A machine learning approach to assess the climate change impacts on single and dual-axis tracking photovoltaic...

Solar tracking system efficiency is affected by climate variability, and adaptive mechanisms must be employed to maximize energy output. Conventional fixed-tilt, single-axis, and dual-axis tracking techniques are not real-time adaptive, resulting in energy loss. This paper introduces COMLAT (Climate-Optimized Machine Learning Adaptive Tracking), an AI solar tracking system that employs climate prediction using CNN-LSTM for climate prediction, XGBoost for estimation of energy yield, and Deep Q-Learning (DQL) for real-time tracking control for solar efficiency optimization. One-year experimental research from January 2024 to January 2025 was conducted at Sitapura, Jaipur, India, with comparative studies of COMLAT and traditional tracking systems for seasonal variations and cloud cover conditions. Results confirm the 55% increase in energy production compared to fixed-tilt installations and 15-20% compared to dual-axis tracking due to its AI-based flexibility. The constructed model achieved 10-day solar irradiance forecasting with an RMSE of 23.5 W/m², outperforming the conventional LSTM and GRU baselines. XGBoost made predictions of energy output with an R² score of 0.94. COMLAT's reinforcement learning controller optimized tracking angles with sub-second latency while minimizing mechanical movement. The integration of hybrid artificial intelligence models allows COMLAT to continuously update its tracking angles in real time and is a scalable and industrially viable solution for smart grids, solar farms, and hybrid renewable energy systems. Increasing computational efficiency, integrating energy storage mechanisms, and optimizing self-learning algorithms will be areas of focus for future research to make it more applicable.