Paper Conference

Proceedings of eSim 2020: 11th Conference of IBPSA-Canada


Potential of the deep learning-based model for prediction of supply air temperature from air-handling units

Md Shamim Ahamed 1,2, Radu Zmeureanu2, José Candanedo3
1 Department of Biological and Agricultural Engineering, University of California, Davis, USA
2 Centre for Net-Zero Energy Buildings Studies, Department of Building, Civil and Environmental Engineering, Concordia University, Montreal, Quebec, Canada
3 CanmetENERGY, Natural Resources Canadá, Varennes, Canada

Abstract: Machine learning-based data-driven models have shown considerable potential for fault detection and diagnosis (FDD) in building energy systems due to their flexibility in model development and the availability of data from the building automation systems (BAS). This paper discusses the potential of deep learning (DL) techniques for the prediction of the supply air temperature of the air-handling unit (AHU) as the first step of FDD. The deep neural networks (DNN) are trained using BAS trend data. The effect of model hyperparameters on the performance of DNN is discussed, and the results are compared with other models. The optimized DNN model provides good predictions with the Mean Absolute Error of 0.17°C compared with the sensor's overall uncertainty of 0.38°C. The DNN model could be employed for the re-calibration of faulty sensors. Keywords: Neural networks, deep learning, hyperparameters, air handling unit, supply air temperature