Towards deep learning methods to improve photovoltaic prediction and building decarbonization in benchmarking study

Abstract

High energy demand, energy transition, energy consumption control are challenges for the future, especially for Building Integrated Photovoltaic (BIPV). There is a great potential to harvest large amounts of photovoltaic (PV) energy on horizontal and vertical surfaces. However, this high potential is often hindered by the slow deployment of these panels, the complex integration into existing buildings, and the possible complex interactions between different factors, such as visualization and active projection of buildings in the decarbonization process. Building Information Modeling (BIM) offers complete and real generative building data that is used in our deep learning methods. Indeed, there is currently no framework for design linking photogrammetry, BIM and PV for BIPV. In this work, we propose artificial learning models, such as Deep Learning, to predict PV energy production for BIPV decarbonization. We determined the optimal prediction of PV production by testing and evaluating different models on a building case study. We compared the PV power generation prediction results with 3D simulation software for solar architecture.