Development of a Comprehensive Intelligent Model for Diagnostics and Forecasting of the Energy Efficiency of Photovoltaic Stations for Agricultural Facilities

This study develops an intelligent real-time diagnostic and forecasting model for photovoltaic (PV) stations operating in the rural regions of Kazakhstan. Field measurements show that in Kazakhstan’s rural conditions, daily temperature fluctuations of 18–27 °C reduce PV output by 10–15%, solar irradiance variations of 800–1100 W/m2 cause an additional 8–12% deviation, and mineral-rich dust accumulation of 20–120 μm results in 3–5% power loss per 10 μm layer. Using the Beer–Lambert–Bouguer law, thermal degradation coefficients, and inverter droop models, the physical–mathematical behavior of PV degradation was described. The hybrid AI model (LSTM + XGBoost + Random Forest) achieved 89–91% accuracy in 24 h forecasting and 88–93% accuracy in fault detection. Overall, the proposed system reduces energy losses by 10–15% and shortens maintenance time by 18–24%, improving the reliability of rural PV stations in Kazakhstan.