Conference Paper
Proceedings of eSim 2018: 10th Conference of IBPSA-Canada
Comparison of iterative-multistep-ahead models for forecasting the supply air flow rate of a HVAC system
Comparison of iterative-multistep-ahead models for forecasting the supply air flow rate of a HVAC system
Jason Runge, Mathieu Le Cam, Radu ZmeureanuAbstract: This paper presents the development of a closed-loop autoregressive neural network (ANN) model for forecasting the supply air flow rate of a heating, ventilation and air conditioning (HVAC) system. Data was provided by the building automation system (BAS) recorded with a 15-minute time interval. The forecast horizon for the ANN model was six hours. The optimal architecture of the ANN model was found through an extensive search algorithm of over 36,750 different design alternatives. The supply air flow rate was forecasted over the testing period with a Root Mean Squared Error (RMSE) of 386 L/s and a CV(RMSE) of 1.8%. The forecasts of the ANN model are then compared with those from the support vector regression (SVR) model which had a RMSE of 463 L/s and a CV(RMSE) of 2.2%. It was concluded that both ANN and SVR forecasting methods provide acceptable forecasts of the target value. In addition, a comparison of different approaches for updating the ANN with the sliding window technique is also presented. Keywords: Forecasting, ANN, SVR, HVAC, sub-hourly, air supply flow ratePages: 50 - 56 Paper:![]()
