Optimising Throughput in the Hunter Valley Coal Chain Using Integer Programming Techniques
Mohsen Reisi Ardali
M.Sc. (Mathematics)
School of Mathematical and Physical Sciences The University of Newcastle
Thesis submitted for the degree of
Doctor of Philosophy in Mathematics
March 2014
Statement of Originality
The thesis contains no material which has been accepted for the award of any other degree or diploma in any university or other tertiary institution and, to the best of my knowledge and belief, contains no material previously published or written by another person, except where due reference has been made in the text. I give consent to the final version of my thesis being made available worldwide when deposited in the University’s Digital Repository, subject to the provisions of the Copyright Act 1968.
Mohsen Reisi Ardali
iii
Acknowledgements
I wish to express my sincere appreciation to those who have contributed to this thesis and supported me in one way or the other during this amazing journey. First and foremost, I am deeply indebted to my supervisors Prof. Natashia Boland and Prof. Martin Savelsbergh whose positive attitude and insightful thinking have guided me throughout this research. Without the stimulating discussions, guidance, and support, this work would not have been possible. Very special thanks to my previous co-supervisor Dr. Faramroze Engineer for his support in the first year of my PhD. My sincere gratitude is reserved for Dr. Hamish Waterer and the Operations Research group at the University of Newcastle by providing frequent feedbacks and opportunities to discuss my research. I would like to thank the support of the Australian Research Council and the Hunter Valley Coal Chain Coordinator, under the ARC Linkage Project grant LP0990739. I appreciate the help of the CSIRO in supplying data, models and documentation that enabled my project to get off the ground. I would also like to take this opportunity to thank Dr Andreas Ernst from the CSIRO for his useful suggestions. I really appreciate personnels at the HVCCC for their feedback and provision of essential data and information. I am thankful for the help I received from all the people in the Department of Mathematics. Dr. Masoud Talebian, Hadi Charkhgard, Rachel Bunder, Simranjit Kaur, Reena Kapoor, thanks for being there for me through the years. Last but not least, I would like to appreciate my lovely wife, Afsaneh, who supported me in every possible way to see the completion of this work. I owe a lot to my parents, who encouraged and helped me at every stage of my personal and academic life, and longed to see this achievement come true.
v
Contents
Chapter 1. Introduction 1 1.1. The Hunter Valley Coal Chain: An Overview4 1.2. Planning and Scheduling in the Hunter Valley Coal Chain7 1.2.1. Simulation8 1.2.2. Throughput Assessment Tool9 1.3. Integer Programming Background 10 1.3.1. Integer programming 10 1.3.2. Symmetry in Integer Programming 14 1.3.3. Implicit Modelling 15 1.3.4. Polyhedral Analysis 17 1.3.5. Lifting 18 1.4. Contributions 19 1.5. Thesis Outline 20
Chapter 2. Demand Driven Throughput Assessment for the HVCC 23 2.1. Introduction 23 2.2. Literature review 23 2.3. Demand Driven Throughput Assessment Models 27 2.3.1. A Train-job Based Model 27 2.3.2. A Component-job Based Model 34 2.3.3. Strategies to Increase Efficiency 36 2.4. A Computational Study 39 2.4.1. Data sets and their key features 39 2.4.2. Efficiency 43 2.4.3. Throughput and System Analysis 45 2.5. Final Remarks 51 Comparison with Singh, Sier, Ernst, Gavriliouk, Oyston, Giles, and Welgama (2012) 52
Chapter 3. Implicit Modeling: Load Point Based Model 55 3.1. Introduction 55
vii viii CONTENTS
3.2. Load point Based Model 57 3.3. Equivalency of Component and Load Point Based Models 60 3.3.1. Proving sufficient conditions for validity of the implicit model 60 3.3.2. Application to the load point model 66 3.4. Implementation 69 3.5. A Computational Study 72 3.6. Final remarks 74
Chapter 4. Production Scheduling with Flexible Deadline Problem 77 4.1. Introduction 77 4.2. One demand with constant capacity 78 4.2.1. The General Model 79 4.2.2. PSFDP-1C and Lot-sizing with Constant Batches 79 4.2.3. Polyhedral Analysis 83 4.2.4. Example 85 4.2.5. Separation 100 4.2.6. A complete reformulation of PSFDP-1C with only facet-inducing inequalities 102 4.3. One demand with varying capacity 112 4.3.1. Separation 125 4.3.2. A complete reformulation of PSFDP-1V with only facet-inducing inequalities 128 4.4. Multiple demands with constant capacity 137 4.4.1. Example 138 4.4.2. Valid inequalities 140 4.4.3. A sequential procedure for a priori generation of inequalities 145 4.5. Various Arrivals 148 4.5.1. Valid inequalities 150 4.6. Implementation 152 4.7. A Computational Study 153 4.8. Final remarks 156
