基于最小二乘支持向量机的高压绝缘子污秽程度评定

研究表明,泄漏电流的各种电气特征量除了与绝缘子表面的污秽状况有关外,还受温度、湿度等环境因素的影响,并且与各因素之间存在着复杂的非线性关系.因此,绝缘子在线监测系统通过单一检测泄漏电流的大小来评定表面污秽状况是不科学的.文中在实验室模拟试验和现场实测数据基础上,利用最小二乘支持向量机这种新的机器学习工具,分别用2种核函数建立了以泄漏电流有效值、泄漏电流峰值、泄漏电流脉冲频度、环境湿度、温度等5个变量作为输入参数,污秽程度作为输出参数的评定模型,超平面参数通过交叉检验的方式确定.实验结果表明,最小支持向量机具有很好的学习、分类和泛化能力,且对于污秽程度评定问题选用RBF核函数相对于多项式核函数有更高的判别准确率.

Assessment of Surface Contamination of High Voltage Insulator Based on Least Squares Support Vector Machine

Investigation shows that the electric characteristics of leakage current (LC) are influenced by the contamination of insulator surface as well as the environmental factors including temperature, humidity and so on. Also the nonlinear relationship between LC and each factor is complicated. Thus it is unreasonable to estimate the contamination of online insulator merely by detecting the root mean square (R. M. S) of the surface LC in the online monitoring system. In this paper, based on laboratory simulation experiments and field data, the LC R. M. S. , the peak value of the LC, the amplitude and times of the pulses of the LC, the temperature and the humidity of temperature are chosen as five input variables, and then the degree of contamination as an output variable. The assessment models using the least square support vector machine (LS-SVM) are built with two kinds of kernel functions. By adopting the polynomial and RBF kernel, the hyper-parameters of classifiers are tuned with cross-validation. Experiment results show that the LS-SVM classifiers are capable of learning quite well from the raw data samples while processing good classification and generalization ability, and the RBF kernel function is more accurate than the polynomial kernel one for the problem of assessing the degree of insulator contamination.