Share

Recommended by the IIR / IIR document

Evaluation of machine learning methods for optimizing the defrosting process of air-to-water heat pumps.

Number: 0117

Author(s) : KLINGEBIEL J., GOBEL S., VENZIK V., MÜLLER D.

Summary

Frost formation decreases system efficiency, so frequent defrosting is necessary to achieve energy-efficient operation. Compared to conventional time-controlled defrosting, demand-controlled defrosting offers a high potential for increasing overall efficiency. This study develops a demand-controlled defrosting strategy based on machine learning. Therefore a detailed simulation model is presented and calibrated using experimental data. Our experimental results indicate that defrosting initiation time is a key parameter for efficient operation: A deviation of 30 minutes from the optimal defrosting time (ODT) results in a 10.7 % reduction in efficiency. We show that self-optimizing control strategies, such as reinforcement learning, can improve the defrosting process due to successfully predicting ODT. Compared to conventional defrosting strategies, the system efficiency increases by a minimum of 5.56 % for a typical 24 h demand profile. The method reduces the number of defrosting operations by avoiding unnecessary defrosting, thus increasing the operating time and thermal comfort.

Available documents

Format PDF

Pages: 11 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: Evaluation of machine learning methods for optimizing the defrosting process of air-to-water heat pumps.
  • Record ID : 30029719
  • Languages: English
  • Subject: Technology
  • Source: 15th IIR-Gustav Lorentzen Conference on Natural Refrigerants (GL2022). Proceedings. Trondheim, Norway, June 13-15th 2022.
  • Publication date: 2022/06/13
  • DOI: http://dx.doi.org/10.18462/iir.gl2022.0117
  • Document available for consultation in the library of the IIR headquarters only.

Links


See other articles from the proceedings (169)
See the conference proceedings

Indexing