Proceedings of the Eastern Asia Society for Transportation Studies, Vol.7, 2009

A Dynamic Traffic Analysis Model for the Korean Expressway System Using FTMS Data

Jeong Whon, YU Mu Young, LEE Professor Researcher Division of Environmental, Civil Graduate School of Engineering and Transportation Engineering Ajou University Ajou University , Korea 443-749 Suwon, Korea 443-749 Fax: +81-31-219-1613 Fax: +81-31-219-1613 E-mail: [email protected] E-mail: [email protected]

Abstract: Operation of intelligent transportation system technologies in transportation networks and more detailed analysis give rise to necessity of dynamic traffic analysis model. Existing static models describe network state in average. On the contrary, dynamic traffic analysis model can describe the time-dependent network state. In this study, a dynamic traffic model for the expressway system using FTMS data is developed. Time-dependent origin-destination trip tables for nationwide expressway network are constructed using TCS data. Computation complexity is critical issue in modeling nationwide network for dynamic simulation. A subarea analysis model is developed which converts the nationwide O-D trip tables into subarea O-D trip tables. The applicability of the proposed model is tested under various scenario. This study can be viewed as a starting point of developing deployable dynamic traffic analysis model. The proposed model needs to be expanded to include arterial as well without critical computation burden

Key Words: Dynamic, Subarea, FTMS Data

1. INTRODUCTION

1.1 Background of the study

Conditions of transportation system change according to the correlation between the physical facilities and the demand, which in turn varies with the passage of time. Until now, static analyses have been conducted for the planning and operation of the transportation system, focusing on the average behavior of the passengers. Meanwhile, the rapid increase in the number of registered vehicles is causing the problem of imbalance between the demand and the supply in the transportation system in Korea. It is not surprising that the concerns regarding dynamic traffic management, which involves the concentration of traffic by time flowing, are increasing (Ran, 1996). The transportation network that uses the ITS (intelligent transportation system) technology facilitates the construction of a time-dependent database for network state (Mahmassani, 2001), and the requirements which come to be high for analysis of the transportation system using this technology indicate the limitations of the current traffic analysis model and the future research trend. Dynamic traffic analysis that is a main subject of this research may be utilized in various traffic field such as real time traffic management, traffic operation, traffic demand management as well as Dynamic Traffic Assignment, utilizing by basis algorithm to prepare in ultramodern traffic system period hereafter. This research has a meaning from the point which is early step of research about dynamic traffic analysis that take advantage of real individual travel data and is expected that more lively research about dynamic analysis field.

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.6, 2009

1.2 Research Contents

Development of dynamic analysis model on the basis of existing static traffic planning model 20 years ago in transportation planning field, is progressing and it is displaying big current of research direction that planning field and research of traffic operation field are widening commonness scope hereafter. In case of static analysis of concept an existing day or during 1 hour, it is a arithmetical and average analysis which follows in O-D traffic volume regarded as intention of travel and network capacity relationships. Also, the distinctive quality according to flowing of traffic to progress with flowing of time can't be reflected, and it has a problem that delay of intersection that is presented in several researches can't be reflected. Currently, travel time, speed, and delay according to Volume-delay Function utilized effect analysis by improvement of transportation system is the standard for choice of route, and it is not a travel time of correct meaning. Time-dependent O-D trip table is constructed every 15 minutes about expressway of whole country, utilizing data of FTMS (Freeway Traffic Management System) in this research, and methodology for converting time-dependent O-D of whole country expressway into time- dependent subarea O-D was developed in order to solve the problem of computation complexity. Subarea analysis must be conducted not only because of its efficiency in terms of operation time but also because the traffic analysis under the advanced transportation system involves the accurate analysis of the subject area instead of the whole network. Analysis of research is taking advantage of TCS data that is each individual real travel data through network of , the error of the O-D traffic volume which is a main input date of traffic analysis could be diminished. and then in case of trip maker using Expressway which is a toll road, he has a strong intention to shorten his own travel time when he chooses a path and this heighten reliability of path selection model. The construction of the time-dependent O-D trip table has been constructed over the - corridor of with model calibration process time zone every 15 minutes. While existing static traffic model is constructed through one calibration process about 1 points, but time-dependent calibration process of (60/15)* 24 =96 about 1 points is accomplished in this research. Also network state analysis by time zone is accomplished through scenario analysis by traffic accidents situation and capacity enlargement plan to confirm application possibility of constructed model.

