A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
An online physical-based multiple linear regression model for building’s hourly cooling load prediction
https://orcid.org/0000-0002-5505-6939
Highlights An novel PB-MLR model has a stronger generalization ability than the BP-ANN model and the MLR model. By using the interpretability of PB-MLR, the total prediction cooling load can be decoupled to obtain the building’s partial cooling loads. When online optimization methods are applied to PB-MLR, the model’s performance has a great improvement. The online PB-MLR model is particularly suitable for buildings of information-poor and operational mode changed.
Abstract Reliable cooling load prediction guides efficient energy supply strategies on the building’s source-sides and is the basis for model predictive control (MPC) of heating, ventilation, and air conditioning (HVAC). To address the insufficient generalization ability of current cooling load prediction models especially in the small sample learning, this paper established a physics-based multiple linear regression (PB-MLR) model, which has the advantages of strong generalization ability under small sample learning, short training time, and strong interpretability. The generalization ability of PB-MLR is much higher than that of the back-propagation artificial neural network (BP-ANN) and the multiple linear regression (MLR) (e.g., MAPE is 36.57% and 11.42% lower than that of them, respectively); The training time of PB-MLR is 625 times faster than that of BP-ANN; By using the interpretability of PB-MLR, the total prediction cooling load can be decoupled to obtain the building’s partial cooling loads, which can provide a reference for building’s energy-saving design. To further improve the performance of PB-MLR, this paper applied the online optimization methods to this model. When the online training and the online calibration work together on the model, the model’s performance is greatly improved, such as its MAPE is reduced by 45.45% compared with its offline model. This shows that the PB-MLR has a large potential for performance improvement and is easy to optimize online. Therefore, the online PB-MLR model (optimal MAPE = 2.64%) is particularly applicable in the scenes of the information-poor buildings with insufficient training samples or the renovation buildings with variable load patterns.
An online physical-based multiple linear regression model for building’s hourly cooling load prediction
https://orcid.org/0000-0002-5505-6939
Highlights An novel PB-MLR model has a stronger generalization ability than the BP-ANN model and the MLR model. By using the interpretability of PB-MLR, the total prediction cooling load can be decoupled to obtain the building’s partial cooling loads. When online optimization methods are applied to PB-MLR, the model’s performance has a great improvement. The online PB-MLR model is particularly suitable for buildings of information-poor and operational mode changed.
Abstract Reliable cooling load prediction guides efficient energy supply strategies on the building’s source-sides and is the basis for model predictive control (MPC) of heating, ventilation, and air conditioning (HVAC). To address the insufficient generalization ability of current cooling load prediction models especially in the small sample learning, this paper established a physics-based multiple linear regression (PB-MLR) model, which has the advantages of strong generalization ability under small sample learning, short training time, and strong interpretability. The generalization ability of PB-MLR is much higher than that of the back-propagation artificial neural network (BP-ANN) and the multiple linear regression (MLR) (e.g., MAPE is 36.57% and 11.42% lower than that of them, respectively); The training time of PB-MLR is 625 times faster than that of BP-ANN; By using the interpretability of PB-MLR, the total prediction cooling load can be decoupled to obtain the building’s partial cooling loads, which can provide a reference for building’s energy-saving design. To further improve the performance of PB-MLR, this paper applied the online optimization methods to this model. When the online training and the online calibration work together on the model, the model’s performance is greatly improved, such as its MAPE is reduced by 45.45% compared with its offline model. This shows that the PB-MLR has a large potential for performance improvement and is easy to optimize online. Therefore, the online PB-MLR model (optimal MAPE = 2.64%) is particularly applicable in the scenes of the information-poor buildings with insufficient training samples or the renovation buildings with variable load patterns.
An online physical-based multiple linear regression model for building’s hourly cooling load prediction
https://orcid.org/0000-0002-5505-6939
Highlights An novel PB-MLR model has a stronger generalization ability than the BP-ANN model and the MLR model. By using the interpretability of PB-MLR, the total prediction cooling load can be decoupled to obtain the building’s partial cooling loads. When online optimization methods are applied to PB-MLR, the model’s performance has a great improvement. The online PB-MLR model is particularly suitable for buildings of information-poor and operational mode changed.
Abstract Reliable cooling load prediction guides efficient energy supply strategies on the building’s source-sides and is the basis for model predictive control (MPC) of heating, ventilation, and air conditioning (HVAC). To address the insufficient generalization ability of current cooling load prediction models especially in the small sample learning, this paper established a physics-based multiple linear regression (PB-MLR) model, which has the advantages of strong generalization ability under small sample learning, short training time, and strong interpretability. The generalization ability of PB-MLR is much higher than that of the back-propagation artificial neural network (BP-ANN) and the multiple linear regression (MLR) (e.g., MAPE is 36.57% and 11.42% lower than that of them, respectively); The training time of PB-MLR is 625 times faster than that of BP-ANN; By using the interpretability of PB-MLR, the total prediction cooling load can be decoupled to obtain the building’s partial cooling loads, which can provide a reference for building’s energy-saving design. To further improve the performance of PB-MLR, this paper applied the online optimization methods to this model. When the online training and the online calibration work together on the model, the model’s performance is greatly improved, such as its MAPE is reduced by 45.45% compared with its offline model. This shows that the PB-MLR has a large potential for performance improvement and is easy to optimize online. Therefore, the online PB-MLR model (optimal MAPE = 2.64%) is particularly applicable in the scenes of the information-poor buildings with insufficient training samples or the renovation buildings with variable load patterns.
An online physical-based multiple linear regression model for building’s hourly cooling load prediction
Chen, Sihao (author) / Zhou, Xiaoqing (author) / Zhou, Guang (author) / Fan, Chengliang (author) / Ding, Puxian (author) / Chen, Qiliang (author)
Energy and Buildings ; 254
2021-10-11
Article (Journal)
Electronic Resource
English
An improved office building cooling load prediction model based on multivariable linear regression
| Online Contents | 2015