Synthetic data generation and energy consumption prediction in district building energy modeling

This study presents a novel district building energy model by integrating the strengths of dynamic and statistical models. This multitier model can generate hourly synthetic energy consumption data for urban building stocks by incorporating building characteristics and local weather data and predict hourly building energy consumption. The methodology involves the following steps: (1) A dynamic model is created, and its key parameters are calibrated according to monthly metered data. (2) The calibrated dynamic model is then utilized to generate synthetic hourly energy consumption data. (3) Finally, statistical models are trained on synthetic data to predict hourly building energy consumption. Once the proposed methodology is tested on a university campus, the calibration reduces the monthly simulation error to an 11.9% Coefficient of Variation of the Root Mean Square Error (CV-RMSE) based on available metered energy consumption data, and the final statistical model predicts the hourly building energy consumption with a 1.5% CV-RMSE. This multitier model offers valuable insights for urban planners in identifying high-demand areas and implementing energy-efficient interventions by generating synthetic hourly energy consumption data.