基于相似矩阵盲源分离与卷积神经网络的局部放电超声信号深度学习模式识别方法

电气设备的故障类型与局部放电现象密切相关,有效提取和分析局部放电信号中的特征信息对故障类型判断和运维检修具有重要意义。针对局部放电超声信号的特点,提出了基于相似矩阵的盲源分离方法对原始超声信号进行预处理,有效提取局部放电的特征量。采用光纤传输的局部放电超声检测平台对4种类型的局部放电信号进行采集,并应用上述方法对信号数据预处理,将处理后的数据作为训练样本用于深度学习模式识别,选用卷积神经网络,最终识别准确率达到90%以上,提高了局部放电类型识别的准确性,为新一代电力系统的设备故障诊断提供了一种新方法。

Pattern Recognition of Partial Discharge Ultrasonic Signal Based on Similar Matrix BSS and Deep Learning CNN

The faults of electrical equipment are closely related to partial discharge (PD) phenomenon. It is of vital importance to extract and analyze the characteristic information in PD signal for faults recognition and operational maintenance in power system. According to the feature of PD ultrasonic signal, this paper proposes a blind source separation (BSS) method based on similar matrix for data pre-processing of original ultrasonic signals, to effectively extract the PD feature. Then, four PD pattern ultrasonic signals are captured with PD ultrasonic detection system based on optical fiber transmission. After signal acquisition, pretreatment is performed as mentioned above. Finally, the processed data is used as training samples for convolutional neural network (CNN) recognition, and the recognition accuracy is up to 90%, thus improving the accuracy of PD pattern recognition. This method could provide reference for new generation of fault diagnosis of power system equipment.