A Modified Evolutionary Multitasking Approach with Efficient Knowledge Transfer for Solving Capacitated Vehicle Routing Problems

In recent years, an evolutionary multitasking framework has emerged within the field of evolutionary computation, exploiting similarities between tasks to share useful knowledge. Real-world vehicle routing problems often involve diverse constraints, including vehicle load capacity. However, the existing mapping relationships only consider customers' geographical locations and neglect capacity requirements, resulting in transferred route solutions that may not satisfy load constraints, thereby reducing transfer effectiveness. In this study, focusing on vehicle routing problems with load constraints, a repair operator is designed for infeasible transferred solutions. Furthermore, an adjusted distance matrix is employed to learn more accurate and realistic knowledge. Moreover, a crowding penalty strategy is adopted for transferred individuals to maintain diversity. To validate the effectiveness of the proposed improvements, empirical studies are conducted on common CVRP benchmarks in multitasking environment.