Machine Learning for Solar Resource Assessment Using Satellite Images

Abstract

Understanding solar energy has become crucial for the development of modern societies. For this reason, significant effort has been placed on building models of solar resource assessment. Here, we analyzed satellite imagery and solar radiation data of three years (2012, 2013, and 2014) to build seven predictive models of the solar energy obtained at different altitudes above sea level. The performance of four machine learning algorithms was evaluated using four evaluation metrics, MBE, R2, RMSE, and MAPE. Random Forest showed the best performance in the model with data obtained at altitudes below 800 m.a.s.l. The results achieved by the algorithm were: 4.89, 0.82, 107.25, and 41.08%, respectively. In general, the differences in the results of the machine learning algorithms in the different models were not very significant; however, the results provide evidence showing that the estimation of solar radiation from satellite images anywhere on the planet is feasible.