Solar energy prediction through machine learning models: A comparative analysis of regressor algorithms.

Area of Science:

• Renewable Energy Systems
• Machine Learning Applications
• Environmental Science

Background:

• Solar energy from photovoltaic panels is a growing global trend with significant environmental and societal benefits.
• The intermittent nature of solar power necessitates accurate forecasting for grid stability, energy storage optimization, and demand-supply balancing.
• Distributed solar systems further increase the need for reliable energy prediction models.

Purpose of the Study:

• To evaluate the performance of five machine learning models for solar energy forecasting.
• To identify the key meteorological factors influencing solar energy generation.
• To determine the most accurate model for predicting solar energy output.

Main Methods:

• Utilized a dataset comprising 21,045 samples.
• Employed five regression models: Gradient Boosting Regressor (GB), XGBoost, K-neighbors Regressor (KNN), LightGBM, and CatBoost Regressor (CatBoost).
• Input features included Humidity, Ambient temperature, Wind speed, Visibility, Cloud ceiling, and Pressure. Model accuracy was assessed using R2, RMSE, and MAE.

Main Results:

• The CatBoost model demonstrated superior performance in solar energy prediction.
• CatBoost achieved an R2 of 0.608, RMSE of 4.478 W, and MAE of 3.367 W during training.
• On testing data, CatBoost yielded an R2 of 0.46, RMSE of 4.748 W, and MAE of 3.583 W. SHAP analysis indicated ambient temperature and humidity as primary influencing factors.

Conclusions:

• The CatBoost model is highly effective for solar energy forecasting.
• Ambient temperature and humidity are critical predictors of solar energy generation.
• Accurate forecasting using machine learning enhances the integration and management of solar energy in power grids.