基于特高频和CNN-LSTM-Attention算法的高压电缆故障诊断方法

随着高压电缆的加速发展和老化，由局部放电(partial discharge, PD)引起的故障问题亟须解决。为此，提出了一种基于特高频(UHF)局放技术与CNN-LSTM-Attention算法的高压电缆故障在线智能诊断方法。首先，对高压电缆的PD产生机理，以及UHF局放技术的实现过程进行描述。其次，利用巴特沃斯(Butterworth)对PD信号进行高通滤波，采用小波变换对信号进行去噪，IPLR算法对PD信号进行降维处理，进而实现特征量的准确提取。最后，建立由CNN-LSTM-Attention算法构成的智能诊断模型。模型中卷积层(CNN)提取轮廓特征，长短期记忆层(LSTM)提取信号时序特征，注意力层(Attention)学习信号重要时序部分。通过实际数据仿真表明：相比传统神经网络方法，CNN-LSTM-Attention神经网络检测方法能够准确识别高采样率的异常放电信号特征，且故障识别准确率明显提高。

High Voltage Cable Fault Diagnosis Method Based on UHF and CNN-LSTM-Attention Algorithm

With the accelerated development and aging of high voltage cable, the fault caused by Partial discharge (PD) needs to be solved. Therefore, A high voltage cable fault on-line intelligent recognition method based on UHF local discharge technology and CNN-LSTM-Attention algorithm is proposed. Firstly, the PD generation mechanism of high voltage cable and the realization process of UHF local discharge technology are described. Secondly, the PD signal is filtered by Butterworth high-pass filter and denoised by wavelet transform, and then the PD signal is dimensionally reduced by IPLR algorithm, so as to achieve accurate extraction of feature quantity. Finally, an intelligent diagnosis model composed of CNN-LSTM-Attention algorithm is established. In the model, the CNN layer extracts contour features, the LSTM layer extracts temporal features of signals, and the attention layer learns important temporal parts of signals. The actual data simulation shows that compared with the traditional neural network method, the CNN-LSTM-Attention neural network detection method can accurately identify the characteristics of abnormal discharge signals with high sampling rate, and the fault identification accuracy is significantly improved.