A Genetic Algorithm Solution for Train Scheduling

Train scheduling or re-scheduling problems are popular among researchers, who have interest in the railway planning and operations fields. Deviation from normal operation may cause inter-train conflicts, which have to be detected and resolved timely. Except very few applications, these tasks are usually performed by train dispatchers. However, due to the complexity of re-scheduling problems, dispatchers utilize some simplifying rules to implement their decisions timely. From the system effectiveness and efficiency point of view, their decisions should be questioned. A genetic algorithm (GA) model for conflict resolutions was developed and evaluated against the dispatchers’ and the exact solutions. The comparison measures are the computation time and the total (weighted) delay due to conflict resolutions. Additionally, an artificial neural network (ANN) model was developed to mimic the decision behavior of train dispatchers so as to reproduce their conflict resolutions. The ANN model was trained and tested with data extracted from conflict resolutions in actual train operations in Turkish State Railways. The GA model developed was able to find the optimal solutions for small size problems in short times, and to reach up to twice as better total delay times than those of the ANN (i.e., trains dispatchers) in reasonable times.