报告主题 ： Data - driven NonparametricOperation of Distribution Systems with Uncertainties
报 告 人 ： Prof . Hongcai Zhang
会议时间 : 4月16日（周六）16:00
会议地点 : 腾讯会议953296275
主办单位 : 重庆大学、输配电装备及系统安全与新技术国家重点实验室、重庆大学溧阳智慧城市研究院
协办单位 : 四川大学、电子科技大学、西南交通大学、成都理工大学、成都中医药大学、四川师范大学、西华大学、西南科技大学、西南大学、重庆邮电大学、重庆科技学院
Personal Profile:
Hongeai Zhang received the B . S . and Ph . D . degree in electrical engineering from Tsinghua University , Beijing , China , in 2013 and 2018, respectively . In 2018-2019, he was a postdoctoral scholar with the University of Califomia , Berkeley , and also a research affiliate with the Lawrence Berkeley National Laboratory , Berkeley , California , USA . Since 2019, he joined the State Key Laboratory of Internet of Things for Smart City of University of Macau , Macau , China , where he is currcntly an Assistant Profcssor in Smart Encrgy . His rcscarch intercsts include intcgratcd encrgy systems , Internet of Things for energy systems , transportation electrification , ete . Hongcai Zhang has published over 40 SCl - indexed journal papers , including 2 ESI highly cited papers . He received the “ Excelent Paper Award ” in EVS34 and “ Best Paper Award ” in iSPEC 2021. He is currently an associate editor of ET Electrical Systems in Transportation , a member of China Electrotechnical Society Young Scholar committee and Secretary - Gcneral of EEE PES China Encrgy and Transportation Nexus Subcommittec .
Abstract:
The opcration of distribution systems is one of the fundamcntal problems in modern powcr systems . The statc - of - the - art strategies are usually based on optimal power flow models that require accurate network parameters ( e . g , the network topology and branch impedances ). However , these parameters are often unavailable in many distribution networks . To address this issue , this alk iniroduces a data - driven nonparametric method for disiribution system operations . It lrains multi - layer percepiron ( MLP ) with ReLU activation functions based on historical metering data to learn power flow models , in which the requirement of network pa Г ameters can be bypassed . The trained MLP can be furthcr cquivalently reformulated into mixed - intcger linear constraints . As a result , the strategic opcration problem of distribution system based on data - driven ( black - box ) MLP models can be effectively solved by the Branch - and - Bound algorithm in off - the - shelf solvers . Considering that distribution neayork operation usualy faces significant uncertainties ( e . g ., renewable generation ), we further extend the proposed method tochan constrained proramming . In this extension , a data - driven quantile regresS1on neural network 1S trained to describe stochas power flow models . Similarly , it can be cquivalently reformulated into tractable mixed - intcgcr lincar constraints . experiments validate the effectiveness of the proposed nonparametric method .
