Share

Hybrid modeling approach for better identification of building thermal network model and improved prediction.

Number: 3489

Author(s) : HAM S. W., KIM D.

Summary

The gray-box modeling approach (i.e., semi-physical thermal network model) has been widely used for prediction applications for buildings such as a model predictive control (MPC). However, applying the modeling approach for practical buildings is still challenging due to unmeasured disturbances such as occupants, lighting, appliances, and in/exfiltration loads. To overcome this problem, several system identification approaches have been proposed by considering the dynamics of unmeasured disturbance. However, the performance of long-term (e.g., one day) zone temperature or load predictions could still be very poor, and this is an important research topic for enabling grid-interactive buildings. In this study, we propose a hybrid modeling approach to improve long-term temperature or load predictions. Several system identification approaches for gray-box models are compared using simulations to understand the limitations. A neural network model that accounts for unmeasured disturbance is developed by considering the limitation of the graybox model and is combined with the gray-box model. This hybrid model approach shows 0.24°C root mean squared error (RMSE) for 1-day ahead temperature prediction on average, while the conventional gray-box model shows 1.1°C RMSE on average.

Available documents

Format PDF

Pages: 10 p.

Available

Free

Download

Details

  • Original title: Hybrid modeling approach for better identification of building thermal network model and improved prediction.
  • Record ID : 30030244
  • Languages: English
  • Source: 2022 Purdue Conferences. 7th International High Performance Buildings Conference at Purdue.
  • Publication date: 2022
  • Document available for consultation in the library of the IIR headquarters only.

Links


See other articles from the proceedings (39)
See the conference proceedings

Indexing