电工技术学报2025，Vol.40Issue(7)：2178-2190,13.DOI:10.19595/j.cnki.1000-6753.tces.240578
基于非对称一致性学习的多类型电动汽车协同参与需求响应方法
Asymmetric Consensus Learning-Based Multi-Type Electric Vehicle Collaborative Participation Demand Response Method
摘要
Abstract
With the widespread adoption of electric vehicles(EVs)and the large-scale integration of renewable energy sources such as wind and solar power into the grid,fully leveraging the potential of EV demand response to address issues such as power fluctuations and poor load stability in the grid is of significant importance.Recently,various control methods for EVs participating in grid demand response have been proposed.However,these existing methods still face several challenges:First,the current methods insufficiently consider the coordination between the autonomous demand response of flexible-contract EVs and the aggregated demand response of fixed-contract EVs.Second,existing optimization methods for aggregated demand response strategies overlook the issue of information asymmetry resulting from differentiated characteristics among entities,leading to slow convergence in aggregate control and higher aggregate output costs.Third,existing optimization methods for autonomous demand response strategies utilize fixed discount rates to guide the learning of agents in flexible EVs but fail to achieve a dynamic balance between immediate rewards and long-term rewards,resulting in poor learning effectiveness.To address these challenges,this paper proposed a multi-type EV collaborative demand response method based on asymmetric consensus learning. Firstly,EVs participating in demand response are divided into flexible-contract EVs and fixed-contract EVs,and a scheduling architecture for multi-type EV collaborative demand response is proposed.Within this framework,the collaboration between flexible-contract EVs and fixed-contract EVs in demand response is manifested in two aspects:1)After flexible-contract EVs autonomously participate in demand response,fixed-contract EVs aggregate demand response based on their autonomous response shortfall,enabling both to jointly meet grid requirements;2)Following the aggregation of demand response by fixed-contract EVs,the aggregated response results are fed back to flexible-contract EVs,prompting them to dynamically adjust their autonomous demand response strategies based on the feedback information. Subsequently,a multi-type EV collaborative demand response strategy based on asymmetric consensus learning is proposed.Specifically,flexible-contract EVs aim to maximize the weighted difference of their income,mileage guarantee,and load curve variance of the power grid.They make autonomous demand response decisions and participate in grid demand response utilizing flexible reinforcement learning.Flexible-contract EVs can dynamically adjust discount rates based on autonomous demand response results,achieving a dynamic balance between immediate rewards and long-term rewards,effectively enhancing EV demand response profits,reducing mileage guarantee costs,and decreasing grid load volatility.Additionally,by fully considering differentiated information such as aggregator contracted capacity,fixed-contract EV quantity,and number of contracts,quantifying the confidence level of aggregator state information,and calculating asymmetric consensus communication weights among aggregators,the asymmetry optimization of fixed-contract EV aggregated demand response strategies is achieved.This enhances convergence speed in EV aggregation control and reduces aggregate output costs.The proposed asymmetric consensus learning algorithm is capable of efficiently handling high-dimensional complex nonlinear relationships,with strong autonomous learning and generalization capabilities. Finally,the effectiveness and rationality of the proposed multi-type EV collaborative demand response method are verified through simulation examples.Simulation results demonstrate that the proposed method can increase autonomous demand response rewards for flexible-contract EVs by 34.42%and improve reward convergence speed by 27.96%.It also enhances fixed-contract EV aggregator incremental cost convergence speed by 36.36%,significantly improving peak shaving and load balancing in the grid while effectively ensuring user comfort.Future research will further explore the impact of real-time pricing incentives on optimizing multi-type EV collaborative demand response.关键词
多类型电动汽车/柔性强化学习/非对称一致性/优化协同/需求响应Key words
Multi-type electric vehicles/flexible reinforcement learning/asymmetric consensus/optimization collaborative/demand response分类
动力与电气工程引用本文复制引用
潘超,汤中卫,廖海君,周振宇..基于非对称一致性学习的多类型电动汽车协同参与需求响应方法[J].电工技术学报,2025,40(7):2178-2190,13.基金项目
国家电网有限公司总部科技项目(52094021N010(5400-202199534A-0-5-ZN))和中国南方电网有限责任公司科技项目(1500002023030103JL00320)资助. （52094021N010(5400-202199534A-0-5-ZN）
