PV DC Yield Determined by Deep Neural Networks: the Case of Building Integrated PV

PV systems can significantly contribute to achieving the climate neutral goal of the Paris agreement. However, they will present additional challenges to the power grids, due to their intermittent nature. This study aims to model the power output of Building Integrated PV systems (BIPV) on the basis of the application of a Deep Neural Network (DNN) to the following input variables: solar irradiance, module temperature as well as time and location. The DNN has been applied to a dataset containing over four years of data of electrical parameters on module level as well as meteorological data, all at a 5-minutes resolution. The results show that the proposed DNN is able to calculate the PV power output accurately with an R2 score of 0.96 and RMSE of 0.04. Though applied to a BIPV system in this case, the method will be applicable to a myriad of other types of monitored PV systems as well.