基于多智能体深度强化学习的智能网联汽车服务迁移优化方法OA

To address the challenges of multi-user resource competition and dynamic changes in edge node availability faced by intelligent connected vehicles during service migration in a highly dynamic Internet of vehicles environment,a service migration method based on multi-agent group relative policy optimization(MAGRPO)was proposed.The ser-vice migration problem was formalized as a long-term multi-user joint optimization problem with resource constraints,and a MAGRPO algorithm that did not require an explicit critic network was designed.A policy update signal was con-structed based on the relative ranking of discounted returns within the group,thereby effectively mitigating training insta-bility caused by severe penalties(e.g.,node overload or failure)and reducing training cost.Simulation results show that the proposed method outperforms existing baseline methods in key metrics such as total service delay,migration energy consumption,and migration success rate.It exhibits stronger robustness and scalability in scenarios where edge node re-sources are limited and their availability changes dynamically.