文献类型：期刊文献

英文题名：A home energy management approach using decoupling value and policy in reinforcement learning

作者：Xiong, Luolin[1];Tang, Yang[1];Liu, Chensheng[1];Mao, Shuai[2];Meng, Ke[3];Dong, Zhaoyang[4];Qian, Feng[1]

机构：[1]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai 200237, Peoples R China;[2]Nantong Univ, Dept Elect Engn, Nantong 226019, Peoples R China;[3]Univ New South Wales, Sch Elect Engn & Telecommun, Sydney, NSW 2052, Australia;[4]Nanyang Technol Univ, Sch Elect & Elect Engn, Singapore 639798, Singapore

年份：2023

卷号：24

期号：9

起止页码：1261

外文期刊名：FRONTIERS OF INFORMATION TECHNOLOGY & ELECTRONIC ENGINEERING

收录：;EI(收录号：20233314539482);WOS:【SCI-EXPANDED(收录号:WOS:001045580300001)】；

基金：& nbsp;Project supported by the National Natural Science Foundation of China (Nos. 62293502, 62293500, 62293504, 62073138, and 62173147), the Fundamental Research Funds for the Central Universities, China (No. 222202317006), and the Nanyang Techno-logical University Startup Grant and MOE Tier 1 (No. RG59/22)& nbsp;

语种：英文

外文关键词：Home energy system; Electric vehicle; Reinforcement learning; Generalization; TP181

摘要：Considering the popularity of electric vehicles and the flexibility of household appliances, it is feasible to dispatch energy in home energy systems under dynamic electricity prices to optimize electricity cost and comfort residents. In this paper, a novel home energy management (HEM) approach is proposed based on a data-driven deep reinforcement learning method. First, to reveal the multiple uncertain factors affecting the charging behavior of electric vehicles (EVs), an improved mathematical model integrating driver's experience, unexpected events, and traffic conditions is introduced to describe the dynamic energy demand of EVs in home energy systems. Second, a decoupled advantage actor-critic (DA2C) algorithm is presented to enhance the energy optimization performance by alleviating the overfitting problem caused by the shared policy and value networks. Furthermore, separate networks for the policy and value functions ensure the generalization of the proposed method in unseen scenarios. Finally, comprehensive experiments are carried out to compare the proposed approach with existing methods, and the results show that the proposed method can optimize electricity cost and consider the residential comfort level in different scenarios.