基于卷积自编码网络的故障电弧多分类识别方法

Multi-classification recognition method of arc fault based on convolutional autoencoder network

针对非线性负载条件下线路正常工作电流波形与故障电弧波形具有相似特征,容易引起故障电弧保护装置误动作的问题,提出一种基于卷积自编码网络的故障电弧多分类识别方法。采用卷积自动编码器进行故障电弧特征提取,优化设计卷积自动编码器网络模型参数,并利用Softmax多分类器建立故障电弧多分类识别网络模型。实验结果表明,所提方法的故障电弧识别准确率达到99.31%,相应负载类型识别准确率达到97.94%,满足故障电弧识别要求。

Under non-linear load conditions,the current waveform during normal operation has similar characteristics as arc faults.The arc fault protection device is prone to malfunction.A multi-class recognition method of arc fault based on convolutional autoencoder network was proposed.A convolutional autoencoder was used to extract arc fault features,the parameters were optimized,and Softmax multi-classifier was used to build the arc fault multi-classification and recognition network model.Experimental results showed that,the arc fault identification accuracy of the proposed method was 99.31%,the identification accuracy of the corresponding load type reached 97.94%.It met the requirements of arc fault identification.