Robust Wavelet–ANFIS Fault Classification for Power‐Transformer Differential Protection

Power‐transformer protection demands rapid and reliable discrimination between internal faults and transients under non‐ideal operating conditions such as magnetizing inrush. This study presents an Adaptive Neuro‐Fuzzy Inference System (ANFIS) that combines fourth‐level Discrete Wavelet Transform (DWT) features with principled feature reduction and imbalance handling for robust classification. Three‐phase current and voltage waveforms in a power system network were simulated in MATLAB/Simulink across healthy operation and seven fault types under both ideal and noisy/disturbed regimes, yielding approximately 178,000 labelled records. From the DWT coefficients, Standard Deviation (SD) and Relative Wavelet Energy (RWE) were computed. Principal Component Analysis (PCA) and feature ranking then produced a compact seven‐feature set. Imbalance in the training partition was addressed using the Synthetic Minority Over‐Sampling Technique (SMOTE). With the ANFIS classifier, the proposed DWT feature set achieved a test accuracy of 97.22% under noise, with an operation time of 10.2 ms. The reliability of the proposed system was tested by simulating it under various internal and external faults, as well as different practical conditions, including magnetising inrush, on‐load magnetising transient, switching transient, and saturation, while varying the transformer tap positions. Finally, the designed system was validated using a secondary dataset and another power transformer rating.