期刊
APPLIED ENERGY
卷 346, 期 -, 页码 -出版社
ELSEVIER SCI LTD
DOI: 10.1016/j.apenergy.2023.121334
关键词
Electric vehicle (EV); Fast charging station (FCST); Dynamic pricing; User satisfaction; deep reinforcement learning (DRL)
With the rapid development of electric vehicles (EVs) and charging infrastructures, the unbalanced utilization rate of fast charging stations (FCSTs) and the long waiting time for charging have become major issues. This paper uses deep reinforcement learning (DRL) to solve these problems by proposing a dynamic pricing strategy for FCSTs. The strategy includes a traffic flow prediction model, an estimation of EV charging requirements, and a dynamic pricing strategy based on deep deterministic policy gradient (DDPG) learning. Simulation results show that the proposed strategy effectively improves the profit of FCSTs, alleviates road congestion, and enhances user satisfaction.
With the rapid development of electric vehicles (EVs) and charging infrastructures, the unbalanced utilization rate of fast charging stations (FCSTs) and the long waiting time for charging have aroused considerable attention. The incurred low operation profit of FCSTs and low satisfaction of EVs impose difficulties on the further development of EV industry. Existing literature ignored the influence of real-time charging price changes on traffic flow variation and EV charging determination during the dynamic price regulating process. This paper focuses on solving these crucial issues in the dynamic pricing for FCSTs with deep reinforcement learning (DRL). Firstly, considering the spatial-temporal interactions of different roads, a traffic flow prediction model is proposed based on the LSTM combined with the GNN-FiLM. Then, the Origin-Destination (OD) estimation is used to estimate the charging requirements of EVs based on the predicted traffic flow, and a charging demand prediction method for FCSTs is developed by converting the EV satisfaction into economic costs with different dimensions. Then, the vehicle-road learning environment is built with the Markov decision process (MDP), and a dynamic pricing strategy based on the Deep Deterministic Policy Gradient (DDPG) learning is proposed to achieve the optimal charging prices of FCSTs with maximum operation profit. Moreover, during the learning process, the real-time charging price is renewed based on the predicted charging demand, and the future charging demand is further predicted under the renewed charging price until the optimal price is achieved. Finally, simulation results validate that the proposed dynamic pricing strategy effectively improves the profit of FCSTs, alleviates the road congestion, and improves the users' satisfaction.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据