Akademik Veri Yönetim Sistemi
2025 9th International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT), Ankara, Türkiye, 14 - 16 Kasım 2025, ss.1-7, (Tam Metin Bildiri)
The increasing integration of renewable energy sources has highlighted the importance of low voltage ride-through (LVRT) capability in power systems. LVRT supports security of energy supply and prevents blackouts by maintaining grid stability. Conventional control methods perform poorly under variable conditions, which reduces system reliability. In particular, unbalanced faults are among the most common problems in power systems and pose greater challenges to ensure the stability of control systems. Therefore, this paper aims to attain LVRT in a two-stage, three-phase grid-connected photovoltaic (PV) system using a deep reinforcement learning (DRL) approach, which offers adaptability under varying conditions such as unbalanced faults. The PV system is modeled in Matlab/Simulink and controlled using soft actor-critic (SAC) and twin delayed deep deterministic policy gradient (TD3) algorithms. The performance of this DRL methods is compared with a conventional proportional-integral (PI) controller optimized by the symbiotic organism search (SOS) algorithm. Key metrics for comparison include grid code compliance, AC current and DC voltage limit, and error values under unbalanced fault conditions. The results show that the TD3 and SAC algorithms give a 66.79% and 9.03% higher error rate compared to the conventional method in the average of the percentage error reduction values for each fault voltage amplitude. However, when looking at dynamic responses, the SOS-PI controller can only generalize one fault, while TD3 agent can generalize three faults and SAC agent can generalize eight faults. The SAC agent stands out in terms of generalization capability.