Multi-filter semi-supervised transformer model for fault diagnosis

Dissolved Gas Analysis (DGA) is the most commonly used method for power transformer fault diagnosis. However, very few reliable and labeled DGA samples are available in substation whilst data without labels easier to obtain, which makes it difficult train detectors high-dimensional input space or select features using wrapper methods. Therefore, order improve diagnosis accuracy limited but more unlabeled data, this paper proposes a novel multi-filter semi-supervised feature selection selecting optimal building effective models. A confidence criterion also proposed high expand training set. Five filter techniques based on different evaluation criteria employed rank features, combination then applied aggregate ranks by multiple filters form lower-dimensional candidate subset. The has been tested IEC T10 dataset compared with traditional supervised diagnostic results show that works well optimizing improving significantly. Besides, robustness of subset validated testing from local utility.