摘要
针对配电网线损异常检测中因数据非平稳、多变量耦合及异常模式隐蔽导致的检测精度不足问题,本文提出了一种融合变分模态分解(VMD)、维度注意力机制长短期记忆网络(DAMLSTM)与卷积自动编码器的混合诊断框架。实验基于某市低压配电网全年实际数据,结果表明:所提方法在RMSE指标上较SVR、RF、GRU、LSTM及EMD-LSTM等主流模型最大降低49.3%,R2系数最高提升30.1%,异常检测误差范围控制在0%~4%以内。该研究为配电网线损管理提供了高精度、可解释的智能诊断方案,对提升电网经济性与安全性具有应用价值。
Aiming at the problem of insufficient detection accuracy caused by non-stationary data,multi-variable coupling and hidden abnormal modes in abnormal detection of line loss in distribution network,this paper proposes a hybrid diagnosis framework that combines variational mode decomposition(VMD),dimensional attention mecha⁃nism long short-term memory network(DAMLSTM)and convolutional autoencoder.The experiment is based on the actual data of a city's low-voltage distribution network throughout the year.The results show that the RMSE in⁃dex of the proposed method is 49.3%lower than that of the mainstream models such as SVR,RF,GRU,LSTM and EMD-LSTM,and the R2 coefficient is increased by 30.1%.The range of anomaly detection error is controlled within 0%~4%.The research provides a high-precision and interpretable intelligent diagnosis scheme for line loss management of distribution network,which has application value for improving the economy and security of power grid.
作者
郭雷
李晓飞
张焕
郭鹏
胡美琳
GUO Lei;LI Xiaofei;ZHANG Huan;GUO Peng;HU Meilin(Beijing Guodentsu Network Technology Co.,Ltd.,Beijing 100085,China;Beijing Tianyi Digital Technology Co.,Ltd.,Beijing 100070,China)
出处
《粘接》
2025年第10期246-250,共5页
Adhesion
关键词
配电网线损
异常检测
变分模态分解
维度注意力机制
长短期记忆网络
distribution network line loss
anomaly detection
variational modal decomposition
dimension atten-tion mechanism
long short-term memory network
