Fault Detection and Diagnosis in Electrical Distribution System Using Generative Adversarial Networks with Deep Belief Networks

In recent years, electrical faults aim to detect early and diagnose to mitigate equipment damage, catastrophic failures and power outages of the distribution network. Traditional approaches for fault detection and diagnosis in electric distribution system had several challenges which include renewable energy integration, more complex due to increasing loads and diverse fault types in sophisticated detection of devices. Therefore, this research proposes Generative Adversarial Networks with Deep Belief Networks (GAN-DBN) for fault detection and diagnosis in electrical distribution system which is employed on Institute of Electrical and Electronics Engineers Node Test Feeder (IEEENTF) dataset. Then, preprocessing is done by using Fast Fourier Transform (FFT) for harmonic distortion analysis, faults and load unbalances detection. After that, features are extracted by using Convolutional Neural Networks (CNN) autoencoder which represents discriminative faults. Next, the classification of hierarchical representation of data is utilized to extract high level features from raw data through GAN-DBN. Finally, GAN-DBN identifies anomaly data for generated distribution in electrical system. The proposed GAN-DBN achieved better results in accuracy (0.9999), specificity (0.9998) precision (0.9995), recall (0.9996), F1-score (0.9997) when compared with existing CNN.