<p>Accurately quantifying energy consumption and carbon emissions from buildings at the urban scale is fundamental for developing effective energy conservation and emission reduction strategies. However, traditional building-by-building simulation methods are often too computationally prohibitive for city-scale assessments, creating a critical barrier to this essential quantification. This study addresses this challenge by developing and validating an urban building energy consumption proxy model that integrates a three-dimensional geographic information system, physics-based energy simulation, and machine learning. Using the Shanhe Bay Valley in Beijing as a case study, a dataset with data from 600 building simulations was generated to train and select the optimal predictive model from four algorithms. The main findings of this study are threefold. First, the validated model effectively quantified the district's energy performance, predicting an average annual operational carbon emission intensity of 74.40&#xa0;kg/(m²·a). Second, a sensitivity analysis identified the five most influential drivers—per capita occupied area, ventilation frequency, solar heat gain coefficient, lighting power density, and equipment power density—providing clear targets for energy-saving interventions. Third, the proxy model demonstrated remarkable efficiency, completing the energy assessment for all buildings within the 13.2&#xa0;km² district in just 4&#xa0;s, a task that could take days with traditional methods. This integrated approach effectively removes the computational barrier, enabling rapid and reliable quantification of urban energy patterns and providing actionable insights for prioritizing optimization strategies in urban planning and policy.</p>