基于声音识别的设备状态检测算法  

State identification algorithm for substation devices based on sound recognition

作　　者：李晶[1] 孙农亮[1] 滕升华[1] 

机构地区：[1]山东科技大学,山东青岛266590

出　　处：《信息技术》2015年第6期94-98,共5页Information Technology

摘　　要：变压器和高抗是变电站中的两种重要设备,设备运行的声音代表了设备运行的状态。但是,目前巡检机器人还不具备设备声音识别的功能,文中利用机巡检器人所带的拾音器采集巡检过程中的设备声音,利用声音信号处理及识别技术对设备声音进行分析与识别,提出了基于声音谐波特征及矢量量化的变电站设备声音识别算法,提取了[0Hz,1000Hz]范围内的21个谐波作为特征,建立数量庞大的样本库,在此基础上利用LBG算法训练得到变压器和高抗设备的码本,最后实现了变压器和高抗设备运行状态的准确识别,达到99%的识别率。The running sound of the devices in power substations is the reflection of their running state,so it can be utilized to identify the device running state. A new state identification algorithm for power substation devices based on sound recognition and inspection robot was proposed. This paper uses the sound recorder of the inspection robot to obtain the running sound samples of the devices. It mainly deals with the two most important devices,that is,transformer and high reactance. Based on large amount of sample analysis,it concludes that the device sound has the harmonics feature,so it adopts the 21 harmonics in [0Hz,1000Hz] as the feature vector,then uses the LBG algorithm to train the optimal codebook for VQ,and it gets the two threshold of the two devices for VQ. This paper experiment results show that a high recognition ratio 99% was achieved.

关 键 词：变电站巡检机器人 谐波特征 矢量量化 LBG算法 

分 类 号：TN912.34[电子电信—通信与信息系统]