A hyper-heuristic with deep Q-network for the multi-objective unmanned surface vehicles scheduling problem

This paper proposes a learning -based hyper -heuristic algorithm for the coverage path planning problem of multiple unmanned surface vehicles (USV). The makespan and coverage of the USVs are considered simultaneously. The proposed method does not need to specify the location of the mapping task points in advance; instead, the task allocation and path planning can be realized only based on the parameter information of the USV and the watershed. Considering the uncertainty in the actual mapping process, the triangular fuzzy numbers are used to represent the mapping time and mapping radius of USVs. In order to solve the multi-USV scheduling problem efficiently, this paper proposes an NSGA-II based on dueling double deep Q -network called NSGA-II-DQN. NSGA-II-DQN integrates ten effective global search operators and five local search operators, which can achieve a good balance between exploration and exploitation. Experiment results in various scenarios demonstrate that NSGA-II-DQN can efficiently search for travel paths covering the entire water area for multiple USVs in a short planning time.