. Moreover, the approach under consideration is capable of providing a set of tradeoff-solutions for the timetable-makers : we deal with a bi-objective optimization of speed tuning and energy saving. This way, the timetable-makers will be able to choose a running time adapted to their needs in the timetabling process.
According to the theory of optimal control, there are four optimal driving regimes defined by application of the Maximum Principle: Acceleration at full power; Cruising at constant speed; Coasting (inertia motion while the engine is stopped); Maximum braking (according to the service braking, softer than emergency braking). Since acceleration is very energy-consuming, the inefficiency of applying unnecessary sequences of braking followed by acceleration is straightforward. Besides, it is a principle of the method that we propose in the paper. In the roadmaps to provide for the drivers, a braking must not be followed by an acceleration.
The optimization problem we deal with consists in determining the order of driving regimes to use and also the positions from which they are switched. These order and positions are set over the train path composed of a sequence of sections. A section is defined by a length and a constant and fixed maximal speed. Consecutive sections always have different maximal speeds. Within each section, a pair of target-speeds is tuned and the speed profile is built in each section successively. The target-speeds are the decision variables that the algorithm must set. Within each section, speed profiling is done in two phases and also in splitting the section in two parts. The main idea is to limit the use of acceleration (the most energy-consuming driving regime) in the first part only. In the second part, energy-friendly (cruising at partial power) or energy-free (coasting or braking) driving regimes are used. The switch-points are determined based on the lengths necessary to reach the target-speeds.
As a complement, a smoothing procedure is applied to the speed profiles produced. Indeed, in some cases due to the steep slopes (descents) and the rules of construction, it may happen that an acceleration follows a braking, which is exactly what we aim to avoid. To this end, the smoothing method replaces sequences (Braking; Acceleration) or (Braking; Coasting with acceleration) by sequences including a cruising phase, in particular.
Given that evolutionary algorithms (EA) are well-suited to multi-objective optimization, our approach is based on a state-of-the-art multi-objective EA: the Indicator-Based Evolutionary Algorithm (IBEA). In a single run, this kind of algorithm is capable of producing a set of distinct solutions. According to the Pareto approach, the solutions produced are said to be non-dominated, i.e. it is not possible to state whether one of these solutions is better than another one, or not. In fact, all the solutions are compromises between the two objectives to optimize and hence, as many several possible choices for the timetable-makers.
Two case-studies are provided for highlighting the interest of such a method. The first one is based on a line proposed by Ko et al. [1] and the second one on the Saint-Etienne to Rive-de-Giers line (France). For both case-studies, speed profiles are produced and compared to their respective reference solution, which corresponds to the lower-bound of running time and also to the upper-bound of energy-consumption. Due to the particular topology of the line, the second case-study shows that the smoothing method is useful to obtain speed profiles corresponding to our aims.
[1] H. Ko, T. Koseki and M. Miyatake, Application of dynamic programming to optimization of running profile of a train, in Computers in Railways IX, 2004, pp. 103-112.
Page 29 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
32 - Measuring Robustness of Timetables in Stations using a Probability Distribution Presenter: Mr. JENSEN, Lars Wittrup (DTU Transport) 1 Background and purpose Stations are often leading to network effects, especially at at-grade junction stations where conflicts between train routes results in interdependencies between the railway lines converging at the station. A train delayed on one line at a junction station may therefore propagate to other lines because of headway constrains and conflicting train routes. Depending of the size of the initial delay, the length and characteristics of the line(s), the initial delay can propagate to the whole railway network.
The interdependencies between lines at a junction station can be reduced by removing conflicts between train routes by changing the track layout, the plan of operation or the timetable. Existing methods can be used to calculate the complexity of a station, which can be used to quantify these changes. Reducing the complexity of a station will reduce the interdependencies or the consecutive delays caused by interdependencies, and result in a more robust operation. Currently three methods to calculate the complexity of station exists: 1. Complexity of a station based on the track layout 2. Complexity of a station based on the probability of a conflict using a plan of operation 3. Complexity of a station based on the plan of operation and the minimum headway times
However, none of the above methods take a given timetable into account when the complexity of the station is calculated. E.g. two timetable candidates are given following the same plan of operation in a station; one will be more vulnerable to delays (less robust) while the other will be less vulnerable (more robust), but this cannot be measured by the above methods.
In the light of this, the article will describe a new method where the complexity of a given station with a given timetable can be calculated based on a probability distribution fitted to empirical data. The method will make it possible to evaluate different timetables and/or timetable variants choosing the most robust one with the least amount of consecutive delays and network effects.
2 Method The method presented in the article is divided in two. Firstly, a study of how delays can be described by a probability distribution is conducted and distribution parameters are estimated for a real case junction station in Denmark. Secondly, the method to calculate the station complexity given a timetable is presented and used on a real case example using an appropriate probability distribution found in the first part.
2.1 Fitting a probability distribution to empirical data In the first part of the paper a study is conducted to reveal the most fitting distribution for non-negative delays in junction stations. Based on this knowledge distribution parameters are estimated based on empirical data about delays provided by the Danish infrastructure manager (Rail Net Denmark) for a small Danish junction station. The parameter estimates will be grouped by arrival, departure and selected trains (Regional, IC, IC Express and EC).
2.2 Measuring robustness of a timetable in a station by calculating the complexity of the station using a probability distribution An improved method to calculate the complexity of switch zones of a station is presented. Based on a matrix containing all conflicts between the dominant train routes in the station, a train list is compiled for every train route. The list contains departure or arrival times for all the trains, within a time period, using the route or the routes conflicting with the incumbent route, ordered by time. For all pairs of consecutive trains in the lists the probability of the second train obtaining a delay is calculated using the probability distribution found. The probability is calculated as a function of the headway time and other relevant parameters (depending on distribution type). The sum of all the probabilities calculated is used to derive a conflict index that is able to represent the complexity of the station and the robustness of the appertaining timetable.
3 Results and perspectives The paper proposes a new method to calculate the complexity of a station based on the timetable, the track layout and a probability distribution. The paper will demonstrate how the method is used and how different timetables, with the same plan of operation, induce different results with varying robustness. Furthermore, the paper will demonstrate how the delays at a junction station can be described by a probability distribution, with varying parameters depending on train type and whether the delay is on arrival or departure.
Using the proposed methodology, it is possible to evaluate timetable candidates more comprehensive in relation to the expected robustness of the timetables. When appraising a set of timetable candidates the expected robustness can be used as a KPI to help planners select the best timetable. This, depending on the priority of KPIs, will result in more robust timetables where network effects also have been reduced.
Page 30 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
33 - Network Reduction and Dynamic Forecasting of Passenger Flows for Disruption Management Presenter: Ms. VAN DER HURK, Evelien (Rotterdam School of Management, Erasmus University) This paper presents a model for guiding passengers in case of a disruption through travel information to optimize passengers’ service level. Passengers’ route choices directly influence both their incurred delay as well as that of others. Therefore guiding passengers is an approach to improve service level that is complementary to other methodologies in the literature that focus on logistic rescheduling. The focus on passengers is enabled by increased information on passengers and opportunities for direct communication with passengers. Results of this research are based on real life data from Netherlands Railways. This research is supported by the NWO Complexity project “Complexity in Public Transport”.
Why passengers are important
Large numbers of people depend on public transport all over the world. Unfortunately disruptions occur regularly in these systems. Malfunctioning rolling stock or infrastructure, accidents and technical problems are just a few examples of the causes of disruptions. A disruption inevitably leads to complex logistic rescheduling problems, as the limitations resulting from the disruption need to be accommodated in the timetable, rolling stock schedule and crew schedule. Disruption Management aims to restore the system and to improve passenger service in such cases.
In recent years, attention has shifted from just finding a feasible solution to finding a feasible solution with attention for passenger service level. Two good examples of this are Nielsen (2011) and Dollevoet (2012). Nielsen focuses on optimizing the rolling stock assignment by using passenger simulation to evaluate the passenger service level. Dollevoet focuses on delay management while taking passengers’ route choices into account.
However, it is not just the logistic schedule that determines service level experienced by passengers, also the passengers route choice decisions themselves are of major influence. Passengers are faced with the complex problem of rescheduling their journey, while often having incomplete information to do so. Their best decisions depend on both the changes in the logistic planning and the decisions of other passengers, as the latter determine whether there is sufficient capacity available for a passenger to board the train of his or her choice.
Consequently, route choice directly relates to the experienced passenger service level. Accurate information enables passengers to make better route choice decisions, and guidance of passengers may reduce or prevent overcrowding of trains. Hence the passenger service level results from the combination of logistic rescheduling and the distribution of passengers over the network.
We show that guiding passengers in a disrupted situation is an important tool to balance capacity and demand. Therefore apart from the logistic rescheduling, guiding passengers in itself can improve passenger service level significantly. To our knowledge, this is the first paper that focuses on actively guiding passengers to improve service level.
Research question
The main question we address is:
How to improve passenger service level by guiding passengers in case of a disruption?
We measure passenger service level as the sum of the delays of all passengers. Guidance consists of providing passengers with a specific travel advice. This travel advice has to consist of a reasonable, possibly dynamically adapted, route from the passengers desired origin to his or her destination, thereby taking into account the limitations caused by the disruption.
Methodology
We present a two level approach where first the journeys that are affected by the disruption are clustered to form a compact set of Origin-Destination flows. The demand for these flows is estimated based on real life smart card data by using an auto regressive regression model.
Secondly, a multi-commodity flow model is used to solve the problem of distributing these passengers over the network. The paths of each commodity are limited to a set of reasonable paths, to ensure a likely compliance of the passengers to the provided guidance. The objective is to minimize delay and optimize seat availability. The objective depends both on the route choice of passengers and the logistic schedule. As the focus of this research is on passengers and not on generating feasible logistic schedules, we assume a logistic schedule defining the rolling stock capacities of the trains to be given. By evaluating multiple schedules, the logistic rescheduling is taken into account in the passenger guidance model.
Case study based on Netherlands Railways
Page 31 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
The results of this study are based on a case study using smart card data of Netherlands Railways, the largest passenger railway transport operator in the Netherlands. Preliminary results show that short term passenger flows are very well predictable based on smart card data. First results indicate moreover that the guidance from the multi-commodity flow model result is a substantial improvement of the service quality in comparison to a approach just focused on the logistic rescheduling proces.
References
Nielsen, L. (2011) The Rolling Stock Rescheduling in Passenger Railways, Phd Thesis, Erasmus University. Dollevoet, T., D. Huisman, M. Schmidt, A. Schöbel (2012) Delay Management with Rerouting of Passengers, Transportation Science, vol 46, nr 1, pp74-89
34 - Developing Capacity Utilisation Measures and Limits for Railway Nodes Presenter: Dr. ARMSTRONG, John (University of Southampton) The station and junction nodes of a railway network typically form the 'bottlenecks' limiting route and network capacity. As noted by UIC (2004) and others, railway capacity is inherently difficult to define and measure. To fully describe and understand the capacity of a railway line, route or network, it is necessary to measure both capacity provision (typically in terms of trains per hour) and capacity utilisation or consumption (in terms of the percentage of potential capacity utilised by a timetable). This requirement reflects the fact that a given level of capacity provision can be timetabled and delivered in different ways, with possible variations in the levels of capacity utilisation, while, on the other hand, a constant level of capacity utilisation may also be achieved by significantly varying levels of capacity provision, again reflecting the effects of different train service patterns and mixes. Where capacity is scarce, the overriding objective of the train planning process is typically to provide as much capacity as possible within acceptable levels of capacity utilisation, to ensure an acceptable level of service quality and reliability. Established Capacity Utilisation measures are available, in the forms of the UIC 406 method (UIC, 2004) and the UK-specific Capacity Utilisation Index (CUI) (Gibson et al., 2002). Recommended maximum values for the two measures are available for network links, for both peak period (75% - 85%) and extended, all-day operation (60%-70%). However, no standard method, and no equivalent recommended utilisation maxima, are available for railway nodes. In the course of the OCCASION project, the CUI method was extended to enable the assessment of railway nodes (Armstrong et al., 2012). Following the development of a standard methodology, historic nodal performance data were reviewed, and were then employed in conjunction with simulation techniques to investigate practical peak and extended operating period CUI limits for nodes, equivalent to those established and published for nodes, beyond which performance is likely to deteriorate to an unacceptable degree.
References
Armstrong, J., Preston, J., Potts, C.N., Paraskevopoulos, D., Bektas, T. (2012). “Scheduling Trains to Maximise Railway Junction and Station Capacity”. Proceedings of CASPT12, Santiago, July 2012.
Gibson, S., Cooper, G., Ball, B. (2002). “Developments in Transport Policy: The Evolution of Capacity Charges on the UK Rail Network”, Journal of Transport Economics and Policy, Vol. 36, Part 2, pp. 341-354, May 2002.
UIC (2004).Leaflet 406: Capacity, Union Internationale Des Chemins De Fer (UIC), UIC, Paris.
Page 32 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
36 - PLANNING SORTING SIDINGS USING BINARY INTEGER PROGRAMMING APPROACH Presenter: Mr. BELOšEVIć, Ivan (University of Belgrade, Faculty of Transport and Traffic Engineering) Stations are of key importance for efficient railways. Great number of problems arises in the areas of capacity of stations and working technology planning, representing main railways optimization problems, in the same time. Multigroup train formation is stated as one of complex marshalling problems thus belonging to the class of railway station working optimization problems. Marshalling process is unique for railways and gives the possibility of adjusting this mean of transport to economy needs. Unfortunately, marshalling has its disadvantages: demands additional sidings and increases wagon layover which results in increase of transport costs and service quality reduction. So far, freight multigroup trains were mainly used in segment of the local transport, in the category of pick-up goods trains. The aim of pick-up goods trains is wagon provision and collection from intermediate stations and their further transport to the next marshalling yard. The starting marshalling yard forms a train by gathering wagons for each intermediate station and arranging these groups in a train according to the disposition of intermediate stations. In the compliance with the new transport concept, railways are changing with the aim of establishing a united transport system. In this system, multigroup trains are gaining on significance and in the segment of long-haul transport. Long-haul multigroup trains have their role in connecting greater number of terminals in the region in order to ensure that small freight flows are gathered on same route and thus form complete block trains. Also, multigroup trains have an important role in the industry railways which represent an important link in the transport chains between industry and public transport. Industry railways are characterized by a developed network of sidings with many handling points. These networks are often segmented into a number of separated areas which are served by the same industry marshalling yards. The role of industry marshalling yard is in forming industry trains where wagons are grouped according to handling points’ disposition with the aim of their faster and simpler distribution. Formulating and classification of methods for forming multigroup trains started in the mid of 20th century in the expert conferences and in the journals dealing with practical problems in railways. Soon after, first scientific papers on marshalling yards and their structure and capacity defining appeared. Following these researches mathematical methods of multigroup train formation are formulated in scientific papers [1, 2]. In some more recent research [3, 5] it is proven that the problem of forming multigroup trains belongs to the class of NP hard problems (nondeterministic in polynomial time). In the paper [5], a concise encoding of sorting schedules for multistage methods is suggested. This way of coding is used in papers [6, 7] for formulation of mathematical programming model which finds out optimal wagon sorting with the aim of minimizing total number of wagon movements. In the papers, the optimization of forming trains for the specific working conditions in the same marshalling yard is performed. Paper [4] introduces heuristic algorithms for optimization of more complex sorting schedules. In this contribution, binary integer programming model based on the above mentioned coding will be given. Different from the papers [4, 6, 7], the proposed model is intended for the phase of planning of work technology and sidings layout, and not only for the optimization of operation of existing marshalling yards. So far research have not paid enough attention to simultaneously consider operation conditions and conditions of marshalling yard designing which sometimes brings into question the possibilities of realization of sorting schedules formulated in theory. The objective is to minimize total system cost while satisfying designing and operation constraints. We consider two types of costs, infrastructure and operation. First group of costs covers all construction and maintenance costs. These costs are directly in function of total track length, number of tracks and the design of track connecting. Operation costs present total shunting costs influenced by number of locomotive’s pull out operation and number of wagon movements. Proposed mathematical model enables consideration of effects of the observed methods application in the phase on planning, i.e. before starting the station building and its exploitation. As this problem is NP problem, in this paper we also present an efficient usage of heuristic techniques to contribute in planning yards for the case of large scale problems. We evaluate algorithms using randomly generated instances and compare them to proposed exact BIP approach.
References [1] Daganzo, C.; Dowling, R.G.; Hall, R.W.: Railroad Classification Yard Throughput: The case of multistage triangular sorting. Transportation Research Part A. 17A, 2(1983), pp. 95-106. [2] Daganzo, C.: Static blocking at railyards: Sorting implications and track requirements. Transportation Science. 20, 3(1986), pp. 189-199. [3] Eggermont, C.; Hurkens, C.; Modelski, M.; Woeginger, G.: The hardness of train rearrangements. Operations Research Letters. 37, 2(2009), pp. 80-82. [4] Hauser, A.; Maue, J.: Experimental evaluation of approximation and heuristic algorithms for sorting railway cars. In: Festa, P. (ed) SEA 2010. Springer (2010). [5] Jocob, R.; Márton, P.; Maue, J.; Nunkesser, M.: Multistage methods for freight train classification. Networks. 57, 1(2011), pp. 87-105. [6] Márton, P., Maue, J., Nunkesser, M.: An improved classification procedure for the hump yard Lausanne Triage. In: Clausen, J., Di Stefano, G. (eds.) ATMOS 2009. IBFI Schloss Dagstuhl, Wadern (2009). [7] Maue, J., Nunkesser, M.: Evaluation of computational methods for freight train classification schedules. Tech. Rep. TR-0184, ARRIVAL (2009).
Page 33 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
37 - Analyzing the Incremental Transition from Single to Double Track Presenter: Mr. SOGIN, Samuel (University of Illinois at Urbana Champaign) Long term demand for rail freight transportation in North America is projected to increase considerably in the coming decades. Additionally, government agencies want to increase the speed and frequency of passenger trains operating on certain freight lines, further adding to demand for new capacity. A significant portion of the routes in the United States are single track with passing sidings. Eventually, the second mainline track will become necessary to maintain network fluidity. However, the full funding for the second track may not be available all at once; subsequently the track can be phased in over time creating a hybrid track configuration. Depending upon the traffic characteristics, traffic will transition from a delay distribution characteristic of single track to a delay distribution characteristic of double track at different points in the progression from single to double track. Rail Traffic controller (RTC) is used to simulate various hybrid track configuration under different operating conditions. In addition to the amount of second main track added, the analysis considered the interaction effects of traffic volume, traffic composition, and the speed differential between train types. Consolidating double track sections can show a marginal improvement over separate double track sections. Additionally, decreasing the bottleneck length in-between sidings is also effective. The benefit of full double track can be realized for high priority trains with partial double-track. However, the low priority traffic may not experience double-track-like performance until nearly the entire 2nd track is installed. These results will facilitate the development of an optimal incremental upgrade model for capacity expansion.
Page 34 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
39 - QUANTIFYING INFRASTRUCTURE CAPACITY CONSUMPTION WHILE TRANSITIONING TO STEADY OPERATIONS DURING DISRUPTIONS IN PASSENGER RAIL TRANSPORT Presenter: Ms. CHU, Friederike (TU Darmstadt, Chair of Railway engineering) --Introduction-- In railway transport, acceptable operational quality levels and transparent passenger information are necessary – even more during larger infrastructural disruptions. One possibility to achieve this objective are disruption programs (DRPs). DRPs are counter measures to maintain operations and passenger information during disruptions. They are planned and negotiated in advance. Since they are already prepared, they are faster to implement and easier to communicate [1]. Whether a DRP is functioning well may only be evaluated after its first implementation during operations (through trial and error learning). Another possibility are simulation studies which are (often too) expensive. The operational quality of a DRP depends on the development of the transition phase. In this context, the transition phase describes the process of stabilizing the operations of a railway network from the beginning of the disruption until steady operations during a disruption – an essential phase of a DRP. The described research proposes an extension of the UIC 406 capacity calculation method [2] which allows modeling this phase. The proposed method helps to evaluate and estimate the feasibility and quality of a DRP in advance. This leads to better DRPs with better operational results and thus to better acceptance and a wider implementation of the concept.
--Why extend the UIC 406 capacity method for disruptions?-- If one considers the available dispatching measures for DRPs, it becomes apparent that –additionally to communication flow problems– the main reasons for delays during the transition phase are capacity issues [3]. Specifically, these are due to the following reasons which arise directly from the nature of DRPs and the transition phase itself: 1. The main elements of DRPs are turnarounds of trains before their scheduled terminal stop. Thus, the trains block these unscheduled turnaround sections much longer during DRP application. 2. After a disturbance occurs, a certain amount of investigation and decision taking time passes before a DRP is applied. This extra time, where some trains just wait, influences the blocking time. Therefore, supplementary blocking time elements have to be considered when calculating the capacity consumption of DRPs. 3. During the application of DRPs the higher capacity consumption by turning trains has to be compensated by a smaller amount of trains passing through these stations. However, during transition phase, the number of trains is not yet fully reduced since that process takes a certain amount of time. Thus, the capacity consumption during the transition phase is higher than during a stable DRP and may lead to congestion while the DRP itself is functioning well. UIC 406 only proposes a method for calculating the capacity consumption of lines but not for nodes [4] and for a regular schedule. In order to describe the transition process mentioned, an extended capacity method is necessary, which specifically considers stations and takes the transition phase and supplementary blocking time elements during disruptions into account.
--Methodology-- The proposed method is an extension of the calculation method of UIC 406. Besides the usual blocking times, additional blocking time elements which occur during the DRP transition phase are taken into account by the extension. Most of these additional blocking times tend to be volatile since they depend on the process quality and on other stochastic influences. As of today, few scientific findings or conclusions on stochastic influences are available [5]. For this reason, different additional blocking times are qualitatively analyzed and described regarding their volatility. For the evaluation of the method, the calculated results are compared to actual operational data of existing DRPs of big German urban railway networks. Furthermore, a heuristic which considers the decreasing number of trains in the system during the transition phase is introduced for the first time.
--Results-- With the extended UIC 406 calculation method, it is possible to calculate the capacity consumption of different DRPs and to estimate if trains queue during transition phase. The transition phase has never been analyzed –neither in research nor in practice. Before, the operational quality of a DRP may have been evaluated in advance, but not more. With the proposed method, it is now possible to evaluate how well the transition to a DRP is feasible. Thus, it complements the development process which today is mostly based on experience only and does not offer much quantitative support yet.
--Implication-- Having a calculation method which describes and explains the operational behavior during DRP-situations and which offers the possibility to optimize the operational quality of a DRP in advance is a new field of research. This approach is complementary to state-of-the-art research, such as Corman et al [6], which focuses on real-time dispatching systems based on methods and heuristics from the domain of Operations Research. Through being able to estimate the behavior in advance, less trial and error learning during the implementation of a specific program is needed. This results in more reliable DRPs from the first implementation on which leads to a shorter learning phase which in turn leads to more acceptance of the concept itself.
Page 35 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
[1] Chu, F., Fornauf, L. (2011): Vom Katastrophen- und vom dynamischen Straßenverkehrsmanagement lernen - Störfallprogramme bei Betriebsstörungen im Schienenverkehr (in German). Proceedings of HEUREKA '11 - Optimierung in Verkehr und Transport. March 2011, Stuttgart, Germany. [2] UIC Code 406 (2004): Capacity, June 2004, Paris, France. [3] Chu, F., Oetting, A. (2013): Towards Steady Operations during Disrupted Situations in Passenger Rail Transport. Submitted to the 13th World Congress on Transportation Research (WCTR 2013), July 2013, Rio de Janeiro, Brazil. [4] Linder, T. (2011): Applicability of the analytical compression method for evaluating node capacity. Proceedings of the 4th International Seminar on Railway Operations Modelling and Analysis (RailRome2011), February 2011, Rome, Italy. [5] Schranil, S., Weidmann, U. (2012): Monitoring des Störgeschehens in Bahnsystemen (in German). Verkehr und Technik, 2012, 3, 83-87. [6] Corman, F., D'Ariano, A., Hansen, I.A. (2011): Intelligent Network Traffic Management in Emergency Situations, Proceedings of the 9th World Congress on Railway Research (WCRR 2011), May 2011, Lille, France.
