Conventional vs. Reinforcement Learning–Based Relays for Power System Protection

This project presents a direct, implementation-level comparison between conventional protection relays and reinorcement learning–based relays for power system protection. Classical distance, differential, and inverse-time overcurrent relays are implemented alongside an RL-based relay using identical voltage and current measurements. All methods are evaluated under the same fault scenarios, datasets, and performance metrics to ensure a fair comparison. The analysis focuses on detection accuracy, operating time, selectivity, and misoperation rates. The project aims to clarify the practical advantages and limitations of learning-based relays relative to established protection logic.