Forecasting of solar radiation for a cleaner environment using robust machine learning techniques.

Area of Science:

• Renewable Energy Systems
• Climate Science
• Machine Learning Applications

Background:

• Growing global demand for renewable energy necessitates accurate solar radiation forecasting.
• Intermittency of solar power requires advanced prediction methods for grid stability.
• Depleting fossil fuels and emissions targets drive research into solar energy solutions.

Purpose of the Study:

• To develop and compare machine learning models for accurate solar radiation forecasting under various climatic conditions.
• To identify the most effective algorithms for predicting Global Horizontal Irradiance (GHI).
• To create a user interface for real-time solar prediction applications.

Main Methods:

• Utilized the National Solar Radiation Database, including features like temperature, humidity, and wind speed.
• Trained multiple machine learning algorithms (MLR, SVR, DTR, RFR, GBR, LSTM) on data segmented by climate type.
• Evaluated model performance based on error approximation and accuracy metrics.

Main Results:

• Long Short-Term Memory (LSTM) achieved the lowest error loss (0.0040) at the 100th epoch.
• For machine learning models, Gradient Boosting Regression (GBR) and Random Forest Regression (RFR) demonstrated high performance.
• GBR outperformed RFR by 2% in hot weather, while RFR was 1% more accurate in cold, autumn, and monsoon climates.

Conclusions:

• Advanced machine learning techniques, especially LSTM, offer highly accurate solar radiation forecasting.
• The developed models provide valuable tools for real-time solar prediction, grid management, and strategic solar energy planning.
• The user interface facilitates practical applications for load operators, engineers, and researchers.