Fault diagnostic of power transformer by CFS-GA Case-Based Reasoning model

To address the problems that knowledge rules are difficult to obtain and inadequate use of fault case experience in the process of transformer fault diagnosis, a CFS-GA case-based reasoning model is proposed. The model uses the correlation- based feature selection (CFS) method to extract key attribute features with strong correlation in power transformer fault cases, and solves the feature redundancy in case searching caused by excessive test data of transformers. Using the reduced attribute features, a genetic algorithm (GA) method is employed to optimize the feature weights that are difficult to determine manually and assign them to the K Nearest Neighbor (KNN) network to realize the diagnosis and final classification of power transformer fault category. The feasibility of the method is verified by experiments. The overall recognition accuracy of fault diagnosis is over 70% and it can reach 84% with the selected attribute feature set and optimized weights. This method also