Intraday Solar Irradiance Forecasting Based on Hybrid Machine Learning Methodology for Photovoltaic Power Applications

Accurate prediction of solar radiation provides significant potential for enhancement of smart grid distribution networks efficiency and electricity management. However, the inherent non-stationary behavior and unpredictability render its estimation a challenging task. In this respect, this paper investigates the potential of ensembled-based machine learning (ML) models in contrast to individual regression models for predicting solar irradiance using meteorological variables. Multiple ensemble models are assessed using simple averaging, combining Artificial Neural Network (ANN), Support Vector Machine Regression (SVMR), and Decision Tree (DT). The comparative results demonstrate that the ensembled model comprising ANN and DT showed improved prediction accuracies for an hour ahead forecasting, surpassing both individual ML models and other ensembled algorithms.