Chapter 5. Conclusion and Future Work 159
Appendix A. Column Generation 163
Bibliography 167 List of Figures
1.1 Hunter Valley Coal Chain2
1.2 A (simplified) schematic representation of a typical coal export system.5
1.3 Different components are piled beside each other to form a stockpile.6
1.4 Reclaiming direction.6
1.5 P 0 is closer to conv(S) than P 11
1.6 Cutting plain αx ≥ β cuts out optimal solution x∗ 12
2.1 Vessel tonnages 39
2.2 Number of cargoes per vessel 40
2.3 Minimum assembly times for the cargoes of a vessel 40
2.4 Frequency distribution of vessel inter-arrival times 41
2.5 Example of 6-clustering 46
2.6 Number of vessels with delay in each terminal on each day 48
2.7 Number of vessels in queue in each terminal on each day 48
2.8 Tonnage of stockpiles that are being built in each terminal on each day 49
2.9 Tonnage of stockpiles that are being built, held, or reclaimed in each terminal on each day 49
2.10Percentage of stockyard used by the stockpiles that are being built, held, or reclaimed on each day 50
2.11Number of trains unloaded at each terminal on each day 51
3.1 Two ways to assign 8 and 6 train-jobs to c1 and c2 56
3.2 An instance with C not sortable 65
4.1 One demand must be met before its deadline 79
4.2 One demand, D = 9,C = 4, m = 2,T = 4 86
4.3 2-demand, D1 = 4,D2 = 5,C = 3, m = 2,T = 4 139
ix x LIST OF FIGURES
4.4 Various arrivals, D1 = 4,D2 = 5,C = 3, m1 = 1, m2 = 2,T = 4 149 List of Tables
2.1 Model parameters 28 2.2 New model parameters 35 2.3 Terminal characteristics 39 2.4 Minimum total delay incurred by the vessels arriving in the main period 42 2.5 Maximum total delay incurred by the vessels arriving in the main period 42 2.6 Results for different combinations of ∆− and ∆+ 44 2.7 Formulations and strategies results 44 2.8 Average number of vessels and the average delay per vessel 47
3.1 Load point based model parameters 57 3.2 An instance with C not sortable 65 3.3 Component based model results 73 3.4 Load point based model results 73 3.5 Separation 74
4.1 Optimal objective values 154 4.2 Problem size 155 4.3 Violated inequalities 156
xi
Abstract
The Hunter Valley Coal Chain (HVCC) is the largest coal export operation in the world. It concerns the transport of coal from mines located in the Hunter Valley to the Port of Newcastle in New South Wales, Australia. Approximately 1700 coal vessels are loaded at the Port of Newcastle and more than 150 million tonnes of coal is exported each year. The HVCC is a complex system involving 14 producers operating 35 coal mines, 27 coal load points, 2 rail track owners, 4 above rail operators, 3 coal loading terminals with a total of 8 berths, and 9 vessel operators. The coal export supply chain is managed by the Hunter Valley Coal Chain Co- ordinator (HVCCC). One of the most important and far-reaching decision problems faced by HVCCC is the planning of the long-term capacity. The demand for coal continues to grow and thus the export through the Port of Newcastle is expected to increase in the future. Therefore, the infrastructure needs to be upgraded and the capacities in the system expanded. As upgrading infrastructure and expanding the capacity is extremely expensive, a careful and thorough system analysis is crucial to ensure that investments are made in the right place and at the right time. HVCCC uses an elaborate and detailed simulation model of the HVCC to analyse and assess the throughput of the system, to detect and identify any bottlenecks in the system, and to investigate and explore the benefits of infrastructure upgrades and expan- sions. As the simulation model is very detailed, it takes a considerable amount of time to run, and as a consequence, a few scenarios can be analysed. We develop an integer programming based decision support tool that quickly assesses the throughput of a coal export supply chain for a given level of demand. The tool can be used to rapidly evaluate a number of infrastructures for several future demand scenarios in order to identify a few that should to be investigated more thoroughly using the detailed simulation model. To make the natural integer programming model computationally tractable, we exploit problem structure to re- duce the number of variables, and we employ aggregation as well as disaggregation
