Iterated greedy algorithms for a real-world cyclic train scheduling problem

In this paper, we develop heuristic algorithms for a complex locomotive scheduling problem in freight transport that arises at Deutsche Bahn AG. While for small instances an approach based on an ILP formulation and its solution by a commercial ILP solver was rather successful, it was found that effective heuristic algorithms are needed for providing better initial upper bounds and for tackling large instances. The main contribution of this paper is the development of heuristic algorithms that strongly improve over the performance of the greedy algorithm used in the previous research efforts. The development process was done on a step-by-step basis ranging from improvements over the initial greedy construction heuristic, the development of a simple local search algorithm, the further extension to an iterated greedy procedure to the adoption of population-based stochastic local search methods. Our computational results show that the iterated greedy algorithm combined with a simple local search is a powerful algorithm for this real-world freight train scheduling problem.