Surrogate models for evaluating HVAC retrofit options in multi-unit residential buildings

This study investigates the use of surrogate modeling to optimize retrofit strategies in multi-unit residential buildings (MURBs), including HVAC systems. A comprehensive synthetic dataset was generated using EnergyPlus simulations, parameterized across a wide range of building and system variables, including ten distinct HVAC configurations. Multiple surrogate modeling approaches were evaluated, including single-output and multi-output artificial neural networks (ANNs) as well as Light Gradient Boosting Machine (LGBM) models. The models were trained to predict key performance metrics: Energy Use Intensity (EUI), Thermal Energy Demand Intensity (TEDI), and Cooling Energy Demand Intensity (CEDI). Results show that multi-output ANN models, with HVAC system as a categorical input, achieved high accuracy (R 2 > 0.997 and RMSE