Solar energy is a sustainable source that is favored in tropical areas where the resource intensity is high. However, developing accurate forecasting models, which is crucial in the optimal design and operation of tropical solar energy systems, has been challenging due to the high-dimensional nature of the system. This study presents a novel forecasting model for such systems using support vector machines. The proposed model was developed using a data-driven methodology. Model optimization and feature selection were applied to improve predicting accuracy. Modeling results show that the medium Gaussian function provided the most desirable balance between the accuracy and speed of the training functions. The previous hour observation was found to be the most significant input, while some variables considered in the initial model caused overfitting. The final model had superior accuracy over the initial models and those developed in the literature, thereby validating the effectiveness of the presented methodology.
Keywords Renewable energy, solar energy systems, energy system modeling, forecasting model, machine learning, support vector regression