A scalable approach for predicting the energy produced by a moving solar panel in the urban environment: a City Scanner case study

Estimating the energy produced by solar panels has been a crucial research field in the last fifty years, driven by the impellent need of finding alternative solutions to fossil fuels. The increasing affordability and technological advancements in this area, along with the exponentially growing number of IoT devices, allowed to create systems disconnected from the electrical grid capable of self-sustaining relying solely on the sun's energy. This research presents a scalable approach for forecasting solar irradiation and yield of a moving solar panel, taking into account dynamic location changes and the influence of the city shading that comes along with these different environments. The presented machine learning model allows predicting the \acrfull{GHI} with an R2 score higher than 0.90, and for each of its output, a Google Street View imaging method is employed to assess the obstruction caused by buildings and trees, to have a more precise measurement at street level. The validation of the selected approach is performed on data collected in The Bronx, NY, during deployment of City Scanner, drive-by sensing platform conceived by Senseable City Lab. For its scalable nature, this approach can be ported in other cities with minimal effort and applied to solar panel equipped fleets of the most diverse types, from drones to electric vehicles, to better assess in advance their energetic potential in an untested environment.