Solving vehicle routing problem by memetic search with evolutionary multitasking

Vehicle routing problem (VRP) is a well-known NP-hard combinational optimization problem. In the literature, existing approaches can be generally classified into two categories: exact methods and metaheuristics methods. The former is only effective for small problem instances while the latter is more suitable for practical applications with larger scale. However, these methods perform evolutionary search only on a single problem form of VRP, while the efficient problem form of a given VRP is often unknown before hand. Keeping the above in mind, this paper proposes memetic search with evolutionary multitasking for solving vehicle routing problem on multiple problem forms. In particular, multiple simplified vehicle routing problems are firstly constructed and acted as the auxiliary tasks of original VRP. The memetic evolutionary search is thus simultaneously performed on simplified vehicle routing problems and original VRP. In this way, the useful routing traits can be shared by knowledge transfer among the tasks to enhance the evolutionary search. To verify the performance of the proposed method, comprehensive empirical experiments have been conducted on commonly used VRP benchmarks.