Genetic programming with surrogate evaluation for the electric vehicle routing problem

Abstract

The focus on environmental sustainability has made the Electric Vehicle Routing Problem (EVRP) an important area of research. Routing Policies (RPs) offer a simple and efficient approach to solving VRPs, providing advantages over methods like metaheuristics by quickly generating solutions. However, designing efficient RPs manually can be time-consuming. Therefore, there is a need to explore hyper-heuristic approaches, particularly Genetic Programming (GP), to automate the design of RPs. However, population-based evolutionary algorithms like GP often require a significant amount of computational resources, especially for fitness calculation. Therefore, surrogate evaluation is essential in enhancing efficiency, especially in GP, where multiple problem instances need to be solved to evaluate each chromosome. In this study, we employ surrogate models within GP to design RPs for EVRP with hard time windows. The experiments show that the RPs designed by GP with surrogate models outperform those produced by standard GP approaches while still requiring less computational time to be generated. Moreover, the RPs designed with GP using surrogate models are also smaller, and consequently, they are also more efficient and easier to interpret.