Improving prediction of solar radiation using Cheetah Optimizer and Random Forest

Area of Science:

• Renewable Energy Systems
• Machine Learning Applications
• Environmental Monitoring

Background:

• Accurate solar radiation (SR) prediction is crucial for efficient renewable energy generation and thermal systems.
• Machine learning (ML) models offer high precision and computational efficiency for complex SR forecasting tasks.
• Existing ML models require optimization for feature selection to enhance prediction accuracy.

Purpose of the Study:

• To introduce an innovative SR prediction model, the Cheetah Optimizer-Random Forest (CO-RF).
• To utilize the Cheetah Optimizer (CO) for optimal feature selection for hourly SR forecasting.
• To evaluate the CO-RF model's performance against established ML techniques.

Main Methods:

• Developed a hybrid model integrating Cheetah Optimizer (CO) for feature selection and Random Forest (RF) for prediction.
• Applied the CO-RF model to two distinct publicly available solar radiation datasets.
• Validated model performance using Mean Absolute Error (MAE), Mean Squared Error (MSE), and R-squared (R2) metrics.

Main Results:

• The CO-RF model demonstrated superior performance in both training and testing phases across both datasets.
• Achieved a low MAE of 0.0365 and MSE of 0.0074 with an R2 of 0.9251 on the first dataset.
• Attained an MAE of 0.0469 and MSE of 0.0032 with an R2 of 0.9868 on the second dataset, indicating significant error reduction.

Conclusions:

• The proposed CO-RF model significantly outperforms Logistic Regression, Support Vector Machine, Artificial Neural Network, and standalone Random Forest.
• The CO-RF model offers a highly accurate and reliable solution for hourly solar radiation forecasting.
• This advancement supports enhanced efficiency and dependability in solar energy systems.