Energy and Operational-Carbon Coupling for Sustainable Beijing Hotel Design: Parametric EnergyPlus Simulation and Machine-Learning Surrogate Analysis

Improving the operational energy and carbon performance of hotel buildings is an important environmental dimension of sustainable building design. This study develops a reproducible EnergyPlus-based framework that combines parametric simulation, interpretable sensitivity analysis, machine learning surrogates, and carrier-resolved operational carbon accounting for Beijing hotel buildings. From 20,000 Latin hypercube candidates, input-side physical and functional screening retained 4640 successful EnergyPlus simulations. The simulated EUI mean was 140.6 kWh/(m2·a), 14.3% above the published Beijing hotel mean; surrogate performance is therefore interpreted as fidelity to the simulator rather than direct measured-building prediction. SRC with bootstrap uncertainty and a SHAP cross-check identified the main domestic-hot-water, building-form, and HVAC drivers. Of 17 models, Poly3-RidgeCV achieved the highest held-out fidelity (R2 = 0.9976; RMSE = 1.72 kWh/(m2·a)). Baseline OCEI averaged 48.20 kgCO2e/(m2·a); EUI and OCEI were strongly correlated (r = 0.954) but not interchangeable, with 71.8% overlap between the top-10% low-EUI and top-10% low-OCEI cases. Emission-factor scenarios showed robust but non-static energy-carbon coupling. The framework supports early-stage comparison of energy-efficient alternatives with comparatively lower operational carbon emissions within the stated accounting boundary, contributing to sustainable hotel design without constituting a whole-life or measured-building sustainability assessment.