Distribution Grid Fault Classification and Localization using Convolutional Neural Networks

This manuscript addresses the critical challenge of fault classification and localization within smart distribution networks, exacerbated by the complex integration of distributed energy resources and the dynamic nature of modern power systems. Traditional methods fall short in accurately and efficiently managing these tasks due to their reliance on linear models and manual inspection, which cannot cope with the data-rich and variable environment of current distribution networks. We introduce a novel data-driven framework utilizing convolutional neural networks to enhance fault classification accuracy and localization precision. Our framework uniquely incorporates an online continual learning algorithm, adapting to system changes and evolving fault patterns without requiring retraining. The method demonstrated significant improvements in fault classification and localization performance. Quantitatively, our CNN-based approach achieved a fault classification accuracy of 98.5% and a fault localization accuracy of 97.9%, outperforming traditional AI models in simulated environments. These results underline the potential of our framework to significantly contribute to the reliability and efficiency of fault management in smart distribution systems, offering a robust solution to the challenges posed by the integration of and the variability of load conditions.