Global Horizontal Solar Irradiance Forecasting Based on Data-Driven and Feature Selection Techniques

Abstract

AbstractWith the rapidly expanding infrastructure of the solar energy system, the need for an accurate solar prediction has become an essential part of the renewable energy sector. Over the past decade, various machine learning (ML) algorithms have been used for this purpose. Although the prediction of solar irradiance forecasting has been discussed in a large number of studies, the use of meta-heuristic optimization techniques has not been explored to select features for the forecasting model. This study comprises two meta-heuristic optimization techniques such as simulated annealing (SA) and ant colony optimization (ACO) for feature selection. The results show that feature selection based on meta-heuristics gave better results than models without feature selection. Amongst the two optimization methods, ACO outperformed SA with some exceptions. For SA, the declining order of performance observed is extreme gradient boosting (XGBoost), random forest (RF), multilayer perceptron (MLP) , decision tree (DT) and support vector regression (SVR), while for ACO the declining order observed is XGBoost followed by MLP, RF, DT and SVR. This manuscript indicates the potential capability of meta-heuristic techniques for accurate prediction of global horizontal irradiance (GHI) given a wide array of feature variables.KeywordsAnt colony optimizationData-driven modelsGlobal horizontal irradianceFeature selectionShort-term forecasting