11th International Conference on ICT for Sustainable Development

Authors - Aneri Shah, Krish Modi, Ansh Soni, Nishant Doshi
Abstract - There is a need for efficient and optimized power distribution for the transformer system. This paper presents a hybrid AI-driven approach integrating Multi-Agent Deep Reinforcement Learning (MARL) and Anomaly Detection for transformer power system optimization. Three key components have been focused: (1) a Predictive Maintenance Model leveraging deep learning to estimate the Remaining Useful Life (RUL) of transformers, (2) an Autoencoder-based Fault Detection Model for anomaly identification, and (3) a Multi-Agent Deep Q-Network (DQN) system optimizing power distribution and balancing loads dynamically. The findings contribute to intelligent power system management, reduced downtime, increased transformer downtime and adaptive optimization framework for future smart grids. Such contributions are aligned with higher sustainability goals by minimizing wastage of resources and minimizing carbon footprints due to inefficient grid operations.