考虑多元用户行为特征的需求侧管理决策方法  

Demand Side Management Pricing Method Considering Multi-user Behavior Characteristics

作　　者：李思维[1,2] 孔祥玉 刘畅[2] 岳靓 曹胜楠 LI Siwei;KONG Xiangyu;LIU Chang;YUE Liang;CAO Shengnan(Key Laboratory of Smart Grid(Tianjin University),Ministry of Education,Nankai District,Tianjin 300072,China;Beijing Fibrlink Communications Co.,Ltd.,Fengtai District,Beijing 100071,China;Marketing Service Center(Metering Center),State Grid Shandong Electric Power Company,Ji’nan 250000,Shandong Province,China)

机构地区：[1]智能电网教育部重点实验室(天津大学),天津市南开区300072 [2]北京中电飞华通信有限公司,北京市丰台区100071 [3]国网山东省电力公司营销服务中心(计量中心),山东省济南市250000

出　　处：《电网技术》2023年第5期1942-1949,共8页Power System Technology

基　　金：国家电网有限公司科技项目(5100-202114296A-0-0-00)。

摘　　要：需求侧管理可有效实现电力负荷的削峰填谷,提高电力系统的稳定性和运行效率。随着电力物联网的发展,不同用户在参与需求响应过程中的行为差异得以凸显,出于对用户隐私的保护,用户用电信息在采集后往往只能就地利用而不能进一步上传,给多元化负荷行为特征分析带来困难。提出了云边环境下基于A3C(asynchronous advantage actorcritic)强化学习算法和长期短期记忆(long short term memory,LSTM)网络的需求侧管理方法,通过强化学习解决需求侧管理决策中前瞻性不足的问题;通过基于LSTM网络的虚拟环境模拟多元用户行为特征,加速学习过程,降低算法实施成本。通过算例分析可知,所述决策方法在保证用户隐私的同时可有效加快学习进程,价格决策时可更准确地把握用户响应行为特征,从而保证决策的经济性。Demand-side management(DSM)realizes the“peak cutting and valley filling”mode of the power load and improve the stability and efficiency of the power system.With the development of information systems,the behavior differences of the users are highlighted in the process of participating in demand response.For the protection of the user privacy,the user's electricity consumption information can only be used locally after being collected without being further uploaded,which brings difficulties to the analysis of the diversified load behavior characteristics.This paper proposes a distributed DSM pricing method for the service providers based on the asynchronous advantage actor-critic(A3C)algorithm and the long/short-term memory(LSTM)network under the cloud-edge environment.The on-site utilization of the user information is realized through the distributed training and the centralized decision-making structure of the A3C algorithm.The training process is accelerated by the LSTM-based virtual environment,which greatly reduces the training cost of the algorithm.Case study results shows that the proposed method is able to make pricing decisions for the DSM service providers under the cloud-edge environment.Moreover,through the combination of the LSTM-based virtual environment and the A3C algorithm,the proposed method requires less historical data than the other methods and improves the profit of the service providers.

关 键 词：需求侧管理 多元负荷 定价决策 LSTM A3C 

分 类 号：TM73[电气工程—电力系统及自动化]