文献类型：期刊文献

中文题名：基于数据驱动与深度神经网络的智能配电网拓扑辨识

英文题名：Topology Identification for Smart Distribution Networks Based on Data-Driven and Deep Neural Networks

作者：杜文凯[1];陈心怡[1];薛栋[1]

机构：[1]华东理工大学能源化工过程智能制造教育部重点实验室,上海200237

年份：2025

卷号：51

期号：3

起止页码：353

中文期刊名：华东理工大学学报(自然科学版)

外文期刊名：Journal of East China University of Science and Technology

收录：;北大核心:【北大核心2023】；

基金：国家自然科学基金(62173147);上海市浦江人才计划(20PJ1403000)。

语种：中文

中文关键词：配电网系统;拓扑结构辨识;数据驱动;深度学习;卷积神经网络

外文关键词：distribution network system;topology identification;data driven;deep learning;convolutional neural network

摘要：网架结构的拓扑辨识是配电系统优化与控制的基础。随着风能和太阳能等可再生能源发电的高比例接入,配电网的拓扑结构变得更加复杂且变化频繁,显著增加了拓扑辨识的难度。为了提高拓扑辨识的准确率,本文结合配电网的结构和运行特点,提出了一种基于自组织映射(SOM)和卷积神经网络(CNN)深度学习框架的配电网拓扑辨识方法。考虑到配电网数据的高维特性,该方法首先利用主成分分析(PCA)对高维电压幅值和有功功率数据进行降维,进而使用SOM提取数据特征,将其转换为二维特征图,并通过CNN学习输入特征与拓扑标签之间的映射关系,从而实现配电网拓扑结构的精准辨识。通过在33、69、123节点配电网算例上进行仿真实验,验证了所提方法的有效性,并且相较于其他方法,该方法在辨识准确率和鲁棒性等性能上具有明显优势。
Topology identification of the network structure is fundamental for the optimization and control of distribution systems.With the high penetration of renewable energy generation,such as wind and solar power,the topology of distribution networks has become more complex and changes frequently,significantly increasing the difficulty of topology identification.To improve the accuracy of topology identification,this paper proposes a distribution network topology identification method based on a deep learning framework that combines Self-Organizing Maps(SOM)and Convolutional Neural Networks(CNN),taking into account the structure and operational characteristics of distribution networks.This method first uses Principal Component Analysis(PCA)to reduce the dimensionality of high-dimensional voltage magnitude and active power data.It then employs SOM to extract data features and transform them into a two-dimensional feature map.Finally,CNN is used to learn the mapping between the input features and topology labels,enabling accurate identification of the distribution network topology.The effectiveness of the proposed method is validated through simulation experiments on 33-,69-,and 123-bus distribution network cases.Compared to other methods,this approach demonstrates significant advantages in terms of identification accuracy and robustness.