基于多维特征提取的紫外局放分级方法及应用  

Method and its application of partial discharge rating based on multi-dimension feature extration

作　　者：刘宇宽[1] 马立新[1] 张建宇[1] 黄阳龙 

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《机电工程》2016年第6期765-769,共5页Journal of Mechanical & Electrical Engineering

基　　金：上海张江国家自主创新重点资助项目(201310-PI-B2-008)

摘　　要：针对局部放电状态无法准确量化分级问题,对采用基于粒子群优化的支持向量机的状态分级方法进行了研究,通过该方法首先完成了多类特征空间在SVM核函数中的映射分类,再利用粒子群选取了最优核参数及惩罚参数。并提出了一种搭载该分级方法的便携式紫外传感电力巡检系统,结合其自身的测距功能,可向终端上位机回传紫外光斑面积、脉冲波形,测量距离、测量角度4种特征量,并以此作为分级判据,以充分利用紫外信号可靠且灵敏的特点。上位机根据已由试验数据建立起的优化分级模型,对设备的异常放电进行了诊断分级。研究结果表明,精度较传统支持向量机显著提高,避免了人为选取参数的盲目性,能够根据现场回传数据准确、实时地完成设备异常放电状态分级。Aiming at the problems of difficulty to accurately quantify the classification for partial discharge status,the new method of PSOSVM classification was investigated. By this method,multiple feature spaces were mapped to different SVM kernel functions,each kernel function and penalty parameters were optimized via particle swarm optimization（ PSO）. A ultraviolet sensing electrical inspection system carried the new method was presented. Combined with range finder and ultraviolet sensor of the system,four kinds of feature data were obtained and returned to the terminal PC. Those data were composed of ultraviolet light spot area,ultraviolet pulse waveform,measured distance and angle. In this way,the model took full advantage of sensitivity of UV signal. According to the classification model set up by the test data,this system can be used to diagnose and rating abnormal discharge of equipment. The results indicate that the new method can complete the abnormal discharge rating accurately according to the data back,and the classification model of PSO-SVM can prevent the blindness of selecting parameters and also has significantly higher accuracy than the traditional SVM.

关 键 词：局部放电 紫外检测 高压设备巡检 PSO-SVM 

分 类 号：TM764.1[电气工程—电力系统及自动化] TP835.4[自动化与计算机技术—检测技术与自动化装置]