Very short-term prediction of photovoltaic energy in the winter building for an automatic energy management system

In the context of the energy transition, renewable energies have an important role to play. This is particularly true of photovoltaic (PV) energy. The use of PV energy in buildings is becoming increasingly common nowadays. Buildings integrated PV (BIPV) represent a major advantage in this respect, thanks to their high PV energy harvesting capacity. However, the amount of PV energy is not fixed, since it depends on the climate. Therefore, it is essential to have effective means of predicting production in advance. This will enable us to arrange before we run out of electricity. In this paper, we propose deep learning models to predict PV energy. These are the Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). We have also used a hybrid model combining these two types of models. We propose a very short-term prediction, corresponding to a prediction every 5 minutes in advance. The advantage of this type of prediction is that we can propose a means of automating the power supply to the building in question according to the predicted result. To this end, we discuss future perspectives with the focus on the automation power supply. This study was applied in a Positive Energy Winter House (PEWH). The data collected from this building enabled us to develop our artificial intelligence models for PV prediction . Our study shows that the combination of models performs better in very short term PV energy prediction for winter buildings using the univariate approach.