Real-Time Fault Diagnosis in Electrical Power Systems Using Neural Networks

Abstract

In this research, we propose a novel approach for real time fault diagnosis of the electrical power systems using a Hybrid CNN - LSTM architecture. The method leverages the strengths of CNNs for spatial feature extraction from time-frequency representations of electrical signals and LSTMs for learning temporal dependencies in fault patterns. First the electrical signals are pre-treated using techniques such as Wavelet transform or Short Time Fourier Transform (STFT) to generate time frequency spectrograms. Then, using Dimensionality reduction methods like Principal Component Analysis (PCA), important features are chosen so that faster model training can be achieved. Through the usage of Tensorflow/Keras with GPU acceleration, the train and evaluation of the hybrid CNN-LSTM model is performed for fast and accurate fault detection in real time. The proposed method improves fault diagnosis accuracy and processing speed greatly, thus, appropriate for deploying in smart grids and wide scope electrical systems, in order to boost the system reliability and minimize the downtime.