Page 36 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
40 - Maintaining tracks and traffic flow at the same time Presenter: Ms. FORSGREN, Malin (SICS) In an ideal world, all railway tracks would be available to trains at all times. In reality, track sections need to be closed every now and again for track maintenance and upgrades in order to ensure a satisfactory level of safety and comfort. During periods when the closed part of the infrastructure is not needed by any trains, the end customer is not negatively affected by track possessions (in this paper, we will use the established term track possession for when a track section is closed for traffic due to maintenance activities). In regions with dense traffic there is however rarely enough train free time in the timetable for all needed track possessions. As a result, track possessions and train paths often have to fight for the same capacity.
Compared with the number of papers published on the train timetabling problem, there are very few academic papers published on planning with both trains and track possessions (see the literature overview in [1]). In addition, most of the papers that do consider both trains and track possessions focus on scheduling one of them while the other is viewed more or less as a side constraint. Our paper describes a model that is capable of dealing with a realistic scenario close in time to the real-time operations, where both trains and track possessions obviously need to be considered. While our model does not give trains and track possessions completely equal treatment, it comes closer to scheduling them simultaneously than we have seen in any previously published paper.
The main idea of this paper is that there always exists a plan that best fulfills the goals for the traffic given a fixed set of track possessions. We assume that the traffic goals are represented fairly by the daily plan that is kept and maintained by the short term planners. This short term plan is the yearly plan rolled out for a specific time period, up-to-date with regard to which trains have been added or canceled.
Provided that the traffic goals can be assumed to be captured by the short term plan, it is obvious that the best production plan for the trains is the one without any kind of further disturbances caused by track possessions. This is the reason for keeping the number of track possessions fixed in the model: As soon as capacity becomes scarce, freeing up capacity by canceling track possessions (or postponing possessions beyond the time horizon of the current production planning period) will always give a better plan for the trains. Since we have not yet found an obviously "correct" way of weighing trains against infrastructure maintenance and/or renewal jobs, we simplify matters by assuming that all given jobs must be carried out within the specified time period. The task then is to optimize their placements in time in relation to the scheduled trains so that the least damage to the traffic is done.
In this paper, we present a MIP model that optimizes a plan with regard to both trains and track possessions, under the following assumptions
1. Track possessions may not be canceled
2. Trains may be canceled
3. Each train is assigned to a path in the geography (a series of consecutive track sections/main signals) and has to adhere to it unless an explicit alternative is provided
4. The goals for a train service are the departure and arrival times at specified locations, the so called important locations
5. Two train paths that have the same departure and arrival times at all important locations are considered equally good
6. Deviations in departure and arrival times at important locations make the affected train lose its commercial value to the RU proportionally to the size of the deviations. At some point that is given as input to the problem, the train loses its commercial value entirely.
The first assumption above may of course be overridden manually by ignoring any selection of track possessions in the input, effectively removing them from the problem. While the model supports the cancellation of trains (assumption no 2), it is also possible to specify which trains (possibly all) that may not be cancelled, although this may mean that there is no feasible solution to the problem. Assumptions 3-5 open up for the possibility of using alternative paths for trains to solve the temporary capacity problem, while assumption no 6 indicates how different solutions are compared to each other in the model.
A prerequisite for the model was that it has to be applicable to real-world cases, and able to provide solutions that need minimal manual post-processing before they can be given to the dispatchers. In addition to putting the model into context and describing it from a mathematical point of view, we briefly present the results on a number of scenarios based on real data. Last but not least, we discuss the practical challenges involved with modeling the capacity consumption of track possessions.
References
[1] G. Budai-Balke, “Operations Research Models for Scheduling Railway Infrastructure Maintenance”, Ph.D. thesis, Erasmus
Page 37 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
University Rotterdam (2009)
41 - A train hierarchical decomposition method to construct cyclic timetables over the PESP model Presenter: HERRIGEL, Sabrina (ETH Zurich) Increasing demands for better and more frequent services, for higher capacity utilization, and for improved reliability make railway planning processes, in particular timetabling, more and more complex. Algorithmic decision support is one promising way to cope with this increasing complexity, and the continuous progress of operations research methods offers a signifiant potential for the railway industry. However, limited experience with real applications, complicated interfaces and long computation times are still obstacles to a wider adoption in practise.
For these reasons, manual timetabling, which relies on the skills and experience of planners to find good solutions, still dominates today's practice. Traditional driving routes, grown over the history of railway companies, are often kept over the years and changed at the most by a few minutes to include new offers. Timetables are constructed sequentially, train by train. There exist corresponding algorithms using sequential timetabling [2], [1], planning a single train, considering that train path fixed, and then planning the next train in a sequence given by the user.
From a mathematical point of view, pure sequential timetabling is a greedy method, always looking for a local best solution. Unfortunately, the construction of timetables is an NP-hard optimization problem, so that finding a globally optimal solution or even a feasible solution using a sequential method is in general very unlikely. Thus, new algorithmic approaches developed for timetable construction over the last decades use a synchronous planning approach, considering all trains simultaneously, which guarantee to find the global optimum (or prove infeasibility) in finite time [6], [4]. But in contrast to the sequential planning, the computational complexity of complete synchronous approaches leads to computation times that are impractical already for medium size problems.
This research concentrates on periodic railway timetabling and a synchronous algorithm based on the periodic event scheduling problem (PESP) [7], which proved to be a suitable method to model a cyclic timetabling problem already in various case studies [3],[5]. We introduce a new decomposition method offering a parameterizable compromise between a strong sequential and a pure synchronous approach and intend to find parameter settings that help to reduce computation time without giving up too much of the quality of timetables.
The approach works by dividing all trains of a given timetabling problem into p different groups (1 <= p <= number of trains), for instance based on their different service priority or on geographical regions. Then synchronous algorithms are used to plan one group after each other, whereby all previous groups are fixed up to a certain time interval of size t_w, which is a fraction of the period length. Varying the number of groups and the degree of freedom from one iteration to the next, we examine the trade-off between computation time and solution quality for different parameter settings.
First test cases show promising results. Setting parameters to a middle ground between the extremes of a purely sequential or a purely simultaneous timetable planning approach turns out to be very effective at reducing computation time while often still providing optimal or close to optimal solutions. Real data sets provided from the Swiss Federal Railways (SBB) and close collaborations with professional planners allow a critical evaluation and analysis of our further results.
[1] P. Hachemane. Evaluation de la capacité de réseaux ferroviaires. PhD thesis, Ecole polytechnique fédérale de Lausanne, Switzerland, 1997. [2] D. Hauptmann. Automatische und diskriminierungsfreie Ermittlung von Fahrplantrassen in beliebig grossen Netzen spurgeführter Verkehrssysteme. PhD thesis, IVE, Universität Hannover, 2000. [3] L. Kroon, D. Huisman, E. Abbink, P.-J. Fioole, M. Fischetti, G. Maroti, A. Schrijver, A. Steenbeek, and R. Ybema. The New Dutch Timetable: The OR Revolution. INTERFACES, 39(1):6–17, 2009. [4] C. Liebchen. Periodic Timetable Optimization in Public Transport. PhD thesis, TU Berlin, 2006. [5] C. Liebchen. The first optimized railway timetable in practice. Transportation Science, 42(4):420–435, 2008. [6] T. Schlechte. Railway track allocation - simulation and optimization. Technical report, Konrad-Zuse-Zentrum für Informatikdienste Berlin, 2011. [7] P. Serafini and W. Ukovich. A mathematical model for periodic scheduling problems. SIAM J. Disc. Math., 2(4):550–581, 1989.
Page 38 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
42 - Consolidation of wagon flows in wagonload traffic Presenter: Mr. VOLL, Robert (Institute of Transport Logistics, TU Dortmund) We consider a decision problem from wagonload traffic which is often referred to as the Railroad Blocking Problem [1, 4, 8]. Due to economic reasons, it is necessary to consolidate freight cars from different relations for certain parts of their routes. Savings obtained by higher train utilization must be equilibrated with higher yard costs arising from additional reclassifications [6].
We present an integer programming model with capacity constraints on trains and yards. Corresponding costs are included in the objective function. Most contributions to this topic consider the North American rail freight market. In contrast, our model takes typical European restrictions on train length and weight into account. These train capacity restrictions imply a higher concentration of wagon flows on central corridors of the network [9]. Moreover, the model formulation overcomes the problem of volatile demand by finding an optimal solution over a set of demand scenarios, which can also be interpreted as particular days of a week [9]. The generated plan is constant all over the planning horizon. Hence, a stable basement for further planning steps is provided[2].
We developed different solution approaches. A column generation scheme with a nice structure is briefly presented [5]. It can keep up with state-of-the-art LP-solvers and even overcome their performance for some instances. Unfortunately, the column generation approach is not able to generate high quality solutions for the integer program. Hence, it was necessary to develop heuristic algorithms, which use ideas from the column generation process. Given a feasible solution, it is possible to compute costs for rerouting a certain relation over a particular arc of the network. Taking these costs into account, it is possible to generate new paths for each relation and hand them over to an IP solver. This idea can be used in constructive and improvement-based algorithms. The heuristic solver provides solutions for real-world instances, which could not be handled before, in short time. We present an overview over our solution efforts and visualize the solution process for an example provided by our industrial partners from Deutsche Bahn.
1. Ahuja, R. K., Jha, K. C. and Liu, J.:Solving Real-Life Railroad Blocking Problems.”, Interfaces, Vol. 37, Nr. 5, pp. 404-419 (2007)
2. Assad A (1980) Modelling of rail networks: Toward a routing/makeup model. Transportation Research Part B: Methodological Vol. 14 12: 101-114
3. Barnhart, C., Jin, H., and Vance, P. H. :Railroad Blocking: A Network Design Application. Operations Research, Vol. 48, Nr. 4, pp. 603-614.(2000)
4. Bodin, L. D., Golden, B. L., Schuster, A. D. and Romig, W.: A model for the blocking of trains. Transportation Research Part B: Methodological. Elsevier, 14(1-2), 115-120 (1980).
5. Clausen, U.; Voll, R.: Column Generation for Multi-Matrix Blocking Problems In: Operations Research Proceedings 2011: Selected Papers of the International Conference on Operations Research (OR 2011), ISBN: 978-3-642-29209-5; ITL-Autoren: Clausen, U.; Voll, R.
6. Fuegenschuh, A., Homfeld H. and Schuelldorf, H.: Single Car Routing in Rail Freight Transport. Dagstuhl Seminar Proceedings 09261, Leibniz-Zentrum Informatik, Germany (2009)
7. Luebbecke, M., Desrosiers, J.: A Primer in Column Generation. Column Generation, GERAD, New York (2005)
8. Newton HN (1996) Network Design Under Budget Constraints with Application to the Railroad Blocking Problem. Dissertation, Auburn University
9. Voll, R., Clausen, U.: A Blocking Model with Bundling Effects Respecting Multiple ODMatrices. Proceedings of the 4th International Conference on Experiments/Process/System Modeling/Simulation/Optimization, Athens (2011)
Page 39 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
45 - Generating optimal signal positions for a two-block signalling system with progressive speed signalling Presenter: Dr. WEITS, Ello (Movares) In the Netherlands railway traffic is growing. As the growth has to be largely accommodated on existing tracks, short headways are increasingly important. Headways are mainly determined by signal positions. Signal positions are severely constrained, by negative constraints (intervals that do not allow a signal), by positive constraints (intervals that require exactly one signal) and by relative constraints (constraints on minimum distances between neighbouring signals). Constraints of the first type derive from e.g. bridges and level crossings. Constraints of the second type are related to points and platform tracks; constraints of the last type are related to braking distances.
Finding a good signal positioning scheme by hand is a time-consuming task and it is impossible to prove optimality. Therefore, an algorithm that generates an optimal signal positioning scheme, taking care of all constraints, has been designed and implemented in a computer program for infrastructure planners. The algorithm calculates for a set of trains the sequence of signal positions that minimises the weighted sum of headways, each pair of trains with a common section yielding possibly two headways.
The first step of the algorithm consists in an exploration of the solution space. The solution space is described in the form of sets of signals, with a number of constraints attached to them, constraints related to one signal as well as constraints related to two signals. Thus in this step the above mentioned positive and negative constraints are fully taken into account. Secondly, a Branch & Bound algorithm is used to quickly further partition and search the solution space. For each (sub)set of the solution space an upper bound is found by calculating the objective function for a specific solution. A lower bound is obtained by letting signal positions vary within the limits of the constraints, approximating the effect on the objective function by expressions that are linear in the shifts of the signal positions. The partitioning of the solution is driven by the need for good approximations.
Deriving the lower bounds for the objective function proves to be difficult due to the fact that in general signal positions not only determine the headways, but also influence the running times. This is caused by the properties of the Dutch NS’54 signalling system as explained below.
The Dutch NS’54 (speed) signalling system essentially is as a two-block signalling system (or: a three aspect signalling system, with the familiar red, yellow and green). See [1], pages 53-56 for a clear description. In principle all blocks are long, i.e. all trains can brake from the local maximum speed to standstill within each block. Occasionally, short blocks are used through progressive speed signalling (see [1], pages 60, 61) or the (local) application of a three-block signalling system (see [1], pages 55-59). In both cases the intermediate speeds are indicated by a number (4, 6, 8 or 13, indicating 40, 60, 80 and 130 km/h, respectively).
In a number of situations signals show yellow or intermediate speeds even when there is no preceding hindering train. In these situations the signal aspects are used to restrain the maximum speed of a train, thereby influencing the running times, to an extent that is dependent on the signal positions.
A validation study showed that the signal positioning schemes produced by the algorithm slightly outperform results found manually, as long as the computer program is restrained to the same number of signals as used in the manual solution. In a number of cases the computer program suggested better solutions using a larger number of signals.
The paper will first discuss the problem definition with an emphasis on the definition of optimality. Next, the Branch & Bound algorithm will be explained. Special attention will be paid to the complication that the signal positions not only determine the headways, but also influence the running times. Finally, results of test cases from the validation study will be given. As the present algorithm is a further development of an earlier, simpler and rather slow algorithm [2], details on calculation times form an important part of the results.
References [1] Pachl, J., Railway Operation and Control, VTD Rail Publishing, USA, 2002. [2] Weits, E.A.G., Van de Weijenberg, D., “Generating optimal signal positions”, In: Ning, B., Brebbia, C.A., Tomii, N. (eds.), Computers in Railways XII, pp. 307-317, WIT Press, Southampton, 2010
Page 40 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
46 - Optimisation of simultaneous train formation and car sorting at marshalling yards Presenter: Ms. GESTRELIUS, Sara (Swedish Institute of Computer Science) Introduction
To maximise the capacity of freight trains in a carload system a hub and spoke network is often operated. The hubs are marshalling yards, where incoming cars are sorted into new outbound trains. Outbound trains have predetermined service routes, and cars are assigned to trains that pass through their destination. Cars thereby often have to be decoupled at intermediate stops, and hence it is desirable that all cars can be decoupled at once at their destination. For this reason, cars should be sorted according to their drop-off location in the train during marshalling. Furthermore, practical limitations such as non-uniform and limited track lengths need to be considered. The paper presents a novel method for freight classification under these circumstances.
Marshalling yards generally consist of an arrival yard, a classification yard and a departure yard. When a train arrives it is parked on an arrival track while its cars are decoupled. The cars are then rolled into the classification yard which consists of classification tracks. Each track used for train formation contains only cars belonging to one train at a time. When cars are rolled into the classification yard they will immediately be directed to the track that has been assigned to their outbound train, if such an assignment exists. If not, the car will be stored on mixed-usage tracks. Cars on mixed-usage tracks will be reclassified at predetermined times. During reclassification, cars whose trains have now been assigned to tracks are pushed to their respective tracks. When all cars of a train have arrived to its classification track, the train undergoes departure preparations and is then pulled to the departure yard where it waits for its departure time.
Related work
Marshalling efficiency and robustness is vital for high quality in carload freight transportation. Early literature in the area mainly considered sorting schemes that transform the input sequence into a sorted sequence regardless of what the input looks like [6], whereas more recent work also consider the “pre-sortedness” of the input in order to minimise the number of pull-out operations [5]. Other variants of the problem have also been studied [3]. A recent survey by Gatto et al. [4] gives an overview of the area.
A common assumption in marshalling methods is that all classification tracks have sufficient length, but in reality, some trains cannot fit on all classification tracks. Classification under these circumstances has been studied by Bohlin et al. [1,2]. In particular, [1] presents a branch-and-price based approach that can find optimal plans for several days within a few minutes of execution time, when any car order is acceptable. In this paper, we propose a novel extension to this model which takes car ordering into account. To the best of our knowledge, simultaneous sorting on both train destination and car order with non uniform track lengths has not been considered before.
Simultaneous train formation and sorting
The particular problem being considered is the classification track allocation, which involves deciding when a track should be assigned to a specific train. In [1], this problem was modelled with train sequences as variables. Each track can then be assigned to an appropriate sequence such that each train is present in one of the assigned sequences. Sequences are generated as needed using column generation. Pricing is done by finding a longest path in an acyclic graph with trains as nodes, and arcs between trains which can be scheduled consecutively on a classification track. The arc weights are calculated from the dual variables and the resulting mixing track usage. To adapt the problem formulation for sorting on both destination and train position, the new model also includes the necessary additional storage on mixing tracks and car movements needed to sort the cars within a train. Also, the method defining the strict partial ordering is updated, as well as the heuristics that determine roll-in, reclassification and roll-out times. We evaluate the new model on historical data from the Hallsberg marshalling yard in Sweden, and test how different levels of ordering affect the solution time and quality.
The paper is structured as follows. Section 1 contains an introduction to the marshalling problem in Hallsberg and previous literature. In Section 2, we formally define the mixing problem with car sorting on both destination and train position, and present the branch-and-price based column generation approach that we use. We also show how to extend this formulation to allow for sorting on both destination and train position, and present the new algorithms for calculating arc weights and determining the strict partial ordering. In Section 3, new pre-processing heuristics are introduced. Section 4 describes the experimental setup and results, and includes an analysis of how simultaneous sorting affects the execution time and solution quality. Finally, Section 5 concludes the paper and outlines future research.
[1] Bohlin, M., Dahms, F., Flier, H., and Gestrelius, S., Optimal freight train classification using column generation. Proc. of the 12th Workshop on Algorithmic Approaches for Transportation Modelling, Optimization, and Systems, Ljubljana, Slovenia, 2012. To appear. [2] Bohlin, M., Flier, H., Maue, J., and Mihalák, M., Hump yard track allocation with temporary car storage. Proc. of the 4th Int. Seminar on Railway Operations Modelling and Analysis, Rome, Italy, p.7, 2011.
Page 41 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
[3] Dahlhaus, E., Horák, P., Miller, M., and Ryan, J. The train marshalling problem. Discrete Applied Mathematics, 103(1–3):41–54, 2000. [4] Gatto, M., Maue, J., Mihalak, M., and Widmayer, P.. Shunting for dummies: An introductory algorithmic survey. Robust and Online Large-Scale Optimization, volume 5868 of LNCS, p.310–337. Springer, 2009. [5] Jacob, R., Marton, P., Maue, J., and Nunkesser, M. Multistage Methods for Freight Train Classification. Networks, Vol. 57(1), 87-105, 2011. [6] Siddiqee, M. Investigation of sorting and train formation schemes for a railroad hump yard. Proc. of the 5th Int. Symposium on the Theory of Traffic Flow and Transportation, p.377–387, 1972.
47 - Capacity, Comfort and Crowdedness: The El Farol Bar in Railway Systems Presenter: Mr. BOUMAN, Paul (Rotterdam School of Management, Erasmus University) In rail transport, one of the factors that determine the comfort experienced by a passenger is the crowdedness of the trains. Many rail systems are overcrowded during peak hours and underutilized during the off-peak hours. As a result, many passengers do not experience the most comfortable journey possible. As flexible working hours become more and more popular, a larger group of passengers has the ability to shift their travel times and evade the most crowded moments of the morning and afternoon peaks. However, finding a good strategy to perform this time shifting is not straightforward: if a large group of passengers decides to perform the same time shift, the situation in the newly chosen train will be just as bad as it was in the old train. Additionally, operators may choose to reduce the rolling stock capacity during the off-peak periods, in order to reduce operational costs. This means that some trains may become crowded faster and should not be used by too many passengers who shift their travel time.
In this paper, we study the interaction of passengers within the trains and its implications for the crowdedness of trains, as well as the choice the operator has in assigning capacities to the trains. To model the behavior of the passengers, we use the notion of ``Minority Games''. Most notably we consider the El Farol Bar Game[1], where the individuals of a population have to decide whether they will go to the bar or stay at home. A visit to the bar only has a positive payoff if less than 60% of the population decides to go to the bar. In this setting, the difficulty lies in the fact that the individuals have no direct means of communication, but can only rely on historic data in order to make a decision. It is easy to see that if all individuals use the same deterministic decision method, either everyone or no one will go to the bar, leading to very bad performance of the system. We argue that that this lack of direct coordination also occurs in a railway setting.
We propose a model where each individual in a population has to choose a time to travel, based on his or her individual preferences. We create different scenarios where we vary the preferences and travel options and we run multiple simulations for these scenarios. In one of the simulations, the passengers are represented by random agents, who are using an approach that can be considered as a mixed-strategy in Game Theory. In another simulation, they are represented by predictive agents, who use historic data to predict the crowdedness of the travel options they have under consideration, similar to the approach used for the El Farol Bar Game. Finally, we calculate an ``operator controlled'' assignment, where the operator assigns passengers to a travel option in such a way that the maximum number of passengers is satisfied. We then compare the performance of the simulations to the upper bound calculated this way.
We propose multiple extensions to our model. One interesting extension considers the information that is available to the predictive agents. In our basic model, the utilizations of all previous train services are available to the agents. In real life, passengers only observe the utilization of the travel options they have used themselves. In some cases, operators can provide some information on crowdedness to the passengers as well. One interesting question is how this informational policy influences the performance of the system.
Another extension is the application of this model on a network level, where passengers have different origins and destinations. In such a setting, different journeys can have some overlap in part of the network, leading to changes in the crowdedness over the course of the journey. While this increases the realism of the model, it poses new challenges to the analysis of the performance of the system and makes the simulation and agent models more complicated.
Finally, we want to consider the impact of changes in the capacity of the train services. A change in the rolling stock schedule makes historic data on utilization of trains unreliable and can therefore hurt the performance of the predictive agents. We want to study the potential decrease in performance and consider whether certain policies for rolling stock allocation and/or policies of informing the passengers about upcoming changes are better than others.
Our model is easy to understand and provides valuable insights in the dynamics of crowdedness, customer satisfaction and capacity allocation. Future versions of our model can assist policy makers in including these aspects in their decision making processes. Also, it gives railway operators valuable insights for the development of informational policies and guidance tools that can allow the passengers to make better and more informed decisions while planning their journeys.
[1] W.B. Arthur. Inductive reasoning and bounded rationality. The American economic review, 84(2):406-411, 1994.
Page 42 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
49 - Validation of Railway network Simulation Software Presenter: Mr. KIRKWOOD, David (University of Birmingham) A range of railway simulation software packages are available that are used for a variety of tasks, from timetable creation and validation to developing an understanding of the behaviour of railway networks and the analysis of the performance of signalling systems. It is essential that there is confidence in the results produced, if the tools are to be useful in the management of railways. The authors of this paper propose a method for validating train timings recorded during simulation runs by comparison with the results logged during real train journeys. This technique is then applied to two different railway operations simulators, namely, Railsys and Hermes.
