University of Tehran Journal of Solar Energy Research 2588-3097 11 2 2026 04 01 Optimising Photovoltaic Power Output Using Hybrid Deep Reinforcement Learning and Real-Time Environmental Adaptation 2923 2933 106599 10.22059/jser.2026.410130.1708 EN Ahmed Al-Maqsoosi Center of Computer and Informatics, Wasit University, Wasit, Iraq Hassan Al-Budairi Center of Computer and Informatics, Wasit University, Wasit, Iraq Abdurrahim Akgündoğdu Department of Electrical and Electronics Engineering at the Faculty of Engineering, Istanbul University-Cerrahpaşa, Turkey Hanan Al Yodaoi Ministry of Trade, Iraq Journal Article 2026 01 27 Development of Artificial Intelligence (AI) systems has transformed the management of renewable energy by resolving long-standing challenges in efficiency, resilience, and responsiveness. Photovoltaic (PV) power generation, highly sensitive to environmental fluctuations, can particularly benefit from AI-based control strategies. This paper proposes a hybrid AI architecture combining model-free Deep Reinforcement Learning (DRL) using Deep Q-Networks (DQN) with Long Short-Term Memory (LSTM) networks to enhance Maximum Power Point Tracking (MPPT) under dynamic conditions including rapid irradiance, temperature, humidity variations, and partial shading. The system employs real-time environmental sensor inputs, namely solar irradiance, ambient and module temperature, relative humidity, and shading indices, as the DQN state. The LSTM processes historical sequences to predict near-future power trends and enable proactive MPPT decisions. Implementation on a low-cost, energy-efficient Raspberry Pi edge computing platform enables decentralised, low-latency control without cloud dependence, suitable for remote or off-grid applications. A 180-day field validation on a rooftop 5.4 kW PV array demonstrated a 37% reduction in convergence time compared with Perturb and Observe (P&O) and 28% relative to Fuzzy Logic MPPT. The system achieved a 12.4% average increase in daily energy yield, rising to 18.7% under sporadic cloud cover and partial shading in real-world operational scenarios contexts. https://jser.ut.ac.ir/article_106599_f369c281f93315b97eba786899f6a47a.pdf