A Selection Hyper-heuristic based on Q-learning for School Bus Routing Problem

School bus routing problem (SBRP) has been studied for decades. Many successful approaches based on heuristics or metaheuristics have been developed for various SBRP problems. However, developing an effective algorithm for SBRP is still a very challenging task. This paper developed a Q-learning-based selection hyper-heuristic to solve basic and open single-school SBRP problems, which both aim to minimize the total travel distance. The proposed algorithm took a Q-learning algorithm as the high-level strategy to select a low-level heuristic from a set of low-level heuristics, which are dependent on the problem domain. The selected low-level heuristic was regarded as an action and then executed to improve the current solution. In each stage of the optimization process, the best action with the best cumulative rewards will be chosen to get better results. The presented algorithm was implemented and some experiments were carried out on some Capacitated vehicle routing problem (CVRP) instances and SBRP benchmark instances. Experiment results on two types of instances demonstrate that our proposed hyper-heuristic algorithm is more competition than existing approaches.