The two constituent parts of an uninterrupted train’s timed journey are the running times between stations and the dwell times at stations. These times are normally calculated by the so-called wheel stop and wheel start times. The proposed validation approach is to collect independent manual recordings of these timings during real train journeys and to compare the results with simulation log files.
In Britain, Network Rail log the movement of trains in the TRUST database. However, this data is not accurate enough for the purpose of comparison as it is logged at track-circuit and axle counter boundaries and at the locations of signal boxes, rather than wheel stop and start timings at stations. The manual collection of timings is thus preferred as its accuracy is assured by the researcher. GPS location data is not currently logged as standard and there are issues with poor satellite visibility where railways operate in built-up areas, cuttings and tunnnels.
A number of conditions must be met to ensure that the collected data is useful for comparison. The station to station runs must be the same as those simulated in the network model being tested and the modelled infrastructure data must correspond with the physical state of the network. Additionally, the train types being compared must be the same, with the correct performance and the trains must take the same route through the network. The trains should not be disturbed by restrictive signalling and must be running on time, according to the working timetable.
In addition to the wheel stop and wheel start times, door open and closed timings and GPS trajectories of the journeys are recorded. If a significant discrepancy is found, further investigation can be made into the source of the problem by using this supplementary information. Once the real-time data has been collected and verified, the corresponding simulation runs are logged and the results are compared in one second steps, using a spreadsheet.
For this paper, the case study involved the northbound route from Finsbury Park to Hadley Wood, on the London North East Route of the Network Rail infrastructure. The demonstration journey was recorded several times on board Inner Suburban services from Moorgate operated with Class 313 trains. The data was recorded as described and then compared with output data from the Railsys and Hermes simulators. In this basic example, the two simulators produced very similar results, within 1% of each others’ timings. This was not surprising since both are based on the application of Newton’s laws and use identical infrastructure data. The real runs produced significantly different results and the discrepancy between the real and simulated runs was investigated further and the decision was taken to conduct further tests with longer journey times and different service patterns.
The proposed technique, repeated for a range of train types and all regularly used routes through network would give a great deal of confidence that most aspects of a simulation are accurately modelled during a run without disturbance to the timetable.
Page 43 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
51 - Simulation-based universal method of evaluation for railway-nodes by dimensioning infrastructure in rail-based transport Presenter: Mrs. LI, Xiaojun (University of Stuttgart) Because of the scarce space for building new tracks in Europe, it is impossible to make big changes to the railway infrastructure. In order to meet the increasing demand of rail-based transport, it is important to improve the efficiency of existing infrastructure by means of in-creasing the performance of the rail operation through process optimization, which is sup-ported by innovative capacity research.
Depending on specific tasks and simulation-tools capacity research can be carried out with different methods on different levels of details. There are three main categories: quality of operation, system performance and bottleneck analysis. Within these methods, it lacks still in the consistent user-oriented combination of macro-, meso- and targeted-oriented microscopic investigation. Additionally, coactions of different parts of railway nodes are limited due to the high computation complexity of nodes. These limits can be overcome with the new developed method in the research project RePlan. Thereby, a continuous user-friendly process of capacity research can be realized with the integration of macro-to microscopic investigation in both synchronous and asynchronous simulation-tools.
In order to evaluate the system performance of infrastructure elements from the microscopic point of view, an appropriate subdivision of an infrastructure is the prerequisite for the correct computation of further results. Because bottlenecks in nodes are resulted from the interaction of infrastructure and operating program, capacity research should consider not only occu-pancy rate but also direction-related hindrance rate arisen from the occupancy of intercon-nected infrastructure elements. However, currently there are not suitable infrastructure mod-els to describe such relationship. For this reason, a new target-oriented model is developed to divide an infrastructure into infrastructure elements with consideration of running directions. Furthermore, with the new algorithm the system performance of microscopic infrastructure elements is evaluated.
Another focus of the evaluation of railway nodes is bottleneck analysis, which is still rudimen-tarily considered by capacity research. In the project RePlan the methodology of bottleneck analysis has been further developed. The potential and effective bottlenecks can be identified by means of evaluating three indicators: bottleneck relevance, bottleneck significance and potential route reserve.
From existing macro- and mesoscopic capacity research, the recommended area of traffic flow and level of delays is derived. In combination with the new method for microscopic ca-pacity research, the innovative universal method is developed for evaluating railway infra-structures and operating programs independently from their complexity and dimensions.
Page 44 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
52 - Simulation and functional requirements of energy-efficient driving Presenter: Prof. RICCI, Stefano (Sapienza University of Rome - DICEA) Railway transport is the most sustainable transport mode from an environmental point of view and more in general the mode with the lower production of externalities. Therefore the Commission promotes a significant modal shift from the road to the rail transport as the basis of a minimum increase of about 10% of railway traffic in each of two decades from 2000 to 2020. With regard to these, due to the little number of accidents and a limited landscape impact in competition with the road mode, energy consumption, air pollution and noise emissions represent the main negative externalities of railway transport. Though, rail transport comes with an inherently high potential for energy efficiency and eco-friendliness, nevertheless an information deficit regarding available technological advancements, and the application of legal requirements, as well as the perception of high additional investment costs have caused rail transport to fall behind other modes of transport in recent years. DICEA Department of Sapienza University of Rome has been facing these topics since many years: within the Project ECORailS, TRAINER (both co-funded by the European Commission and the IEE - Intelligent Energy Europe Programme) and other relevant own research activities. The energy consumption of a train [1], [3] strongly depends on the energy consumption of the traction system, of auxiliaries and of comfort function equipment, rather depending on the real efficiency of instantaneous operational parameters and maintenance conditions. Therefore the energy consumption of a train depends on characteristics of train and conditions of the line (i.e. service profile [1] [3]), environmental conditions, comfort parameters, planned operation and different driving styles, that could cause relevant variation (up to 49%) among different drivers operating the same traction unit and timetable [6]. The energy consumption calculation needs extended variable data bases and parameters to be kept under control. The UIC UNIFE TecRec 100_001 [3] shows how to calculate and measure the energy consumption of a train in a specific service profile or in a standard service profile. Driving style can play a rewarding role for the energy saving in itself and it is an important reference for the evaluation of technologies and operational measures aiming at energy saving [5]. DSB (Danish State Railway) decided to implement in all rolling stock a system that handles all information required to perform the mathematical calculations necessary to arrive on time and save energy, such as position, timetable and speed limitations. A detailed report [6] has been produced to determine the potential energy savings by introducing this system in Denmark. By a series of tests and results analysis, the report states that DSB will save up to 15% on the traction energy while improving punctuality. On the other side RFF (French infrastructure manager), SNCB/NMBS (Belgian TOC) and others infrastructure managers and operators decided to adopt energy meters on board the trains. This covers a lack of information about the real energy consumption of trains, though the authors worked on realistic and suitable methodological approaches for data handling in function of clearly defined objectives. In fact such data seem difficult to be useful if not well managed. The research activity is now aiming at providing a simplified approach for energy consumption evaluation useful to determine the more suitable driving style to save energy and respect the timetable in a specific line with a specific train. The proposed methodology is based on a simplified set of data: characteristics of the line usually provided by the infrastructure manager to operators and from them to their drivers, the rolling stock characteristics usually provided by the supplier as design documentation. The methodology is based on the simulations and is strictly related to the driver prospective and its already available documentation (timetable sheets for drivers) and regulation. This research is aiming at define driver assistance procedure and systems that let the driver focus on safety related aspects acceptable from the National Safety Authorities. The following driving styles for energy efficient driving can be applied: cruising (DSB strategy) or coasting (French drivers strategy) [7], reducing maximum speed, using valleys and hills. The rewarding driving strategy depends also on the on board power of the train [8]. The paper will describe the first results of the proposed simplified approach to determine the most suitable driving styles to save energy and respect the timetable constraints by line section and timetable structure. The assessment of strategies are based on analytical calculations and simulation of drivers behaviors on the Italian railway network.
Key references [1] S.Ricci, E.Cosciotti, A.Baldassarra, Deliverable 7, “Integration of technological feedback from the User Platform and the consortium into the guidelines”, ECORailS Project www.ecorails.eu [2] B. Spiegel, “Energy metering an overview of the complete system”, Energy Efficiency day in Tours, France, 23-26 September, 2009 [3] UIC UNIFE TecRec 100_001, 11 March 2010 [4] TRAINER Project - TRAining programmes to INcrease Energy-efficiency by Railways http://w3.disg.uniroma1.it/Trainer/ [5] Baldassarra A., Cosciotti E., Ricci S., “Energy Efficiency and Environmental Criteria in the awarding of railway vehicles and services: methodologies of implementation and monitoring”, Poster presentation of the World Congress on Railway Research, Lille, 2011 [6] P. Buchwald, “GEKKO. Guide to Energy Efficient Driving”, Energy Efficiency Days, September 24th, 2009 [7] D. Vastel, “SNCF Energy Savings Program”, Energy Efficiency Days, September 23rd, 2009 [8] H. Rohrer, “Impatto dello stile di guida e della programmazione dell'orario sul consumo energetico”, Conference Per un uso attento dell’energia nel trasporto su ferro, Milan, 25 January 2010
Page 45 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
53 - The value of enhanced service reliability, case: transformation from bus to light rail Presenter: Dr. VAN OORT, Niels (Goudappel Coffeng Mobility Consultants) The value of enhanced service reliability, case: transformation from bus to light rail
Dr. Niels van Oort Goudappel Coffeng mobility consultants NvOort@Goudappel.nl
Service reliability is an important quality characteristic in public transport. However, in cost-benefit analyses (CBA), this quality aspect is rarely taken into account explicitly. It is more common to calculate vehicle indicators (e.g. punctuality) instead of passenger focused metrics. In a CBA however, the latter is required to illustrate the potential benefits of a project (Li et al. (2010)). Figure 1 shows the results of a quick scan of randomly selected public transport projects in the Netherlands. It is demonstrated that the attention to calculating service reliability effects is limited. Most of the time, a qualitative assessment or expert judgement is used, while proper calculations would be more appropriate since most public transport projects aim at improving service reliability. In our research (Van Oort 2011), we presented the main impacts of vehicle variability on passengers, being additional waiting time, a distribution of passenger travel time and crowding. In this paper we describe how to calculate these effects and how to take them into account in a CBA. We demonstrate the added value of the method in an actual case study: a Dutch project of transforming an overcrowded bus line into light rail.
Figure 1: Results of quick scan service reliability in CBA
The Dutch government required a CBA to financially support the construction of a light rail line in Utrecht, the Netherlands, between the central station and the Uithof, where the hospital and university are located. Nowadays, about 30.000 passengers travel here by double articulated buses. Demand forecasts (Goudappel Coffeng 2011) show a growth towards 45.000 passengers per day in 2020. To facilitate reliable service, plans are made to shift from bus to light rail services. Figure 2 shows this line, which is about 8 km long and operates about 23 x per hour per direction during the morning peak.
Figure 2: Proposed route of light rail line Central station - Uithof vice versa
The main benefit transferring the bus line into a light rail line is, next to less direct emissions, that service can be provided by less vehicles than in the case of bus operations. And since less vehicles are needed, the hindrance for crossing traffic (i.e. car and bike traffic) is less and more important, the probability of bunching of vehicles is limited. Growth of demand is expected to be larger in the light rail case than in the bus case due to the “rail bonus”. Bunschoten (2012) presents an additional growth of about 5% due to this factor. However, the construction and operation costs of light rail may be higher than bus operations.
In the cost benefit analysis of this case we calculated the reliability benefits of transferring the existing bus system into a light rail system. We compared 2 future situations (in 2020): 1 Reference case: No infrastructure will be constructed and operations will be similar to the operations nowadays (i.e. partly right of way). Capacity of infrastructure is limited. 2 Light rail case: In this case the service is operated by trams with own right of way operations.
We calculated the passenger effects concerning the reduction of waiting time, distribution of travel time and the increase in the probability of finding a seat. For these calculations we used AVL data of the existing bus services. We simulated the new APC and
Page 46 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
AVL data, adjusting the dwell times and the level of bunching. The predicted AVL and APC data enabled us to calculate the passenger effects. In the reference case, the level of service will be very low due to high passenger demand and insufficient bus infrastructure. In case of the light rail line, sufficient infrastructure is provided and besides, light rail services require fewer vehicles, thereby reducing the probability of bunching. We calculated the additional travel time per passenger, the distribution of travel time and the probability of having a seat as shown in table 1. Due to the high level of reliability in the light rail case, the passenger effects are neglectable.
Table 1: Passenger effects of unreliability of services in reference and light rail case Reference case Light rail case Average additional travel time per passenger due to unreliable services 4,9 min 0 min Distribution of travel times (st. dev) 2,4 min 0 min Probability of having a seat 32% 58%
After the calculation of these values, the monetary values of these effects were calculated, using values of time, values of reliability and values of the probability of having a seat (Rand 2005). Table 2 shows the total costs and benefits of the project (Ecorys 2011), showing the substantial contribution of improved reliability to the positive score of the cost benefit analysis. This ratio convinced the Dutch Minister of Infrastructure and Environment to support the project by €110 million. We demonstrated that our framework concerning calculating benefits of service reliability is valuable and may be applied directly in practice
Table 2: Additional costs and benefits of light rail line compared to reference case Value compared to reference case (millions in 2011) Investment costs -€222 Operating costs €66 Total costs €288
Additional ticket revenues €40 Increased travel time €67 Service reliability effects - Less waiting time €123 - Reduction in distribution €78 - Increased probability of finding a seat in the vehicle €4 External effects (emissions, safety, etc.) €8 Total benefits €336
Benefits-costs +€48 Benefit cost ratio 1,2
References
Bunschoten, T. (2012). To Tram or not to Tram. MSc. Thesis, Delft University of Technology
Ecorys (2011), CBA Uithoflijn, Results report (in Dutch)
Goudappel Coffeng (2011), Demand forecasting and service reliability analysis Uithoflijn Utrecht (in Dutch)
Li, Z., D.A. Hensher and J.M. Rose (2010), Willingness to pay for travel time reliability in passenger transport: A review and some new empirical evidence, Transportation Research Part E, 46, pp. 384-403.
RAND Europe and AVV (2005), The value of reliability in transport: Provisional values for The Netherlands based on expert opinion, Leiden/Rotterdam.
Van Oort, N. (2011), Service Reliability and Urban Public Transport Design, T2011/2, TRAIL PhD Thesis Series, Delft
Page 47 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
54 - A recoverable robust solution approach for the real-time railway crew rescheduling problem Presenter: Mr. VEELENTURF, Lucas Petrus (Rotterdam School of Management, Erasmus University) In this paper we study real-time railway disruption management and we focus on crew rescheduling under uncertainty. Every day railway operations have to deal with disturbances and disruptions in their operations. Small-scale disturbances, often caused by minor delays, do not require a significant change of the schedules other than letting propagate (and hopefully diminish) the delays. This research however focuses on large-scale disruptions, often caused by failures of the infrastructure (e.g., broken overhead lines or defect signals), weather conditions or the unavailability of resources (e.g., defect rolling stock or delayed crew members). Such disruptions make the planned timetable, rolling stock and crew schedules infeasible, which inevitably requires a major adjustment of the schedules. Mind that the exact duration of the disruption is generally not known upon its start. For example, recovery works on a broken railway line segment may take 2 hours in an optimistic scenario, however they may stretch up to 4 hours in a pessimistic scenario.
Effective disruption management is key to a good operational performance of a train operating company. We refer to Jespersen-Groth et al. (2009) for a detailed description of the disruption management process. In the ideal situation the rescheduling of the timetable, rolling stock and crew schedules is integrated into one approach. However, this leads to problems of unsolvable complexity, and therefore most research focuses on rescheduling the resources sequentially: first the timetable, then the rolling stock, and the crew as last, even if this leads to some sub-optimality.
Current approaches to reschedule the crew (e.g. Rezanova and Ryan (2010) and Potthoff et al. (2010)) use a deterministic approach to deal with the duration of the disruption: based on an initial estimation of the duration, the rescheduling step is carried out again and again when new information about the duration of the disruption becomes gradually available.
In this paper we consider methods that take the uncertainty in the duration of the disruption explicitly into account such that, if the disruption takes longer than expected, less rescheduling effort is necessary, and less trains are cancelled due to lack of crew. We assume that the timetable and rolling stock schedule have already been adapted based on the expected duration of the disruption. The primary criterion in assessing the quality of a schedule is to count the number of trains which cannot be covered by any crew. If trains have to be cancelled due to lack of crew, the rolling stock must also be rescheduled again. Other costs like operational and process costs are also taken into account, but have less priority.
We propose a quasi-robust optimization approach, which is built upon the recently introduced concept of recoverable robustness (see Liebchen et al. (2009)). The main idea is to compute a good schedule for an optimistic duration of the disruption in such a way that it can easily be turned into a feasible schedule if the disruption takes longer than expected in the optimistic scenario. This is achieved by requiring that the computed crew duties have an alternative for trains for which it is uncertain that they will be operated.
We call a duty of a crew member recoverable robust if the duty can also be performed by the crew member the disruption appears to take longer than expected. This means that if the approach generates recoverable robust duties for every crew member, no duty have to adapted if the disruption has a longer than expected duration. However, the approach admits to balance the robustness and the operational costs by requiring a certain percentage of the rescheduled duties to be recoverable robust. Since not all duties have to be recoverable robust we call this approach a quasi-robust approach. The quasi-robustness is rather easy to incorporate into existing crew scheduling algorithms without substantially raising their running time.
We demonstrate the value of our approach by computational tests carried out on large-scale instances of Netherlands Railways (NS), the main operator of passenger trains in the Netherlands. For almost every instance, against some slightly higher cost for the optimistic scenario (without additional uncovered tasks), we can reduce the number of uncovered tasks in stage 2 if rescheduling was necessary since the pessimistic scenario took place. This is an important feature for the application in practice, since it means that against some slightly higher cost, we can prevent future cancellations of trains due to lack of crew if the disruption takes longer than expected. Moreover, our algorithm can explore the consequences of several robustness levels, and thereby help the decision makers to find the best balance between robustness and operational cost.
The contributions of this paper are summarized as follows: 1) We describe disruption management methods for railway crew rescheduling under uncertainty; 2) We develop a framework for dealing with the uncertainty about the duration of the disruption; 3) We evaluate the proposed approach on large-scale crew rescheduling instances of NS.
References
J. Jespersen-Groth, D. Potthoff, J. Clausen, D. Huisman, L.G. Kroon, G. Maróti, and M.N. Nielsen. Disruption Management in Passenger Railway Transportation. In R.K. Ahuja, R.H. Möhring, and C.D. Zaroliagis, editors, Robust and Online Large-Scale Optimization, p. 399-421. Springer, New York, 2009.
Page 48 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
C. Liebchen, M.E. Lübbecke, R.H. Möhring, and S. Stiller. The Concept of Recoverable Robustness, Linear Programming, and Railway Applications. In Robust and Online Large-Scale Optimization, p. 1-27. Springer, New York, 2009.
D. Potthoff, D. Huisman, and G. Desaulniers. Column Generation with Dynamic Duty Selection for Railway Crew Rescheduling. Transportation Science, 44: 493-505, 2010.
N.J. Rezanova and D.M. Ryan. The train driver recovery problem - A set partitioning based model and solution method. Computers & Operations Research, 37: 845-856, 2010.
60 - Development of Rolling Stock Assignment System for Taiwan High Speed Rail Presenter: Dr. LAI, Yung-Cheng (National Taiwan University) Railway rolling stock is one of the most expensive assets of a railway agency or company. Therefore, efficient utilization of rolling stock is a one of the most important objectives pursued in practice. The rolling stock assignment process in the operational level aims to assign appropriate equipments to cover a given set of utilization schedule with considerations of practical requirements, such as maintenance, station capacity, and rules for circulation. Due to its complexity, this task is still a manual process at Taiwan High Speed Rail Corporation (THSRC). Experienced railway practitioners can generally create a good and feasible plan; however, there is no guarantee of the optimality of the solution. Also, considering only short-term process may lead to a myopic decision away from the global optimum. Consequently, there is a need to develop a decision support tool to automatically generate the optimal rolling stock assignment. In this research, we developed an operational rolling stock assignment model by using network optimization techniques. This model can help THSRC utilize their rolling stock efficiently and assist the rolling stock investment plan. Using this decision support tool will help railways with similar characteristics maximize their return from rolling stock investment and also provide reliable service to their customers.
61 - Development of Base Train Equivalents for Headway-Based Analytical Railway Capacity Analysis Presenter: Dr. LAI, Yung-Cheng (National Taiwan University) A conventional railway system usually has multiple train types with various service patterns operating on the same line. Differences in train characteristics lead to varied capacity impacts to the system. Analytical rail line capacity models commonly define capacity as the maximum number of trains that can be operated on a section of a track with an expected level of service within a given time period. However, a particular unit (trains/hour or trains/day) does not reflect the train type the unit refers to. In this study, we propose using the concept of base train equivalent (BTE) to standardize different train types in accordance with a specific base train type selected by the user. This concept is similar to the passenger car equivalent in highway transportation, which converts trucks to passenger car units. Due to there are many different types train in operation we will promote the method for computing BTE of multiple types. A headway-based approach is also developed to determine BTE based on the result obtained from analytical railway capacity models. This method was implemented and validated by using data obtained from the conventional railway systems in Taiwan. Using the proposed method, capacity measurements from different lines or systems can be compared and evaluated, resulting in meaningful and useful attributes.
Page 49 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
62 - Timetable synchronization for large-scale urban rail transit network Presenter: Ms. WANG, Yuanyuan (School of Transportation and Logistics,Southwest Jiaotong University,China;Institute of Railway Systems Engineering and Traffic Safety,Technische Universität Braunschweig,Germany) Improving service quality is a very important way to increase the modal split of urban rail transit system. Designing a timetable that enables smooth transfer with short waiting time for the passengers is a crucial factor for good service. There is a wealthy of literature on timetable synchronization for public transit. As the requirement of some Chinese urban rail transit operation companies, however, the already existing model in the literature can’t be put into practice directly. The cross-platform interchanges between different lines allow passengers to transfer without changing to another platform. Some companies require good coordination of different lines' timetable at cross-platform interchanges to make sure that the passengers could enjoy “immediate”transfer. The above coordination is even required to have the absolute priority. The majority of the existing models just aim to minimize the total transfer waiting time, and the cross-platform coordination is not taken into account.
The paper describes a multi-objective mixed integer programming model and heuristic algorithm for optimizing timetable synchronization in large-scale urban rail transit network. The proposed model and algorithm is based on the following considerations: (a) The model should first aims at offering minimum transfer waiting time for passenger on the cross-platform interchanges. (b) The model can minimize the “just miss” cases, which describe the phenomenon that when the passenger arrive at the platform, the train just leave.(c)The model and algorithm should be generally applicable to the design and optimization of a wide range of practical urban rail transit network, and should not favor particular network configurations.(d)Solutions obtained from this article should given reasonably good result in a reasonable amount of time, as permitted by the current computing power.
An important innovation in the model is the three layers of objectives.The first layer is to minimize the total passenger transfer waiting time. The second layer is to minimize transfer waiting time for the cross-platform interchanges.And the third one is to minimize the “just miss” case. The heuristic algorithm is based on the genetic algorithm, and the tabu search and hill climbing method are embedded to enhance the local search capability.
