基于内点策略优化的受约束电动汽车充放电策略

Constrained Electric Vehicle Charging and Discharging Strategy Based on Interior-point Policy Optimization

电动汽车(electric vehicle,EV)作为传统化石燃料动力汽车的环保替代品,其接入电网时可以作为柔性负载或分布式储能单元参与电网调控。当EV参与具有时变电价信号的需求响应时,可以通过优化其充放电时间来降低充电成本。考虑到EV的出行与充电模式存在随机性,亟待解决如何在满足充电需求的前提下确定最优充放电方案。为此,该文将EV充放电决策问题描述为一个约束马尔可夫决策过程,提出一种基于内点策略优化的无模型方法来确定EV充放电最优策略,无需随机性建模的先验知识,直接通过神经网络生成充放电状态时间表。神经网络结构包括一个提取时序电价特征的循环神经网络和一个生成受约束充放电行为的深度神经网络。最后通过算例分析验证了该方法的有效性。

Electric vehicles(EV),acting as the environmentally friendly alternatives to the traditional fossil fuel-powered vehicles,may participate in the grid dispatching as the flexible loads or the distributed energy storage units when they are integrated into the grid.When the EVs participate in the demand response with the time-varying electricity price signals,the charging costs can be reduced by optimizing the EV charging and discharging schedules.Considering the uncertainty of the EV travelling and charging patterns,it is difficult to determine the optimal energy management scheme while satisfying the charging requirements.In order to address this problem,we describe the charging and discharging scheduling problem of the electric vehicles as a Constrained Markov Decision Process(CMDP).To solve the CMDP,this paper proposes an Interior-point Policy Optimization(IPO)based model-free method to determine the optimal policy,which generates a timetable of charging and discharging states directly through a neural network without knowing the detailed modelling and the uncertainties of the EV charging.The network structure of the proposed SRL consists of a recurrent neural network for the extraction of time series electricity price features and a deep neural network for generating the constrained charging and discharging behaviors.Finally,case studies demonstrate the effectiveness of the proposed method.