Ant Colony Optimisation for Dynamic and Dynamic Multi-objective Railway Rescheduling Problems
Recovering the timetable after a delay is essential to the smooth and efficient operation of the railways for both passengers and railway operators. Most current railway rescheduling research concentrates on static problems where all delays are known about in advance. However, due to the unpredictable nature of the railway system, it is possible that further unforeseen incidents could occur while the trains are running to the new rescheduled timetable. This will change the problem, making it a dynamic problem that changes over time. The aim of this work is to investigate the application of ant colony optimisation (ACO) to dynamic and dynamic multiobjective railway rescheduling problems. ACO is a promising approach for dynamic combinatorial optimisation problems as its inbuilt mechanisms allow it to adapt to the new environment while retaining potentially useful information from the previous environment. In addition, ACO is able to handle multi-objective problems by the addition of multiple colonies and/or multiple pheromone and heuristic matrices. The contributions of this work are the development of a junction simulator to model unique dynamic and multi-objective railway rescheduling problems and an investigation into the application of ACO algorithms to solve those problems. A further contribution is the development of a unique two-colony ACO framework to solve the separate problems of platform reallocation and train resequencing at a UK railway station in dynamic delay scenarios. Results showed that ACO can be e ectively applied to the rescheduling of trains in both dynamic and dynamic multi-objective rescheduling problems. In the dynamic junction rescheduling problem ACO outperformed First Come First Served (FCFS), while in the dynamic multi-objective rescheduling problem ACO outperformed FCFS and Non-dominated Sorting Genetic Algorithm II (NSGA-II), a stateof- the-art multi-objective algorithm. When considering platform reallocation and rescheduling in dynamic environments, ACO outperformed Variable Neighbourhood Search (VNS), Tabu Search (TS) and running with no rescheduling algorithm. These results suggest that ACO shows promise for the rescheduling of trains in both dynamic and dynamic multi-objective environments.
- PhD