基于卷积神经网络的放电声音故障检测

Fault Detection of Discharge Sound Based on Convolutional Neural Network

设计了一种基于卷积神经网络的放电声音检测方法,针对电力系统中设备绝缘老化引起的局部放电现象,提出通过终端边缘节点的声信号检测方法实时监控设备正常工作、局部放电和发生故障的三种状态,并将异常状态通过边缘计算专网反馈给运维中心。该系统通过设备终端的边缘节点采集发生故障时放电音频数据,这些故障包括:正常工作、局部放电和故障已发生的状态。并进行信号预处理和提取能够反映故障状态的音频特征。然后,将处理后的数据作为卷积神经网络的输入。实验表明所提方法与经典的深度神经网络相比,平均识别率提高了约2%。

A discharge sound detection method based on convolution neural network is proposed.Aiming at the partial discharge phe-nomenon caused by equipment insulation aging in power system,the acoustic signal detection method of terminal edge node is proposed to monitor the normal operation,partial discharge and fault status of the equipment in real time,and feed back the abnormal state to the operation and maintenance center through the edge calculation network.The system collects the audio data of discharge when the fault occurs through the edge node of the device terminal.These faults include normal operation,partial discharge and the state of the fault being occurred.The signal preprocessing and extraction can reflect the fault state of audio features.Then,the processed data are used as the input of the recognition model constructed by the convolutional neural networks.Experiments show that the average recognition rate of the proposed method is about 2%higher than that of classical deep neural network.