Surrogate-Accelerated Evolutionary Algorithm for Solving Electric Vehicle Routing Problems

The electric vehicle routing problems (EVRPs) have recently emerged as a focal point within the transportation field, highlighting the need for the simultaneous optimization of customer service orders and vehicle recharging schedules. This dual-optimization requirement poses a significant challenge for traditional evolutionary algorithms (EAs), which often encounter difficulties in achieving fast convergence and high computational efficiency when addressing such complex problems. To address these EVRPs more effectively, this article introduces a surrogate-accelerated EA. At the core of this approach is a Transformer network-based surrogate model, designed to directly predict the total driving distance for customer-exclusive routes. This innovative design bypasses the need for charging optimization processes and the explicit placement of charging stations, thereby significantly streamlining the optimization of recharging schedules within routing plans. Moreover, the incorporation of a masked embedding layer paves the way for the algorithm to adeptly handle multiple EVRP variants. To evaluate the effectiveness of the proposed algorithm, a comprehensive comparison was conducted with state-of-the-art heuristic methods and EAs across various EVRPs. The experimental results demonstrate that the proposed algorithm scales well to different types of EVRPs, offering high computational efficiency.