基于多维时频特征的变压器故障声纹识别方法研究  

Research on Transformer Fault Acoustic Identification Method Based on Multidimensional Time-Frequency Features

作　　者：唐冬来 李擎宇 龚奕宇 钟声 陈泽宇 聂潇 TANG Donglai;LI Qingyu;GONG Yiyu;ZHONG Sheng;CHEN Zeyu;NIE Xiao(Sichuan SGIT Technology Co.,Ltd.,Chengdu 610047,China;State Grid Sichuan Electric Power Company,Chengdu 610041,China)

机构地区：[1]四川思极科技有限公司,四川成都610047 [2]国网四川省电力公司,四川成都610041

出　　处：《自动化仪表》2023年第11期11-14,19,共5页Process Automation Instrumentation

基　　金：国家重点研发计划基金资助项目(2019YFB2103000)。

摘　　要：为解决变压器绕组故障检查难度大、辨识准确率低的问题,提出了一种基于多维时频特征的变压器故障声纹识别方法。首先,通过傅里叶变换与分帧法提取变压器噪声,过滤电晕、风机、环境干扰数据。其次,将变压器全寿命周期声纹库与当前变压器声纹进行比较,判断是否存在异常。在此基础上,通过熵权法调整多维时频特征评估权重,从而辨识变压器绕组故障声纹。最后,在某省电科院进行变压器故障声纹仿真试验。该方法识别准确率为95.9%,能够精准辨识变压器绕组故障。To solve the problems of high difficulty in transformer winding fault inspection and low recognition accuracy,a transformer fault acoustic identification method based on multidimensional time-frequency features is proposed.Firstly,the transformer noise is extracted by Fourier transform and split-frame method,and the corona,fan and environmental interference data are filtered.Secondly,the transformer full life cycle acoustic library is used to compare with the current transformer acoustic to determine whether there is any abnormality.On this basis,the weight of multidimensional time-frequency feature evaluation is adjusted by entropy weighting method,to recognize the transformer winding fault acoustic.Finally,the transformer fault acoustic simulation test is carried out in a provincial electric academy.The recognition accuracy rate of this method is 95.9%,which can accurately recognize the transformer winding fault.

关 键 词：变压器 多维时频特征 故障声纹 熵权法 分帧法 绕组故障 振动传播 绝缘老化 失真度 

分 类 号：TH87[机械工程—仪器科学与技术]