Differential Protection of Power Transformers Using Discrete Wavelet Transform and Convolutional Neural Network

This research paper presents a hybrid approach for differential protection of transformer using discrete wavelet transform (DWT) and convolutional neural network (CNN). A simulation of a three-phase transformer is conducted to capture various current conditions such as normal, fault, and inrush currents. A statistical tool DWT is employed to decompose the input signal according to the mother wavelet Daubechies level. Further, it is used to extract required features such as mean, norm, standard deviation. The preprocessed data are feed into CNN, to discriminate between normal, fault, and inrush conditions accurately. In the training process, CNN demonstrated very good performance by achieving accuracy of 98%. To validate the effectiveness of the proposed method, a detailed analysis of the correct classification with actual true labels is carried out by employing performance metrics. The results indicate the potential of this hybrid DWT-CNN framework, to provide reliable and efficient differential protection of transformers that significantly improves reliability of the system.