Optimization of Tripping Times in Adaptive Overcurrent Protection Coordination for Distribution Networks with Distributed Generation Using Deep Reinforcement Learning

Modern electrical distribution systems face increasingly complex protection coordination challenges due to variations in power flows under different operating conditions. In this context, this work proposes an adaptive coordination scheme based on deep reinforcement learning (DRL) to reduce the operating times of overcurrent relays while maintaining sensitivity, selectivity, and speed requirements. The methodology was implemented in Python 3.13 using the IEEE 33-bus distribution network with distributed generation (DG), and the coordination problem was addressed using the Deep Deterministic Policy Gradient (DDPG) algorithm, which adjusts the protection settings from previously calculated fault currents. The results show that the DDPG-based approach reduces fault-clearing times compared with the conventional methodology, achieving reductions between 17.01% and 77.5% for three-phase faults and between 18.5% and 74.1% for single-phase-to-ground faults across the analyzed scenarios, without compromising coordination between primary and backup relays. In addition, a comparison with a PSO-based offline optimization approach was included as an additional benchmark, showing that the proposed method provides competitive operating times while preserving its adaptive learning-based nature. These findings show that the proposed methodology is a viable option for adaptive protection coordination in modern distribution networks.