基于最优正则极限学习机的变压器故障诊断

针对基于极限学习机的变压器故障诊断模型隐含层神经元个数较多时,存在过拟合、稳定性差以及精确度不高的问题,提出了一种最优正则极限学习机的变压器故障诊断方法。方法收集了变压器油中溶解气体作为故障指标,将采集的数据集合随机分成训练集、验证集和测试集。首先通过训练集对极限学习机故障诊断模型进行训练;其次,将验证集输入已构建的模型,利用验证集精度与训练精度的差值进行反馈,并引入最优正则系数对模型参数进行惩罚性调整;最后,利用更新后模型对测试集进行故障诊断。通过算例分析与比较可以得出,最优正则系数极限学习机比极限学习机稳定性强,精确度高,并且方法简单,计算速度快,可有效实现变压器故障诊断。

Transformer fault diagnosis based on optimal regular learning machine

In order to solve the problem of over-fitting, unstable stability and low precision for transformer fault diagnosis model based on extreme learning machine while the number of hiddenneurons in extreme learning machine is relatively large , a transformer fault diagnosis method based onoptimal regular limit learning machine was proposed. Data of dissolved gas in transformer oil wascollected as failure indexes, and the data sets were randomly divided into training set, verification setand test set. Firstly, the extreme learning machine fault diagnosis model was trained by the trainingset. Secondly, the verification set was input into the already constructed model, then the output ofverification set was compared with the training output error for the purpose of over-fitting judgment.When the model is over-fitting, the optimal regular coefficient was introduced to punish the modelparameters. Finally, the adjusted model was tested by using test data set. Comparing with the extremelearning machine through the calculation and analysis of the example, the optimal regular coefficientextreme learning machine is more stable and accurate , and the method is simple and the calculationspeed is fast. It can effectively realize transformer fault diagnosis.