Transfer Learning for the Prediction of Energy Performance of Water-Cooled Electric Chillers: Grey-Box Models Versus Deep Neural Network (DNN) Models

Hongwen Dou and Radu Zmeureanu ()
Additional contact information
Hongwen Dou: Centre for Zero Energy Building Studies, Department of Building, Civil and Environmental Engineering, Gina Cody School of Engineering and Computer Science, Montreal, QC H3G 1M8, Canada
Radu Zmeureanu: Centre for Zero Energy Building Studies, Department of Building, Civil and Environmental Engineering, Gina Cody School of Engineering and Computer Science, Montreal, QC H3G 1M8, Canada

Energies, 2024, vol. 17, issue 23, 1-11

Abstract: The development of data-driven prediction models of energy performance of HVAC equipment, such as chillers, depends on the quality and quantity of measurement data for the model training. The practical applications always struggle with the credibility of results when the training dataset of an existing chiller is relatively small. Moreover, when the energy analyst needs to develop a reliable predictive model of a new chiller, the manufacturer’s proprietary data are not always available. The transfer learning method can soften these constraints and can help in the development of a predictive model that captures the knowledge from the available chiller, called the source chiller, using a small dataset, and apply it to a new chiller, called the target chiller. The paper presents the successful application of transfer learning strategies by using grey-box models and DNN models for the prediction of chillers performance, when measurement data are recorded at 15 min time intervals by the building automation system (BAS) and used for training and testing. The paper confirms the initial hypothesis that both the grey-box models and DNN models of the source chiller from July 2013 predict well the energy performance of the target chiller with measurement datasets from 2016. The DNN models perform slightly better than the grey-box models. The pre-trained grey-box models and DNN models, respectively, are transferred to the target chiller using three strategies: SelfL, TLS0, and TLS1, and the results are compared. SelfL strategy trains and tests the models only with the target data. TLS0 strategy directly transfers the models from the source chiller to the target chiller. TLS1 strategy transfers the models, pre-trained with an extended dataset that is composed of training dataset of D s and training dataset of D t . Finally, the models are tested with another set of testing data. The difference in computation times of these two types of models is not significant for preventing the use of DNN models for the applications within the BAS, when compared with grey-box models.

Keywords: chiller; performance; transfer learning; measurements; grey-box; neural network (search for similar items in EconPapers)
JEL-codes: Q Q0 Q4 Q40 Q41 Q42 Q43 Q47 Q48 Q49 (search for similar items in EconPapers)
Date: 2024
References: View references in EconPapers View complete reference list from CitEc
Citations:

Downloads: (external link)
https://www.mdpi.com/1996-1073/17/23/5981/pdf (application/pdf)
https://www.mdpi.com/1996-1073/17/23/5981/ (text/html)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:gam:jeners:v:17:y:2024:i:23:p:5981-:d:1531540

Access Statistics for this article

Energies is currently edited by Ms. Agatha Cao

More articles in Energies from MDPI
Bibliographic data for series maintained by MDPI Indexing Manager ().