Performance Evaluation of a Parallel Ant Colony Optimization for the Real-Time Train Routing Selection Problem in Large Instances

The real-time Train Routing Selection Problem (rtTRSP) is the combinatorial optimization problem of selecting, for each train in a rail network, the best routing alternatives. Solving the rtTRSP aims to limit the search space of train rescheduling problems, highly affected by the number of routing variables. The rtTRSP is modelled as the minimum weight clique problem in an undirected k-partite graph. This problem is NP-hard and the problem size strongly affects the time required to compute optimal solutions. A sequential version of the Ant Colony Optimization (ACO) for the rtTRSP has been proposed in the literature. However, in large instances the algorithm struggles to find high-quality solutions. This paper proposes the performance evaluation of a parallel ACO for large rtTRSP instances. Specifically, we analyze the performance of the algorithm by standard parallel metrics. In addition, we test two parallel local search strategies, which consider different solution neighbourhoods. Computational experiments are performed on the practical case study of Lille Flandres station area in France. The results show a significant speedup of the rtTRSP search space exploration and more promising diversification patterns. Both these improvements enhance the rtTRSP solutions.