Fault-MTL: A Multi-task Deep Learning Approach for Simultaneous Fault Classification and Localization in Power Systems

Power system interruptions, though brief, carry significant costs and adverse consequences. Ensuring a reliable power supply necessitates accurate fault detection and swift resolution through fault classification and localization. The existing individual and joint approaches for fault classification and localization involve comprehensive expert-guided feature engineering and pre-processing algorithms depending on the selected classifier. This research proposes a fully automated, integrated multi-task deep learning-based framework called the Fault-MTL (multi-task learning) model for performing fault classification and localization at the same time. The multi-task feature interdependency accelerates the conduction of tasks with better performance. To the best of our knowledge, this is the first-of-its-kind study where the multi-task learning (MTL)-based deep model is developed for power system studies. To evaluate the model, we tested it on three state-of-the-art transmission line topologies, each with various initial conditions and simulated fault data variations in terms of fault type, distance, inception angle, and resistance. The tenfold cross-validation is used to measure the performance. The simulation experiments exhibit an exemplary performance by attaining the perfect validation in fault classification and the minimal mean average error of 1.59% in fault localization, across all three adapted topologies. For investigating the feasibility of the model, the ablation studies including the effect of noise, comparison with the existing state-of-the-art methods and the performance comparison of the work with the studies from the literature are performed in individual and combined simulation experiments of fault classification and fault localization.