Schedule-based modeling approach to the planning and design of railway networks

Abstract

Two approaches to modeling the planning and design of railway networks, frequency-based and schedule-based, are commonly adopted. The schedule-based modeling approach is more flexible and more accurately describes the railway operation and passenger behaviors. This thesis aims to develop a schedule-based model for application in more realistic and complicated situations. The thesis first focuses on integrating train timetabling and passenger assignment. In the model developed at this stage, passenger demand is elastic and passengers choose only their optimal itineraries. The train timetabling determines not only the departure and arrival times, but also the tracks used by the trains at each passing station. Three solution methods for this model are presented and compared in a computational efficiency test. The iterative method with decomposition performs the best with large-scale problems. Line planning, i.e., deciding the number of scheduled trains and the stopping patterns, is then integrated into the model. The uncertainty of passenger itinerary choice is also included. The problem is formulated as the minimization of passenger itinerary costs and solved using the iterative method with decomposition. In the iterative method with decomposition, passenger uncertainty is handled by a new itinerary choice model, the mixed itinerary-size weibit model, which addresses the independently and identically distributed assumptions and heterogeneity in passengers’ perceptions of itineraries. A hypothetical network is used to show the capability of the model and method. Finally, the integrated model is extended by adding rolling stock scheduling, which accounts for the various vehicle types and the inventory resource problem. The extended model has two objectives, minimizing the operators’ costs and minimizing the passengers’ costs, and is resolved by an ε-constraint method that adopts the iterative method with decomposition to find solutions to its generated subproblems. The rolling stock scheduling and the Pareto frontier outputted by the ε-constraint method are presented and discussed in a hypothetical example. This thesis also applies the above models to problems with the planning and design of southern China’s high-speed railway network, using collected train data and assumed passenger data. Reasonable results are found and interpreted, indicating the workability of the proposed models and corresponding solution methods.