Optimal Self-Consumption Scheduling of Highway Electric Vehicle Charging Station Based on Multi-Agent Deep Reinforcement Learning

Abstract

Due to the randomness of renewable energy and electric vehicles (EVs) in highway charging stations, it is difficult to ensure the consistency of renewable energy supply and EVs demand. Considering the randomness of EVs charging and renewable energy power generation, an optimal self-consumption scheduling of a highway EV charging station based on multi-agent deep reinforcement learning (MADRL) is proposed to realize the economy, self-consumption, low-carbon operation and ensure reliability of power supply. In day-ahead, the traffic flow prediction model based on the CNN-BiLSTM and the queuing model based on user psychology are built to predict the charging load. The 24-hour optimal charging price is obtained by solving the incentive price optimization model and guides the orderly charging of EVs. In intra-day, considering the prediction errors of day-ahead and the diversity of charging levels, an optimal scheduling based on the MADRL is proposed. Regarding the multi-objective scheduling of the highway charging station, the multi-objective nonlinear and non-convex problem is transformed into multi-agent Markov game model. Finally, the effectiveness and optimality of the proposed method are verified on a highway charging station The results show that the proposed method can realize the economy, self-consumption and low-carbon operation of the charging station.