Page 50 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
63 - Analysis of delays on saturated railway line Presenter: Mr. RICHTER, Troels (Rail Net Denmark) Background The classical approach to analysis of train delays is either based on analysis of delays over a certain threshold and their reasons or consists of an analysis of all delays per measuring point irrespective of cause. On the lines with the densest traffic on the Danish railway network, these classical approaches have not been able to yield sufficient information on which factors that triggers delays and poor performance. In some cases small delays of one train cause an accumulation of delays of the following trains where as apparently identical delays in other cases do not cause this accumulation. Also, some train departures in the fixed interval timetables (departure minute) perform much worse than others. However, it is often not clearly understood which systematic parameters influence this. The timetabling process is today based on the experience from the timetablers and the feedback from the CTCs after scrutiny of the suggested timetables. However, there is only a limited feedback loop from the actual performance of the existing timetable to the timetablers. Many inaccuracies between the actual performance and the timetables are absorbed by the timetable supplements, thus preventing them from absorbing delays occurring during operations. This leads to a lack of timetable stability and punctuality. A typical problem in the railway system is that it is difficult to find the triggers of poor on-time performance and that symptoms are then attempted treated instead of the actual root causes. An identification of the root causes of poor performance can thus lead to an improved punctuality and increased capacity.
Method This paper demonstrates three novel methods to analyse train run performance on saturated networks in order to identify trains that frequently cause escalating delays, and locations where delays frequently originate. Based on this knowledge, it is then possible for other functions within the Danish Railway system, such as the “Precision Groups”, to make in-depth investigations of why the pinpointed specific trains / train systems or locations perform poorly and to take corrective action. The first method is a process-based method. Defining the arrival at a station as the outcome of the process, the correlation between this and a number of input parameters such as delay at arrival to the line, delays obtained during passenger exchange, and signalling aspect on the line is tested. Trains are grouped based on their position in the fixed interval timetable (departure minute). As expected, some of these groups of trains perform significantly worse than other. In this way, the factors explaining delays are identified and corrective actions can be taken in order to improve on-time performance. On most parts of the Danish railway network, data is available describing the occupation of block sections and important track circuits. This information is used to construct an enriched train run history describing the time-wise distance between trains as well as their delays and to deduce the most likely signalling aspect on a block section level. This way, it is possible to describe both the development of delays at a specific measuring point or for a specific train. The second method demonstrated uses this enriched train run history at block section level to investigate occurrence and triggers of queues. Based on the signalling aspect, trees of trains causing restrictive signalling are constructed. An analysis of the trains at the head of the tree is then carried out finding common factors, such as location, time of day and train position in the fixed interval timetable. Analysis is also carried out of where on the line geographically time is lost and delays occur. With this knowledge, it is possible to identify the bottlenecks in the infrastructure with greater detail than previously possible. Furthermore, it is possible to identify the train that most frequently cause queues. Having indentified these trains or train systems, apparently identical situations are identified and the differences between them are investigated. This allows for a precise identification of the root causes of the queuing. Finally, a method using statistical correlation between train delays at certain measuring points over a longer period of time is described. A high degree of correlation between trains that on average are delayed indicates that there may be a planning error in the timetable. In some cases, the correlation is between a train and the subsequent train, where as the link in other cases is less obvious. None the less further analysis of these cases reveal the connection between these trains and corrective measures can be taken. In most cases this consists of timetable adjustments.
Results The three methods presented all help identifying trains and trains systems with inferior performance and locations where delays occur more frequently than other. Having this knowledge, it is then possible for other functions within the Danish Railway system to make in-depth analysis and to take corrective actions. The corrective actions can consist of timetable supplement adjustment, timetable changes, adjusted train and crew plans, changes in shunting plans as well as infrastructure adjustments. The methods presented in this paper, thus makes it possible to focus the punctuality improving efforts to achieve higher punctuality and capacity.
Key references Richter, T., Data aggregation for detailed analysis of train delays, COMPRAIL 2012 13th International Conference on Computer System Design and Operation in the Railway and other Transit Systems.
Page 51 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
64 - Synchronal algorithms for determination and evaluation of conflict resolution scenarios for real-time rescheduling Presenter: Prof. OETTING, Andreas (DB Netz AG) 1 Introduction and requirements
One of the important objectives in railway operation is to manage deviations in the timetable and thereby reduce the delays. In order to achieve this DB Netz AG developed algorithms for real time rescheduling KEKL (conflict detection and conflict solution). These are implemented in pro-totyping software and tested in real operation. This should help the line dispatchers in the control centers to make their decisions in the case of block section conflicts between trains. This re-search in the project FreeFloat 1 was partially funded by the German Federal Ministry of Econom-ics and Technology. A requirement of DB Netz AG is the possibility to use these algorithms both for railway lines and in bigger nodes. Additional objectives besides the punctuality are better use of capacity and more stability in the operational timetable especially in congested areas of the railway network. Solutions have to be calculated in a very short time in order to produce useful conflict solutions in real-time operation. KEKL is a so-called online optimization. In compliance with experts from the control center it is agreed that the solution time is limited: - 1 second for solutions without alternating tracks - 2 seconds in 95% of cases for solutions with alternating tracks - 4 seconds as maximum solution time Another requirement of the practical experts is that KEKL is a decision supporting system. KEKL does not have to only give the best conflict solution. KEKL must offer various solutions (see Chapter 3) to enable the dispatcher to choose the best solution based on his experience. The algorithms have to be non-discriminatory and traceable in order to take the binding EU regu-lations into account.
2 Conflict detection
It is important to calculate on a very microscopic network since KEKL is a live system for rail-way traffic. Thus KEKL works with the same data base as the other control center systems. KEKL receives relevant information as live data from the control center: - all infrastructure information (e.g. tracks, signals, stopping places, platform length) - positions of all trains - current prognosis of the time-distance graphs - train characteristics (e.g. engine power, train mass, train length) Using this information KEKL is able to calculate the blocking times of all trains and the so-called blocking stairway. So KEKL detects in the prognosis when trains will be blocking the same track section at the same time. This overlapping time is the criteria for a conflict. The block with the first overlapping time is the conflict position. Conflicts are relevant when the conflict time is between a reaction time and a conflict detection horizon. The conflict detection horizon is use-ful because a prognosis of the time-distance graph spreads with the time of forecast. Therefore the accuracy of a detected conflict with its impacts decreases. Another positive effect is that KEKL does not need to identify all overlapping times and the dispatcher can focus on the next relevant conflicts. In addition to conflicts between trains KEKL also detects infrastructure conflicts, e. g. when tracks are closed due to maintenance work or an accident and the train must be rerouted.
3 Conflict solution
In every approach for rescheduling a decision is necessary whether to focus on approaching the global optimum or reaching a low computation time in microscopic networks. The solving time, non-discrimination and traceability mean that we can only accomplish the task with heuristics. By approximating further conflict solutions using a specific evaluation we can get close to the global optimum. We decided in the first version of KEKL to handle two-train conflicts with synchronal solving methods only. This is possible because in the process of selecting a conflict and choosing a con-flict solution the dispatcher can always interact. Practice test results demonstrate that the majority of line dispatching conflicts are two-train conflicts and remaining conflicts with three or more trains are mostly solvable by decompose them into two-train conflicts. Furthermore, in every KEKL con-flict solution further conflicts with other trains are taken into consideration using an evaluation. Different solving heuristics have to be used for different conflict situations and solving strategies in order to meet the requirement of short solution times. In this respect it is very important to limit the solution space regarding feasible und useful solutions. Thus KEKL is able to solve the problem much faster and help to use multiprocessing applications. The conflict situations are divided into: - both trains running along the same line in the same direction - both trains running along the same line in the opposite direction - trains sharing only one common signaling block
Page 52 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
- the conflict is only located in one common track in a station Different standard solving strategies are implemented in KEKL to achieve fast, standard and good conflict solutions: - maintaining the train order (referring to the conflict position) - altering the train order (referring to the conflict position) - changing the route of one or both trains (opposite tracks and detour lines) Note that the solution space grows in a linear way by the number of alternative tracks in one sta-tion in the case of changing the train order, but explodes in a combinatorial way by changing the route. This is the greatest challenge for our algorithm development. KEKL includes special driving dynamic calculations for bending of train paths (adjusting of the speed). The train paths may be bended either for the entire common itinerary (option 1) of the two trains investigated for several sections of the common itinerary (option 2) separately. Option 1 may lead to a high capacity consumption. In case of option 2 it may be difficult for the driver to follow the calculated train path. The developed algorithms consider both aspects by limiting the number of changes of the speed per train kilometer and the speed difference. For the evaluation of different solutions KEKL uses a disutility function considering three elements: - additional delay of both trains - penalties for changing platform (for passenger trains) - approximation of further conflicts (impact on delays of third trains) For solving infrastructure conflicts the algorithms are similar, but KEKL solves these conflicts by changing the route (opposite track and detour lines).
4 Conclusions and future prospects
With KEKL the division FreeFloat of DB Netz AG has produced a prototype application for real-time train rescheduling taking practice requirements into consideration. Concerning the usability on lines and big nodes we are currently testing the algorithms and the intermediate results shown effects in reducing delays and making better use of capacity. The so-lution time complies with the given constraints. Only in the case of multiple rerouting over many very large stations we see a need for development. Also our algorithms are non-discriminatory and traceable in respect to EU regulations, what is very important for the DB Netz AG. We will also look for new extensions. One extension is the special handling of conflicts with more than two trains, especially in big disturbances. Another area of research examines conflict solu-tions with only one feasible choice to be solved automatically including only a final check by the dispatcher (one click solution).
Page 53 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
65 - Route Oriented Waiting List Passenger Management in Indian Railways Presenter: Mr. MISHRA, ABHYUDAY (Indian Institute of Technology, Kharagpur (India)) Indian Railways (IR) is operating ten thousand trains everyday and passengers from seven thousand different originating stations reaches their destination round the clock. But even with such a huge fleet of carriage (passenger coaches) satisfying the need of different economic segments of passengers, IR is not able to supply adequate number of berths to cater the demand. Demand pattern is also not homogenous throughout the train travel route. Volume of Demand in different Origin-destination (O-D) pairs is also not smooth for the entire train route. This paper presents a system approach which will match capacity to demand in particular O-D pairs and then suggests a model by introducing a new concept of GLOBAL waiting list Number (GWL).Utilizing aggregate demands method to a set of trains in chosen OD pairs, not confirmed waitlisted passengers are offered an alternate accommodation. A Global waiting list number is provided in addition of existing waiting list status, in printed train tickets which will be valid for all the train of this route till confirmation or completion of a day. GWL increases chance of getting reserved Seat by provisioning accommodations in alternative trains of the same route on FCFS basis. GWL concept is very useful in clearance of seasonal passenger rush caused due to peak demand pattern in a particular OD pairs. It also overcomes the hardship of passengers from standing in tedious long queue for cancellation of unconfirmed ticket within prescribed time limit. The concept of Global waiting list number will also help railway operator in planning new or special trains to carry the crowd in particular O-D pairs.GWL provides a real time data enabling divisional engineers to take decision for extension of trains in other travel legs or attaching extra coaches to transport crowd in certain O-D pair.
Indian Railways runs more than 10 thousand trains every day but still there are overcrowding in every trains.
For long distance IR trains have three classes for reserved class passengers: Air conditioned coaches with reserved berth, non AC coaches with reserved berth and Non- AC unreserved coaches with provision of seating purpose only. Reserved AC coaches are further categorized for different segment as First class AC with independent coupe, second class AC coaches and three AC coaches. Berths of each class are booked in advance to get the advantage of first come first served (FCFS) policy of IR. Late-decision travel can book berth through TATKAL product. Tatkal reservation facilities are provided to meet the urgent travel requirement of passengers who plan their journey at short notice. Tatkal tickets are sold at10-30% extra charges.
Not getting confirmed journey ticket is the main issue raised in this paper. The booking of advance ticket is open for 90 days which can be brought from authorized windows, internet or through SMS. If a seat is not available, then the ticket is given a wait listed number; else the ticket is confirmed, and a berth number is printed on the ticket. A person receiving a wait listed ticket must wait until there are enough cancellations to enable him to move up the list and obtain a confirmed ticket. If his ticket is not confirmed on the day of departure, he may not board the train.
Thus presently Indian Railways passenger reservation system generates two types of waiting list tickets: Tatkal waiting list ticket and normal waiting list tickets. Allotment of accommodation is done by the computer as per pre-defined logic. During final train chart preparation Tatkal waiting list passengers are given priority over normal waiting list passengers. Even after final chart printing, vacant berths/seats are booked by Train Ticket Examiner (TTE) i.e. train conductor on train also. Some tickets are up gradated to higher class against vacant berth during final printing out of chart. Very few research works has been done on Indian Railways passenger’s management system. Bharill and Rangaraj (2008) in his paper studied the passenger service in premium segment of Rajdhani Express. Gopalakrishnan and Narayan,(2010) discussed a model to allocate train capacity among multiple travel segments on an Indian Railways train route with several stops. He used a linear programming model and data preprocessing and post processing to determine the optimal capacity allocation on multiple travel legs.
Our aim is to benefit maximum number of passengers by seat confirmation in chosen OD pairs so that passengers originated at origin station can be transported to at least destination by any available trains in the chosen route. For this system thinking approach (practical studies of Railway industry) is utilized, which will view Problems as a part of overall system rather than reacting to specific part. It is based on the belief that the component parts of a system can best be understood in context of relationships with each other and with other systems, rather than in isolation. System thinking focuses on cyclical rather than linear cause and effect. We will not consider individual train to transport passenger in chosen train instead a set of trains passing the selected OD pairs will be chosen and their whole berth inventory is taken as integrated entity. The passenger with10-13 trains in their desired OD booking individually for any trains will lose some of the advantages which are inherent in having one ticket for a set of trains. We will not consider individual train to transport passenger in chosen train instead a set of trains passing the selected OD pairs will be chosen and their whole berth inventory is taken as integrated entity. The passenger with10-13 trains in their desired OD booking individually for any trains will lose some of the advantages which are inherent in having one ticket for a set of trains.
Page 54 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
66 - Expected Passenger Travel Time as Objective Function for Train Schedule Optimization Presenter: Mr. SELS, Peter (Katholieke Universiteit Leuven, Belgium) 1 Necessity of an Objective Function
Most literature in train schedule optimization concentrates on the constraints of the problem. A few examples are [4, 5, 6, 7, 8]. Very little is found on the optimisation criterium: the objective function [2, 3]. We believe one of the main reasons is that in generating schedules that are robust against delays, this is the harder part of the mathematical modelling. Indeed, unlike the hard constraints, it requires the use of statistics, namely delay statistics. It also requires accurate measurement of these delays or derivation from train logs. We consider total passenger travel time as the most important criterium for judging the quality of a passenger train schedule [1, 9, 10, 11], so this is what we want to minimize as the objective function. Approaches to train scheduling that lack an objective function like this are unable to make an informed decision about the size of supplements. These methods usually resort to inserting buffers on a train traject of up to some maximum percentage of the total train travel time [5]. Such approaches do not take into account the typical delays at a certain location, nor the number of passengers sitting in, entering, leaving or transferring to or from this train at that location. Therefore, it is impossible to calculate the travel time increase or decrease resulting from a buffer on all passenger streams. In our objective function, we use passenger numbers instead of trains as weight factors. While optimizing a schedule, we focus on decreasing this objective function and this implies that no or few planned transfers are missed, especially if many people are transferring. Not only transfers, but also ride and dwell actions should be planned with supplements that are neither too small such that they are unable to absorb any practical delay, nor too big such that they create unnecessary idle time in practice. Thanks to our detailed objective function, we can make this trade-off and determine optimal buffer times which result in better schedules, even optimal ones for passengers.
2 Derivation of the Objective Function
In our paper, we analytically derive the stochastically expected passenger time on each location throughout the network. Firstly, we decomposed a general train network action graph into actions of five distinguished types: ride and dwell train actions and the source (embarking), transfer and sink (alighting) passenger actions and then identified four types of subsequent action-pairs, representing separate passenger streams: departing, through, transfer and arriving passengers. Then, we also take knock-on delays into account, where only the passenger numbers on the second train matter. Thanks to this, we were able to analytically derive the probability and cost functions for each of these local passenger flows. This includes the probabilistic trade-off for catching or missing a next action. These local functions of one or two supplement variables, indicate the cost trade-off and can be used to evaluate a given schedule on quality of these local supplements and even determine the values of the optimal local supplements. In order to do all this analytically, we assume a general negative exponential distribution for the actions’ durations. Thirdly, summing these local functions for the whole schedule results in a global cost function that can be used by an optimizing solver directly. Minimizing the total objective function over all supplements, respecting all schedule constraints, effectively makes the trade-off between all local supplements.
3 Use of the Objective Function
We show that the constructed objective function can quickly evaluate an existing schedule. Evaluation, even of passenger train timetables for a whole country, takes less than a second. For optimization, we demonstrate that the Belgian schedule with all 288 passenger trains during a morning peak hour, solver times are only about twenty minutes, when knock-on delays between all train pairs per common track sections are ignored, and up to a few hours, when these knock-on delays are considered.
4 Main Contributions
We believe that all factors relevant to determine optimal buffers and supplements for a whole
Page 55 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
network are present in our objective function and that this is a unique research result. The application of this objective function to optimize the passenger train schedule for a whole country also shows its value in practice.
References
[1] Dewilde, T., Sels, P., Cattrysse, D., Vansteenwegen, P., “Defining Robustness of a Railway Timetable”, In: Proceedings of 4th International Seminar on Railway Operations Modelling and Analysis (IAROR): RailRome2011, 2011. [2] Goverde, R.M.P., “Synchronization Control of Scheduled Train Services to Minimize Passenger Waiting Times”, In: Proceedings of the 4th TRAIL Annual Congress, part 2, TRAIL Research School, Delft, 1998. [3] Goverde, R.M.P., “Improving Punctuality and Transfer Reliability by Railway Timetable Optimization”, In: Proceedings of the 5th TRAIL Annual Congress, TRAIL Research School, Delft, 1999. [4] Kroon, L., Dekker, R., Vromans, M.J.C.M., “Cyclic Railway Timetabling: A stochastic Optimization Approach”, Geraets, F., Kroon, L., Sch¨obel, A., Wagner, D., Zaroliagis, C.D. (eds), Algorithmic Methods for Railway Optimization. Lecture Notes in Computer Science, pp. 41-66, 2007. [5] Liebchen, C., “Periodic Timetable Optimization in Public Transport”, Operations Research Proceedings, Volume 2006, Part II, 29-36, 2007. [6] Nachtigall, K., “Periodic network optimization with different arc frequencies”, Discrete Applied Mathematics, Vol. 69, pp. 1-17, 1996. [7] Schrijver, A., Steenbeek, A., “Spoorwegdienstregelingontwikkeling (Timetable Construction)” Technical Report, CWI Center for Mathematics and Computer Science, Amsterdam, (in Dutch), 1993. [8] Serafini, P., Ukovich,W., “A Mathematical Model for Periodic Scheduling Problems”, SIAM Journal on Discrete Mathematics, Vol. 2, pp. 550-581, 1989. [9] Sels, P., Dewilde, T., Cattrysse, D., Vansteenwegen, P., “Deriving all Passenger Flows in a Railway Network from Ticket Sales Data”, In: Proceedings of 4th International Seminar on Railway Operations Modelling and Analysis (IAROR): RailRome2011, 2011. [10] Vansteenwegen, P., Van Oudheusden, D., “Developing railway timetables which guarantee a better service”, European Journal of Operational Research”, Vol. 173, pp. 337-350, 2006. [11] Vansteenwegen, P., Van Oudheusden, D., “Decreasing the passenger waiting time for an intercity rail network”, Transportation Research Part B: Methodological, Vol. 41, pp. 478-492, 2007.
Page 56 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
69 - HELRA – The Improvement of Helsinki Railway Yard’s Functionality Presenter: Mr. PITKäNEN, Jukka-Pekka (Ramboll Finland Oy); Mr. CAPEK, Karel (Ramboll Finland Oy) BACKGROUND
Problems with rail capacity, caused by a rapidly increased demand for railway traffic, are common to many infrastructure managers. The Helsinki Railway Yard, as the busiest and most important railway yard in Finland, has been a significant capacity bottleneck during the last several years. The traffic in the railway yard has increased 40 % during the last 10 years and will continue to increase in the future after the completion of the two major railway infrastructure projects. The yard serves about 200 000 daily passengers with approximately 100 long-distance and 850 short-distance train departures a day from 19 tracks. The yard is extremely sensitive to disturbances, particularly during the winter season. The disturbances mainly originate from problems related to the infrastructure, the rolling stock and from various technical problems during the peak hours.
The Helsinki railway yard hasn’t been redesigned during its previous enhancements, only built and enhanced upon the existing infrastructure, and this has resulted in a complicated and ineffective switch and track layout regarding the shunting movements of trains and the rail capacity. This has effectively made it impossible to optimize the railway yard’s space for running the current traffic model functionally. The only way to get around the problems described previously and to run even more trains, in order to meet the future traffic demands, is the complete redesigning and partial rebuilding of the switch and track layout between the Helsinki and Pasila stations.
In order to find an optimal solution to the capacity increase of the Helsinki Railway Yard, the Finnish Transport Agency has launched a project called HELRA with the following key objectives:
1. to find an optimal track layout model for the railway yard to fulfill the future traffic demands 2. to increase the capacity of the railway yard 3. to reduce the sensitivity to disturbances 4. advance decision making processes and serve the forthcoming planning stage
This paper describes how the HELRA project solves the current main problems of Finland’s busiest and most important railway yard, how the local conditions shape the planning and operations of railway traffic in Finland and also show how the railway traffic simulation, combined with innovative theoretical and empirical research, can come up with functional solutions to the problems described.
METHODOLOGY The HELRA project combined use of OpenTrack simulation tool with a set of indicators to compare several proposals to improve the track layout of the yard.
The role of the Open Track simulation as a tool was to test the proposed layouts with the predefined traffic demand model for peak hours operation. The simulations were carried both with and without pre-defined disturbances. Simulation results were evaluated using the set of developed indicators. The evaluation results had impact on the layout planning via a feedback loop. This helped to specify which of the proposed track layouts, including switches and the signalling, is the most effective and what features should possibly be optimized further.
The aforementioned indicators, which have been developed with the help of empirical data and known causalities, include such commonly used factors as
- capacity consumption - the number of trains - the heterogeneity of traffic
but also some innovative new ones that have been shown to correlate strongly with train delays in our latest studies, which include
- overtakings/meeting ratio - the proportion of pairs of trains driving to opposite directions - the possibility to add trains into timetable structure.
The first two of these new indicators have been verified to correlate with train delays by empirical data. The third new indicator was verified in the following way: by using a method for defining the index for traffic heterogeneity (as defined in the previous paragraph) in a given timetable, and knowing the positive, proven correlation between the index and expected delay sum of trains, we first counted the heterogeneity index for various timetable. Following this, we were able to verify a negative correlation between the increasing potential of trains and expected delay sum of trains.
Page 57 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
Another project, besides HELRA, which will benefit from the use of the previously described indicators, deals with a common problem in northern latitudes: the frost. The conditions at the Finnish Railways network can be characterized by frost during the winter season and a low proportion of double tracks. As a result, railway traffic is prone to many primary and secondary disturbances during the winter season. In addition, rail capacity diminishes in many parts of the railway network due to the lowered speed limits (“frost limits”) during winter. This is not ideal as the capacity consumption should optimally be at the same time smaller than during the summer, spring and autumn to prevent the decrease of train punctuality.
Evaluating the harm caused by frost is a challenging and multi-layered question. The goal of the project is to find a measure with which the level of harm caused by frost in different destinations could be evaluated and therefore arranged in the order of the severity of harm inflected on the traffic. This measure, The Frost Harm Index, will help guide the decision making process of how to prioritize projects and direct financial resources. It will also provide some additional value to the HELRA project.
