Forecasting photovoltaic energy for a winter house using a Hybrid Deep Learning Model

As part of the energy transition, controlling energy consumption is a challenge for everyone. To this end, a number of sustainable solutions are being proposed, notably for BIPV (Building Integrated Photovoltaics) buildings. In addition, artificial intelligence (AI) is an effective tool for analyzing photovoltaic (PV) energy production and consumption data. It will then be possible to predict the PV energy production of a BIPV building or any other system integrating PV panels. This paper presents the implementation of artificial learning models for the prediction of the very short-term energy production of PV panels in a Positive Energy Winter House (PEWH). These are methods based on multivariate time series, including Long Short- Term Memory (LSTM), Convolutional Neural Networks (CNN) and a hybrid model. In the case of a winter house with a long period of snow, accurate prediction of solar power output in the very short term is needed to face the fluctuation that can be caused by climate. In this study, we are working on a winter house located in Poschiavo. The proposed method is applied to data recorded in real time by PEWH's photovoltaic solar panels, and the results are compared and tested over period of time. The results confirm the validity of each proposed model in forecasting PV Energy.