AI-Based Solar Thermal Cooling Optimization for Large-Scale Data Centers: A Sustainable Approach

This study investigates the performance of three cooling configurations for a large-scale (1 MW) data center in a high solar irradiance region. The configurations include: (i) a conventional electric chiller system, (ii) a solar thermal cooling system with PID control, and (iii) an AI-optimized solar thermal cooling system using Deep Reinforcement Learning (DRL). A dynamic co-simulation environment was developed integrating TRNSYS, EnergyPlus, and a Python-based PPO agent to evaluate the trade-off between energy efficiency and water sustainability. The results demonstrate that the AI-based system reduces grid electricity consumption by 79% and water usage by 92% compared to the conventional baseline. Furthermore, it significantly improves thermal stability, achieving a Root Mean Square Error (RMSE) of 0.24°C relative to the setpoint, and shortens the response time to load disturbances to 11 seconds. This work validates that AI-enhanced solar cooling is a scalable and sustainable solution for data centers in water-stressed regions.