基于小波时频图和AlexNet网络的电弧故障诊断方法

Fault diagnosis method of series arc fault based on wavelet time-frequency diagram and AlexNet

针对故障电弧电流频域特征难以提取,检测率不高的问题,提出一种基于小波时频图和深度学习卷积神经网络的低压交流故障电弧诊断方法。该方法对串联型故障电弧电流信号进行小波时频变换,利用小波对高频突变信号的放大能力,捕捉不同负载的差异特征,得到的时频图经过压缩统一至适当大小后作为AlexNet卷积神经网络的输入特征图,利用AlexNet网络提取小波时频图特征,通过全连接层拟合数据实现故障电弧的智能诊断。建立故障电弧实验平台,分别采集阻性负载、感性负载、阻感负载下的电弧电流2 000组,利用该数据集完成模型训练。选取新负载及10~60 dB信噪比研究模型的泛化能力和鲁棒性。实验结果表明,所提方法能够有效识别故障电弧,识别率达到96.7%,且具有良好的稳定性。

Aiming at the problem that the frequency domain characteristics of fault arc current are difficult to extract and the detection rate is not high, a low-voltage AC fault arc diagnosis method based on wavelet time-frequency map and deep learning convolution neural network is proposed. This method carries out wavelet time-frequency transformation on the series fault arc current signal, and uses the amplification ability of wavelet to the high-frequency sudden change signal to capture the differential characteristics of different loads. The obtained time-frequency diagram is compressed and unified to an appropriate size as the input characteristic diagram of AlexNet convolution neural network. The wavelet time-frequency diagram characteristics are extracted by AlexNet network, and the intelligent diagnosis of fault arc is realized by fitting the data of full connection layer. Establish a fault arc experimental platform, collect 2 000 groups of arc currents under resistive load, inductive load and resistive load respectively, and use the data set to complete the model training. A variety of new loads and 10~60 dB signal-to-noise ratio are used to study the generalization ability and robustness of the model. The experimental results show that the proposed method can effectively identify the fault arc, the recognition rate is 96.7%, and has good stability.