xiii xiv ABSTRACT to strengthen the linear programming relaxation. We use the tool in a computa- tional study in which we analyse system performance for different levels of demand to identify the potential bottlenecks. Afterwards, we discover a symmetry in the integer programming model which contributes to a computational issue. By aggregating certain variables, we suggest an aggregated (implicit) integer programming model that can break the symmetry. Applying the Halls marriage theorem and the model structure, we provide the re- quired constraints which make the solution of the aggregated model feasible for the original model. As the number of required constraints is considerably large, and most of them are redundant at the optimal solution, a separation algorithm is de- signed to find those constraints that are necessary and add them in the search tree. A computational study proves the advantage of the aggregated integer model when it encounters with challenging scenarios. Inspired by the HVCC problem, we introduce Production Scheduling with Flex- ible Deadline Problem. The problem is to satisfy the customer demand with pro- duction capacity before its deadline which has to be decided (flexible deadline). We analyse its polyhedron and provide all facet-inducing inequalities which together provides a complete formulation for the cases that the there is one customer and the capacity is either constant or varying. We also generalise the problem to in- clude more demands, and to capture varying arrivals, so that it can resemble a sub-problem of the HVCC problem. We provide two classes of valid inequalities for the generalised problems and explain the way to implement those in the HVCC problem. A computational study shows that these inequalities lead to a stronger formulation for the HVCC problem that shows itself with the higher LP relaxation optimal objective values. CHAPTER 1
Introduction
Since the global population is growing, and living standards are improving in developing countries, the demand for energy is increasing. Coal is still consid- ered as a cheap and safe source of energy to satisfy future demand. Over the past 10 years, Australia’s coal exports to countries, such as Japan, China, Ko- rea, India and Taiwan have increased by more than 50 percent. Demand for coal is still expected to increase dramatically over the next 10 years in China and India (http://www.australiancoal.com.au). The black coal industry has been creating significant employment in Australia, and it plays a major role in Australia’s social and economic development. More than 54 percent of the electricity used by business, industry and households in Australia is generated by black coal. Furthermore, 4.7 million tonnes of coal are used by the Australian steel industry each year, and coal is an essential ingredient in making cement. Coal deposits are found in most states of Australia, but are particularly abundant in New South Wales. The Hunter Valley area, located in New South Wales, is a rich source of coal with mostly open cut mines, while the mines in other parts of New South Wales are usually underground mines. The Hunter Valley Coal Chain (HVCC) is the largest coal export operation in the world. This operation includes transporting coal from over 35 mines located in the Hunter Valley area to the port of Newcastle in New South Wales. The Port of Newcastle serves approximately 1700 vessels to export more than 150 million tonnes of coal per year. The HVCC refers to the inland part of the coal chain in the Hunter Valley. The HVCC follows the Hunter River path from mining areas in the Hunter Valley to Newcastle. Most of the coal mines in the Hunter Valley area are open cut mines. When the coal is mined, it is stored either at a coal loading facility shared by several mines, or at a railway siding located at the mine. Whilst some coal is transported by truck to the terminals at the Port of Newcastle, the majority of coal is transported by train to the terminals. Afterwards, the coal is unloaded at the terminal dump stations and stacked as stockpiles on the stockyard. Once the vessel for which the
1 2 1. INTRODUCTION coal is meant arrives at the berth, the coal is loaded onto the vessel. Figure 1.1 gives an overview of the Hunter Valley Coal Chain.
