With the rapid development of the energy Internet, new power services such as vehicle-network interaction have increasingly stringent requirements on service quality, which brings many challenges to the power terminal access network. To address the problem of resource waste and network performance degradation caused by the overlapping coverage of multiple communication technologies and a wide range of communication methods in terminal access network, a Communication Method Selection algorithm based on Random Forest(RF-CMS) is proposed, which intelligently categorizes a large number of diverse new electric power services and selects the most suitable communication methods for them through Random Forest mode. Then, the Multi-Agent Proximal Policy Optimization(MAPPO) algorithm is employed to dynamically allocate routes for the power services from the viewpoint of traffic loading and communication quality to ensure that various terminal service data (e. g., measurement information, control information) can be transmitted timely and accurately in the access network. The effectiveness of the proposed algorithm is validated by comparing it with the routing algorithm based solely on MAPPO in terms of average end-to-end delay and load balancing degree.