Efficient Multi-Objective Evolutionary Algorithms for Railway Train Platforming Rescheduling Under Arrival Delays

This paper studies the railway train platforming rescheduling problem (TPRP). Rescheduling train arrival/departure time and routes are conducted when a train arrives with delays. A bi-objective optimization problem is formulated to minimize the total train delays and rescheduling costs. Multi-objective evolutionary algorithms (MOEAs) are applied to solve the problem. An integer-permutation-based encoding method is developed to represent the platform track assignment and train departure priority at the station. The arrival and departure times of the trains are obtained through a decoding method, which minimizes train delays and handles the constraints. Based on the above encoding and decoding methods, three state-of-the-art MOEAs are adopted, that is, nondominated sorting genetic algorithm-II (NSGA-II), MOEA based on decomposition (MOEA/D), and multi-objective particle swarm optimization with crowding distance and elitist learning strategy (MOPSO-CD-ELS). The experimental results show the effectiveness of the encoding and decoding methods in obtaining the true Pareto front (PF). Among the compared algorithms, NSGA-II and MOPSO-CD-ELS outperform MOEA/D, nearly covering the entire true PF.