CONCLUSION
This paper demonstrated a combination of microscopic simulation and a set of indexes to evaluate impacts of several track layout variants to capacity of railway yard. In addition, this paper gives an overview on the research of capacity indicators used and developed in Finland during the past decade. In their master’s thesis and studies after the graduations the authors have studied, evaluated and developed new methods to evaluate rail capacity and the quality of Timetables in Finnish circumstances. The Finnish rail network is less dense than most of the rail networks in Europe and filled with single track lines, thus, many commonly known methods needs to be calibrated / modified before using them in Finland
Page 58 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
70 - Linking Theory and Practice to the Efficient Delivery of Railway Line Capacity Presenter: Mr. WARDROP, Alex (WorleyParsons) Academic research and analysis of railway line capacity tends to focus on the abstract. This is partly because of the researcher’s need to deconstruct the elements of line capacity. However, it is also partly because of the difficulty of access to real railway alignment, track configuration, signal layout and/or train performance data necessary for physical analysis. The result is that researchers often develop analytical models driven only by observable data, such as signal clearance times, and thus are unable to fully explore all the factors controlling line capacity. This is particularly so in North America. The European approach to measuring, although not necessarily fully interpreting, line capacity has culminated in the UIC 406 process. Based on observable data and block compression, this process can take into account a diversity of train types. However, this process relies on trains being able to run between refuging locations and appears to enshrine German signalling practices. As such, it cannot take into account the impact of multiple aspect signalling as applied in North America, UK, France, Australia, New Zealand, etc. In any case, this process does not give any guidance as to how signalling should be laid out to deliver a particular line capacity outcome – it only measures how a specific layout performs for an input set of trains. A more desirable approach would be to have a process which assembles a set of performance specifications, accommodates multiple train types (and their performances) where appropriate and can take into account different signalling codes. It must produce metrics on whether, or not, the performance specifications have been met. Finally, it should provide desktop means of stress testing signalling layouts. We have developed such a process in Australia in response to its particular geographic and rail traffic characteristics. It primarily relies upon the joint application of train performance modelling and signal system simulation. It can be supplemented with high-level infrastructure and train flow modelling. Australian rail networks are long and thin with relatively little route diversity and limited multiple trackage. Rail passenger traffic is overwhelmingly commuting (average passenger trip lengths are approximately 20 kilometres). Rail freight traffic is generally trainload bulk commodity or intermodal. Often rail passenger and freight traffic has to share corridors, in which case there are train performance disparities which can consume valuable line capacity. This is particularly so in the Sydney region which has a single gauge railway network and, by Australian standards, relatively generous levels of multiple track. Behind all this is an imperative to provide 24/7 network access (not currently deliverable) to both passenger and freight trains while keeping the necessary level of infrastructure economical. Two case studies are presented: reconfiguration of an existing joint passenger and freight corridor to handle future traffic growth and to provide 24/7 access for all traffics; and designing the signalling layout for a new suburban passenger railway with challenging alignment and train performance issues. Steep gradients, train braking characteristics, train performance disparities, headway delivery and future migration to ETCS are features common to both case studies. The geographic and temporal distributions of traffic, track configurations and the resolution strategies are points of difference. The purposes of these case studies are to identify the service specifications, uncover and discuss the specific operating issues (particularly train braking), identify line capacity resolution strategies, explain and illustrate the design processes and discuss specific outcomes. Access to railway alignment, track configuration, signal layout and train performance data is the key to successful application of our process.
References [1] ARTC, Engineering Standard – NSW: Signalling: Braking Distance – SDS 03 (formerly RIC Standard: SC 00 13 01 03 SP), Australian Rail Track Corporation Ltd, Sydney, March 2005 [2] ATSB, Rail Investigation Report 2001/002: Collision between Suburban Electric Passenger Train 6369 and the Empty Express Electric Train 6371, Australian Transport Safety Bureau, Civic Square, ACT, November 2001 [3] ATSB, Rail Investigation Report 2002/001: Collision between Suburban Electric Passenger Train 1648 and the Suburban Electric Empty Train 1025, Australian Transport Safety Bureau, Civic Square, ACT, April 2003 [4] Cambridge Systematics, National Rail Freight Infrastructure Capacity and Investment Study, technical report, Association of American Railroads, 2007 [5] Comfort N, The Channel Tunnel and its High Speed Links, pp 75-77, The Oakwood Press, Usk, Mon, UK 2006 [6] Forster ADJ, The Capacity of Rapid Transit Railways, Sydney University Engineering Society, Sydney, NSW, 29 November 1918 [7] Jones JCM, Walker AE, The Application of Models of Single Railway Track Operation to Evaluate Upgrading Alternatives, Rail International, 1973 [8] Kontaxi E, Ricci S, Techniques and Methodologies for Carrying Capacity Evaluation: Comparative Analysis and Integration Perspectives, Ingegneria Ferroviaria, 12.2009 [9] Lang AS, Soberman RM, Urban Rail Transit: Its Economics and Technology, MIT Press, Cambridge Mass USA, 1964 [10] Mackenzie S, Train Timetabling on a Complex Network, Conference on Railway Engineering (CORE) 2000, Adelaide, SA, 2000 [11] OPENTRACK: http://www.opentrack.ch/opentrack/opentrack_e/opentrack_e.html [12] Pachl J, Application of Blocking Time Analysis for Specific Signal Arrangements, Transportation Research Board, 84th Annual Meeting on 9-23 January 2005 Compendium of Papers [13] Parkinson T, Fisher I (ed), TCRP Report 13: Rail Transit Capacity, Transportation Research Board, National Research Council, National Academy Press, Washington DC USA, 1996 [14] Pudney P, Howlett P, Mackenzie S, Harris D, Wardrop A, R3.104: Corridor Capacity Analysis, CRC for Rail Innovation, Brisbane, Qld, 2009
Page 59 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
[15] Pudney P, Wardrop A, Generating Train Plans with Problem Space Search, Proc 9th CASPT, San Diego Ca USA, 2004 [16] RailCorp, Standard Working Timetable 2009: Freight Services, All Days, Rail Corporation of NSW, Sydney, NSW, October 2009 [17] RailCorp, Standard Working Timetable 2009: Passenger Services, Mondays to Fridays, Rail Corporation of NSW, Sydney, NSW, October 2009 [18] RailSYS: http://www.rmcon.de/english/uber_railsys.html [19] UIC Leaflet 406, Capacity, 2004 [20] Wardrop AW, Simulation of Railway Operations, Institute of Railway Signal Engineers (IRSE) Australian Section, Technical Meeting, Sydney, NSW, 3 November 1972 [21] Wardrop AW, Development and Use of the MTRAIN Train and Signal System Performance Program, Proc CompRail 90, Rome, 1990
Page 60 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
71 - An optimal rescheduling algorithm from passengers' viewpoint based on mixed integer programming formulation Presenter: Prof. TOMII, Norio (Chiba Institute of Technology) 1. Introduction Trains in Japan are known to be very punctual but sometimes disruptions occur due to accidents, natural disasters, malfunctions of facilities. When trains are delayed, a series of changes are added to train schedules in order to prevent increase of passengers’ inconvenience. This task is called (train) rescheduling. Major methods used in rescheduling are: change of departing orders of trains, change of tracks at stations, change of assignment of train-sets, (partial) cancelation of trains and so on. Train rescheduling is currently done by human experts, called train dispatchers. Nowadays, aids by computers for rescheduling such as prediction of future train operation are emerging, but the task to make a rescheduling plan is still done by dispatchers and aids by computers are longed for. In order to develop a computer aid for rescheduling, we have to develop an efficient algorithm to make rescheduling plans. But to develop such algorithms is extremely a difficult task especially in busy railway lines because of the following reasons [1]. - Trains are running densely and we have to deal with a large-sized problem. - The problem becomes a complicated combinatorial problem because we have to take various kinds of methods to change train schedules into account. - We need a quick algorithm because we have to solve the problem on-line. - We have to take passengers’ dissatisfaction into account.
We believe passengers’ dissatisfaction should be explicitly included in the criteria. In some past reports, only delays are considered but this does not suffice especially in busy railway lines because: - If we cancel trains, obviously delays will decrease but on the other hand congestion of trains and stations will increase a lot and the situation might become chaotic. In these cases, trains must not be cancelled even if the delays increase. - In urban railway lines, it is more important to transport massive amount of passengers. In these lines, to equalize the intervals of trains is sometimes more important than to reduce delays.
2. Related works There have been published a lot of works for an algorithm to make a rescheduling plan (see [2] ). D’Ariano [3] has proposed a formulation based on an alternative graph model and introduced an algorithm using meta-heuristics and showed its effectiveness for actual data. But this paper does not explicitly deal with dissatisfaction of passengers. Tomii [4] has proposed an algorithm based on PERT formulation to minimize passengers’ dissatisfaction. In this paper, various kinds of situations when passengers might complain are sorted out a priori. These situations are considered to be constraints and the problem is regarded as a kind of constraint satisfaction problem and an algorithm to give a rescheduling plan based on meta-heuristics is introduced. One of the drawbacks of meta-heuristics, however, is that it is usually difficult to get an optimal solution and it is almost impossible to know how close the obtained solution is to the optimal solution. Recently, a mixed integer programming (MIP) approach is attracting the attention. One of the merits of MIP approach is that it is always possible to get an exactly optimal solution. Törnquist proposed a MIP formulation to manage disturbed traffic conditions by means of train reordering and rerouting [5]. Gely also introduced a MIP approach for rescheduling algorithm [6]. In these papers, however, passengers’ dissatisfaction is not included in the objective functions. Although Schӧbel introduced MIP formulation which minimizes sum of all delays over all passengers [7], it is assumed that passengers take a fixed route and that they wait for the next train which arrives after a cycle time when a connection is missed. Chigusa has introduced a rescheduling algorithm based on MIP formulation which aims at decreasing loss of passengers [8]. But the timetables and track layouts dealt in the paper is too simple and the sizes of the problems are too small. In addition, only limited types of rescheduling methods are implemented (e.g., train cancellation is not considered) and passengers are assumed to change trains at most once.
3. An optimal rescheduling algorithm from passengers’ viewpoint based on mixed integer programming In this paper, we introduce an algorithm which produces a rescheduling plan optimal from passengers’ viewpoint based on MIP formulation. We have introduced a mathematical programming model for passenger’s behavior. The objective function of our algorithm is the total sum of each passenger’s loss, which is defined as “difference between the travel time in the rescheduled timetable and the travel time in the normal timetable.” The basic idea is similar to [8] but we have succeeded in dealing with more complicated timetable and more general assumption for passengers’ behavior. We have implemented the formulation using CPLEX and confirmed that the obtained solutions are reflecting passengers’ demand. In our paper, we will describe the formulation of train traffic and passenger behavior in detail and give results of the numerical experiments.
References [1] N. TOMII, ed.: Techniques to restore disrupted train traffic (in Japanese), Ohmsha (2010)
Page 61 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
[2] J. Törnquist: Computer-based decision support for railway traffic scheduling and dispatching: A review of models and algorithms, ATMOS 2005 (2005) [3] A. D’Ariano and M. Pranzo: An advanced real-time train dispatching system for minimizing the propagation of delays in a dispatching area under severe disturbances, RailHannover07 Hannover (2007) [4] N. TOMII et al: Train Rescheduling Algorithm which minimizes Passengers' Dissatisfaction, Innovations in Applied Artificial Intelligence, Lecture Notes in Artificial Intelligence 3533, Springer (2005) [5] J. Törnquist and J. A. Persson: N-tracked railway traffic re-scheduling during disturbances, Transportation Research, Part B, 41 (3):342–362 (2007) [6] L. Gely, G.Dessagne and C. Lerin: Modelling Train Re-scheduling with Optimization and Operational Research Techniques: Results and Applications at SNCF, Proceedings of WCRR08, Seoul (2008) [7] A. Schöbel: Optimization in Public Transportation. Springer (2006) [8] K. Chigusa, K. Sato and T. Koseki: Passenger-Oriented Optimization for Train Rescheduling on the Basis of Mixed Integer Programming (in Japanese), Trans. IEEJ, Vol. 132, No.2 (2012)
Page 62 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
72 - Development and application of Danish Key Performance Indicators for Railway Timetables Presenter: Mr. SCHITTENHELM, Bernd (DTU Transport & Rail Net Denmark) 1. Background and purpose Company managers want to be able to measure and evaluate the processes taking place in their companies to evaluate performance levels. This is also the case for the railway sector. Politicians demand a higher efficiency from infrastructure managers (IM) and train operating companies (TOC) by requiring improved products with reduced costs at the same time. Introducing Key Performance Indicators (KPI) can help achieve a higher efficiency. Creation of a feasible and attractive railway timetable is the most important process for both TOC and IM. When measuring the quality of the produced timetables, one also measures the success of the applied timetabling process.
KPI for railway timetables most be based on a common accepted set of timetabling criteria. This paper begins with a brief description of the process that lead to the creation of a common Danish list of prioritized timetabling evaluation and optimization criteria. It included individual interviews and a joined timetabling criteria workshop with the most important stakeholders. See below for the result:
* High prioritization: Consumption of capacity on railway line sections & Systematic timetables are preferable
* Medium prioritization: Robustness of the timetable & Societal acceptance of the timetable
* Low prioritization: Travel time of trains & Attractive train transfer options
2. Process The paper proposes a set of KPI for each criterion from the common Danish list of railway timetabling criteria. Today some KPI are already being used. In the paper some of these KPI are reused, others are improved and most are new developments. For each KPI a practical application is presented.
2.1. Consumption of capacity on railway line section The paper recommends the continued use of the UIC 406 methodology. An example of a network wide UIC 406 analysis based on the Danish national timetable from 2010 is presented. This analysis has been the basic scenario for several capacity investigations for future timetable scenarios.
2.2. Systematic timetables are preferred Using the Swiss timetable regularity index, the concept of timetable patterns is defined and introduced, leading to two new KPI: One to be used in case of big differences between timetable patterns and one to be used for small variances. Both KPI focus on the length of the time span a timetable pattern is applied. This approach is tested on two examples: With big and small differences between timetable patterns.
2.3. Robustness of the timetable The normal timetabling approach to ensure a robust timetable is to add running and dwell time reserves. Agreed upon timetable planning rules between IM and TOC include time reserves. This paper recommends the degree of deviation from the planning rules as KPI: Comparing timetabled running and dwell times with planning rule times. Negative deviations can give a less robust timetable and positive deviations can be seen as scheduled waiting time.
Robustness of the timetable depends highly on the traffic complexity. The paper recommends using a “Conflict Risk”-Index for stations, comparing the number of potential conflicts with the number of high risk conflicts. The later depends on a predefined minimum headway between two potential conflicting train paths.
By introducing the concept of timetable fix points, measuring the complexity of railway traffic can be done in a more varied way. Timetable fix points are e.g.: Crossing stations, overtaking stations, level railway junctions and stations where trains split up or couple. Recommended KPI include: Number of fix points per train path, the average for a group of train paths, geographical distribution of fix points which leads to a risk profile for a train path.
2.4. Societal acceptance of the timetable A successful timetable must be accepted by society, both by politicians and customers. Measuring the societal acceptance level can be done with customer satisfaction surveys. The paper recommends doing this as a KPI. A minimum score must be achieved in the following subjects: Punctuality levels, travel time, frequency and attractive train connections. It is important that these surveys are carried out regularly, by only one and independent organization. Inspiration can be taken from “Passenger Focus” in United Kingdom.
2.5. Travel time of trains The paper recommends using the degree of travel time prolongation as KPI: Comparing the shortest timetabled travel time with the
Page 63 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
shortest possible travel time (direct non-stop trains) according to the planning rules.
2.6. Attractive train transfer options A train transfer will often cause a longer travel time. In Denmark each station has been given a predefined minimum train transfer time. This is the minimum time between arriving and departing transfer trains for a planned transfer. The paper recommends the degree of transfer time prolongation as KPI: Comparing the timetabled transfer time to the predefined minimum transfer time.
A train transfer is most attractive if it takes place at the same platform. To measure the attractiveness of train transfers the paper recommends to use the degree of optimal transfer conditions: Comparing the number of planned train transfers taking place at the same platform with the total number of planned transfers at a station.
3. Results and perspectives Based on a common accepted list of Danish railway timetable evaluation and optimization criteria, this paper presents a set of KPI for railway timetables.
Introducing the concept of timetable patterns to measure how systematic a given timetable variant is, has shown great potential. Two KPI are presented that use the time span of the most used timetable pattern to measure how systematic a timetable is. One KPI is needed for big differences between timetable patterns and one for small.
Applying the new concept of timetable fix points for measuring the level of traffic complexity in regards to timetable robustness is very promising. Such an analysis can cover everything from a single train path to all train services, all trains running on a given railway line section to the entire railway network and time wise from one hour to an entire operational day. Identification of timetable fix points needs a high level of knowledge about timetabling and railway infrastructure and therefore is done manually. An automated approach could greatly improve the quality control of an entire timetable.
To measure the societal acceptance level of an implemented railway timetable, one must ask the railway customers and the traffic political decision makers. In United Kingdom the independent non-departmental organization “Passenger Focus” conducts half yearly satisfaction surveys amongst train passengers. This concept could be applied. A future improvement is to include railway freight customers, both existing and potential future ones.
Measuring the degree of travel time prolongation in a railway timetable as a KPI is useful in a socio economic context. When preparing future timetables the timetable planner must ensure a correlation between the degree of travel time prolongation for a travel relation and the number of affected passengers: Minimizing the scheduled waiting time in passenger minutes.
Train transfers are most often reason for prolongation of travel time. Calculating the degree of timetabled transfer time prolongation is important for a socio-economic evaluation. Numbers of affected passengers must be considered. Calculations are made complicated since it is manual work to identify which train transfer options are relevant for an arriving train. A future improvement would be to make the predefined minimum needed train transfer time dependent on the platform track usage of the involved trains.
Page 64 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
73 - Application and validation of a timetabling algorithm to a large Italian network Presenter: Dr. MEDEOSSI, Giorgio (University of Trieste) The timetable represents a key element for the competitiveness of railways, since it allows exploiting the existing infrastructures at their maximum. To support timetable planners in meeting the variety of requirements given by operators and customers, since the early 90s a rich literature has appeared on the Train Timetabling Problem (TTP), that is the problem of determining, for each train, the arrival and departure time at each of the stations the train visits (see, e.g., the recent survey [1]). For the TTP it is possible to distinguish two main classes of models: the ones based on the solution of the Periodic Event Scheduling Problem [2,3] and the ones based on the multicommodity flow models [4,5]. For the above models both exact and heuristic solution approaches have been proposed. In particular, heuristic solutions are implemented when the rail network considered is of medium-large size. Despite of the wide literature, few applications of the TTP under real conditions can be found. This work presents a large-scale application of the heuristic solution approach to a multicommodity flow model. The model is tested on a real network and the results are compared to the timetables created by timetable planners using conventional methods in order to assess its concrete applicability. The model is based on a mesoscopic infrastructure, which allows a significantly higher accuracy compared to the macroscopic models used in most scientific work. The mesoscopic model allows a realistic estimation of the headway times and of the conflicts on lines and stations as well as a calculation of running times and time-losses performed with the same detail enabled by simulation models. In order to maximize the accuracy in the definition of the timetable, various parameters can be defined for each train, including the buffer times, the priority and the allowances. The model is applied to the rail network of the North-East of Italy. It is tested under different realistic demand conditions, for example considering an increase of the demand for freight slots or a different structure of regional services. Moreover, it is used to obtain a rough estimate of the maximum capacity for freight trains combined to fixed passenger services and the effects of infrastructure improvements. In addition, the limits of the mathematical modeling of a complex system and the consequent difficulties of introducing decision support system in the real word operations are discussed. [1] R. M. Lusby, J. Larsen, M. Ehrgott, and D. Ryan. Railway track allocation: models and methods. OR Spectrum, 33:843-883, 2011. [2] P. Serafini and W. Ukovich. A mathematical model for periodic event scheduling problems. SIAM Journal of Discrete Mathematics, 2:550-581, 1989.
[3] L. Kroon, D. Huisman, E. Abbink, P.-J. Fioole, M. Fischetti, G. Maróti, A. Schrijver, A. Steenbeek, and R. Ybema. The new Dutch timetable: The OR revolution. Interfaces, 39(1):6-17, 2009.
[4] A. Caprara, M. Fischetti, and P. Toth. Modeling and solving the train timetabling problem. Operations Research, 50(5):851- 861, 2002.
[5] R. Borndörfer and T. Schlechte. Solving railway track allocation problems. In J. Kalcsics and S. Nickel, editors, Operations Research Proceedings 2007, pages 117-122. Springer Berlin/Heidelberg, D, 2008.
Page 65 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
74 - Blocking times vs headway times to support timetable planning Presenter: Dr. MEDEOSSI, Giorgio (University of Trieste) Railway timetables comply with a number of technical and operational bounds, which reflect on one side the layout and the equipment of lines and stations and on the other the requirements of the Train Operating Companies. While the use of blocking times to support timetable planners by visualizing these operational bounds has become widespread in various countries, in others pre-defined headway times are still in use. The use of the headway times might appear rough when compared to the blocking times, which allow higher accuracy in representing the effective infrastructure usage simply based on the microscopic network model and do not require pre-estimating the headway times. However, while the blocking times are calculated automatically, the estimation of the buffer times, which is fundamental for maintaining a high reliability level without hindering the capacity of the network, is still performed empirically by most Infrastructure Managers. As a result, despite of the accuracy of blocking times a fundamental element is still estimated roughly on the basis of empirical rules. On the other hand, when using the headway times, it is possible to estimate them with the support of stochastic simulation, in order to include the buffer times required to keep the system stable under realistic perturbations. Thus, a relatively accurate estimation of the operational bound to be considered in timetable planning is performed. Moreover, the simple data model used is more suitable for optimal timetabling algorithms. This paper presents a method for estimating the headway times, including the buffers on large terminal stations and on their branch lines, where the conflicts between trains arriving and departing strongly influence the capacity. The method improves the approach used on the Italian Network by the Infrastructure Manager Rete Ferroviaria Italiana. It includes the operational bounds in the so-called Specifiche Tecniche (Technical Specifications), which must be considered as technical rules to be respected in timetable planning. Moreover, they are the starting point for the capacity estimations based on compression (UIC 406): instead of the blocking times. While some results are valid in general, in this work the approach is applied to the main station of Venice (Venezia S.Lucia), a 22-track terminal station that represents the critical point in the network of North-Eastern Italy. A timetable defined upon the results of the method is compared to a another one defined from scratch using the blocking times in order to assess the differences between them in terms of number of usable capacity and corresponding robustness, estimated using stochastic micro-simulation.
Key References:
Ciuffini, F. (2007). - Capacità di una stazione elementare di testa, Ingegneria Ferroviaria, 10, CIFI, Roma Ciuffini, F. (2011) - Qualità trasportistica dell’orario ferroviario, Ingegneria Ferroviaria, 6, CIFI, Roma Huerlimann D., Longo G., Medeossi G., (2009) “Stochastic micro-simulation as a timetable robustness estimation tool” In 3rd International Seminar on Railway Operations Modelling and Analysis. Zurich. Hansen, I. A., Pachl J., (2008) Railway Timetable and Traffic. Eurail press Hamburg.
