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

Area of Science:

• Renewable Energy Systems
• Geospatial Analysis
• Meteorological Forecasting

Background:

• Accurate solar potential assessment is crucial for solar power system development, installation, and operation.
• Existing methods may lack the precision and global scope required for comprehensive site evaluation.
• Integrating diverse data sources is key to improving solar energy potential predictions.

Purpose of the Study:

• To propose a novel method for evaluating solar energy potential.
• To develop a global application for assessing solar capacity.
• To enhance the accuracy and efficiency of solar site assessments.

Main Methods:

• Utilized integrated geospatial, meteorological, and infrastructural multidimensional data.
• Trained an XGBoost machine learning model on historical solar irradiance and meteorological data (1980-2015).
• Validated the model using simulated weather data (2015-2099) and conducted case studies in the Philippines, Mongolia, and Greece.

Main Results:

• The XGBoost model demonstrated superior performance in handling complex nonlinear interactions and temporal weather patterns.
• Achieved low error metrics: Root Mean Square Error (RMSE) = 0.97 kWh/m² and Mean Absolute Error (MAE) = 0.76 kWh/m².
• The developed application provides precise solar power estimates and financial viability assessments within minutes.

Conclusions:

• The proposed methodology effectively forecasts solar energy potential globally.
• The XGBoost model is highly effective for long-term solar energy planning and site assessment.
• The application enables rapid, effortless, and accurate site evaluations worldwide.