Forecasting Minute Averaged Solar Irradiance Using Machine Learning for Solar Collector Applications

Abstract

Challenges in utilising fossil fuels for generating energy call for the use of renewable energy. This study focuses on modelling and forecasting solar energy and optimum tilt angle of solar energy acceptance using historical time series data collected from one of the South African radiometric stations, USAid Venda station in Limpopo province. In the study we carried out a comparative analysis of Random Forest and Bayesian linear regression in short-term forecasting of global horizontal irradiance (GHI). To compare the predictive accuracy of the models, k-Nearest Neighbors (KNN) and Long short-term memory (LSTM) are used as benchmark models. The top two models with the best performances were then used in hourly forecasting of optimum tilt angles for harvesting solar energy. The performance measures such as MAE, MSE, and RMSE were used and the results showed RF to have better performance in forecasting GHI than other models, followed by the LSTM and the third best model was the KNN whereas the BLR was the least performing model. RF and LSTM were then used in modelling and forecasting the tilt angles of optimal solar energy acceptance and as thus, the LSTM outperformed the RF by a small margin.