Application of a large smart thermostat dataset for model calibration and Model Predictive Control implementation in the residential sector

Electrification of space heating combined with electricity generation through renewable resources has a significant potential to reduce greenhouse gas emissions, especially when coupled with advanced control strategies that utilize building energy flexibility. This study analyses data from approximately 7,800 houses in Ontario and Québec, Canada to investigate the most suitable data length, data interval and calibration horizon of building models for assessment of Model Predictive Control strategies on a large scale. Overall, models with calibration horizon of 24 h, data length of 7 days and time interval of 15 min provided the best balance between accuracy and computational resources. The calibrated models were used to compare a simple deadband controller to a Model Predictive Controller that minimizes the cost of electricity for a cold day in January. The results showed that the Model Predictive Controller can systematically reduce the high-price energy consumption, while improving thermal comfort, by successfully preheating before the high-price periods. Preheating shifts the peak power consumption, reducing it by 15 % for Ontario and 30 % for Québec. If just 1,000 houses adopt optimal strategies, the grid could see an average reduction in energy consumption of up to 15 MWh and 11 MWh during high-price periods in Ontario and Québec respectively. •Data from 7,800 houses were used to calibrate building energy models for one-day ahead predictions.•The effect of training data length, data interval, and calibration horizon was studied on building models.•15-minute data intervals, 7 days of data and calibration horizon of one day showed the best results.•Model Predictive Control reduced median high-price peak power demand by 60 % in Ontario and 70 % in Québec.•Model Predictive Control also reduced cost of electricity by 16 % in Ontario and 31 % in Québec.