Page 66 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
75 - MULTI-STRATEGY BASED TRAIN RE-SCHEDULING DURING RAILWAY TRAFFIC DISTURBANCES Presenter: Mr. IQBAL, Muhammad Zeeshan (Blekinge Institute of Technology) Introduction Railway networks with heterogeneous traffic and a high capacity-utilization are often sensitive to disturbances. A smaller disruption can propagate and result in large consecutive delays. The dispatchers then may need to re-schedule the traffic, which concerns e.g. changing train orders, altering the assignment of tracks and platforms, and changing departure times. This problem has received an increasing attention in literature, and extensive studies are presented in e.g. [1, 2, 3, and 4]. In our previous research, the re-scheduling problem is addressed by proposing a parallel algorithm [6] based on a greedy Depth-First-Search (DFS) branch-and-bound method [5]. The parallel algorithm applies parallelization at time T0 (i.e. the time when the initial disturbance occurs). The approach uses a master-slave paradigm where a candidate list, NC, is constructed. NC contains the next event to execute for each train and is sorted according to earliest start time first in the initial configuration. The number of slaves (i.e. workers) created corresponds to the number of candidates at T0. Each worker performs a DFS branch-and-bound search and informs the other workers if improved solutions are found. The ability to find a good re-scheduling solution is dependent on the selection of a suitable candidate to re-schedule in each iteration. How the candidates in the list are sorted has shown to affect the results significantly. Recently (see [7]), we evaluated the effect of having the greedy algorithm using a selection of different sorting strategies: (i) s0: earliest start time, (ii) s1: earliest track release time with two versions s1α and s1β, (iii) s2: smallest buffer time, and (iv) s3: shortest section runtime. These sorting strategies complement each other and no single strategy is superior to the others for all disturbance scenarios. Therefore, this paper proposes a parallel multi-strategy based re-scheduling approach.
Parallel Approaches In order to evaluate how to design the parallelization scheme and use of different sorting strategies, we decided to evaluate three different approaches. Each worker is assigned one of the strategies mentioned previously and sorts its candidate list according to that: • Approach 1 – 1 strategy and 1 worker/candidate: In this approach, the number of workers equals the number of candidates at T0 (i.e. minimum 50). All workers are assigned the same sorting strategy. All five strategies are evaluated, but separately. • Approach 2 - 5 different strategies and 1 worker/strategy: Here, we create five workers which are assigned one strategy each. • Approach 3 – 5 workers/candidate: Here, we create five workers per candidate at T0 (i.e. minimum 5x50 = 250 workers) and each of these five workers is assigned one strategy each.
Experimental results and conclusions We have considered a dense traffic area of Sweden as shown in Figure 1 and 20 disturbance scenarios of three categories: Scenario 1-10 have initially a temporary single source of delay; in scenario 11-15, a train has a ’permanent’ malfunction resulting in increased running times on all line sections it is planned to occupy; and in scenario 16-20, the disturbance is an infrastructure failure. The sequential and parallel algorithms are implemented in Java, and all experiments are conducted on a server with two quad-core processors. The results in Table 1 show the final delay for all trains at their destination. The parallel approaches succeed to find improved solutions. In order to assess the influence of using parallelization, we compare the results from Approach 1 with the sequential algorithm using the same strategy. In several scenarios, the parallelization shows a positive effect. Furthermore, analysing the results for Approach 2 it is evident that combining different strategies is very effective. Using parallelization enables this since there is no need for deciding when to select which strategy, which would have been necessary in a sequential hybrid set-up. If we only analyze the ability to find improved solutions, Approach 3 can be considered a combination of Approach 1 and 2. However, when the number of workers exceeds the number of processors available, the time limit and CPU resource allocation may influence the results. These results will be discussed further.
Based on our results, we can see that the parallel hybrid algorithm finds near to optimal solutions within the given time limit (30 s) but often a lot quicker than that. We can conclude that parallelization and a combination of different sorting strategies is quite successful to achieve better solutions than the individual single-strategy based sequential algorithm.
Page 67 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
76 - Evaluating the robustness of a railway timetable: a practical approach Presenter: Mr. TASSENOY, Sven (INFRABEL – Network, Belgium) Thorsten Büker, VIA Consulting & Development GmbH, Germany Karolien Scheerlinck, INFRABEL - Rail Access, Belgium Sven Tassenoy, INFRABEL – Network, Belgium Sabine Verboven, INFRABEL – Network, Belgium & Department of Mathematics and Computer Science, University of Antwerp, Belgium
Keywords: Robust timetabling, simulation, railway operations, …
With the growing traffic density in modern railway networks, it is no longer sufficient to make a timetable that is conflict free. In fact, a robust timetable is required. Robustness is in a railway network environment related to resistance of delays. Since it is impossible to operate a railway network without any primary delays, robustness has become an important property in the timetabling domain. Indeed, depending on the timetable’s structure, the amount of secondary delays caused by primary delays can vary. The manner in which the timetable can cope with these primary delays is determined by its robustness.
We define a robust timetable as a timetable that is as conflict free as possible and on which disturbances on the network have a minimal impact on the trains. A timetable with a large robustness reacts better to primary delays, and therefore has less secondary delays than a timetable with a small robustness. Translating this definition into a measure demands a series of simulation tests both on macroscopic and microscopic scale.
We are studying 3 different tests based on macroscopic and microscopic simulation. The tests will assess different characteristics of the timetable. Each of them will give a quantitative output, which will be combined into a measure of robustness.
On macroscopic scale, the tool OnTime propagates primary delays through the network. The result of OnTime is then a calculation of the total delay in the network at macroscopic detail. Using OnTime, it is thus possible to compare the delays originating from different timetables. A macroscopic description of the network is present but finding detailed data of the primary delays is less straightforward. Therefore, before we can use OnTime, a complex calibration of the parameters describing the primary delays is necessary.
The microscopic simulations are performed with the tool LUKS®, a software package which is able to simulate a timetable and operations in high detail. The main parameters of the simulations are the probabilities of primary delays. The more primary delays, the more perturbed the timetable will get. Fine-tuning these parameters involves many trial and error runs.
The three tests will provide knowledge on the timetable’s reaction to normal disturbances, its ability to recover and its reaction on above average disturbances. • Baseline test: Simulating the timetable with a fixed calibration (fixed probability of delay) • Recovery test: Giving the timetable above than average disturbances for a fixed time and then letting itself recover from it. • Breakdown test: Performing series of simulation scenarios, each time increasing the probability of delay, thus increasing the amount of primary delays.
Compared to the literature, our practical approach on testing robustness of a timetable has not been tackled before. The most innovative part is the breakdown test in which we will analyze the relation between the primary delays and the secondary delays. We believe that this relation holds interesting information on the robustness of a timetable. This ongoing study will also include a visualization of the results that will help practitioners to decide between different timetables. Therefore, the results from all tests will be summarized into one 2-dimensional graph that gives a comprehensive view on the robustness of the timetable.
References
1. Robustness and recovery in train scheduling – a simulation study from DSB S-tog a/s, M. Hofman, L. Madsen, J. J. Groth, J. Clausen, and J. Larsen, ATMOS 2006, http://drops.dagstuhl.de/opus/volltexte/2006/687. 2. Robustness in Railway Transportation Scheduling, M. A. Salido, F. Barber, and L. Ingolotti Proceedings of the 7th World Congress on Intelligent Control and Automation, 2008, pp. 2880 - 2885. 3. SIMONE: Large scale train network simulations, D. Middelkoop and M. Bouwman Proceedings of the 2001 Winter Simulation Conference, pp. 1042 – 1047.
Page 68 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
77 - Rolling Stock Rescheduling in Passenger Railway Transportation by Using Dead-Heading Trips Presenter: Mr. WAGENAAR, Joris (Rotterdam School of Management, Erasmus University) More than a million people make use of the railway network of the Dutch Railways (NS) on a daily basis. The NS is constantly focussing on improving the quality of its services. An important measure for this quality is the ability to react on unforeseen events. Two kinds of unforeseen events are of interest to railway operators: disruptions and disturbances. When an incidents requires a change in the planned schedules, it is called a disruption, otherwise it called a disturbance. A disturbance is, for instance, a train conductor who is one minute too late for his task because he had to go to the toilet and a disruption is, for example, a broken rolling stock unit on a track. During a disruption the planned timetable, rolling stock schedule and crew schedule may no longer be feasible, while during a disturbance the delay is most of the time absorbed by the slack in the system.
During a disruption the planned schedules have to be adapted in order to secure the feasibility of the schedules. The first step is to update the original timetable. In the Netherlands, more than a thousand different, so called, contingency plans exist to adapt the timetable. These contingency plans contain a number of rules stating which trains have to be cancelled, rerouted or delayed in case of a specific disruption. Secondly, based on the new timetable, the original rolling stock circulation has to be adapted. At last, with the new timetable and the new rolling stock schedule as input, the crew schedule will be adapted.
There already exist models for the adaption of the rolling stock circulation, however these models are currently not applied in practice. This is partly due to the fact that the models do not take all details of the real world into account.
In this paper we will discuss the current models for solving the Rolling Stock Rescheduling Problem (RSRP) and introduce an extension on the current models. Fioole et al. [1] formulated a scheduling model to assign the rolling stock to the timetable. This model is able to handle complicated line structures, such as combining and splitting of trains. This model is used by NS to generate the rolling stock schedules since 2004. However, this model is not capable of scheduling rolling stock in a real-time environment.
Nielsen [3] extended the model of Fioole et al. [1] to cope with the rescheduling of rolling stock. He formulated an integer programming problem to solve the RSRP. The inputs of this model are the adjusted timetable and the original rolling stock schedule and it gives the adjusted rolling stock schedule as output. The goal of the new rolling stock schedule is to differ as little as possible from the original rolling stock schedule, so that the passengers are least affected by the disruption.
In Nielsen et al. [2] a rolling horizon is used to solve the RSRP. The idea behind the rolling horizon is that at the beginning of the disruption not all the information is known, this information becomes gradually available. The rescheduling is periodically performed within a limited rolling horizon length, possibly taken new information into account. At each time instant when an updated timetable becomes available, or when a certain amount of time has passed without any update, the MIP of Nielsen [3] is solved for the next time window. This model is tested on instances from NS and solutions with small deviations of the original plan are found in a short time.
The major objective of the new rolling stock schedule is to differ as little as possible from the original schedule. In other words, to cancel as little trains as possible and to use trains that have enough capacity for all passenger demand. However, the rolling stock is scattered throughout the whole country. Therefore, during a disruption it is likely that certain stations have a surplus of train units while other stations have a shortage of train units to execute the new timetable. A shortage of train units can lead to possible cancelled trains, because there is no available rolling stock to appoint to the trip. Also, a shortage of train units can lead to using trains with too little capacity for the demand on the trip. Some passengers will still have to wait for the next train in that case. That is why the NS has the possibility to send empty trains (dead-heading trains) from one station to another to increase the local inventory during a disruption. By using dead-heading trains the NS wants to decrease the total number of cancelled trains and increase the customer satisfaction during a disruption. In this paper we will adapt the current RSRP models, by adding the possibility to use dead-heading trains during a disruption. We will solve the model with the rolling horizon proposed by Nielsen et al [6], ensuring up to date information about disruptions.
[1] P. Fioole, L. Kroon, G.Maróti, and A.Schrijver. A rolling stock circulation model for combining and splitting of passenger trains. European Journal Of Operational Research, 174: 1281-1297, 2006.
[2] L. Nielsen, L.G. Kroon, and G. Maróti. A rolling horizon approach for disruption management of railway rolling stock. European Journal of Operational Research, 220: 496-509, 2012.
[3] L.K. Nielsen. Rolling stock rescheduling in passenger railways. Ph.D thesis, Erasmus University, Rotterdam, The Netherlands, 2011
Page 69 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
[4] G. Budai, G. Maróti, R. Dekker, D. Huisman, and L. Kroon. Rescheduling in passenger railways: the rolling stock rebalancing problem. Journal of Scheduling, 13: 281-297, 2010.
[5] L.G. Kroon, G. Maróti, and L. Nielsen. Rescheduling of railway rolling stock with dynamic passenger flows. Technical Report, No. ERS-2010-045-LIS, Erasmus Research Institute of Management, Rotterdam, 2010.
Page 70 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
79 - Timetable Heterogeneity, Reactionary Delay and the Calculation of the Congestion Charge in Britain Presenter: Mr. HAITH, John (Univeristy of Leeds, UK) Britain’s railways are recognised as being some of the busiest in the world. In the last ten years annual passenger kilometres have increased by 44% to 57.3 billion and the amount of freight moved by rail has increased by 18% to 22.92 billion net tonne kilometres per year. To meet this demand total train kilometres have grown by 15.4%. However, this traffic growth is on a network that has shrunk during the same period by 4.5%.
Congestion is seen as a real issue with significant financial implications. Solutions are often extremely expensive and can take years to implement. In the mean time the costs of congestion continue to rise. Within the rail industry congestion costs arise in the performance payments that the infrastructure operator (Network Rail) makes to train operators due to reactionary delay. This is the delay that an already late train causes to a following train. In order to recover some of these costs and provide an incentive to introduce solutions that increase timetable robustness, a capacity charge was developed and applied. This is in the form of a simple tariff that varies according to time band and location.
The charge was calculated based on the theoretical link between capacity utilisation and the level of reactionary delay. An established method for measuring capacity usage called CUI (The Capacity Utilisation Index) was adopted. This works on the principle of ‘compressing’ the timetable for a given period into the smallest possible duration based on the relevant planning rules. The difference between the ‘compressed’ time period and actual time period gives the CUI.
In July 2012 Network Rail started consultation on its proposals for recalibrating the capacity charge as part of the 2013 Periodic Review. In the consultation document they explained that income from the existing capacity charge in 2010/11 was £158 million.
In the document, Network Rail revealed that they intended to broadly follow the approach adopted in the original work and use CUI. This was because it was a proven methodology which produced reasonable results. However, because “there is mounting anecdotal evidence suggesting that the capacity charge is no longer fulfilling objectives as well as it could be” they were minded to ask consultants to develop a methodology for calculating capacity usage at junctions and stations which were not included in the original analysis.
This paper explains how the calculation of capacity usage at junctions and stations is technically feasible and refers to published work where this has been done. However, the complexity of accurately calculating CUI for all the nodes on the rail network in the timescales proposed suggest that it is inappropriate to attempt to do this for the calibration of the capacity charge.
In this paper we propose an alternative methodology to CUI, which takes as its starting point a published study of reliability on the Dutch rail network which concludes that performance is as much to do with how capacity is used rather than just how much. In other words the heterogeneity of the timetable is an important factor. The theory is that the level of reactionary delay will be determined by the minimum gaps that exist between trains. Ironically these are the gaps in the timetable which CUI eliminates as part of its measurement process. This paper outlines the process used to develop a new measure of heterogeneity of railway timetables which adopts the concepts the authors of the Dutch study proposed.
A small but complex sample network was used to test the various methodologies and the reasoning behind this is explained. The handling of the timetable data is described as is how the various alternate measures of capacity usage were applied. The analysis of the reactionary delay data obtained is also described and it is explained how this information furthers our understanding of why this type of delay occurs.
The regression analysis that has been carried out is described. The paper discusses the apparent anomalies in the results and the importance of understanding why they occur. For example, an off peak hour at one location has more reactionary delay than that incurred in the equivalent peak hour. The steps taken to identify the cause of this apparent contradiction are explained.
The results of the regression analysis show that CUI in its original form is indeed a reasonable predictor of reactionary delay. However, the methodology based on measuring the heterogeneity of the timetable consistently performs much better. The paper describes the work required to test the transferability of the heterogeneity method and gives examples of off-peak periods in the current British timetable where the phenomenon of ‘bunching’ of trains, which the concept of heterogeneity suggests will lead to more reactionary delay, can be seen.
Finally, the paper describes the implications of the findings from the work carried out so far. The significant economic impacts in terms of how the capacity charge works now and how it might work in the future are outlined. We also describe the important implications for the structure of timetables. It is explained why the findings described in this paper, are of major importance which apply globally rather than just to the British rail network.
Key References
Page 71 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
Armstrong, J., S.Blainey, J. Preston and I. Hood (2011) Developing a CUI Based approach to Network Capacity Assessment – 4th International Seminar on Railway Operations Modelling and Analysis (Rome 2011)
Gibson, S., G.Cooper and B.Ball (2002) The evolution of capacity charges on the UK rail network Journal of Transport Economics and Policy Vol.36 Part 2 May 2002 pp 341-354.
Network Rail (2012) Periodic Review 2013 – Consultation on the Capacity Charge (Downloaded from the Internet 24/08/12)
Office of the Rail Regulator (2012) – Official Statistics (Obtained from the Internet 24/08/12)
Vromans, M.J.C.M., R.Dekker, and L.G.Kroon, (2006) Reliability and heterogeneity of railway services European Journal of Operational Research 172 2006 pp 647-655
Page 72 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
81 - Robust railway station planning: an interaction between routing, timetabling and platforming Presenter: Mr. DEWILDE, Thijs (University of Leuven, Belgium) Railway robustness is a broad concept tackled by many authors. During the IAROR conference in Rome, we have introduced our vision on robustness: “A railway timetable that is robust minimizes the real total travel time of the passengers, in case of frequently occurring small delays.”[1] Thus, focusing on the passengers, we want to decrease the travel times in practice, which include delays and missed transfers. In this paper, we consider complex, busy stations whose limited capacity is one of the main reasons of delay propagation. Our goal is to improve, during the planning phase, the robustness of a complex station by considering the potential of the available capacity. The main feature of our approach is the interaction between routing decisions, timetabling, and platform assignments. By altering one of these, slack can be created to allow improvements by the others as well. As an objective function, maximizing the spread of the trains is used.
In a first step of our study, the interaction was considered between the train routing through a station and a (limited) modification of the timetable, using only shifts and a simple swap in the order of trains [2]. Now, we present the results achieved with an extended and significantly better version of this approach.
Only changing the order of trains may not increase the robustness unless other (related) timetable adjustments are considered as well. This is what we call the potential of a change. During the scheduling phase, the potential of changing the order of trains or deviating from the preferred platform is investigated intensively. This is done by allowing rerouting and retiming actions to take place simultaneously. The same principle is applied when one of the changes would result in a conflict. As a result, a new schedule arises in which a large number of impending conflicts between trains are avoided.
Initially, this approach was used to improve the performance on the network connecting the three major stations in Brussels, Belgium’s capital. The Brussels area truly is an interesting case study to evaluate our approach as it contains three of the country’s four busiest stations regarding passenger numbers. It includes the largest station with respect to platforms and a true, physical bottleneck since 19 platforms are connected through a 6-track tunnel, containing the most busiest station, with the 12 platforms of the station on the other side.
A microscopic simulation tool is built to estimate the impact of our approach on the Brussels area. Due to confidentiality reasons, a timetable of a few years ago is used as reference timetable. Compared to that timetable, for which spreading was not an issue, we were able to improve that spreading with a factor above 80 percent. If we compare with the results of our previous algorithm, the improvement is about 35 percent. The improvement in robustness of our previous algorithm was 5% and now it is 8%. The fact that this is an order of magnitude smaller than the improvement in spreading is not surprising since the spreading only influences the robustness indirectly, through the avoidance of conflicts and the resulting decrease of knock-on delay. Currently, we are, together with the Belgian railway infrastructure manager Infrabel, evaluating the applicability of our results on a larger scale to confirm its positive impact on the robustness.
To show the general applicability of our approach, we now also consider a second case study that is significantly different from the first one. In the station zone of Antwerp, the two major stations are connected through three corridors allowing trains to arrive at three different levels in the Central station. The zone of Antwerp is an interesting case study because of the high capacity usage by a heterogeneous fleet of trains; international trains mix up with slow and fast local trains and with freight trains. Other challenges are that all but four platforms in the Central station are dead-ending and that, in comparison with the study for Brussels, there are more restrictions on the combinations between inbound and outbound lines and the allocated platforms.
Summarizing, we outperform our previous algorithm and significantly improve the robustness for Brussels. We can now fully exploit the potential of an integrated optimization of routing decisions, train sequences and schedules, and platform allocations. Moreover, a new and totally different case study, the highly used railway network around the city of Antwerp, is introduced. This illustrates that our approach can be applied successfully to any type of station. Some extensions related to this case study are presented as well, for instance, the interference with heavy freight traffic and improving turnovers.
Acknowledgement: This research is made in close cooperation with the Belgian railway infrastructure manager Infrabel and is partly funded by a Ph.D. grant of the Agency for Innovation by Science and Technology (IWT).
References
[1] T. Dewilde, P. Sels, D. Cattrysse, P. Vansteenwegen. Defining Robustness of a Railway Timetable. In: Proceedings of the 4th International Seminar on Railway Operations Modelling and Analysis (RailRome), 2011.
[2] T. Dewilde, P. Sels, D. Cattrysse, P. Vansteenwegen. Improving the robustness in railway station areas. European Journal of Operational Research (under review).
Page 73 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
Page 74 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
82 - Assessing Timetable Robustness using Critical Points Presenter: Ms. ANDERSSON, Emma (Dept of Science and Technology, Linköping University) Introduction A global tendency seen for quite some time is the growing demand for railway capacity which no longer can be accommodated. This causes congestion, delays and insufficient punctuality. A previous empirical study of the Swedish railway traffic show that punctuality of different trains differs significantly despite similar prerequisites (Andersson et al., 2011). In order to maintain certain robustness, we have, in collaboration with Trafikverket (the Swedish Transport Administration), identified a need for systematic and effective strategies concerning how to increase the timetable robustness. There is also a need for relevant robustness measures that can describe the robustness quantitatively. One main purpose of this paper is to study the effect of existing timetable robustness measures on a sample timetable which contains known delay-sensitive situations. There is also a purpose to define a robustness measure applicable in the practical timetabling process to support the development of robustness strategies.
Problem description The most common approach used today when measuring timetable robustness is to measure punctuality, i.e. the amount or percentage of trains running on time. However, measuring the traffic performance is not a sufficient approach when creating robust timetables. Figure 1 illustrates that one can view robustness measures from two perspectives: 1) From a timetable construction perspective, analyzing the timetable characteristics with respect to headways, margins etc., and 2) with respect to the traffic performance given different disturbances.
Figure 1. Two categories of robustness measures used when analyzing timetable robustness
In this paper we focus on timetable characteristics since it is during the design of the timetable that robustness initially can be inserted. We need to know which timetable characteristics that affect the robustness and how to measure them.
Related research Through the study of related research we are able to identify a number of indicators and measures previously used when analyzing robustness. Salido et al. (2008) give example of three methods to obtain a robust timetable; insert runtime margins, decrease the theoretical track capacity and decrease traffic heterogeneity. Kroon et al. (2007) and Fischetti et al. (2009) use insertion of margins implicit when they optimize timetables with the objective to minimize delays. They use Weighted Average Distance (WAD) as a measure of how margins are allocated. Carey (1999) has developed some measures for robustness (he uses the term reliability) for both individual trains and complete timetables with and without probability information. He studies the headway and how to reduce knock-on delays. Vromans et al. (2006) study how to get a timetable more homogenous. They also introduce the Sum of Shortest Headway Reciprocals (SSHR), a measure of the headway size combined with traffic heterogeneity.
Critical points By studying timetables and traffic performance, we have identified some weaknesses in the timetables affecting the traffic performance negatively, but which were not captured by existing measures. These weaknesses can be interpreted as locations in the network where two trains, scheduled with short headway, interact on the same track. If one train is delayed it will easily end up after other slower train and get even more delayed. We refer to these points as critical points and they can e.g. appear during overtaking or when the first departure of a train is scheduled close to another train. Figure 2 shows two scenarios, where a delayed fast train enters a critical point (station B). If the train dispatchers choose to decide as in the second scenario, the delay for the fast train will increase and the train has no possibility to use its margins for recovery. In the first scenario the fast train will instead recover from some of its delays and even though it gets delayed at station B, the slow train will arrive to the end station on time.
Figure 2. Two different scenarios when the fast (train 1) runs either before or after the slower commuter train (train 2). The continuous lines are the planned timetable and the dashed/dotted lines represent the actual run with delays.
Critical points are defined in both time and space and the timetable robustness depends on the number of points, their location and magnitude (how critical the points are). We therefore define a measure called Robustness in Critical Points (RCP) which returns a robustness value for every critical point. We illustrate critical points using a small fictive example and using an example from a real timetable. We also compare the proposed RCP measure with existing measures discussed in literature.
