Assessment of predicting hourly global solar radiation in Jordan based on Rules, Trees, Meta, Lazy and Function prediction methods

Abstract

Nowadays, predicting solar radiation is widely increased to maximize the efficiency of solar systems globally. Meteorological data from metrological stations is used to implement the intelligent prediction systems. Unfortunately, uncertainty in the used solar variables and the selected prediction models would increase the difficulties in using intelligent models to predict solar radiation. Several studies perfectly estimated solar radiation using only time and date variables. The main objective of this study is to review different prediction methods in predicting the solar radiation of Jordan. To achieve this target, five main methods including Rules, Trees, Meta, Lazy and Function Methods are selected, and then the most important and used algorithms in each method are selected to build a prediction model. The study shows that M5Rule, Random forest, Random committee, Instance Based Learning with Parameter K and multi-layer perceptron are the best algorithms in Rules, Trees, Meta, Lazy, and Function Methods respectively. Random forest algorithm performed better than other algorithms in predicting global solar radiation. The results of the analysis found that the accuracy of prediction depends on the used category, training algorithm and variables combinations.