Fault diagnosis method for transformers based on feature selection of dissolved gas in oil

Dissolved gas analysis is important for the early warning and diagnosis of transformer faults. Aiming at the problems of numerous types of features for dissolved gas in oil and the insufficient analysis of fault associated features, a new fault diagnosis method for oil-immersed transformers based on feature selection of dissolved gas in oil is proposed. Firstly, the derivation of original features for dissolved gases is completed. The optimal combination of features is selected by calculating the importance of features for fault diagnosis based on random forest (RF). Then, the tree-structured parzen estimator (TPE) is used to realize the parameter optimization of the RF model, and the TPE-RF diagnostic model is obtained. Combined with the various evaluated indicators, the proposed method is proved to be able to diagnosis the transformer faults accurately. Finally, the TreeSHAP model is introduced to analyze the importance of the features corresponding to each fault, and the specialized features for each fault are selected. According to the case of transformer in operation, the applicability of the method in the power system is discussed, and the effectiveness of the method is verified.