Optimización de los tiempos de disparo en la coordinación adaptativa de protecciones de sobrecorriente en redes de distribución con generación distribuida mediante aprendizaje por refuerzo profundo

Abstract

This research aims to design an adaptive protection coordination model based on optimizing the starting current (Is) and time- switching time (TMS) of overcurrent relays (OCRs). This was achieved using a deep reinforcement learning algorithm, specifically the Deep Deterministic Policy Gradient (DDPG). This algorithm takes values calculated according to the IEC 60909 standard and selects two ideal settings for each relay. This is accomplished through algorithm learning, ensuring the model fulfills its objective function while respecting the algorithm's constraints and penalizing excessive operating times. The result is a faster, more sensitive, and more selective response time compared to the traditional method. The proposed methodology was implemented in a 33-bus IEEE system with distributed generation (DG) using Python and integrating PandaPower, Topological changes to the system were also considered to verify the model's efficiency.