Addressing the challenges posed by climate change and meeting urban energy demands is of utmost importance in today's world. Building Integrated Photovoltaics (BIPV) emerges as a crucial solution for energy conservation and carbon emissions reduction in urban environments. However, traditional methods of assessing solar radiation on buildings using physical models are often computationally intensive and time-consuming. This paper introduces a novel hybrid approach that integrates physical model-based solar radiation calculation with machine learning techniques to analyze Solar Radiation Potential (SRP) across city-wide building infrastructure. The proposed approach precisely evaluates the SRP of representative blocks by leveraging computing-intensive physical models integrated with 3D building data. Subsequently, two machine learning models are developed to effectively predict the SRP of building roofs and facades across the entire city. To validate the efficacy of this approach, an experiment was conducted in Shenzhen, China, yielding insightful results. The findings reveal that Shenzhen has a huge potential for BIPV solar power generation, with mean annual total building roof and facade solar radiation values of 9.22*10⁷kwh and 2.47*10⁸kwh, respectively. It can be further observed that relying solely on rooftop installations is insufficient to meet electricity demand. This study not only provides an innovative alternative for city-wide SRP estimation by combining physical modeling and machine learning but also offers valuable insights for fostering low-carbon urban environments and informing data-driven and model-driven urban planning and management strategies.