Resumen de Railway traffic optimization
José Luis Espinosa Aranda
Currently railway systems are one of the most important means of transport used by passengers or for freight transportation in developed and developing countries. This activity has grown in importance and complexity in urban and inter-urban contexts because of the increasing population and the continuous change of demand patterns. One of the problems that must be solved in a railway system is the scheduling of the trains. This problem is known as the timetable setting problem and can be tackled following two main approaches: i) an on-line approach, which is focused on restoring the schedule when a disruption occurs in the system in real-time using rescheduling techniques, and ii) an off-line approach focused on the developing of a timetable from scratch or from an updated timetable. This thesis deals with the train scheduling problem, proposing two new approaches for the on-line and off-line scheduling of a railway system. With respect to the on-line context an Intelligent Transportation System composed of a conflict detection and a conflict resolution module is proposed. The conflict detection module is defined as a discrete event simulation model which represents the behaviour of the trains considering the planned schedule. The conflict resolution module models the railway system by means of the alternative graph concept and implements the First Come First Served and Avoid Most Critical Completion algorithms of the literature and a novel approach to solving this problem based on demand. Computational experiments for the real case study of the RENFE Cercanías Madrid network have been performed in order to test this methodology. In the off-line context, a mesoscopic model is proposed for High-Speed Rail systems, following a nominal approach and attempting to maximize the benefits of the company. This model represents the supply of the railway network as a discrete event simulation model, and the behaviour of passengers based on a random utility framework, which leads us to propose a novel constrained nested logit model, CMNL, which considers that a passenger will select a service depending on the timetable, price, travel time and seat availability. CMNL adopts the entropy-maximizing approach to include non-linear constraints for a decision-maker in its formulation. We propose the use of Reproducing Kernel Hilbert spaces to represent the utilities. This issue is essential to dynamic choice modelling such as departure time modelling. The calibration of CMNL is also discussed. Due to its high computational cost, the complete model is solved using various metaheuristic methods, proposing a hybridization approach which combines the advantages of the basic algorithms. Computational experiments for the real case study of the Madrid-Seville corridor have been developed to test this methodology.
