An integrated framework for assessing solar photovoltaic potential of building surfaces at city scale using parametric simulation and optimized machine learning models

Efficient utilization of building-integrated photovoltaics is an important pathway for achieving sustainable urban development. However, existing methods for assessing solar photovoltaic potential of large urban building surfaces suffer from issues such as coarse modeling, low prediction accuracy, and incomplete assessments. To address these challenges, this study proposes a novel solar photovoltaic potential assessment method for large cities, which builds a high-accuracy NSGA-II-ANN predictive model using Ladybug Tools and Machine Learning techniques to predict and optimize photovoltaic panel parameters. Taking Xi'an, China, as an example, the study calculates and evaluates solar photovoltaic potential of building surfaces from multiple dimensions. The main findings are as follows: (1) The solar photovoltaic potential of building surfaces in Xi'an is significant, with all roof shading rates below 15 %, and the average solar radiation intensity reaching 1020.42 kWh/m², offering great utilization value. (2) The NSGA-II-ANN predictive model constructed for photovoltaic panel parameters has R² values greater than 0.960, MSE values less than 0.04, and the loss curve demonstrates clear convergence characteristics. (3) After optimization, the maximum solar radiation potential of rooftop photovoltaic systems in Xi'an can reach 59.398 TWh, with 25.534 TWh in summer and 14.055 TWh in winter. (4) The maximum photovoltaic generation capacity for building surfaces in Xi'an ranges from 18.27 to 22.84 TWh, potentially meeting 46.88 % of the city's annual electricity demand or up to 175.99 % of residential electricity consumption. This research framework and findings provide a more practical assessment of solar photovoltaic potential in large cities, offering recommendations and strategies for urban photovoltaic utilization.