A novel method of complex PQ disturbances classification without adequate history data

Abstract

Classification of power quality (PQ) disturbances has become an important issue due to the increasing of disturbing loads in electric power system and sensitivity to these disturbances. However, there are still some challenges in this issue such as lack of adequate training data to match disturbances with their corresponding types, especially for some multiple disturbances. In this paper, a new semi-supervised based method with S-transform is presented for solving classification problem without adequate history data. First, S-transform is chosen for PQ disturbances detection as it enables exaction of adequate time-frequency characteristics of the PQ events for later classification under noise, and features are obtained with S-matrix. Then, some unlabeled PQ features data (without class information of PQ) are used in building adaptive regularization classifier with their underlying manifold for improving the classification performance under rare history data. Later, seven binary classifiers are constructed with the regularization ruler to identify seven single PQ disturbances. Finally the multiple PQ disturbances are classified by composition of above classification results. The proposed method is validated through simulation studies using a database based on IEEE-1159 standard. Also, performance comparison with some conventional supervised classification method, such as SVM and ANN, is given out with different number of labeled PQ data.