Figure 1 Procedure of the research

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.6, 2009

Chapter 3 of this paper introduces a methodology for converting the dynamic origin- destination (O-D) trip tables established using the FTMS (Freeway Traffic Management System) data into the subarea O-D trip tables. In Chapter 4, the applicability of this model is verified via scenario analysis, and the necessity of dynamic traffic analysis model, which is a future traffic analysis trend, is presented. In Chapter 5, the results of this study and the aim of the future study are presented.

2. Relevant Literature

The researches on dynamic traffic analysis that have thus far been conducted used either of two kinds of techniques: the analytical and the simulation techniques. Among the analytical techniques, the approach that uses variational inequality is known to be useful in dynamic assignment. Friesz (1993) presented the problem of dynamic user equilibrium (DUE) by converting it into a variational inequality, and the link-based variational-inequality model was developed by Ran et al. (1996). The network model, which is a constraint of the dynamic model, must be accurately established so that it could describe the traffic with the passage of time. To describe the congestion in the network, the point queue model, wherein the travel time of the link is represented by the constant travel time and the delayed queue (Kuwahara, 1997), was recently studied. The horizontal traffic network load model, which horizontally expresses the physical location of a car and which is an improved version of the vertical traffic load model that expresses the queues by piling them up just in front of the crossroad, was also studied. Meanwhile, the models that focus on describing the dynamic characteristics of the traffic flow in a manner similar to the real world are called “dynamic network loading models.” These models, wherein the single link is divided into travel and queue sections and is described by dividing the dynamic condition of the link into travel and waiting, have a limitation in describing the diverse dynamic characteristics of the traffic flow. To address this problem, Cremer (1999) developed a moving cell-based simulation model. In the dynamic analysis model, the travel time changes with the passage of time, unlike the current static shortest-route selection algorithm, which has a fixed-link travel time regardless of the passage of time. The time-dependent, shortest-route selection algorithm was studied by Cooke and Halsey (1996) in full scale. Ziliaskopoulos (1993) developed a time-dependent, shortest-route algorithm from all origins to a specific destination, based on Bellman’s principle of optimality. Dynamic travel assignment has the limitations of dynamic changes, including the queue length of each link, dynamic travel time and traffic inflow/outflow, and route selection. When a constraint of the objective function is added, the objective function can be nonconvex and the optimal solution cannot be derived through the mathematical-optimization technique (Carey, 1992). The simulation loading algorithm, which is suitable for route-based travel assignment, was proposed for the driver’s dynamic route selection. Mahmassani (2001) developed a simulation-assignment methodology for dynamic travel assignment. The vehicles generated by the time-dependent O-D traffic are assigned using four methods. The travel pattern by time zone for the next travel assignment is generated through simulation

