基于掩码自编码技术的变压器故障声纹诊断方法研究

Research on Voiceprint Diagnosis Method of Transformer Faults Based on Mask Self-coding Technology

电力变压器作为电力系统中电能转换与能量传输的核心设备,其可靠运行对电力系统的安全运行有着极为重要的意义。近年来随着边缘计算在电力设备诊断方面的应用不断盛行,受限的硬件算力对算法的计算量提出了新的要求。对此提出一种基于声纹识别与掩码自编码技术的变压器故障诊断方法。首先,将采集的变压器辐射声音信号进行梅尔频谱计算和归一化处理得到声纹特征;其次,将声纹特征运用掩码自编码器进行训练,得到可用于对特征降维的编码器;最后,利用卷积神经网络对编码后的变压器声纹特征进行识别和分类。实验结果表明,该方法在掩码比例为40%的时候能达到93.75%故障识别精度,识别准确率高于对比算法。此外,相对于没有采用掩码自编码的分类算法,该方法在提升17.26%准确率的同时,将计算量缩减为不到原来的1%,可有效降低变压器检测的计算量。

The electric power transformer is a core device for energy conversion and transmission in the power system,and its reliable operation is of great significance for the safe operation of the power system.In recent years,with the increasing popularity of edge computation in the diagnosis of power equipment,limited hardware computation power has put forward new requirements for algorithmic computation.In this paper,a transformer fault diagnosis method based on voiceprint recognition and mask autoencoder technology is proposed.Firstly,the mel-frequency spectrum calculation for the collected transformer radiation sound signal is performed and the normalization processing of the results is carried out to obtain voiceprint features.Secondly,the voiceprint features are trained using a mask autoencoder to obtain an encoder that can be used for feature dimension reduction.Finally,the convolutional neural network is used to recognize and classify the encoded transformer voiceprint features.The experimental results show that when the mask ratio is 40%,the proposed method can achieve a fault recognition accuracy of 93.75%,which is higher than that of the comparative algorithm.In addition,compared with the classification algorithm without using mask autoencoder technique,the proposed method reduces the calculation amount to less than 1%of the original while improving accuracy by 17.26%,effectively reducing the calculation amount of transformer detection.