Photogrammetry and deep learning for energy production prediction and building-integrated photovoltaics decarbonization

Abstract

Building-Integrated photovoltaics (BIPV) have emerged as a promising sustainable energy solution, relying on accurate energy production predictions and effective decarbonization strategies for efficient deployment. This paper presents a novel approach that combines photogrammetry and deep learning techniques to address the problem of BIPV decarbonization. The method is called BIM-AITIZATION referring to the integration of BIM data, AI techniques, and automation principles. It integrates photogrammetric data into practical BIM parameters. In addition, it enhances the precision and reliability of PV energy prediction by using artificial intelligence strategies. The primary aim of this approach is to offer advanced, data-driven energy forecasts and BIPV decarbonization while fully automating the underlying process. To achieve this, the first step is to capture point cloud data of the building through photogrammetric acquisition. This data undergoes preprocessing to identify and remove unwanted points, followed by plan segmentation to extract the plan facade. After that, a meteorological dataset is assembled, incorporating various attributes that influence energy production, including solar irradiance parameters as well as BIM parameters. Finally, machine and deep learning techniques are used for accurate photovoltaic energy predictions and the automation of the entire process. Extensive experiments are conducted, including multiple tests aimed at assessing the performance of diverse machine learning models. The objective is to identify the most suitable model for our specific application. Furthermore, a comparative analysis is undertaken, comparing the performance of the proposed model against that of various established BIPV software tools. The outcomes reveal that the proposed approach surpasses existing software solutions in both accuracy and precision. To extend its applicability, the approach is evaluated using a building case study, demonstrating its ability to generalize effectively to new building data.