基于关系图卷积神经网络的新能源配电台区拓扑识别方法OA北大核心CSTPCD

A proposed topology identification method,based on a relational-graph convolutional network,addresses the challenges faced by traditional methods in adapting to the complex electrical coupling characteristics of low-volt-age distribution substation areas with a high proportion of distributed photovoltaic(PV)integration.This paper an-alyzes the influence mechanism of distributed PV integration on the identification of line-user relationships in low-voltage substation areas,and presents an adaptive method for identifying line-user relationships in distribution sub-station areas with high-penetration distributed PV integration.This method achieves line-user relationship identifi-cation through voltage Pearson correlation coefficient matrix modeling and global adaptive clustering.Considering the topology association characteristics of distribution substation areas with high renewables,the paper classifies and matches distribution network node associations into separated,hierarchical,parallel,and PV node acceptance rela-tionships.An adjacent matrix model of substation area topology is established which is specifically adapted to dis-tributed PV integration.This paper proposes a distribution substation area topology generation algorithm based on a relational-graph convolutional network.By extracting voltage measurement data to create node feature portraits for the substation area and employing graph link prediction to uncover potential node association relationships,the dis-tribution substation area topology is progressively generated.A case simulation comparison validates the effective-ness of the proposed topology identification method,which improves identification accuracy by more than 4.3％when compared to traditional algorithms.