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.6, 2009

3. METHODOLOGY

3.1 Dynamic Traffic Analysis Model

Traffic assignment process for dynamic traffic analysis is divided by path selection process and dynamic loading process. At existing static traffic assignment process, while suppose travel time of link by fixing price of average concept for shortest path choice, in case of path selection algorithm of dynamic traffic assignment, shortest path (Time-dependent Shortest Path Algorithm) is selected considering change of link travel time according to traffic volume change. Also, if had applied network loading process by User Equilibrium principle based on mathematical arithmetic price that do not consider the movement of vehicles to describe traffic network utilization pattern of the O-D traffic volume in existing static model, apply dynamic loading process called simulation load techniques for delineation by time-dependent location of vehicle in dynamic methodology. Micro-simulation of the time-dependent O-D trip tables is accomplished through time zone of every 15 minutes Stochastic Shortest Path Algorithm instead of traffic assignment process that utilize time-dependent shortest path algorithm for dynamic state analysis of traffic network in this research and this is based on assumption which path selection process according to network state change may not change within 15 minutes making allowances for Expressway which is the target of this research. Also, accomplishing simulation that achieve traffic description based on Car-following Theory, it can reflect network loading according to time- dependent location of each individual vehicles.

3.2 Subarea Analysis Model

The time-dependent O-D trip tables was constructed using the FTMS data of Korea Expressway Corporation, and network provided by KTDB (Korea Transportation Database) was used as a one of the input data after modifying. The time-dependent O-D traffic was loaded on the network according to the origin and starting time at the every simulation interval in the model. In this study, a network for the micro-simulation of the national expressway was constructed using the commercial program Trans-Modeler v1.5. The network was formed with 2,219 links and 1,958 nodes. The total traffic for the simulation was 1,939,393 vehicles/day, and run-time for one day simulation would have lasted for 16 hour or more. It was thus adjudged to be beyond the scope of the model run-time. In this study, a subarea network was constructed on the Seoul-Daejeon section of Gyeongbu Expressway, considering the simulation time for the analysis.

3.2.1 Subarea Network

The section that was analyzed in this study covered the Seoul-Daejeon section of Gyeongbu Line. The link which is connected Seoul-Daejeon section of Gyeongbu Line is included in subarea to reflect traffic flow in and out of the analyzed segment. : Dongsuwon-Maseong section (Yeongdong Line), Songtan-Namanseong section (Chungju-Pyeongtaek Line), JC-Namcheonan IC section (Cheonan-Nonsan Line), Nami JC-Seocheongju section (Jungbu Line), Cheongwon JC-Muneui IC section (Dangjin-Sangju Line), and Hoideok JC- Bukdaejeon IC section (Honam Branch Line). Digital maps and literature were used for the accurate expression of the diverging/merging area of the network, which were important points in the traffic flow.

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.6, 2009

Figure 2 Subarea network

Table 1 Subarea traffic analysis zone 1 Seoul 8 15 Namanseong 22 Jungbu Line Cheonan- 2 Suwon 9 Cheongwon 16 Seocheongju 23 Nonsan Pyeongtaek- 3 Giheung 10 Sintanjin 17 Muneui 24 Chungju Yeongdong 4 11 Daejeon 18 Bukdaejeon 25 (EB) Yeongdong 5 12 Dongsuwon 19 Gyeongbu 26 (WB) Honam Branch 6 Cheonan 13 Songtan 20 Line Cheongwon- 7 Mokcheon 14 Seoanseong 21 Sangju

3.2.2 Subarea O-D

The end points of subarea become imaginary centroid and there are two problems in the calculation of the traffic at the end points of subarea, in converting the O-D trip table constructed over the national expressway into the subarea O-D trip table. First, the travel time for the traffic starting from each toll gate up to the subarea network must be considered. Second, although the O-D trip tables are known by the current TCS data, there is a problem in the route choice of each O-D trip. To solve this problem, total volume of the imaginary centroid were calculated using the VDS(Vehicle Detection System) data, and virtual trip distribution obtained from the static travel assignemnt method (user equilibrium) is applied.

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.6, 2009

Figure 3 Two problems in converting O-D table

3.3 Methodology for the Time-dependent O-D Construction

For the construction of the time-dependent O-D tirp tables for the whole network, TCS data were used, and a methodology for converting them into subarea O-Ds was developed. The four-step procedure for converting the O-D trip tables that had been constructed over the national expressway into subarea O-D traffic is shown in Fig. 4.

