Photovoltaic power forecast for the next 24 hours with an analytical model and a FFNN model

Photovoltaic (PV) self-consumption installations have increased by 101.84% in Spain, from 2020 to 2021. In this context, developing PV generation forecasting tools can contribute to increase the PV self-consumption ratio, boosting the use of renewable energies. On this matter, this paper presents two PV generation forecasting models for the next 24 hours. These are an analytical model developed in OpenModelica software and a model based on artificial intelligence (AI), specifically a feedforward neural network (FFNN). Both models use measured meteorological data obtained from a weather station 3km from the PV installation. This work analyses how the use of different data as input information affects the prediction of the FFNN. It was found that adding a time vector as an input of the FFNN improves the prediction, thus compensating the fact that it is not a recurrent network. Furthermore, the behaviour of both models has been compared. Both the analytical model and the FFNN obtain a correlation coefficient r of 0.941 and 0.94, respectively. Even so, the MAE and RMSE metrics highlight how the analytical model has a higher error and level of dispersion.