BOGGABRI
WHITEHAVEN TO TAMWORTH GUNNEDAH Gunnedah Curlewis
New GUNNEDAH Werris Creek South WERRIS CREEK ULAN Wales Quirindi NEWCASTLE Willow Tree
Ardglen BICKHAM Murrurundi
Parkville Scone TO BRISBANE
Aberdeen Gloucester DARTBROOK ULAN BENGALLA STRATFORD Ulan ANVIL HILL
TO GULGONG Coggan’s Ck Kerrabee Muswellbrook Sandy Hollow DURALIE WILPINJONG MT ARTHUR St. Helliers DRAYTON Grasstree Ant ene LIDDELL i NEWDELL HUNTER VALLEY MT. OWEN RAVENSWORTH MACQUARIE GENERATION CAMBERWELL UNLOADER WAMBO LOOP RIXS CREEK ASHTON Singleton MTCL 1
MTCL 2 BULGACessnock Maitland KOORAGANG ISLAND AUSTAR PT WARATAH BLOOMFIELD NEWCASTLE TERALBA NEWSTAN Broadmeadow ERARING POWER Morisset STATION UNLOADER
VALES PT. TO SYDNEY UNLOADER
Figure 1.1. Hunter Valley Coal Chain
As mentioned above, the majority of the coal is transported by trains from mines to the terminals. The trains with 6 locomotives and up to 148 wagons with length of up to 2 kilometers can carry about 8,500 tonnes of coal. Employing such a long train is beneficial in terms of costs and service efficiency, but requires high-speed loading and unloading facilities and large storage capacity. The coal transportation is de- pendent on a shared rail system. The railway corridor which is used in the HVCC is part of the Main North railway line. The Hunter Valley rail infrastructure is man- aged by the Australian Government through the Australian Rail Track Corporation (ARTC). Any authorised rail operator can use the track as it is an open access track. The other infrastructures, such as the load points that are associated with the coal transportation are owned by a mine or coal loader. Pacific National and Aurizon are two major above-track (rolling stock) operators which use the Hunter Valley rail track. Both operators use the track to transport coal, agricultural and industrial products, and other freight. Pacific National transports approximately 80% of the coal in the Hunter Valley and Aurizon the remaining 20%. In addition, passenger services are operated by CityRail on parts of the track as the Hunter line. As mentioned above, the main corridor which is a large section of the rail track, is shared with the passenger and freight rail operations. The movement of the trains through the main corridor is restricted by pre-determined time and space paths. Outside the main corridor, the movement of the trains is not as restricted as the 1. INTRODUCTION 3 main corridor. Train paths are divided into up-paths which travel from the load points to the terminals, and down-paths which travel from terminals to the load points. The Kooragang Coal Terminal (KCT) on Kooragang island and the Carrington Coal Terminal (CCT) in the suburb of Carrington are two terminals at the port of Newcastle, located on either side of the Hunter river. The terminals are shared facilities that are owned by 15 different coal mining companies in the Hunter Valley. The coal export facilities at these terminals are operated by Port Waratah Coal Services Limited (PWCS). In each terminal, there are facilities for unloading and storing the coal on the stockyard and reclaiming and loading the coal onto the vessels. The Newcastle Coal Infrastructure Group (NCIG) terminal is a new one that has been established recently to accommodate the export growth. The stockyard at the KCT has four pads, each 2.5 km long and 56 m wide with the capacity of 2.8 million tonnes of coal. The stockyard at the CCT also has four pads, each 1 km long and 40 m wide with the capacity of 0.6 million tonnes of coal. The KCT can load three vessels at the same time with the ship loading capacity of 88 Mtpa. The KCT can only accept coal delivery by rail, while the CCT can accept coal by either rail or road. The CCT has the capacity to load two vessels at the same time with the ship loading capacity of 25 Mtpa. The HVCC is constituted by the mining companies, the track owners, the rail operators, the PWCS, and the NCIG. Coordinating the entire system is necessary to increase the efficiency of the supply chain. The Hunter Valley Coal Chain Coordi- nator (HVCCC) is responsible for providing an integrated planning and scheduling center for the HVCC. Their job is to support infrastructure expansion with a strate- gic modelling, and to schedule activities at the operational level. As the integrated system is very complex, optimisation technology is highly valuable to support the HVCCC in their goal to maximise the efficiency of the system. The HVCCC plans and schedules all movements of coal in the system, but it is not responsible for the plan execution. The “inbound” plan that is created by the HVCCC covers the transportation of coal from the mines to the stockyards including the load points activities, train scheduling, dump stations, and stacker activities. The HVCCC also creates “outbound” plans which cover the movement of coal from the stockyard onto the vessels including the reclaimers and ship loaders activities. The “Live Run” team at PWCS is responsible for executing the plans that are provided by the HVCCC. The inbound plan covers 36 hours of operations 4 1. INTRODUCTION and is ready 24 hours before its execution time. The outbound plan also covers 36 hours of operation and is released 7 hours before it goes into effect. In this thesis, the 2008 data are used for the computational studies. At that time, the NCIG terminal was not built, and two pads of the KCT terminal were only 1.3km long each. Therefore, we study only the CCT and KCT terminals with the capacities of 2.2 million tonnes of coal in KCT and 0.6 million tonnes of coal in CCT. However, the models in this thesis can easily accommodate the NCIG and the KCT terminals at their full capacity.
