Summary: | This thesis investigates cost functions for evaluating and optimising the performance of a timetable with mixed train services. Specifically, the performance considered herein includes crowdedness, journey time, punctuality and waiting time. To examine the implications of optimising using these cost functions, a multi-objective optimisation algorithm is developed to derive an optimised timetable for mixed train services. The optimisation algorithm consists of three stages: a Genetic Algorithm (GA) is used to determine the optimal sequence of train runs, followed by Dijkstras shortest path algorithm for determining the optimal schedule based on the sequence determined by GA, and finally an iterative Hill-Climbing procedure for determining the optimal number of train runs in the system. Experiments were carried out on the Brighton Main Line and examined the effect of different timetabling parameters. The first series of experiments showed that the cost of the timetable can be driven down simply through resequencing the trains such that trains exiting the network quickly are more evenly distributed through the time period examined. This occurs since trains exiting early create a buffer which can absorb delays, preventing their propagation. The experiments have also shown that different demand levels influence the number of trains to be scheduled. The optimal number of trains to schedule though relies on the equilibrium between the crowdedness and punctuality cost function. Scheduling additional trains leads to a non-linear reduction in the marginal gains in terms of the crowdedness function while, on the other hand, the cost of punctuality increase exponentially. Finally, we derive the Pareto Frontiers for different combinations of cost functions. This research contributes to the state-of-art of railway system analysis and optimisation.
|