基于1D-CNN-PSO-SVM的电力变压器故障诊断

Fault Diagnosis of Power Transformer Based on 1D-CNN-PSO-SVM

针对变压器故障诊断过程中人工提取特征泛化性差,诊断正确率低的问题,提出了一种基于一维卷积神经网络(1D-CNN)和粒子群优化支持向量机(PSO-SVM)的故障诊断模型。首先构建一个1D-CNN作为特征提取器,以变压器油中溶解气体原始数据作为输入进行训练,逐层自适应的学习与故障类型相关性更高的深层抽象特征。训练完成后,用分类性能更优的PSO-SVM代替传统1D-CNN中的Softmax分类器实现变压器故障类型的识别。仿真结果表明,经1D-CNN提取特征后,不同故障类型的样本间具有很高的区分度;利用PSO-SVM对提取得到的特征进行分类识别,相比于采用Softmax分类器时,诊断准确率得到了进一步提高,验证了所提方法的有效性。

Aiming at the problem of poor generalization and low diagnostic accuracy of artificial feature extraction in the process of transformer fault diagnosis,a fault diagnosis model based on one-dimensional convolutional neural network(1D-CNN)and particle swarm optimization support vector machine(PSO-SVM)is proposed.Firstly,a 1DCNN was constructed as a feature extractor,the original data of dissolved gas in transformer oil was used as input for training,and the deep abstract features with higher correlation with fault types were adaptively learned layer by layer.After the training was completed,the PSO-SVM with better classification performance was used to replace the Softmax classifier in the traditional 1D-CNN to realize the identification of transformer fault types.The simulation results show that after extracting features by 1D-CNN,the samples of different fault types have high discrimination.Using PSOSVM to classify and recognize the extracted features,compared with using Softmax classifier,the diagnostic accuracy has been further improved,which verifies the effectiveness of the method proposed in this paper.