摘要
提出了一种基于多通道自适应加权神经网络的配网电气拓扑识别算法,构建了多通道一维卷积神经网络(1DCNN)模型,以电压、电流、功率和功率因数4种采集数据作为各通道的输入数据,通过两层叠加的卷积模块实现特征提取;同时,提出了一种自适应加权的特征融合方案,通过神经网络自适应学习各通道重要性特征。实验采用真实用电数据制作数据集,并针对通道数、数据种类、数据维度等参数进行了多组实验。实验结果表明,该算法融合多种用电数据特征,配网电气拓扑辨识准确率达到99.772%。
his paper proposes a distribution network electrical topology identification algorithm based on a multi-channel adaptive weighted neural network.The algorithm builds a multi-channel 1DCNN(one-dimensional convolutional neural network)model,and uses four types of data:voltage,current,power and power factor,to make the datasets.Feature extraction has been realized through two CNN layers stacked;Meanwhile,an adaptive weighted feature fusion is proposed,it can learn the importance of each channel's feature through neural network adaptively.Datasets collect real consumption data,and multiple sets of experiments are conducted with the number of channels,data types,data dimensions and other parameters.Results show that the proposed algorithm can integrate the advantages of multiple data features,the accuracy of electrical topology identification can reach 99.772%.
作者
刘丽娜
王韬
周一飞
程志炯
李方硕
张昱航
徐杰
LIU Lina;WANG Tao;ZHOU Yifei;CHENG Zhijiong;LI Fangshuo;ZHANG Yuhang;XU Jie(State Grid Sichuan Electric Power Corporation Metering Center,Chengdu 610045;School of Information and Communication Engineering,University of Electronic Science and Technology of China,Chengdu 611731)
出处
《电子科技大学学报》
EI
CAS
CSCD
北大核心
2023年第2期247-253,共7页
Journal of University of Electronic Science and Technology of China
关键词
自适应加权
深度学习
配网电气拓扑识别
特征融合
多通道模型
adaptive weight
deep learning
distribution network electrical topology identification
feature fusion
multi-channel model
