Solar potential assessment using machine learning and climate change projections for long-term energy planning

This work proposes a novel method for evaluating solar potential, essential for the development, installation, and operation of solar power systems. The approach forecasts solar energy potential for specific sites by utilizing integrated geospatial, meteorological, and infrastructural multidimensional data. A new application has been released to assess the solar capacity globally. The study evaluated various machine learning methods, ultimately selecting an XGBoost model for training on historical sun irradiance and meteorological data spanning from 1980 to 2015. This model demonstrates significant promise for handling complicated nonlinear interactions and simulating temporal weather patterns affecting solar irradiance. Preliminary results indicate a strong capacity for worldwide predictions on the potential of solar energy, utilizing simulated weather data from 2015 to 2099. The application delivers precise solar power estimates and financial viability, enabling rapid and effortless site assessments from any location within minutes. The results demonstrate that the XGBoost model outperforms other ML algorithms, by achieving lower values of RMSE = 0.97 kWh/m² and MAE = 0.76 kWh/m², respectively, for solar energy potential. Furthermore, to evaluate the impact of the proposed methodology, three case studies were conducted in Mindanao (Philippines), Gobi-Altai (Mongolia), and the Peloponnese (Greece). The results demonstrate the efficacy of the proposed method in long-term solar energy planning.