Enhanced Global Horizontal Irradiance Forecasting: A Hybrid Machine Learning Approach

Background The integration of solar power into national and municipal energy grids is a key strategy in the global transition toward sustainable energy sources. Reliable solar irradiance forecasting is fundamental to this integration, ensuring the stability, efficiency, and cost-effectiveness of solar power systems. Global Horizontal Irradiance (GHI), a core parameter in assessing solar potential, is influenced by various meteorological and environmental factors, making its prediction a complex task. Traditional statistical methods often fall short in capturing the nonlinear and dynamic patterns of solar irradiance. Machine learning (ML) techniques, especially ensemble and hybrid models, have shown promising results in this context. This study focuses on enhancing GHI prediction accuracy through a hybrid ML model. Methods To design the proposed forecasting model, historical GHI data was used for training and testing the models. A hybrid model combining K-Nearest Neighbors (KNN) and Extreme Gradient Boosting (XGB) was designed to leverage the local approximation power of KNN and the robust, nonlinear learning capability of XGB. The model's performance was benchmarked against individual traditional ML models. In addition, various input feature combinations were tested to assess the impact of feature selection on forecasting performance. The models were evaluated using standard metrics: Root Mean Square Error (RMSE) and the coefficient of determination (R²). Results The proposed kNN-XGB hybrid model significantly outperformed all the individual algorithms tested. It achieved an RMSE of 24.18% and an R² value of 99.94%, indicating exceptional prediction accuracy. Feature selection analysis revealed that selecting the most relevant features can significantly improve model performance, while improper feature selection can degrade prediction accuracy, emphasizing the importance of optimizing input variables. Conclusions This study confirms that hybrid models for GHI forecasting outperform traditional models and demonstrates that selecting the most relevant features is critical for enhancing model performance and prediction accuracy.