A district-scale spatial distribution evaluation method of rooftop solar energy potential based on deep learning

Abstract

Accurate evaluation of rooftop solar potential is increasingly important in sustainable urban development. However, accurately evaluating the solar photovoltaic (PV) potential of rooftops in urban areas is a challenge due to the diversity of urban rooftop outlines and rooftop obstacles. This study proposes a generic framework for evaluating the potential of urban rooftop solar PV that integrates deep learning and geographic information systems (GIS). GIS is used to extract information about land use types and classify buildings based on land use types. A deep learning-based method for calculating the rooftop available area of multi-type buildings is proposed.To validate the proposed methodology, an area in Wuhan containing a variety of building features was used, combined with the utilization of the available rooftop area to estimate the solar photovoltaic potential, and solar irradiance measurement experiments were set up for validation. The results show that the classification of rooftops by building type can effectively improve the accuracy of rooftop availability identification, and the classification process improved the identification of rooftops by 12.68% and obstacles by 12.42% overall. The annual solar potential of urban rooftops in Hanyang District is 8864.37 GWh/year.