Fault classification and location of a PMU-equipped active distribution network using deep convolution neural network (CNN)

Accurate fault detection and localization play a pivotal role in the reliable and optimal operation of electric power distribution networks. However, the integration of intermittent distributed generation (DG) brings distinctive challenges to traditional fault diagnosis schemes, requiring more robust approaches. This paper proposes an intelligent and robust hierarchical framework for a fault diagnosis approach in power distribution grids integrated with intermittent DG. The approach employs deep convolutional neural networks (CNN), removing the need for hectic signal processing tools for feature extraction. The process starts with modeling a typical distribution network in the real-time digital simulator (RTDS) rack by incorporating the uncertainties of DG generation, load demand, and fault information through various probability density functions. Then, three-phase current signal of two cycles (fault and pre-fault) from PMUs are collected by applying faults in different feeder locations. These generated signals are inputted into the proposed CNN models for fault classification, section identification, and localization. The method achieves an accuracy of 99.46% for fault classification, 99.39% for fault section identification, and an error of 0.9% for fault location. Furthermore, the extensive and comparative performance analysis with state-of-the-art fault diagnosis techniques reiterates the effectiveness of the proposed strategy in successful fault diagnosis. •A hierarchical framework based on three convolution neural networks (CNN) is proposed for full fault diagnosis.•The method considers the uncertainty associated with DG generation and load.•The method shows robustness against the number of measurements.•The proposed method outperforms other methods in the literature.