Conclusions and future research Based on the results we can conclude that there are critical points appearing in timetables, which have not been captured by previous research. It is important to have a measure that can identify these weaknesses and indicate where margins need to be inserted to improve the robustness. Most of the previously known measures are comparative measures and we believe that with critical points and RCP we have a new robustness measure that can be applicable in the practical timetabling process.
Page 75 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
In future research, we intend to use optimization to maximize the timetable robustness as expressed by the RCP measure. The optimization models can e.g. be designed to; restrict the magnitude of the critical points, maximize the margins in the most critical points alternatively minimize the number of critical points of the timetable.
References Andersson E, Peterson A, Krasemann J T (2011) Robustness in Swedish railway traffic timetables. In: Ricci et al. (eds.) Proceedings of RailRome 2011, University of Rome La Sapienza and IAROR, Doc. No. 2
Carey M (1999) Ex ante heuristic measures of schedule reliability. Transportation Research Part B 33:473-494
Fischetti M, Salvagnin D, Zanette A (2009) Fast approaches to improve the robustness of a railway timetable. Transportation Science 43:321-335
Kroon L, Dekker R, Vromans M (2007) Cyclic railway timetabling: A stochastic optimization approach. Railway Optimization LNCS 4359: 41-66
Salido M A, Barber F, Ingolotti L (2008) Robustness in railway transportation scheduling. In: 2008 7th World Congress on Intelligent Control and Automation, Chongqing, China: 2880-2885
Vromans M, Dekker R, Kroon L (2006) Reliability and heterogeneity of railway services. European Journal of Operational Research, 172:647-665
Page 76 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
84 - Cost benchmarking of railway projects in Europe – can it help to reduce costs? Presenter: Mrs. TRABO, Inara (DTU Transport) Past experiences in the construction of high-speed railway projects show either positive or negative financial outcomes of the actual project’s budget. Usually some uncertainty value is included into initial budget calculations. Uncertainty is related to the increase of material prices, difficulties during construction, financial difficulties of the company or mistakes in calculations, etc. Such factors may finally change the initial budget estimates and cause budget overruns. According to the research conducted by Prof. B. Flyvbjerg, related to investigation of budget overrun causes in large transport infrastructure projects, 9 projects out of 10 came out with budget overruns. Mette Skamris Holm in her PhD thesis observed 58 different rail projects with an average cost overrun of 44,7%, ranging from -46% to +200%. Whereas the new research conducted by UNITE [UNcertainties In Transport project Evaluation] project group considered transport projects in Scandinavia and UK and found out the sample of 19 rail with cost escalation ranging from -40% to +60% and the mean of 13%. In practice causes of budget overruns are divided into 4 groups and explained by technical, economic, psychological and political mistakes, where the economic and political mistakes are most common among the transport infrastructure projects. As an example of cost overruns is the British project Channel Tunnel Rail link, the railway line between London and the British end of the Channel Tunnel. The project was delayed for 11 months and final construction costs were escalated to 80%, later on it was investigated that initial calculations and passenger forecasts were overestimated deliberately in order to get financial support from the government and perform this project. The case study in this research paper is the first Danish high-speed railway line “The New Line Copenhagen-Ringsted”. The project’s aim is to avoid cost overruns and even make lower the final budget outcomes by looking for the best practices in construction and implementation of other high-speed lines in Europe and learning from their experiences in order to become best-in-class project in 2018, when there will begin railway operations on the new line. Apart from experiences with cost overruns there are also many projects with positive financial outcomes. French, Dutch, Italian projects have productive experiences in constructing and operating high-speed railway lines. Belgium has completed its high-speed railway network in 2009 and it is open to share the knowledge and experience with current and future railway projects.
The methodology of this paper is based on international benchmarking of construction costs and all information related to the construction (i.e. construction companies and construction materials). Benchmarking procedure is essential to improve particular project areas and reduce costs. For high-speed railway projects benchmarking is important for the comparison of unit cost per major cost drivers (i.e. cost of tunnels, bridges, km of track, etc.) For the Danish project there were selected 23 high-speed railway projects from France, Italy, Spain, Belgium, The Netherlands, Sweden, UK, suitable for detailed comparison. The selection criteria was based on project’s allocation along the motorway, on the flat terrain and not filled up with complex structures (i.e. long tunnels and bridges, construction in the urban and densely populated places). Benchmarking was conducted on the three levels, where the initial level considered the comparison of average price per kilometer among selected project. The next level provided the detailed budget distribution to the main cost disciplines and unit cost analysis. The final step was related to the observation of industry companies and material prices. The outcomes of benchmarking process will bring additional knowledge to the case project management group, they will be able to compare their initial estimates for major cost drivers with the benchmarking results and take decisions if they find out that they are above the average estimates from comparable projects. The other infrastructure managers will find this methodology useful as well. They will obtain basic knowledge about construction cost values per different cost disciplines among other European railway projects. Afterwards they will be able to conduct an internal benchmarking of relevant cost positions and compare them with other projects and find out whether they can implement this knowledge to further project policy and planning.
References Nicolaisen, M. S. (forthcoming), Forecasts: Fact or Fiction? Uncertainty and Inaccuracy in Transport Project Evaluation. Ph.D. thesis, Aalborg University, Aalborg, 2012 Trafikstyrelsen, Udenlandske baner, Sammenligning af anlægspriser, 2009 Trafikstyrelsen, Capacity extension Copenhagen-Ringsted - Adding an extra track or constructing a new double-track railway, 2009 UIC, Infracost the cost of railway infrastructure, Final report, 2002 BSL, Comparison of High Speed Lines' CAPEX, Final report, 2009 Infrabel, HSL, 300 kph on the Belgian rail VOL . 2 The history of a challenge, 2009 Barlett School of Planning, Channel Tunnel Rail Link, Centre for Mega Projects in Transport and Development Mette K. Skamris Holm, Economic Appraisal of Large Scale Transport Infrastructure Investments, PhD Thesis, Aalborg University, 2000 Bent Flyvbjerg,Megaprojects Policy and Planning: Problems, Causes, Cures, 2007
Page 77 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
85 - Comparing railway capacity allocation models Presenter: Dr. SCHLECHTE, Thomas (Zuse Institute Berlin) This paper addresses a major real life optimization problem of the railroad industry -- the train pathing or train dispatching problem. The problem asks for an integrated routing and scheduling decision for a set of trains that run through the same infrastructure network and thus have to be scheduled carefully to avoid collisions, see Lusby et. al. [2009] for a comprehensive survey. The goal is to minimize the total cost of the schedule, which depends on the train route and the associated departure and arrival times.
In general the objective function includes: the total delay, the scheduled deviance at some intermediate nodes and at the destination node, and the unpreferred track time usage.
We compare different fundamental mathematical formulations for railway timetabling on a common set of sample problems, representing both multiple track high density services in Europe and single track bidirectional operations in North America. One formulation enforces against conflicts by constraining time intervals between departing trains, see Schlechte[2012]. While the another formulation, see Harrod [2011], monitors physical occupation of controlled track segments in a hypergraph. In addition, we present experiments with a disjunctive program formulation to solve the scheduling problem in case of fixed train routes. The computational results demonstrate that both models return comparable solutions in the aggregate.
Page 78 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
87 - Evaluation of facility improvements from the viewpoints of service level robustness for passengers Presenter: Mr. KUNIMATSU, Taketoshi (Railway Technical Research Institute) The goal of our research is to improve the robustness of transportation service during disrupted train operations, as well as increase effective capacity of rail network. We establish a method to estimate the effectiveness of facility improvements for train operation. The facility improvements include addition of turn tracks to provide temporal shuttle services in front of the blockage section under operational disturbances, and platforms to prolong dwell time at stations. Our approach is based on micro simulation for estimating train operation and passengers’ behaviour. We also adopt Monte Carlo method to represent variations of disrupted train operations. This enables us to evaluate effects of new facilities from the viewpoints of passengers considering various operational conditions. We can also compare several deployment plans by that method.
When an accident or operational disturbance occurs on a railway line in Japan, it is often the case that all trains on that line are stopped. It may significantly decrease the transportation service level for passengers, because even passengers who need not go through the blockage section are forced to wait until the cause of suspension is removed. Deploying additional turn tracks and providing shuttle service can localize the effect of any accident that might occur, which will minimize the reduction of the service level of passenger transport in case of disruption. Since a few turn tracks are installed on typical Japanese railway lines, significant amounts of investment is needed to realize the intended purpose. Recently, as the frequency of suspensions is increasing in Japan, the deployment of additional turn tracks is often discussed.
On the other hand, although railway companies operate as many trains as possible at peak hours in Japanese commuter lines, chronic train delay has occurred in Japan. One reason for that is the extension of dwell time at a certain station by congestion, but the short interval of train also encourages the chronic delay. That is, the delay which occurred on a certain train easily propagates a successive train. One way for decreasing chronic delay is to improve facilities to increase capacity at stations or junctions. For example, when we install an additional platform at the bottle neck station and assign successive trains to two platforms mutually, we can prolong the planned dwell time at the station without decreasing transportation capacity, and moderate delay propagation from a former train to the latter.
So, it can be said that additional turn tracks or platforms may moderate transportation inconvenience under disrupted timetables or daily chronic delay. Moreover, in some cases, we can use additional platforms for turn tracks to provide shuttle services when we face operational disturbances. But, as it is very costly to deploy them, we have to estimate the cost effectiveness of investments appropriately.
In this research, we propose two evaluation methods for facility improvements, one is for additional turn tracks under timetable disruption, and the other is for additional platforms under chronic train delay. The common concepts of the two methods are that, we use micro simulation to evaluate from the viewpoints of passengers. We use “Train operation and passenger behaviour simulator,” introduced in RailZurich2009. The simulator use timetable data and passengers’ Origin-Destination data to estimate both train operation and passengers’ behaviour from their origin station to destination. The simulator outputs each passenger’s detailed behaviour or experienced services during their journey. So, we can calculate each passenger’s evaluation for the train operation. We use the index that represents passenger utility calculated from the passengers’ experience of the congestion, amount of time to reach the destination, the number of transferring. It is possible to have the value of the index for all passengers since the simulation can trace each passenger appeared in the system. By summing up each passenger’s value, we can get overall evaluation for whole train operation.
We added some functions to that system. First, to represent passengers’ detour behaviour, taking other transportation methods to their destinations under suspension, we incorporate passengers’ detour choice model. Second, we calculate the running time between stations or dwell time necessary at stations stochastically, using average, standard deviation, and normal random number. This enables us to represent daily deviation of stoppage time under the same congestion or that of driving time among drivers. By these upgrades, we can estimate train operation or passengers’ behaviour with more accuracy.
In the evaluation for turn tracks, we first prepare two train rescheduling patterns for the premised operational disturbance. One is in case without additional turn tracks, providing limit shuttle services until the cause of suspension is removed. The other is with them, providing extended shuttle services. Second, we make several disturbance scenarios by stochastic means reflecting the frequency of them and deviation of occurrence time, location, and duration. Then, we work the simulator for each disturbance scenario with/without turn tracks, and calculate evaluation values. Finally, we consider differential of evaluation values as the effects of turn tracks, summarize for each scenario, and calculate expected value reflecting frequency rate of each suspension.
In the evaluation for platforms, we first prepare two timetables and operational facility data before or after facility improvement. Second, using the simulator, we estimate train operation and passengers’ behaviour for sufficient times, and calculate evaluation values for each case. Finally, we summarize differential of average of evaluation values for each precondition to take expected value of effects.
Page 79 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
The two methods are a kind of Monte Carlo method, which represents deviation of operational disturbances, delay occurrence or delay propagation.
We have applied our method for the imaginary facility improvements in an actual urban commuter line in Japan. As a result, we conclude that the proposed methods can evaluate facility improvements appropriately from the viewpoints of passengers. We are going to show numerical evaluation results for the line.
89 - Fast optimal solutions to the Train Dispatching problem Presenter: Mr. LAMORGESE, Leonardo (SINTEF ICT); Prof. MANNINO, Carlo (SINTEF ICT) In a first, simplified picture, a rail network may be viewed as a set of stations connected by tracks. Each train follows a specific route in this network, namely an alternating sequence of stations and tracks. The trains run their routes trying to agree with the production plan, which, in principle, ensures that no two trains will occupy simultaneously the same railway resource or different but incompatible resources. However, due to unpredictable events, the actual train schedules can deviate from the official ones, and conflicts in the use of resources may arise. As a consequence, re-routing and re-scheduling decisions must be taken in real-time so as to minimize a measure of the delays. In other words we are dealing with an optimization problem and we refer to it as the Train Dispatching problem (TD). In spite of its relevance and computational complexity, train dispatching is basically still all in the hands of human operators, which have to take, in a few seconds, decisions that affect the entire network. Not surprisingly, this problem can be effectively tackled by suitable optimization techniques, leading to faster and better solutions than those provided by dispatchers. In short, the TD problem amounts to establishing, in real-time, a route and a schedule for each controlled train so that no conflicts occur with other trains and some function of the deviation from the official timetable is minimized. As such, the TD problem falls into the class of job-shop scheduling problems, which can be represented by suitable disjunctive formulations. In turn, such formulations can be transformed into Mixed Integer Linear Programs (MILPs) by associating a binary variable with every pair of (potentially) conflicting operations and, for any such variables, a pair of alternative precedence constraints representing different orderings amongst operations. In [1], we introduced a new modelling approach to the TD problem and a solution methodology which allowed us to overcome some of the limitations of the standard MILP formulations. Indeed, we were able to solve to optimality the MILPs corresponding to a number of real-life instances in single-track railways within the stringent running times imposed by the application. The methodology is based on a structured decomposition of the TD into two sub-problems, the Line Dispatching problem (LD) and the Station Dispatching problem (SD), which tackle different aspects of the TD problem and give raise to distinct sets of variables and constraints in the model. All problems in this decomposition can be formulated and solved by means of Integer Linear Programming. This approach proved to be decisive for successfully tackling the large instances arising from some railway lines of the Italian railway system. The major advantages of the decomposition were shown to be the drastic reduction of the number of variables and constraints with respect to the standard, natural formulation and the freedom in modelling the SD problem with suitable complexity, as the (general) SD problem we proved to be NP-hard. Actually, a semi-automated route setting system based on such decomposition, but with the exact integer programs replaced by simple local heuristics, has been already put into operations in several Italian railway lines in different regions. In the decomposition approach presented in [1], the LD problem acts as master problem and the SD problem (slave problem) is solved by exploiting certain properties of the interval graph associated with the arrivals and departures in each station. Master and slave(s) are then coupled through a suitable cut generation. The cuts prevent solutions to the (so extended) LD problem result in configurations which are infeasible for the associated SD problems. In [1] the LD problem is modelled by associating decision variables with pairs of trains and every point where such pair can possibly meet or pass. We are currently developing a new representation for the LD problem which is based on a recursive decomposition of the line. Rather than establishing the exact point in which trains should meet or pass, the new variables are associated with entire sub-regions. Through a suitable column generation mechanism, new variables are generated only if needed, i.e. if trains are actually in conflict in the sub-region assigned for their meet/pass event. Also, valid sets of strong linear inequalities were derived, improving the formulation and allowing tighter bounds to be found. Our preliminary computational results with the new approach appear to be very promising. Indeed we were able to solve to optimality previously unsolved hard instances and in all cases respected the timeframe acceptable for dispatchers, largely reducing computation times. Due to these encouraging results, we are currently preparing tests on the national network in cooperation with the Norwegian rail operator.
[1] L. Lamorgese, C. Mannino, An exact decomposition approach for the real-time train dispatching problem, Technical Report N. A23274, SINTEF ICT, Norway, 2012, submitted.
Page 80 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
90 - Heterogeneity measures and secondary delays on a simulated double-track Presenter: LINDFELDT, Anders (KTH) Anders Lindfeldt Division of Traffic and Logistics Department of Transport science Royal Institute of Technology SE-100 44 Stockholm Sweden [email protected]
Abstract Background and introduction The ever increasing demand for transportation on railways makes it important to understand how railway operation reacts to increased capacity utilisation. One factor that is of great importance when capacity is discussed is traffic heterogeneity. Heterogeneity can be used to describe two different properties of the timetable. The first is how evenly distributed the train movements are over a given period of time and the second is associated with speed differences between trains. In heterogeneous timetables, trains use the infrastructure unevenly over time with great difference in average speed. Besides reducing the number of trains it is possible to schedule, high heterogeneity also increases the risk for delay transfer, i.e. secondary delays. In the first case, the buffer times between trains are unnecessary small. In the second the speed difference implies that faster trains risk catching up on slower trains and slower trains are forced to stand aside for unscheduled overtakings. Objectives Many different definitions of heterogeneity can be found in literature. The main objective of this paper is to analyse some of these heterogeneity measures and how they correlate to secondary delays on a simulated double track. For example, what creates most delays, heterogeneity in headways or speed? Other objectives are to determine the effect of heterogeneity when primary delay levels and distance between overtaking stations are varied. Method The analysis is performed in RailSys, a tool for timetable planning and microscopic simulation of train operation. The analysed infrastructure models represent fictive double-track lines with overtaking stations at regular intervals. Two infrastructure models with overtaking stations spaced at different intervals are used. For each infrastructure variant, a large number of timetables is generated where traffic density as well as the mix of slower and faster trains is varied. Each timetable is characterized using the different definitions of heterogeneity and is simulated in RailSys. In the simulation, primary delays are modelled as stochastic processes and include entry delays, running time extensions and dwell time extensions. The empirical distributions are compiled from real data from the Swedish rail network. From the simulation results, secondary delays on line sections and at stations are estimated with high accuracy. A correlation analysis of secondary delays and the different heterogeneity measures of the corresponding timetables is performed. Expected results Among the expected results is a deeper understanding about the effect of heterogeneity on secondary delays in double track operation. Other questions that are expected to be answered are: Which of the analysed measures of heterogeneity correlates best to secondary delays and is therefore most important to consider when more robust timetables are designed? Does the size of the primary delays or the distance between the overtaking stations affect which measure is best? Key references Vromans M., Dekker R., Kroon L., (2006) Reliability and heterogeneity of railway services. European Journal of Operational Research 17, p647-665 Gorman, M. (2009), “Statistical estimation of railroad congestion delay”, Transportation Research Part E. Huisman, T., Boucherie, R., (1999), “Running times on railway sections with heterogeneous train traffic”, Transportation Research Part B 35, 2001, 271-292 Lindfeldt, A, 2011. Investigating the impact of timetable properties on delay propagation on a double track line using extensive simulation. Presented at Railway Engineering 11:th International Conference and Exhibition, London.
Page 81 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
92 - OPTIMIZING TRAIN TRAFFIC: DEMONSTRATING BENEFITS IN A CASE STUDY Presenter: Mr. MIDDELKOOP, Dick (ProRail) On the Dutch railway network about 6500 trains run on a daily basis nowadays. To construct a timetable that incorporates all trains of all train operating companies is a challenging process year by year. Because capacity resources like railway infrastructure, rolling stock and crew are expensive and scarce, timetable planners have to search intensively for feasible solutions. A feasible solution should meet design standards and criteria that measure commercial attractiveness, capacity and safety aspects. At this point the difference between the planning and the operation of the timetable should be as small as possible. This means that train drivers and traffic controllers should be aware of the targets that are set by the timetable and how they can reach them given their actual situation. To investigate the potential benefits of an enhanced level of their situational awareness ProRail, the Dutch rail infrastructure manager, has developed a combined simulation and optimization tool. Two existing applications: FRISO a railway simulator based on a microscopic infrastructure model and TMS a traffic management and decision support system, are coupled in a scalable architecture.
In a case study in the Den Bosch area the potential benefits on punctuality, non-commercial stops and safety were shown. In the Den Bosch station large construction works will reduce the capacity temporarily. The case study shows the benefits of decreasing the scattering of timetable events in the operation. Deviations in the operations are minimized. In a series of simulation experiments where a comparison of several traffic controlling strategies has been made, the TMS-approach shows good/the best results. The TMS monitors actual train positions, train status/speeds and performance targets in the control area. It predicts conflicts and gives advisory speeds, new schedules and alternative routes to trains to prevent upcoming conflicts. The TMS optimizes the train traffic in large areas and provides decision support information that may be distributed to train driver and traffic controllers. Following experiments are needed to show effects/influence, of for instance the real-time advisory speeds, on the behavior of train drivers and traffic controllers. By using the IEEE High Level Architecture for coupling FRISO and TMS, the combination may be extended with more applications or connected to real-life systems. As a first extension, an interface for train controllers, already is available. It will be followed by an interface for train drivers.
The TMS internal model is based on the alternative graph scheme, properly adapted to the rail traffic scheduling problem. Over the years, the TMS model has been continuously extended and applied to many simulation studies. To assure higher adherence to the real world and to reach significant performance results, new features and capabilities have been added to TMS. Among them, a new optimization scheme has been developed and integrated in the graph model, to allow the efficient and effective choice of alternative routes. Alternative routes improve the flexibility of TMS when looking for solutions to prevent heavy conflicts or to avoid traffic congestion. Now, TMS is able to deal with many details of the infrastructure, safety system, rolling stock and timetable: signal relations, speed boards, specific rolling stock characteristics, train relations concerning passengers and rolling stock, possible alternative routes, route setting rules for phased route setting, etc. This ability is crucial for a real-world implementation of a TMS.
TMS optimizes all trains simultaneously over a certain time horizon; current objective function is the sum of future delays over all trains and total energy consumption. An important characteristic is that a global optimization is performed over all trains and the whole area, as opposed to local optimization strategies. TMS also shows the capability to deal with unpredictable events, compensating them with short term forecasting.
The paper explains the TMS architecture and algorithms, describes the simulation experiments for the case study Den Bosch and quantifies the benefits of using the TMS in this case. Note that infra, timetable, rolling stock and disturbances all reflect the actual situation in great detail. This study therefore provides a highly realistic example.
Page 82 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
94 - Strategic freight train routing in macroscopic rail networks Presenter: KLUG, Torsten (Zuse Institute Berlin) The rail transport volume in Germany increases for years, while corresponding expansion of the infrastructure is rather small, since the changes of the infrastructure are always capital-intensive and long term projects. Germany as a transit country in central Europa faces a great challenge in the next years. In particular, this applies for the rail freight traffic. Based on data of 2009, recent estimates assume an increase by 80 percent up to 213 billion ton kilometers by 2025. Therefore, it is necessary to analyze the existing network to estimate and make best use of the available capacity. In this context our project partner Deutsche Bahn AG focuses on an strategic planning in a simplified transport network. We develop model for the rail freight train routing model in such macroscopic networks. Here macroscopic refers to an aggregation of complex real-world structures into fewer network elements. For instance, a station with all its tracks and switches corresponds to a single node in the macroscopic network. Moreover, the departure and arrival times of freight trains are approximated.
The freight train routing problem is then defined as follows: Given a transport network and a set of freight train requests, each defined by an origin and destination station pair and a release time. The major goal is to determine routes for freight trains by taking into account the available railway infrastructure and the already planned passenger traffic. The determined routes should minimize the sum of all expected delays and the subordinate criteria running time and path length. Our macroscopic routing is the input for later planing steps where detailed timetable are generated. By minimizing the sum of delays we increase the chance that timetables with small delays can be produced.
The routing of freight trains is quite different from passenger trains since departure and arrival times windows are less strict and routes are not limited by several intended intermediate stops. Nevertheless, we have to consider the railway system as a whole including railway passenger transport and infrastructure to provide reasonable strategic prospectus. Previous research concentrated on microscopic train routings for junctions or inside major stations. Only recently approaches were developed to tackle larger corridors Schlechte et al. (2011) or even networks Caimi (2009).
