Machine Learning for Fault Diagnosis in Active Distribution Networks

The distribution network links the customer to the power supplier in power systems. One fundamental responsibility of the distribution network is to provide consumers with high-quality and reliable electricity. Faults in distribution grids cause severe burdens and introduce financial losses to the customers relying on them. The uncertainty about the location and the type of fault amplifies the issue as it complicates the maintenance process and causes additional losses. This paper used the short-time Fourier transform to extract features from a simulated active distribution network's measurements. The features extracted were then fed to feedforward neural network models, which we trained for fault detection, classification, and localization. Results demonstrate that the developed models accurately detect and classify the faults in the active distribution network, demonstrating the reliability and effectiveness of the proposed models. Also, the proposed approaches were able to locate the faults in the simulated network accurately. Eventually, suggested models could elegantly handle load variation, renewable energy resources generation, and fault information ambiguity.