Fast Evaluation of Rooftop and Façade PV Potentials Using Backward Ray Tracing and Machine Learning

• Published in: 2021 IEEE 48th Photovoltaic Specialists Conference (PVSC) pp. 0294 - 0299
• Main Authors: Bredemeier, Dennis, Schinke, Carsten, Gewohn, Timo, Wagner-Mohnsen, Hannes, Niepelt, Raphael, Brendel, Rolf
• Format: Conference Proceeding
• Language: English
• Published: IEEE 20.06.2021
• Subjects: Buildings; Computational modeling; Machine learning; Photovoltaic systems; PV potential analysis; Radiation effects; ray tracing; Support vector machines; Urban areas
• DOI: 10.1109/PVSC43889.2021.9518660

The accurate calculation of annual insolation for a large amount of building surfaces in urban environments is a key prerequisite for a targeted increase of the installed PV capacity. In this study, we build a digital city model upon publicly available geographic data. We use this model in combination with machine learning for performing PV potential analyses on large scales and at low computational costs. We train the machine learning algorithm using ground truth values for insolation, which we determine from forward ray tracing calculations for the city of Hanover, Germany. We find that our machine learning approach is able to predict the annual insolation for all types of building surfaces independent from their tilt and orientation. The RMSE relative to the mean ground truth value is 3.2% for rooftops and 8.8% for facades. The calculation time is reduced by a factor of 20 compared to the forward ray tracing approach.