A framework for a general class of network design problems is presented in Kim and Barnhart (1997) and applied to the blocking problem in railroad traffic in the US, see Barnhart et al. (2000). Integrated service network design for rail freight transportation in the US is considered in Zhu et al. (2009); Ahuja et al. (2007); Jha et al. (2008). The railroad blocking problem can be formulated as a very large-scale, multicommodity, flow-network-design and routing problem with billions of decision variables, see Jha et al. (2008) and Barnhart et al. (2000). Ahuja et al. (2007) presented an algorithm using an emerging technique known as very large-scale neighborhood search to support major US railway companies that transfers millions of cars over its network annually. The authors report that their heuristic approach is able to solve the problem to near optimality using one to two hours of computer time on a standard workstation computer.
In the case of road traffic Köhler et al. (2009) present mathematical theory on flow depended cost functions. A major difference is that in road traffic the routing is decentralized, arbitrarily partitionable, and assumed to be selfish. In contrast to that railway systems are centralized and we are aiming for a system optimum. In addition the train flow can not be partitioned arbitrary and thus the routing and timetable is more rigid system in comparison to flow of cars.
We give a mixed-integer nonlinear programming (MINLP) formulation for freight train routing problem, which is a multi-commodity flow model on a time-expanded graph with additional routing constraints. The model’s nonlinearities are due to an algebraic approximation of the delays of the trains on the arcs of the network. The MINLP is reduced to an equivalent mixed-integer linear model (MILP) by piecewise linear approximation. Finally, we will present com- putational results for data from our project partner Deutsche Bahn AG. We use different state of the art MILP solvers and compare their results with a greedy-type construction heuristic.
References
Ahuja, R. K., K. C. Jha, and J. Liu (2007). Solving real-life railroad blocking problems. Interfaces 37 (5), 404–419.
Barnhart, C., H. Jin, and P. H. Vance (2000). Railroad blocking: A network design application. Oper. Res. 48 (4), 603–614.
Caimi, G. (2009). Algorithmic decision support for train scheduling in a large and highly utilised railway network. Ph. D. thesis, ETH Zurich.
Jha, K. C., R. K. Ahuja, and G. Sahin (2008). New approaches for solving the block-to-train assignment problem. Networks 51 (1), 48–62.
Kim, D. and C. Barnhart (1997). Transportation service network design: Models and algorithms. In N. H. M. Wilson (Ed.), Proc. of the Seventh International Workshop on Computer-Aided Scheduling of Public Transport (CASPT), Boston, USA, 1997, Volume 471 of Lecture Notes in Economics and Mathematical Systems, pp. 259–283. Berlin, Heidelberg: Springer-Verlag.
Page 83 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
Köhler, E., R. H. Möhring, and M. Skutella (2009). Traffic networks and flows over time. In J. Lerner, D. Wagner, and K. A. Zweig (Eds.), Algorithmics of Large and Complex Networks: Design, Analysis, and Simulation, Volume 5515 of Lecture Notes in Computer Science, pp. 166–196. Springer.
Schlechte, T., R. Borndörfer, B. Erol, T. Graffagnino, and E. Swarat (2011). Micro-macro transformation of railway networks. Journal of Rail Transport Planning & Management.
Zhu, E., T. G. Crainic, and M. Gendreau (2009). Integrated service network design in rail freight transportation. Research Report CIRRELT-2009-45, CIRRELT, Montr ́al, Canada.
Page 84 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
95 - An Optimization Model for Simultaneous Periodic Timetable Generation and Stability Analysis Presenter: Mr. SPARING, Daniel (Delft University of Technology) Introduction
One way to offer more capacity on a train service is a frequency increase. More frequent trains, however, can decrease the reliability of the timetable execution, not only because of reduced slack times between trains on the line in question, but also because of possible new dependencies with other lines. Homogenizing services, i.e. running trains with very similar stopping patterns, can increase slack time between trains and make the frequency increase possible, at the cost of higher planned travel times for some passengers.
Given different stop patterns, frequencies and given train orders, the stability of the timetables are compared and it is shown, under what conditions is a certain frequency increase feasible with a given minimum stability margin.
The theoretical capacity of a railway line can be estimated using the UIC-406 standard (International Union of Railways, 2004). Further capacity calculation methods are also described in Landex (2008). The effects of additional train services on a network has been already explored e.g. using linear regression on recorded train run data on a line (Flier, 2008). We use an analytical mesoscopic model of a network of multiple lines to assess the timetable stability.
Methodology
The railway network is modelled on the mesoscopic level as a discrete event dynamic system (Goverde, 2007). Such an analytic model does not include every signal and track section and therefore runs much faster than a micro-simulation, but including infrastructure constraints as minimum headways at station entrances and junctions, it provides a better approximation of railway capacity than macroscopic models. Furthermore, for key infrastructure sections, it is possible to calculate minimum headways accurately using a microscopic simulation tool and later use these values in the higher-level model. Other mesoscopic models available include NEMO (RMCon, 2012).
Candidate timetables of given frequencies are constructed using the minimum process times known from earlier timetables and potentially assuming new locations where overtaking of trains is possible. Homogenized stop patterns are introduced by defining new stops for Intercity services, scheduling only local trains or introducing alternating stopping patterns.
The stability of a given periodic timetable can be analyzed calculating the maximum disturbance duration which can happen before any event of the otherwise delay-free network without delaying the same event in the following period. This recovery time, as well as the behaviour of knock-on delays, can be calculated efficiently using max-plus algebra (Goverde, 2010).
Case Study
The railway network in the Netherlands is facing two conflicting problems at the same time: demand is exceeding current capacity at peak hours, and the reliability of the train network is already perceived unsatisfactory due to the high number of trains and the interconnected train operations.
In the highly urbanized areas of the Netherlands, current railway service supply is characterized by most train services running four times an hour at high speeds, with double-deck trainsets of maximum platform length running at peak hours on the most heavily used routes. Higher capacity is expected to be achieved by a further increase of train frequencies on key corridors. However, as real-life tests have shown, these very high frequencies make the network even more vulnerable to knock-on delays.
Numerous approaches have been suggested to improve the reliability of the current train network, including driver assistance systems (Van den Top, 2009), installing more signals (Weeda and Hofstra, 2008) and the timely implementation of the European Rail Traffic Management System (ERTMS) (Goverde et al., 2012). Another proposition is the simplification of the network by shorter lines, less route variations and more homogeneous rail traffic by fewer different stop patterns and no freight traffic at peak hours. Homogenizing the line pattern also increases capacity, at the cost of increased travel times for some routes.
The City Region of Amsterdam (Stadsregio Amsterdam) proposes to introduce both higher frequencies and more homogenous train traffic at peak hours on the commuter lines to and from Amsterdam (Wiebes, 2012). The homogenization is expected to offset the decrease in reliability due to higher frequencies. Furthermore, heavily differenciating between peak and off-peak hours means that the advantages of long, direct express train lines are maintained for most of the day, at the expense of braking the pattern of running virtually the same train schedule through the whole day. A different peak schedule was already proposed for the Dutch train network earlier by Bruijn and Kieft (2004).
Page 85 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
The above model and methodology makes it possible to explore the effects of increased frequencies and homogenized train traffic on timetable stability, and determine, to what extent can homogenization make a reliable higher frequency timetable possible.
Bruijn, I. A. A. A., & Kieft, D. S. C. S. (2004). Net als in Japan: aparte spitsdienstregeling voor NS? (in Dutch), Colloquium Vervoersplanologisch Speurwerk, Zeist, 2004. Flier, H., Graffagnino, T., & Nunkesser, M. (2008). Planning Additional Trains on Corridors. Goverde, R. M. P. (2007). Railway timetable stability analysis using max-plus system theory. Transportation Research Part B: Methodological, 41(2), 179-201. doi:10.1016/j.trb.2006.02.003 Goverde, R. M. P. (2010). A delay propagation algorithm for large-scale railway traffic networks. Transportation Research Part C: Emerging Technologies, 18(3), 269-287. Elsevier Ltd. doi:10.1016/j.trc.2010.01.002 Goverde, R. M. P., Hansen, I. A., Corman, F., D’Ariano, A., & Trinckauf, J. (2012). Innovatie op het spoor en mogelijkheden van ERTMS in Nederland (in Dutch). International Union of Railways. (2004). UIC Code 406: Capacity. International Union of Railways (UIC). Landex, A. (2008). Methods to estimate railway capacity and passenger delays. PhD Thesis, Technical University of Denmark. RMCon. (2012). Net evaluation with NEMO. Retrieved September 6, 2012, from http://www.rmcon.de/en/products/railsys-product-family/railsys-enterprise/net-evaluation-with-nemo.html Van den Top, J., Laube, F., Van Luipen, J. (2009). Controlling the Normal, not the Exception. 3rd International Seminar on Railway Operations Modelling and Analysis, Zürich, 2009. Wiebes, E. (2012). Beter OV voor de Stadsregio Amsterdam (in Dutch). Weeda, V. a., & Hofstra, K. S. (2008). Performance analysis: improving the Dutch railway service. Computers in Railways XI, 463-471. doi:10.2495/CR080451
Page 86 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
97 - Railway System Resilience Analysis and Evaluation Presenter: Ms. DAI, Linsha (University of Birmingham) Introduction Though rail performance has steadily improved in recent years, delays are still inevitable in any railway system. Disturbances to railway traffic can be caused by a range of factors. On many lines where there is high capacity utilisation, a small disruption of a service for a few minutes, can lead to a chain of knock on delays that affect other services. It follows that minimising the propagation of delays is an important task to ensure the smooth running of the railway.
This work presents a case study analysing a variety of junction control methods and compares their effectiveness in reducing the influence of knock on delays in a range of scenarios. A general definition of the resilience of a railway has been developed for this work; it is divided into three levels according to the size of the disturbance and strategies to absorb delays [1-2]. It can therefore be used for disturbances of any severity on the railway. An analysis method that allows the visualisation and measurement of the propagation of delays is used to compare the response of the different junction control methods to a range of delayed scenarios.
Classification of disturbances and resilience In this paper disturbances are considered in three main categories: (1) perturbations – that is, delays that are caused by natural variances in the system, such as dwell times at stations, signal sighting times, etc.; (2) minor disruptions – that result from knock-on delays, junction interactions, degraded train performance, non-timetabled train sequencing, etc.; (3) major disruptions – that are caused by, for example, train breakdown, power system failure, accidents, signalling failure, rail breaks, and extreme weather conditions, which result in the cancellation of trains and/or the closure or blocking of lines [3].
The term resilience is used here to represent the ability of the railway system to deal with disturbances. Generally, railway systems can deal with disturbances in three ways: (1) systems can absorb the perturbation automatically without active train rescheduling (i.e. no action being required by the traffic controller); in this paper this is referred to as stability; (2) active train rescheduling or reordering strategies can be implemented to prevent the propagation of minor disruptions and aid the recovery of the system, called robustness here; and (3) operational management measures such as train cancellation, rolling stock re-allocation etc. can be put into place when major disruptions occur, referred to as recoverability.
Methodology The Graffica HERMES rail simulation platform is used to model disturbances and the subsequent running of the rolling stock over a section of the railway network. A series of disturbed scenario simulations of varying severity and at different locations is systematically performed. The series of simulations is repeated with different junction control rules in place. Thus, information can be derived about the propagation of delays and overall resilience of the system in response to the different strategies.
Before evaluating the degree of resilience provided by different junction control approaches, a quality visualisation of resilience is needed. The visualisation shows the lateness, and change in lateness, of all trains in the system with respect to time, as well as the individual lateness of the most severely delayed trains. To evaluate the outcomes, some performance measures are needed to indicate the performance. Three measures, shown in Figure 1, are used; they are maximum lateness, time to recover and integral of delay.
Figure 1 Visualisation of system resilience
Case study The case study presented simulates part of East Coast Main Line (ECML) timetable, based on published data [4]. A series of disturbed scenarios representative of perturbations in the system (e.g. extended dwell times of up to 120 seconds), and of minor disruptions (e.g. degraded train performance, or longer dwell time delays) are simulated throughout the network with different junction control strategies in place. These are summarised in Table 1.
Table 1 Junction control strategies used in ECML case study
Example The following shows a disrupted scenario and the propagation of delays with two different junction control strategies, FCFS (Figure 2) and rule-based (Figure 3), in place. A single train heading for London King’s Cross is delayed for 2 minutes from 07:26 to 07:28 at Welwyn Garden City.
Figure 2 Recovery with FCFS junction control
Figure 3 Recovery with rule-based junction control
Table 2 Performance measures for FCFS and rule based junction control
Page 87 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
Figures 2 and 3 show the subsequent propagation of delays and Table 2 summarises their performance measures. It can be seen from the visualisation that the resilience of the system is enhanced when the junction control method is FCFS compared to rule-based.
The example shows that the resilience of the system is affected by changing the junction control method and demonstrates the usefulness of the visualisation approach. The paper will present a comprehensive evaluation of the resilience of the ECML case study with respect to the different junction control approaches described in Table 1.
References 1. L. Climent, M. A. Salido and F. Barber, Robustness in dynamic constrain satisfaction problems. International Journal of Innovative Computing, Information and Control ICIC International, 2011, 8(4) pp 2513-2532. 2. C. Roberts et al., D1.2 - A framework for developing an objective function for evaluating work package solutions, ON-TIME ONT-WP01-D-UOB-025-01, 2012. 3. J. Yuan, Stochastic modelling of Train Delays and Delay Propagation in Stations. PhD thesis, Delft University of Technology. 2006, pp 27-32. 4. UIC Leaflet 406: Capacity, International Union of Railways, UIC, 2004.
Page 88 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
98 - Defining Railway Capacity through Quality of Service Presenter: Ms. LU, Menglei (University of Birmingham) Background Railways worldwide are experiencing unprecedented growth and are searching for new approaches to improve capacity. This problem is drawing increasing attention from both government and industry.
Even though capacity is viewed as a very significant problem by the railway industry, different stakeholders (e.g. infrastructure managers, operators, funding agencies) view capacity in different ways. This results in there being no single agreed definition of the nature of railway capacity and its management. There is therefore still doubt about how to identify remaining capacity, making it very hard to identify which approaches to use when attempting to make capacity improvements on railway lines. Furthermore, different stakeholders have varying perceptions of which are the most important factors affecting capacity.
Definition of Capacity Railway capacity is a diverse concept. A wide variety of capacity definitions and methods of assessment have been published, for example (UIC 1983; Krueger 1999; Moreira, Garcia et al. 2004; UIC 2004; AEATechnology 2005; Albrecht, Brunger et al. 2008). However, the existing definitions and measures of capacity are often focused only on traffic volume or infrastructure occupation, both of which are quite different from one another. At a system level, the Quality of Service is actually the measure considered by customers when they decide whether or not to use rail.
Measures to Improve Capacity and its Utilisation According to Pachl (2002) and Abril, Barber et al. (2008), the capacity of a railway line or a part of a network could be improved by changing: (1) infrastructure parameters, such as block section lengths, the signalling system and track layout; (2) rolling stock and traffic parameters (e.g. train performance and mix); and (3) operating parameters (e.g. system variability, such as dwell times ).
Additionally, many other subsystems within the railway system affect railway capacity and its usability, such as the power supply, train door arrangements, junction characteristics and passenger management capabilities. The impact of these subsystems is often not considered in conventional measures as they are hard to quantify, although in certain cases they can have a significant impact (Roberts, Schmid et al. 2010). The paper presents a number of simulation case studies, based on real-world scenarios, which show how the different conventional measures of capacity used by different stakeholder groups are affected in different ways by varying subsystem parameters.
Defining ‘Quality of Service’ In order to develop a measure that addresses the key requirements of all stakeholders, a measure of Quality of Service is developed and defined in this paper. This research has been influenced and developed together with a group of infrastructure managers, operators, suppliers and academics working within the EC FP7 ON-TIME project (ON-TIMEConsortium 2012).
In the paper, Quality of Service is defined as a combination of the capability and dependability of the railway system. The aim has been to develop a comprehensive concept that covers the main attributes of the railway system, considering parameters as diverse as traffic volume, journey time, connectivity, punctuality, resilience, passenger comfort, energy, and resource usage. Mathematical expressions are developed for the various parameters. A high level view of the Quality of Service measure can be found in Figure 1.
Figure 1: Factors considered in the Quality of Service measure
From Figure 1, it can be seen that from the point of view of the railway system, the factors affecting Quality of Service can be broken down into capability and dependability. Capability covers all the “static” components that are relatively hard to change, which define the underlying ability of the railway system to perform its function, such as rolling stock type, infrastructure systems, timetables and operational rules. Dependability includes all the “dynamic” components of the system, which determine the performance of the system given the static components, such as, traffic management strategies, operational management, human factors, maintenance strategies and environmental factors. Dynamic components can be modified over a relatively short term with moderately low cost. It is shown in the paper how the railway system’s Quality of Service can be influenced by changing parameters relating to the railway systems components.
Conclusions will be drawn on the best way to define Quality of Service, and how the railway system can be improved to respond better to customer needs.
References Abril, M., F. Barber, et al. (2008). "An assessment of railway capacity." Transportation Research Part E: Logistics and Transportation Review 44(5): 774-806. AEATechnology (2005). Capacity Utilisation Indices Methodology. Albrecht, I. T., O. Brunger, et al. (2008). Railway Timetable & Traffic: Analysis, Modelling, Simulation, Eurailpress. Krueger, H. (1999). Parametric modeling in rail capacity planning. Simulation Conference Proceedings, 1999 Winter.
Page 89 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
Moreira, N., L. Garcia, et al. (2004). Network capacity. Computers in Railway Six. 15: 35-43. ON-TIMEConsortium. (2012). "ONTIME." from http://www.ontime-project.eu/home.aspx. Pachl, J. (2002). Railway operation and control, VTD Rail Pub. Roberts, C., F. Schmid, et al. (2010). A new railway system capacity model: 82. UIC (1983). UIC Leaflet 405-1: Method to be Used for the Determination of the Capacity of Lines, International Union of Railways. UIC (2004). UIC Leaflet 406: Capacity, International Union of Railways.
101 - Modeling Train Delays in Rail Networks with Large Disturbances Presenter: MILINKOVIć, Sanjin (University of Belgrade) ABSTRACT Railways are complex systems consisted of subsystems equipment (infrastructure, rolling stock, safety and control systems, etc.) that interact with each other. It is often a case that disturbances (technical failures, human factor, and external factor) in railway networks upset the balance of subsystems and cause primary train delays. Further, primary delays can propagate thru timetable to generate additional knock on delays. Stations and junctions are the bottlenecks of railway network as places where train routes conflict and interact according to the timetable. At stations, arrival delays have a complex structure as they can be primary or secondary delays. A detailed delay analysis enables the identification of the influencing factors, delay dependencies and creation of a model for predicting train delays. For the management of delays, it is necessary to develop a model for prediction of train delays. A model for calculating train delays can be used in the process of railway operations and timetable planning, and operational management. Variations and dependencies of train delays can be examined by analytical and simulation methods. Statistical analysis of arrival delays data suggests that many factors influence train delays and that prediction of the train delay is very difficult because of the many factors involved., Analytical models for calculating arrival delays (Yuan) can give excellent results for railway systems with small disturbances. When there are large and frequent disturbances, (like in Serbian Railways), there is a need for a train delays model that can be easily adaptable to any specific station. We suggest an approach based on soft computing techniques (Karlaftis) i.e. an Artificial Neural Networks (ANN) model for calculating arrival delays. Owing to their capabilities of learning and generalization from observation data, ANN has been widely accepted by engineers and researchers as a tool for processing of experimental data. Neural networks are massively parallel, distributed processing systems representing a new computational technology built on the analogy to the human information processing system. The network is adjusted, based on a comparison of the output and the target, until the network output matches the target. This system requires many pairs of data to train the network. By comparing inputs and outputs and thus training the network, ANN is extracting knowledge from the available data, and has a capability of learning from examples. A neural network model for calculation of train delays is developed to predict train delays in rail network with large disturbances. The ANN model is developed using the historical data of train delays in two stations in Belgrade Railway Node in Serbian Railways. We have analyzed the data of train’s delays during years 2010, 2011 and first half of 2012 in stations Rakovica and Novi Beograd for passenger trains. Previous and preliminary analysis made for station Rakovica in July 2010 has shown that many factors affect train operations. In Serbian Railways, passenger trains operate by the timetable, daily repeating the same arrival and departure pattern. This pattern is affected by additional freight trains that operate in time intervals between regular passenger trains according to the transport demand. Due to the poor maintenance, many sections of Serbian Railway network are in such condition that, for safety reasons, there are speed restrictions. Restricted speed sections reduce the capacity of railways and generate train delays. Even with the low number of trains and the low utilization of railway line capacity, timetable is not able to compensate many primary delays. A combination of the restricted speed sections, lack of locomotives and the timetable design is generating train delays and producing low punctuality and low reliability of Serbian railway system. Large disturbances frequently influence system in such way that the time supplements in the timetable cannot compensate delays, and traffic control measures are necessary. Because of the specific conditions, it is important to analyze arrival delays by stations on train line level and on individual train level. After the analysis of traffic conditions and data collected during the previous research, we defined the input data for the ANN model for train delays has four parameters: Train category; Period of the day (time of the arrival at station); Train traveled distance (percentage of the distance that is completed when the train arrived in station); and Infrastructure influence (weighted numerical description calculated for each train route depending on the specific conditions on rail line where the train was running). The ANN model is trained using the historical data of train delays as target data and predefined input data for each train. The ANN model is verified and compared with the statistical multiple regression model developed using the same pairs of input-target data. Application of models is possible for the stations with known historical data of train delays where relation between timetable and infrastructure can be described and adapted to create a model input.
Page 90 5th International Seminar on Railway Operations Modelling and Analysis - RailCopenhagen / Book of abstracts
105 - Integrated Optimization of Rolling Stock Rotations for Intercity Railways Presenter: Mr. REUTHER, Markus (Zuse Institute Berlin) Deutsche Bahn Fernverkehr AG provides the largest intercity railway service in Europe. In order to implement their timetables rolling stock rotations are built to operate passenger trips by rail vehicles, which are among the most expensive and limited assets of a railway company. The main challenge arising in planning the operations of the rolling stock is to integrate the treatment of different technical aspects. The major requirements can be summarized as vehicle composition rules, maintenance constraints, infrastructure capacities, and regularity stipulations. Each of these requirements is already complex in its own right. Moreover, it is almost impossible to treat them sequentially, i.e., a step by step approach leads to infeasibilities or inefficient results. The most important requirements are the following: One main characteristic of nearly all railway systems is that rail vehicles can be combined to form trains. Therefore detailed rules, e.g., for vehicle compositions and for coupling activities must be considered. The rolling stock has to be maintained frequently. This leads to several maintenance constraints. We consider cumulative time and distance resources which are constrained by predefined bounds. To comply to those bounds rail vehicles must be maintained in periodical intervals. Maintenance and also parking activities usually consume infrastructure capacity. This capacity is limited and therefore it is integrated in our model. A common structure of railway timetables is that they are almost periodic. Only a few of the given passenger trips differ from day to day. Another requirement for the rolling stock rotations is that they should utilize the periodicity of the timetable; this objective is called regularity.
The paper contributes a new generic hypergraph based mixed integer programming approach to optimize rolling stock rotations. This model is able to handle all requirements of Deutsche Bahn w.r.t. rolling stock rotations in an integrated manner. We show how our model can be instantiated to deal with very complex and detailed technical rules arising in industrial rolling stock applications. In addition, we present a new integrated algorithm that is a combination of a special column generation method, a local search heuristic, and an adaption of the rapid branching search scheme. Comprehensive computational results prove that our model and algorithm produces high quality and implementable results. Rotation planners of Deutsche Bahn validated the resulting rolling stock rotations from a detailed technical and operational point of view.
Page 91