基于声纹识别的变压器工况检测方法及验证系统

Detection Algorithm and Verification System of Transformer Working Condition Based on Voiceprint Recognition

通过对大型电力变压器(电抗器)声纹信号采集、处理与特征分析,提出了一种基于声纹识别技术的变压器工况检测方法及验证系统,实现对变压器工作状态的判断检测。首先,通过声纹采集传感器实地采集获取73组变压器音频,共约1800 min;其次,分别运用分段、分帧、加窗音频预处理方法对所采集的变压器声纹进行去噪处理;再次,综合运用能量特征、频率特征、梅尔频率倒谱系数、频率压缩方法提取变压器声纹特征并进行有效融合;最后,针对变压器工况声音由稳定工作状况发出的声音与不稳定的瞬时杂音加性叠加问题,提出一种基于余弦相似度算法实现声纹叠加的分离检测,同时建立一套变压器工况检测与验证分析系统。

Through the acquisition,processing and characteristic analysis of voiceprint signal of large power transformer(reactor),this paper proposes a detection and recognition algorithm model and system of transformer working condition based on voiceprint recognition technology,which can judge and detect the working state of transformer.Firstly,73 groups of transformer audio including 1800 minutes are collected by voice acquisition sensor.And then the noise of the collected transformer voiceprint is processed by segmenting,framing and windowing pretreatment methods.Secondly,the features of transformer voiceprint are extracted and effectively fused by using energy feature,frequency feature,Mel frequency cepstrum coefficient and frequency compression method.Finally,in order to solve the problem of additive superposition of sound from stable working condition and unstable instantaneous noise of transformer,this paper proposes a separation method of voice print superposition based on cosine similarity algorithm,and establishes transformer working condition detection and verification analysis system.