Solar radiation forecasting using MARS, CART, M5, and random forest model: A case study for India.

Area of Science:

• Renewable Energy
• Atmospheric Science
• Data Science

Background:

• Solar radiation is a variable input for solar power generation.
• Accurate solar power output forecasting requires reliable solar radiation predictions.
• Short-term solar radiation forecasting aids utility and Independent System Operator (ISO) bidding processes.

Purpose of the Study:

• To perform hourly solar radiation forecasting from 1 to 6 days ahead.
• To evaluate the performance of multiple machine learning models for this forecasting task.
• To identify the most effective model for short-term solar radiation prediction.

Main Methods:

• Utilized historical solar radiation data from Gorakhpur, India.
• Employed Multivariate Adaptive Regression Splines (MARS), Classification and Regression Trees (CART), M5, and Random Forest models.
• Conducted hourly forecasting for lead times of 1 to 6 days.

Main Results:

• The Random Forest model yielded the best forecasting accuracy.
• The CART model exhibited the poorest performance among the evaluated models.
• Model performance varied across different forecasting horizons.

Conclusions:

• Random Forest is a highly effective model for short-term solar radiation forecasting.
• The study highlights the importance of model selection for accurate solar energy prediction.
• Findings support improved grid management and energy market participation for solar power plants.