Modern Method to Discriminate High Resistance Internal Winding Faults and Inrush Condition/External Faults in Power Transformer

The conventional method relies on measuring the second harmonic components to identify the magnetizing inrush current, which is a result of an internal transformer fault. This method has become unsuitable for modern power transformers, which produce less harmonic-rich inrush currents. This research introduces an innovative hybrid technique integrating discrete wavelet transform (WT) and probabilistic neural networks (PNN) to address this issue. The DWT performs feature extraction by decomposing the power transformer currents, generating feature details that are then used as input to the PNN. Specifically, we train the PNN algorithm to distinguish internal faults, such as high resistance, from magnetizing inrush currents. The algorithm selects the suitable Daubechies wavelet to optimize the proposed technique and determines the appropriate number of decomposition levels. Diverse simulations conducted on a 230/66 kV, 100MVA, 50Hz 3-phase transformer thoroughly assess and validate the method's efficacy. The results show that the suggested protection method works better at finding high-resilience internal faults and stopping fast currents from coming in. Additionally, comparing the proposed PNN technique to other methods like the Radial Basis Function Neural Network (RBFNN) and Back Propagation Neural Network (BPNN) approaches shows that it is better.