AI-Driven Fault Detection in Electrical Grids: Comparative Analysis Between Gradient Descent with Momentum and Adaptive Learning and Resilient Propagation Algorithms

Reliable fault detection is critical for ensuring the stability and safety of electrical grids. This study applies artificial intelligence (AI) techniques to fault diagnosis by comparing two optimization algorithms: Resilient Propagation (RProp) and Gradient Descent with Momentum and Adaptive Learning (GDM-AL). A deep learning-based fault detection model was developed using voltage and current signals as input features, simulating various fault and non-fault conditions in an electrical network. The dataset was processed and trained in MATLAB, where both algorithms were evaluated based on convergence speed, classification accuracy, and adaptability to different fault scenarios. The modelling approach involved training a feed forward artificial neural network (ANN), optimized using RProp and GDM-AL, to classify electrical faults. RProp adapts weight updates independently for each parameter, leading to faster convergence but limited handling of highly non-linear conditions. In contrast, GDM-AL integrates momentum and adaptive learning rates, improving stability and reducing oscillations in fault detection. A detailed confusion matrix analysis confirmed that both methods achieved high classification accuracy, with GDM-AL minimizing false positives in complex scenarios. Additionally, mean squared error (MSE) evaluations demonstrated the effectiveness of both techniques in detecting faults with minimal prediction errors. These findings contribute to the advancement of AI-driven predictive maintenance in electrical grids, supporting the development of intelligent fault detection systems. Future work will explore hybrid optimization techniques that leverage the strengths of both methods to enhance detection accuracy, adaptability, and real-time performance.