The traditional approach to differential protection relies on detecting the 2nd harmonic to discern between internal faults within a transformer and the initial magnetizing inrush currents upon energization. However, this conventional method is becoming less reliable, particularly with the advent of modern power transformers that produce less harmonics-rich inrush currents. Consequently, distinguishing between these current types becomes challenging, leading to erroneous tripping events. Such false trips disrupt the power system and pose risks of economic losses and potential damage to transformers. To address this issue, this study introduces an innovative hybrid technique integrating wavelet transform (WT) and Levenberg–Marquardt backpropagation neural network (LM-BPNN). The WT serves for feature extraction by decomposing differential currents and generating spectral energy details utilized as input for the LM-BPNN. Specifically, the LM-BPNN is trained to differentiate internal faults from other occurrences. To optimize the proposed technique, the study computes the Pearson correlation coefficient and signal-to-noise ratio to select the suitable Daubechies wavelet and determine the appropriate number of decomposition levels, respectively. The efficiency of the proposed method is thoroughly assessed and validated through diverse simulation scenarios involving a three-phase transformer rated at 132/33 kV, 63MVA. The results underscore the superior performance of the proposed protection approach in accurately detecting high-resistance internal faults while effectively restraining inrush currents. Furthermore, comparative analysis with alternative methods, such as variable learning rate backpropagation and the BFGS quasi-Newton approach, corroborates the proposed technique’s superiority
