A deep reinforcement learning-assisted large neighbourhood dynamic robust algorithm for dynamic robust traveling salesman problems
Xia Ji, Chunjian Pan, Moduo Yu, Qingchao Jiang, Qinqin Fan · Journal of Control and Decision · 2025
Solving dynamic traveling salesman problems (DTSPs) is a challenging task due to the constantly changing freight volumes and demands between different city nodes. To alleviate this issue, a deep reinforcement learning-assisted large neighborhood dynamic robust algorithm (LNDRA-DRL) is proposed in the present study. In the LNDRA-DRL, an end-to-end method is used to produce a high-quality initial robust individual, which is then refined using a large neighborhood dynamic robust algorithm to find final robust solutions. To demonstrate the performance of the proposed LNDRA-DRL, five novel dynamic robust traveling salesman problems (DRTSPs) are constructed based on the TSPLIB benchmark test suite, and three competitive algorithms are used for comparison in experiments. Experimental results demonstrate that the proposed LNDRADRL can find a set of satisfactory robust solutions that are nearly optimal in different dynamic environments. Furthermore, it can reduce the switching times of solutions within an acceptable threshold when environmental conditions change.