Fault Classification Method Based On Convolutional Neural Network And Random Forest Algorithms For Recognize The Power Transformer Condition

Abstract

In the power system, the power transformer plays an important role in transmitting voltage at various levels. Therefore, to enhance the lifespan of the power transformer, the fault recognition is done at the early stage. In order to achieve this goal, the fault detection is done using the signal processing and machine learning algorithms by analyzing the voltage and current data. In this paper, we have presented a fault classification method based on convolutional neural networks and random forest algorithms for recognizing the power transformer condition using the ensemble learning approach. Initially, in this method, the IEEE 14-bus system is designed in the MATLAB Simulink model for fault generation in the power system by measuring the voltage and current signals for different faults. After that, the preprocessing of the data is done to prepare it for the CNN and RF algorithms by performing the Butterworth filtering, normalization, and feature extraction using wavelet decomposition algorithms. Then, RF and CNN algorithms are trained and tested for fault classification, and final classification is determined by performing the voting classifier algorithm. The result indicates that the proposed method accomplishes the accuracy value of 0.9983, F1-score value of 0.9867, and MCC value of 0.9860. The key finding of the proposed method is that it determines the type of fault and helps in timely maintenance to prevent the major breakdown in the power transformer.