AI-driven solar air heating: Reinforcement learning optimization of thermoelectric systems for enhanced thermal-electrical performance

This study presents a novel reinforcement learning (RL) framework for optimizing thermoelectric generators (TEGs) in solar air heaters to enhance both thermal and electrical efficiency while reducing carbon emissions. A comprehensive mathematical model integrating energy balance equations across system components was developed and coupled with deep RL algorithms. The Soft Actor-Critic (SAC) algorithm optimized TEG placement while Deep Deterministic Policy Gradient (DDPG) controlled mass flow rate (MFR) dynamically. Results demonstrated that optimal TEG positioning at 0.5 normalized distance from the inlet maximized power generation, with temperature differentials of 28.4°C. The RL-TEG system achieved 10.8% higher thermal efficiency and 81.0% greater electrical output compared to conventional designs, reducing LCOE by 24.4% (from 9.0 $/kWh to 6.8$/kWh). Weather prediction integration maintained 96.5% performance during cloud events versus 62.8% for fixed-rate systems. Federated learning was implemented to enable knowledge sharing across distributed installations. This approach proved effective, with federated models achieving 85.2% accuracy compared to 73.8% for local models. Environmental analysis revealed carbon emission reductions of up to 803 kg/m 2 /year in high-insolation regions with carbon-intensive grids, with payback periods ranging from 1.2 to 2.6 years. Field validation across three distinct climate zones confirmed the system's adaptability, demonstrating 18.5% performance improvement in variable continental climates.