1.1. The Hunter Valley Coal Chain: An Overview
The HVCC is a pull system in which activities are driven by the customers’ demands, on the contrary to the push system that first produces and then tries to find the customers for the products. In the HVCC, the demand is specified in the form of a shipping stem, i.e., a stream of vessel arrivals. Each vessel arrival, referred to as a trip, is characterised by an arrival time that is a date and time at which the given trip will arrive at the port. The terminal at which the vessel is to be loaded is specified in advance. In addition, a cargo-profile is associated with the trip, which specifies the various brands of coal and their tonnage that constitute the vessel’s cargo. Since coal is a blended product, each brand comes with its tonnage and a brand-recipe which is a combination of mines and their percentage contributions. For example, one recipe for brand X might be 25 percent from mine A and 75 percent from mine B. A component is the material in a cargo that comes from a particular mine, so a cargo made up of 25 000 tonnes from mine A and 75 000 tonnes from mine B, for example, has two components. A complex infrastructure supports and sustains the coal export supply chain in the Hunter Valley. A simplified view of the infrastructure presented in Figure 1.2. Coal moves from the load points at the mines along the rail tracks to the dump stations at the terminals (going through a number of rail junctions), where conveyors take the coal to stackers that assemble the stockpiles on pads in the stockyard. Blended coal is subsequently reclaimed from the pads in the stockyard by reclaimers and transported via conveyors to the ship loaders at the berths to finally be loaded onto waiting vessels. The following list describes the infrastructure components and their character- istics in more details:
• Load point: a facility where trains are loaded. In many cases, there is a one-to-one correspondence between a mine and a load point, but in some 1.1. THE HUNTER VALLEY COAL CHAIN: AN OVERVIEW 5
Figure 1.2. A (simplified) schematic representation of a typical coal export system.
cases a load point serves more than one mine. Each load point is served by trains of a specific size (referred to as the loaded coal tonnage) and using a specific type of wagon. Each load point has a specific load rate per hour. • Junction: a junction connects two rail track segments and is used to model rail track capacity. Specifically, a junction capacity limits the number of trains that can pass through the junction in a day. • Stockyard: the area at a coal terminal where coal is stacked and reclaimed, i.e., where coal blends are assembled. • Pad: a specific area in the stockyard where stockpiles of coal are assembled. Each pad has a length (meters). • Dump station: a facility where trains are unloaded. Each dump station has a dumping rate (tonnes per hour) and a number of working hours per day. When a train is being unloaded at a dump station, the coal is immediately transported to the stockyard by a conveyor belt. • Stacker: a machine that is used to stack coal on a pad in the stockyard of a terminal. Each stacker has a stacking rate (tonnes per hour) and a number of working hours per day. • Reclaimer: a machine that is used to reclaim stockpiles from pads in the stockyard. Each reclaimer has a reclaimer prepare time (hours), a reclaim- ing rate (tonnes per hour), and a number of working hours per day. • Ship loader: a machine that loads coal onto a vessel. Each ship loader has a loading rate (tonnes per hour) and a number of working hours per day. • Berth: a place where a vessel can be loaded. 6 1. INTRODUCTION
As mentioned earlier, coal is a blended product that is assembled in the form of a stockpile on a pad. The purpose of building a stockpile is to store the coal until its vessel arrives and to mix the coal at the time of reclaiming. A stockpile takes the entire width of a pad and a certain amount of its length, which can be derived from the tonnage of the stockpile.
Figure 1.3. Different components are piled beside each other to form a stockpile.