基于深度强化学习的住宅区电动汽车充电策略  

A Deep Reinforcement Learning-based Charging Strategy for Electric Vehicles in Residential Areas

作　　者：王保义[1] 祝郑凌 张少敏[1] WANG Baoyi;ZHU Zhenging;ZHANG Shaomin(Department of Computer Engineering,North China Electric Power University,Baoding 071003,China)

机构地区：[1]华北电力大学计算机系,河北保定071003

出　　处：《电力科学与工程》2024年第6期19-27,共9页Electric Power Science and Engineering

基　　金：河北省科技厅科技项目资助(Z2012077)。

摘　　要：为降低电动汽车(Electricvehicle,EV)充电成本、缓解用户里程焦虑,提出了一种基于深度强化学习算法的EV充电控制策略。首先,使用数学模型定量分析EV用户的充电成本、里程焦虑以及电池老化;然后,将EV充放电问题转化为状态转移概率未知的马尔可夫决策过程,使用强化学习算法得到离散的充放电决策。与传统的基于模型预测控制算法相比,在用该策略得到充电策略时可以不依赖充电环境的精确模型以及EV用户行为等不确定性因素的预测结果。实验结果表明,应用该策略可以在满足用户充电需求的同时降低充电成本、保护电池,其控制性能相比基于模型预测控制的充电策略也更为优异。In order to reduce the charging cost of electric vehicle(EV)and alleviate the users’range anxiety,a EV charging control strategy based on deep reinforcement learning algorithm is proposed.Firstly,a mathematical model is used to quantitatively analyse EV users’charging costs,range anxiety and battery aging.Then the EV charging and discharging problem is transformed into a Markov decision-making process with unknown state transition probabilities,and the reinforcement learning algorithm is used to get the discrete charge-discharge decision.Compared with the traditional model-based predictive control(MPC)algorithm,this strategy can obtain charging strategies without relying on accurate models of charging environment and prediction results of uncertain factors such as EV user behaviour.The experimental results show that the proposed strategy can reduce the charge cost and protect the battery while satisfying the charging demand,and the control performance of the proposed strategy is better than that of model predictive control.

关 键 词：电动汽车 住宅区 充放电控制 深度强化学习 

分 类 号：TM73[电气工程—电力系统及自动化]