73 / 2022-03-10 14:36:31

A novel feature selection method for power transformer fault diagnosis based semi-supervised learning

semi-supervised learning,Dissolved Gas Analysis,feature selection,transformer fault diagnosis

   Collecting labeled Dissolved Gas Analysis (DGA) data is difficult because the determination of the transformer fault is time-consuming and expensive in the transformer substation, but DGA data without labels is easier to obtain. In order to make full use of DGA data with few labels and a large number of unlabeled data to improve the transformer fault diagnosis rate, it is an important to study the DGA fault diagnosis method based on semi-supervised learning for solving practical problems in the field.  The paper proposed a novel filter-based semi-supervised feature selection method  for selecting optimal DGA features. Different from other time-consuming filter-based methods, the method uses different filter criteria for feature ranking after semi-supervised iterative, and then uses majority voting strategy (MVS) and complement linear aggregation (CLV) for selecting optimal feature after several rounds. The method is test by using IEC T10 dataset and DGA samples from the local power utility and compared with traditional supervised diagnostic models.The results show that the proposed method works in optimizing DGA features and has strong robustness in solving small sample classification problems.