Automatic mapping of solar panels and generation of solar forecasts through aerial imagery and machine learning
Solar Photovoltaics (PV) is rapidly deployed in Sweden, mostly in terms of small decentralized systems on buildings. So far it has been difficult for authorities in Sweden to keep records on how much and where PV is installed. Currently new systems are reported to the Swedish Energy Agency (SEA) by the distribution system operators (DSO). While the reporting system matures, SEA suspects older systems are incorrectly reported or not at all. However, for both new and old systems, the exact location is not reported, only in what electric network area. This information is useful for various authorities, e.g., fire departments or the Swedish Civil Contingencies Agency.
Therefore, this research project focuses on the development and evaluation of methods for identifying solar energy systems (PV and thermal) from aerial or satellite images using machine learning. Statistics about installed systems could the be updated after each aerial/satellite image survey, and thus this automatic process will help authorities who need this information.
Also, by estimating the slope, azimuth and shading of a system using LiDAR (Light Detection and Ranging) it is possible to get an idea of the size or installed capacity and possibly also improve forecasts of PV power generation. This is useful both for the DSO and the transmission system operator as PV penetration increase, which may pose challenges for maintaining power quality and balance of power.
Swedish Energy Agency
2020-08-03 - 2022-12-31
Johan Lindahl (Becquerel Sweden)
David Lingfors (Uppsala University)
Âzeddine Frimane (Uppsala University)