Multi-agent reinforcement learning-driven genetic algorithm for solving the soft time window electric vehicle routing problem
HAN Yuyan
AN Junyu
YANG Xiaoyu
WANG Yuting
LI Huan
TIAN Xinru
Abstract：For the path optimization problem considering constraints such as the energy consumption of e-lectric vehicles,order time windows,and vehicle loads,a mixed-integer programming model with the min-imization of path cost as the core optimization objective is first constructed.Then,based on the problem characteristics,a multi-agent reinforcement learning-driven genetic algorithm(MRLGA)is proposed for solving it.In the MRLGA,the multi-agent reinforcement learning method is used to dynamically optimize the mutation and crossover probabilities in the genetic algorithm and intelligently proxy the selection and crossover operations to enhance the algorithm's search efficiency.By introducing the Shannon diversity in-dex to measure population diversity,premature convergence is avoided,and effective maintenance of popu-lation diversity is achieved.The 2-opt search algorithm is adopted to enhance local search capabilities,and the discrete Lévy flight strategy is introduced to improve global search capabilities,achieving efficient ve-hicle path planning.Finally,experiments are conducted with 120 test cases,and the results show that the proposed MRLGA algorithm can effectively reduce path costs under complex constraints,verifying the ef-fectiveness and feasibility of the algorithm.
Keywords：vehicle routing problemgenetic algorithmshannon diversity indexreinforcement learning
Publication Date：2026-02-25
Online Publishing Date：2026-08-28(First online date of this platform, not the publication date of the document)
Pages：12( 32-43 )