Machine Learning Based Modeling For Effective And Accurate Prediction Of Solar Radiationin Saudi Arabia

Abstract

With the increase in population size, the demand for energy is expected to rise drastically. As a result, more resources will be needed to be utilized to meet increased electricity production demands. Therefore, the use of renewable energy resources like solar energy captured by photovoltaic (PV) panels is being considered as an alternative to non-renewable resources. However, solar radiation is subjected to changing weather conditions which introduces uncertainty in the amount of electricity that can be generated. To overcome this uncertainty, this thesis compares the performance of Decision Tree (DT), Artificial Neural Network (ANN), and ensemble methods including Random Forest (RF) and Gradient Boosting (GB) in predicting the Global Solar Radiation (GSR) of the next 30 minutes intervals. National Renewable Energy Laboratory (NREL) solar radiation dataset of Jeddah and Qassim cities was used. A comprehensive literature review of related studies on solar radiation prediction was performed. In addition, two data partitions methods for the 5-year data were used; 4 years for training and 1 year for testing, and a random split of 70% for training and 30% for testing, and their performance was compared. The parameters of DT, ANN, RF, and GB were tuned to get the optimized performance. In addition, correlation-based feature elimination was used for feature selection. Moreover, an analysis of how well the proposed solar radiation prediction models generalize to other locations was performed. Finally, from the experiments it was found that the ensemble models (RF and GB) performed better than the single machine learning models (ANN and DT).