基于极限学习机的变压器故障诊断方法研究  

Study of Transformers Fault Diagnosis Based on Extreme Learning Machine

作　　者：苑津莎[1] 张利伟[1] 王瑜[1] 尚海昆[1] 

机构地区：[1]华北电力大学电气与电子工程学院,河北保定071003

出　　处：《电测与仪表》2013年第12期21-26,共6页Electrical Measurement & Instrumentation

基　　金：中央高校基本科研业务费专项资金资助项目(13XS26);中央高校基本科研业务费专项资助金资助项目(13MS69)

摘　　要：针对基于传统智能学习方法的变压器故障诊断存在训练速度慢、需调整的参数多及参数确定困难的问题,提出了基于极限学习机(Extreme Learning Machine,ELM)的变压器故障诊断方法。文中根据变压器故障的特点选取输入特征向量,分析了激活函数、隐含层节点数目对诊断性能的影响,并与基于BP神经网络和SVM的诊断方法进行了对比。实验结果表明,文中提出的变压器故障诊断方法性能明显优于BP神经网络,与SVM的诊断正确率相当,需要预先设置的参数更少,训练速度更快,更加便于工程应用。Transformer fault diagnosis based on conventional learning methods faces some drawbacks like slow learning speed, trivial tuned parameters and difficult parameter determination. A transformer fault diagnosis based on extreme learning machine（ELM） is proposed in this paper to overcome these drawbacks. Input feature vector is selected according to the characteristic of transformer fault, and then the influence of active functions and hidden layer node number to the diagnosis performance are studied in detail. Comparison between the diagnosis based on BP neural network（BPNN） and SVM are performed, and the experimental results show that, the proposed diagnosis method is better than diagnosis based on BPNN, similar to SVM at correct diagnosis rate but more convenient to engineering application with quicker learning speed and less human tuned parameters.

关 键 词：变压器 故障诊断 极限学习机 ELM神经网络 激活函数 

分 类 号：TM411[电气工程—电器]