文献类型：期刊文献

英文题名：Multi-objective dispatch of integrated renewable power systems leveraging robust optimization in deep reinforcement learning

作者：Ma, Aoqun[1,2,3];Li, Zhi[2,3];Shen, Feifei[3];Peng, Xin[2,3,4];Liu, Yurong[2,3,4];Zhong, Weimin[1,2,3];Qian, Feng[2,3]

机构：[1]East China Univ Sci & Technol, State Key Lab Chem Engn & Low Carbon Technol, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai 200237, Peoples R China;[3]East China Univ Sci & Technol, Sch Informat Sci & Engn, Shanghai 200237, Peoples R China;[4]East China Univ Sci & Technol, State Key Lab Ind Control Technol, Shanghai 200237, Peoples R China

年份：2025

卷号：201

外文期刊名：COMPUTERS & CHEMICAL ENGINEERING

收录：;EI(收录号：20252318548228);WOS:【SCI-EXPANDED(收录号:WOS:001509586300001)】；

基金：This work was supported by the National Key Research and velopment Program of China (2022YFB3304701) , National Science Foundation of China (62303186) , Shanghai Pujiang Program (24PJD021) , Major Science and Technology Project of Xinjiang 2022A01006-4) , Postdoctoral Fellowship Program of CPSF under Number GZC20240469 and the State Key Laboratory of Industrial Control Technology, China (Grant No. ICT2024A23) .

语种：英文

外文关键词：Integrated energy systems; Energy management; DQN method; Robust optimization; Multi-time scale

摘要：The time-varying nature of wind speed and solar radiation introduces significant intermittency and uncertainty to the grid integration of renewable energy. We propose a robust optimization method with a dynamic framework based on reinforcement learning to address this challenge. Our approach considers economic cost and carbon emissions as objective functions in a multi-objective robust optimization model. Day-ahead scheduling results and real-time renewable energy forecasting are used to dynamically adjust the uncertainty set in dispatch, employing support vector clustering and Deep Q-Network. The dynamic framework aims to achieve feasible and cost-effective dispatch by accounting for real-time prediction errors and penalty costs associated with energy spillage and load curtailment risks. A practical industrial system case study shows that the proposed algorithm significantly reduces average day-ahead costs by 21.43% compared to static solutions and exhibits better environmental performance compared to the counterparts.