Assessment of surface contamination condition of insulator based on attribute reduction algorithm of rough sets and least squares support vector machine
Shangbin Jiao, Ding Liu · 2010 Sixth International Conference on Natural Computation · 2010
A model of the assessment of surface contamination condition of insulator was investigated by the method of combining rough sets (RS) theory and least squares support vector machine (LS-SVM). According to the lab and field data, a attribute decision table is built up and the redundant attributes of the data is deducted by means of attribute reduction algorithm, thus the kernel factors of assessment contamination condition is determined and a new decision table is formed. The table was acted as a learning sample to train and construct the LS-SVM multi-classifier, thus the mapping relationship between the contamination classes and the electric character variables of the leakage current (LC) and the environment factors was formed and contamination condition assessment was realized by the trained LS-SVM multi-classifier. Experiment results show that the method of combining RS and LS-SVM is faster and more accurate for the assessment of surface contamination condition of insulator in compare with the traditional LS-SVM classifiers, and the RBF kernel function has more accurate than polynomial kernel function for the problem of assessment contamination condition of insulator.