Improved Error Correcting Output Codes Support Vector Machines And Application In Power Quality Evaluation

Power quality evaluation is an important content of power quality monitoring systemwhich is used to divide the indices of power quality comprehensively and can be attributed asa classification of power quality. At present, support vector machine (SVM) algorithm is oneof the most important methods for the classification of power quality. However, it often exitsinseparable regional issues when the traditional SVM algorithms are used to classify the realpower quality data. Then, an improved error correcting output codes support vector machines(ECOC-SVMS) which is named as based on agglomerative hierarchical clusteringECOC-SVMS algorithm is presented in this paper. By improving the design of coding matrix,this algorithm not only can solve inseparable regional problems of multiple classification, butalso reduces the training and testing time of model effectively and improves the availability ofencoding matrix etc.Additionally, in the practical application of power quality classification, it often needs todecide the category of sample and compute its probability of belonging to the category.Therefore, based on the results of SVM classification, the paper presents a B-spline leastsquare algorithm fitting probability density function. The algorithm uses the B-spline leastsquare methods to fit the classification resluts of SVM method and calculate the probability oftest sample belonging to category. The two-dimensional classification dataset fourclass inLibsvm is used to test the efficiency of this algorithm. The experimental results show that themethod can well solve the problem of low dimension probability classification.Finally, the improved ECOC-SVMS algorithm is applied to the actual evaluation ofpower quality in this paper, and its performance is tested on the power quality data. Theexperimental results show that the algorithm is superior to the classical ECOC-SVMS in thetraining time, testing time and testing accuracy. On the basis of the algorithm, a new empiricalprobability method is presented for high dimension probability classification. It is applied tothe actual power quality data, and ideal results are obtained.