基于DCAE-SVM的变压器故障诊断

TRANSFORMER FAULT DIAGNOSIS BASED ON DCAE-SVM

为了充分利用变压器油中溶解气体含量的特征信息,提高故障诊断准确率,提出基于深度收缩自编码器(DCAE)与支持向量机(SVM)相结合的故障诊断方法。使用无标签样本数据对收缩自编码器进行逐层训练学习,获取到高层特征表示并确定网络结构参数;通过用有标签样本数据对DCAE-SVM网络进行微调,确定最优网络参数并完成故障诊断。实验结果表明,该混合模型故障诊断精度较高、鲁棒性强。

In order to make full use of the characteristic information of dissolved gas content in transformer oil and improve the accuracy of fault diagnosis,a fault diagnosis method based on deep shrinkage auto-encoder(DCAE)and support vector machine(SVM)is proposed.The labeled sample data was used to train and learn the shrunken self-encoder layer by layer to obtain the high-level feature representation and determine the network structure parameters.The labeled sample data was used to fine-tune the DCAE-SVM network to determine the optimal network parameters and complete the fault diagnosis.The experimental results show that the hybrid model has high accuracy and robustness in fault diagnosis.