摘要
为了提高小样本条件下配电网故障辨识准确率,提出一种门控循环注意力网络模型.首先,通过注意力机制赋予故障相中关键周期较高权重,通过加权运算使得模型更加关注上述关键信息.其次,利用门控循环网络处理波形序列,该网络利用门控信号控制记忆传递过程,并借由记忆传递建立序列中不同阶段输入波形和故障类别概率间的关系,从而提升识别准确率.基于仿真数据和实际数据的实验均表明:所提方法在小样本条件下的可靠性和准确率远优于同等条件下支持向量机、梯度提升决策树、卷积神经网络等常用分类模型,为配电网故障辨识技术提供了一种新思路.
To improve fault identification accuracy in power distribution systems,a model named gated recurrent attention network is proposed.First,a higher weight is put on the key cycles of fault phase based on the attention mechanism,making the model focus more on these key messages by weight assignment.Then,the gated recurrent network is adopted,which controls the memory transmission with gate signal and constructs the relationship between input waveform and probability of events at different stages to process the waveform sequence,thereby improving recognition accuracy.Experiments based on both simulation and field data show that the proposed method,under the small-sample-learning condition,is much better than other commonly-used classification models,such as support vector machine,gradient boosting decision tree,and convolutional neural network,providing new insights into fault identification technology in power distribution systems.
作者
陈昊蓝
靳冰莹
刘亚东
钱庆林
王鹏
陈艳霞
于希娟
严英杰
CHEN Haolan;JIN Bingying;LIU Yadong;QIAN Qinglin;WANG Peng;CHEN Yanxia;YU Xijuan;YAN Yingjie(School of Electronic Information and Electric Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;State Grid Qinghai Electric Power Company,Xining 810008,China;State Grid Beijing Electric Power Company,Beijing 100031,China)
出处
《上海交通大学学报》
EI
CAS
CSCD
北大核心
2024年第3期295-303,共9页
Journal of Shanghai Jiaotong University
基金
国家电网有限公司科技项目资助(52020121000C)。
关键词
配电网
故障辨识
注意力机制
门控循环网络
power distribution systems
faults identification
attention mechanism
gated recurrent units(GRU)