Share

IIR document

Fault diagnosis of heat pump system with different transfer tasks based on cross domain deep transfer learning.

Author(s) : GUO Y., LIU Y., JIN Z., WANG Y., DU C., LIU C., LI W.

Type of article: IJR article

Summary

Fault diagnosis can effectively reduce the energy consumption caused by system faults in building applications. However, conventional machine learning methods require large labeled datasets and suffer from limited generalization. This study proposes a cross domain deep transfer learning (CDDTL) model aimed at enhancing the applicability and effectiveness of fault diagnosis in building scenarios with limited labeled data. A cross- system carbon dioxide heat pump fault diagnosis model based on TL has been established for two different TL tasks in building environments. Feature selection is conducted from dual perspectives of thermodynamic theory and machine learning through theoretical analysis and random forest importance ranking. Transfer learning models are optimized via hyperparameter tuning, significantly improving fault diagnosis accuracy. The results indicate that the CDDTL method outperforms the other four TL methods in both tasks, achieving an optimal accuracy of 91.55%. Through feature variable screening, the accuracy of the correlation alignment fault diag nosis model for the transfer task from the water heater to the air conditioner is improved by 43.44%. After model parameter optimization, the CDDTL method significantly improves fault diagnosis effectiveness, particularly in the water heater to air conditioner transfer direction within building applications. This approach achieves a fault diagnosis accuracy of 91.55%, representing a 9.07% improvement over the baseline (82.48%).

Available documents

Format PDF

Pages: 14 p.

Available

  • Public price

    20 €

  • Member price*

    Free

* Best rate depending on membership category (see the detailed benefits of individual and corporate memberships).

Details

  • Original title: Fault diagnosis of heat pump system with different transfer tasks based on cross domain deep transfer learning.
  • Record ID : 30034487
  • Languages: English
  • Subject: Technology
  • Source: International Journal of Refrigeration - Revue Internationale du Froid - vol. 181
  • Publication date: 2026/01
  • DOI: http://dx.doi.org/10.1016/j.ijrefrig.2025.10.022

Links


See other articles in this issue (33)
See the source

Indexing