基于SMOTE与Bayes优化的LSTM网络变压器故障诊断  

Fault Diagnosis of LSTM Network Tansformer Based on SMOTE and Bayes Optimization

作　　者：张宏杰 陈贵凤 闫宏伟 杨晓龙 侯天仁 张伟[3] ZHANG Hongjie;CHEN Guifeng;YAN Hongwei;YANG Xiaolong;HOU Tianren;ZHANG Wei(State Grid Ningxia Electric Power Co.,Ltd.,Yinchuan 750001,China;Beijing Kedong Electric Power Control System Co.,Ltd.,Beijing 100192,China;North China Electric Power University,Beijing 102206,China)

机构地区：[1]国网宁夏电力有限公司,宁夏银川750001 [2]北京科东电力控制系统有限责任公司,北京100192 [3]华北电力大学,北京102206

出　　处：《中国电力》2023年第10期164-170,共7页Electric Power

基　　金：国家自然科学基金资助项目(51877061)。

摘　　要：随着电力信息化的提高,智能算法结合历史数据进行变压器故障诊断的方法越来越受到关注。在溶解气体分析法基础上借助少数类样本过采样(SMOTE)算法合成新样本,实现样本多维度扩充,并以贝叶斯优化算法寻找长短期记忆(LSTM)网络模型参数的最优设置值,以降低训练集错误率,进而建立了变压器故障诊断模型。结果表明:样本扩充后的变压器故障诊断模型过拟合度降低约20%,测试集准确率提升约10%。With the improvement of power informatization,the method of transformer fault diagnosis based on intelligent algorithm and historical data has been paid more and more attention.On the basis of dissolved gas analysis,synthetic minority oversampling technique(SMOTE)algorithm was used to synthesize new samples,realize multi-dimensional expansion of samples,and use Bayes optimization algorithm to find the best setting value of long short term memory(LSTM)network model parameters to reduce the error rate of training set,and then establish transformer fault diagnosis model.The results show that the overfitting degree of the transformer fault diagnosis model after sample expansion is reduced by about 20%,and the accuracy of the test set is increased by about 10%.

关 键 词：变压器 故障诊断 采样 长短时记忆网络 

分 类 号：TM41[电气工程—电器] TP18[自动化与计算机技术—控制理论与控制工程]