基于小波变换与逻辑斯蒂回归的混合式配电变压器故障辨识

Fault Identification of Hybrid Distribution Transformer Based on Wavelet Transform and Logistic Regression

混合式配电变压器(HDT)在智能配电网中能够代替传统配电变压器实现无功功率补偿、谐波治理和电压调节等功能。为了区分当HDT发生故障的时候,是变压器内部故障还是电力电子故障,该文首先通过仿真得到HDT不同故障工况下的大量一次侧、二次侧、三次侧、四次侧电流特征量数据,然后借助小波变换理论,对得到的数据进行四层离散小波变换,从中抽取小波域下数据的归一化能量、归一化能量矩和样本熵作为电流特征量数据的特征值。利用机器学习的方法,构建逻辑斯蒂回归分类器,将特征值组成的特征矩阵作为分类器的输入、训练模型,得到受试者工作特征(ROC)曲线和混淆矩阵表现很好的分类器模型。最后多次随机抽取数据,统计训练得到的机器学习模型的HDT故障识别的准确率均在90%左右。

Hybrid distribution transformer(HDT)can replace traditional distribution transformers in intelligent distribution networks to achieve reactive power compensation,harmonic control,voltage regulation and other functions.In order to distinguish between the internal fault of the transformer and the power electronic fault when the HDT fails,this paper first obtains a large amount of current characteristic data of primary side,secondary side,tertiary side and quaternary side under different fault conditions of HDT through simulation.Then,with the help of wavelet transform theory,four-layer discrete wavelet transform is performed on the obtained data and the normalized energy,the normalized energy moment and sample entropy of the data in the wavelet domain are used as the eigenvalues of the current characteristic data.Using machine learning technology,a logistic regression classifier is constructed,and the feature matrix composed of feature values is used as the input of the classifier,and the model is trained to obtain a classifier model with good receiver operating characteristic(ROC)curve and confusion matrix performance.Finally,the data was randomly extracted many times,and the accuracy of the HDT fault recognition of the machine learning model obtained by training was about 90%.