3.3.1 Time-dependent O-D Construction of the Nationwide-Expressway-Network(Using TCS Data)

A time-dependent O-D trip tables were constructed for the 248 closed-type toll gates of Korea Expressway Corporation. Considering that the expressway is the subject of this study, it was expected that the change in the network states within 15 min would not be large. Consequently 96 ((60/15) *24) O-D trip tables were obtained.

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.6, 2009

Figure 4 Construction of the subatea O-D

3.3.2 Conversion into Subarea O-D(Using VDS Data)

The O-D table of the subarea network is shown in fig. 5. The traffic volume between the inner zones of subarea can be constructed using TCS data (AREA[1]). Also time-dependent total volume of inner zones can be obtained from TCS data.

Figure 5 Subarea O-D matrix

VDS data were used for the calculation of the time-dependent total volume in zones 19-26 (AREA 2), which were the end points of subarea (imaginary centroids). Time-dependent total inflow of imaginary centroids can be created using VDS data directly, but a problem was encountered about time-dependent outflow of imaginary centroids: the constructed time- dependent O-D trip tables(from TCS data) was based on the starting time, whereas VDS data was based on arrival time. Accordingly, considering a less-than-one-hour travel time from Seoul TG to Cheonan TG, it was assumed that the total outflow of imaginary centroids at time t was proportional to the total outflow of imaginary centroids until t ~ (t+2).

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.6, 2009

= traffic inflow of the toll gate i at time t = total volume of subarea O-D table at time t

= VDS traffic volume of imaginary centroid i at time t

3.3.3 Derivation of the Travel Distribution

Traffic volume between end points of subarea(imaginary centroid) and Subarea inner zone is impossible of direct creation from FTMS data ([3] AREA). Trip distribution between each zone is needed for solution of this. That is, as time-dependent production/attraction volume of each zone is created from TCS data, draw virtual trip distribution ratio(using User Equili - brium Assignment) and calculate the traffic volume between each zone. At this time, in the case of Subarea inner zone, traffic volume among subarea inner zone takes advantage of TCS data(time-dependent true traffic volume). The total volume at end point of subarea (imaginary centroid) is created from VDS data of that location. This trip is happened in Subarea outside area not to happen at Subarea end point. and trip distribution must be reflecting pattern between Subarea outside zone(trip generated truly) and inner zones of subarea, and it is not a trip distribution pattern among imaginary centroid and inner zones of subarea. In this study, it was adjudged that the rates of the routes selected by trip makers would not greatly change whereas the traffic volume between origins and destinations would change with the passage of time. A virtual travel pattern that resulted from user equilibrium traffic assignment with the concept in which the user selects the route that minimizes the travel time was also derived and employed. The subarea O-D was created with the counting station set as the end points of subarea. Accordingly, the rate of the travel distribution of the traffic passing end points of subarea (imaginary centroids) could be obtained.

3.3.4 Matrix Balancing

Matrix balancing was conducted to satisfy the production/attraction volume constraint of the constructed O-D trip table. To fix the traffic volume between the inner zones of the subarea network, which could be known using TCS data, 3D matrix balancing was used. The balancing model can be expressed as follows.

for each O-D pair(p, q) (1)

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.6, 2009

(2)

(3)

(4)

These are the processes that can be employed for finding the optimal αp, βq that satisfies the total volume preservation constraints (2) and (3) for each origin p and destination q. During these processes, the traffic volume between inner zones in the subarea network (gpq, where p<18 and q<18) was kept constant at the TCS true data (cpq). The algorithm for finding α and β verifies if the results of steps 1 and 2 satisfy the convergence condition of step 3, and it is expressed as follows:

Figure 6 Balancing algorithm

4. MODEL EXPERIMENT

4.1 Model Construction

Micro-simulation network is constructed for dynamic traffic analysis in this research and utilized 'Transmodeler 1.5' that is common use program The network that was constructed over the Seoul-Daejeon corridor of Gyeongbu Line was formed with 341 links and 304 nodes. 18 inner toll gate of subarea and 8 imaginary zone(end

