基于数据特征增强和残差收缩网络的变压器故障识别方法

Identification Method of Transformer Fault Based on Data Feature Enhancement and Residual Shrinkage Network

为增强深度残差收缩网络对变压器故障特征的学习能力从而提高故障识别精度,文中研究构建了故障特征气体向量配合改进的深度残差收缩网络来识别变压器故障。首先,构建可变软阈值函数消除恒定偏差的影响,利用快速回溯算法加快阈值确定速度的同时确保输出结果的完整性。然后,提出带可变权重的交叉熵函数降低误识别对网络精度的影响,并将构建的特征气体向量作为网络输入,保证网络学习并识别更多故障因素的特征。最后,以过热故障和电弧放电故障为样本的实验结果验证了该方法的有效性。与传统方法相比,所提方法的识别精度高,而且适用于电力系统多特征故障识别。

In order to enhance the ability of the deep residual shrinkage network to learn the features of transformer faults and improve the accuracy of fault identification,the fault feature gas vector with an improved deep residual shrinkage network is constructed to identify transformer faults.First,a variable soft threshold function is constructed to eliminate the effects of constant deviations,and the fast back tracking algorithm is used to speed up the threshold determination and ensure the integrity of the output results.Then,a cross-entropy function with variable weights is proposed to reduce the effects of misrecognition on the network accuracy.The constructed feature gas vector is used as the input to ensure that the network learns and recognizes the features of more fault factors.Finally,taking overheating faults and arcing faults as samples,the experimental results verify the effectiveness of the method.Compared with the traditional method,the recognition accuracy of the proposed method is higher,and it is suitable for the identification of multi-feature faults in the power system.