Deep Reinforcement Learning-Based Autonomous Control Architecture for Renewable Energy-Integrated Smart Grids
4th Cognitive Models and Artificial Intelligence Conference, AICCONF 2026, Prague, Çek Cumhuriyeti, 24 - 25 Nisan 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/aicconf69182.2026.11600688
- Basıldığı Şehir: Prague
- Basıldığı Ülke: Çek Cumhuriyeti
- Anahtar Kelimeler: Adaptive Control Systems, Autonomous Energy Management, Deep Reinforcement Learning, Renewable Energy Integration, Smart Grids
- Atatürk Üniversitesi Adresli: Evet
Özet
The rapid proliferation of renewable energy sources such as solar and wind power has introduced significant variability and operational uncertainty into modern power systems. Conventional grid control strategies, largely based on predefined rules or deterministic optimization, struggle to maintain stability and efficiency under highly dynamic renewable generation conditions. In this study, a novel deep reinforcement learning-based autonomous control architecture is proposed for renewable energy-integrated smart grids. The framework enables real-time, self-adaptive decision-making for energy dispatch, storage coordination, and grid interaction without relying on fixed control policies. The proposed architecture integrates a deep reinforcement learning agent with a grid-aware state representation that incorporates load demand, renewable generation, battery state-of-charge, and frequency deviation indicators. A multi-objective reward formulation is designed to simultaneously minimize operational cost, mitigate frequency instability, and preserve battery health. Unlike traditional approaches that treat forecasting and control separately, the presented method establishes an end-to-end adaptive control mechanism capable of learning optimal strategies directly from stochastic system behavior. The performance of the proposed system is evaluated within a simulated renewable-integrated microgrid environment under varying solar irradiance and load fluctuation scenarios. Comparative analyses against rule-based and model predictive control strategies demonstrate improved operational efficiency, enhanced grid stability, and reduced renewable curtailment. The results indicate that deep reinforcement learning can provide a scalable and robust foundation for next-generation autonomous smart grid control systems.