Improving prediction of solar radiation using Cheetah Optimizer and Random Forest

Abstract

In the contemporary context of a burgeoning energy crisis, the accurate and dependable prediction of Solar Radiation (SR) has emerged as an indispensable component within thermal systems to facilitate renewable energy generation. Machine Learning (ML) models have gained widespread recognition for their precision and computational efficiency in addressing SR prediction challenges. Consequently, this paper introduces an innovative SR prediction model, denoted as the Cheetah Optimizer-Random Forest (CO-RF) model. The CO component plays a pivotal role in selecting the most informative features for hourly SR forecasting, subsequently serving as inputs to the RF model. The efficacy of the developed CO-RF model is rigorously assessed using two publicly available SR datasets. Evaluation metrics encompassing Mean Absolute Error (MAE), Mean Squared Error (MSE), and coefficient of determination (R2) are employed to validate its performance. Quantitative analysis demonstrates that the CO-RF model surpasses other techniques, Logistic Regression (LR), Support Vector Machine (SVM), Artificial Neural Network, and standalone Random Forest (RF), both in the training and testing phases of SR prediction. The proposed CO-RF model outperforms others, achieving a low MAE of 0.0365, MSE of 0.0074, and an R2 of 0.9251 on the first dataset, and an MAE of 0.0469, MSE of 0.0032, and R2 of 0.9868 on the second dataset, demonstrating significant error reduction.