Building Energy Surrogate Modelling Methodology For A Detached Single-Family Century Home Archetype In Toronto, ON

A surrogate model was developed for a detached archetypal home in Toronto, ON. EnergyPlus was used to perform 1500 simulations within a design space defined by 23 input parameters with ranges based on field study data. Elastic net regression was used to create a surrogate model to predict annual energy use and to perform embedded feature selection. An analysis comparing house size to model performance found that including both small and large homes did not decrease the model accuracy. The final regression model predicted energy use with an average R² of 0.946 and MAPE of 6.1% using nested cross- validation. A case study predicted actual annual energy use of two homes in Toronto within 10% error of utility bill data. A preliminary optimization analysis found that several weeks of simulation time could be saved and more optimal solutions could be discovered compared to a brute-force forward stepwise selection optimization.