A Scalable and Coordinated Energy Management for Electric Vehicles Based on Multiagent Reinforcement Learning Method

Abstract

The electric vehicle (EV) has been popular in recent years, which also brings huge challenges to the distribution network due to its energy instability. In order to consider the economic factors of dispatching these distributed renewable resources, the voltage variation is also important. A novel model-free method is put forward for collaborative management of EV resources of aggregators in the distribution network. The economic costs and physical network constraints for this energy management issue are considered at the same time. A Multiagent Deep Deterministic Policy Gradient (MADDPG) algorithm is applied to learn the cooperative energy control strategies. A transfer learning technique is used to fine-tune the trained policy when more aggregators join in the network. The proposed method can achieve close results to the traditional optimization methods, while it takes less than one second to take control actions, making it is more suitable for real-time online energy management. Compared to other advanced reinforcement learning (RL) models, numerical simulations conducted on IEEE test cases greatly illustrate the effectiveness and superiority of the proposed method.