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.6, 2009 point of subarea) is established to centroid in this study. The October 13, 2008 FTMS data were used for the construction of the time-dependent O-D. The national-expressway traffic volume consisted of 1,939,265 vehicles. 4 case can be divided as following picture. Inner - inner (P1), outside - inner (P2), inner - outside (P3), outside - outside (P4). As see in picture in case of travel from subarea network outside zone to outside zone, you can divide travel which is accomplished outside Subarea network and pierced subarea network.

Figure 7 4-type travel pattern

The traffic volume in the subarea consisted of 614,695 vehicles, which accounted for 32% of the whole traffic volume. The exclusive and the ETCS utilization rate were set at 10% and 30% of the O-D flow.

Figure 8 Subarea O-D(October 13)

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.6, 2009

Table 2 Time-dependent traffic volume of subarea Time Volume Ratio Time Volume Ratio 1 9,533 1.5% 13 33,055 5.2% 2 6,562 1.0% 14 35,465 5.6% 3 4,817 0.8% 15 37,340 5.9% 4 4,313 0.7% 16 38,848 6.2% 5 5,535 0.9% 17 38,720 6.1% 6 11,295 1.8% 18 39,675 6.3% 7 24,565 3.9% 19 39,155 6.2% 8 36,193 5.7% 20 32,568 5.2% 9 37,683 6.0% 21 27,129 4.3% 10 37,187 5.9% 22 23,621 3.8% 11 37,958 6.0% 23 18,821 3.0% 12 36,382 5.8% 24 13,311 2.1%

4.2 Application of the Model and It’s Analysis

To verify the applicability of the constructed model, time-dependent network analysis was conducted, according to the diverse situations that could occur. On the supply side, the network state with the passage of time was analyzed through the analysis of the travel time for the Daejeon toll gate-Seoul toll gate in the case of the lane expansion of the Giheung toll gate–Suwon toll gate section (Gyeongbu Line). It showed an average travel time reduction of 0.33 min, and of 5 min for the vehicle starting at 6:00. The travel time reduction by lane expansion at the non-peak period was insignificant. It even increased in some cases due to the simulation error.

Figure 9 Analysis of the network state in the case of road expantion

The change in the network was analyzed assuming a traffic incident between Singal JC and the Seoul toll gate (Gyeongbu Line, towards Seoul) starting from 14:00, for 40 min. The time- dependent results of the analysis within the Suwon toll gate–Seoul toll gate section showed that the delay continued for 90 min, until 15:30. The vehicle that departed from the Suwon

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.6, 2009 toll gate at 14:20 showed the largest increase (13.4 min). The travel time increase of the vehicle that passed the Suwon toll gate was relatively high at the initial stage of the incident occurrence, which indicates that the early detection and handling of an incident are necessary to provide high-quality service in the advanced traffic system.

Figure 10 Analysis of the delay caused by an accident

For the analysis in the case of traffic increase caused by social and economic reasons, the travel time for the Daejeon toll gate–Seoul toll gate section was analyzed with a 4% increase of the time-dependent O-D traffic volume. The average travel time increase was 1 min, and the vehicle that started at 17:00 showed a travel time increase of 7 min. The results of the analysis of the travel time influenced by the increase in the O-D traffic volume showed that the impact at the peak time is more important than that at the non-peak time. Diverse transportation plans, including feasibility evaluation and priority decision, are currently based on the average daily-network data. In this study, a macroscopic analysis of the daily traffic for reasonable and practical traffic plans was conducted, and its results indicate the need to consider the traffic data and the network characteristics by time zone.

