摘要
输电设备配有振动信号测量系统;在不同机械与电气故障状况下的振动特性,可分析OLTC状况与振动信号之间的关系.但是,低温环境会导致设备材料的脆化及润滑剂凝固,改变设备振动和声纹特征,增加异常振动声纹识别的复杂性.提出一种寒冷环境下的输电设备复杂异常振动的声纹识别方法.将经验模态分解(Empirical Mode Decomposition,EMD)和斯坦无偏估计(Stein's unbiased estimate,SURE)相结合,采集输变电设备的振动声纹信号,并对其进行去噪处理;将去噪处理后的声纹信号划分为多个声纹片段,并转换为语谱图.在特征提取阶段,将正常语谱图作为输入,利用长短期记忆网络(Long Short-Term Memory,LSTM)进行训练,以分类输入的输变电设备语谱图声纹样本,并确定异常样本,实现对输变电设备异常振动声纹的准确识别.实验结果表明:所提方法可以精准识别输变电设备异常振动声纹,识别率均在96％以上,识别耗时 94.47 ms.
Abstract
The transmission equipment is equipped with a vibration signal measurement system;The vibration characteristics un-der different mechanical and electrical fault conditions can be analyzed to determine the relationship between OLTC conditions and vi-bration signals.However,low-temperature environments can lead to embrittlement of equipment materials and solidification of lubri-cants,changing the vibration and voiceprint characteristics of the equipment,and increasing the complexity of identifying abnormal vi-bration voiceprints.Propose a voiceprint recognition method for complex abnormal vibrations of transmission equipment in cold envi-ronments.Combining Empirical Mode Decomposition(EMD)with Stein's unbiased estimate(SURE)to collect vibration voiceprint signals of power transmission and transformation equipment,and denoise them;Divide the denoised voiceprint signal into multiple voiceprint segments and convert them into spectrograms.In the feature extraction stage,the normal spectrogram is used as input,and a Long Short Term Memory(LSTM)network is used for training to classify the input spectrogram voiceprint samples of power trans-mission and transformation equipment,and determine abnormal samples to achieve accurate recognition of abnormal vibration voice-print of power transmission and transformation equipment.The experimental results show that the proposed method can accurately i-dentify abnormal vibration voiceprints of power transmission and transformation equipment,with recognition rates of over 96％and rec-ognition time of 94.47 ms.
维普
万方数据