Machine learning techniques for estimation of BIPV production

Abstract

This study evaluated the performance of nine ANNs to estimate the electricity production of a BIPV system from meteorological data that are generally accessible from ground-based meteorological stations. Solar irradiance was proved to be the most adequate input variable to predict PV production with a simple ANN. However, the accuracy of this simple model was notably improved with the inclusion of the temperature and relative humidity. Wind speed and direction were less relevant as their statistical indicators highlighted. Indeed, all reviewed ANN structures shown good performance as the ܴܯܵܧ and ܧܤܯ kept relatively low. Besides, the largest number of neurons does not lead to a better performance.