Figure 11 Analysis of the network state in the case of O-D trip increase

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.6, 2009

4. CONCLUSION

In this study, a dynamic traffic analysis model for expressway network using FTMS data was developed, reflecting the need for the dynamic analysis of the traffic system. The need for dynamic traffic analysis, currently the main traffic analysis trend, was likewise emphasized. Thus far, traffic analysis has been a mere analysis of the daily average using static data. In this study, in order to solve simulation computation complexity, dynamic traffic analysis model which was utilizing subarea analysis was developed. A methodology for converting the dynamic O-D of the national-expressway network into subarea O-D was developed using TCS data, and the network conditions were tested by time zone.

Diverse scenarios were analyzed in the aspects of supply and demand to verify the applicability of the model. This study is important in that it is an early-stage dynamic-traffic- analysis study and in that it uses FTMS data, which are actual individual travel data. Studies on the construction of time-dependent O-D trrip table, which can be applicable to networks, including other roads as well as expressways, must be conducted. Diverse researches must also be conducted to address the computation complexity, including the development of a hybrid model and a dynamic analysis program.

The requirement level of the traffic system service will continue to increase. In addition, the dynamic analysis model will serve as a theoretical foundation for the intelligent transportation system (ITS), and data creation through the dynamic analysis of the traffic system is a reliable way of improving the current analysis and service. It is expected that more scientific and practical studies will be performed using the dynamic analysis model proposed in this study.

6. REFERENCE

1. Adamo V., V. Astarita, M. Florian, M. Mahut, J. H. Wu, 1999, Modeling the Spill-Back Congestion in Link Based Dynamic Network Loading Models : A Simulation Model with Application, 14th International Symposium on Transportation and Traffic Theory, Elsevier. 2. Carey, M., 1992, Nonconvexity of the dynamic traffic assignment problem, Transportation Research 26B. 3. Gang, J. G. et al., 2004, Construction of a predictive multiple-layer dynamic distribution m odel, and development of its algorithm. Journal of Korean Society of Transportation, vol.2 2. 4. Kuwahara, M., T. Akamatsu, 1997, Decomposition of the reactive dynamic assignment with queues for a many-to-many origin-destination pattern, Transportation Research 31B, 1-10. 5. Mahmassani, H. S., 2001, Dynamic network traffic assignment and simulation methodolog y for advanced system management application, Networks and Spatial Economics, 267-292. 6. Lam, W. H. K., H. J. Huang, 1995, Dynamic user optimal traffic assignment model for ma ny-to-one travel demands, Transportation Research 29B. 7. Park, J. S. et al., 2001, A dynamic assignment model for dynamic traffic management during the AM peak. Journal of Korean Society of Transportation, vol19. 8. Ran, B., D. E. Boyce, 1996, Modeling the dynamic transportation network: An intelligent-tr ansportation-system-oriented approach (2th ed). Springer.

Proceedings of the Eastern Asia Society for Transportation Studies, Vol.6, 2009

9. Ran, B., Hall, R. W., D. E. Boyce, 1996, A link-based variational-inequality model for dyna mic departure time/route choice, Transportation Research B, 30. 10. Sheffi, Y., 1985, Urban transportation networks: Equilibrium analysis with mathematical p rogramming methods. Prentice-Hall. 11. Wie, B. W., T. L. Friesz, R. L. Tobin, 1990, Dynamic user optimal traffic assignment on a congested multidestination network, Transportation Research 24B, 431-442. 12. Xu, Y. W., J. H. Yu, M. Florian, P. Marcotte, D. L. Zue, 1999, Advances in the continuous dynamic network loading problem, Transportation Science, 33(4). 13, Zhou, X., S. Erdogan, H. S. Mahmassani, 2006, Dynamic origin-destination trip demand e stimation for subarea analysis, Transportation Research Record 1964, 176-184. 14. Ziliaskopoulos, A. K., H. S. Mahmassani, 1993, Time-dependent, shortest-path algorithm for real-time intelligent-vehicle-highway-system applications, Transportation Research Rec ord 1408.