Development of dynamic modeling framework using convolution neuron network for variable refrigerant flow systems.

Number: 2515

Author(s) : WAN H., CAO T., HWANG Y., CHIN S.

Summary

Modeling the air conditioning system provides an excellent tool for system design, control, operation, and fault diagnosis. Such models were developed as either steady-state and transient models or knowledge-based and physics-based models. Most of the current studies mainly concentrated on physics-based models or steady-state models. Knowledge-based dynamic models were rarely discussed. In this paper, a knowledge-based dynamic model using a Convolutional Neural Network was developed for the air conditioning system. Instead of using operating parameters at a time point, we used the numbers in a time window as input data. We conducted a case study of the variable refrigerant flow system with field tests in an office building to validate this approach. It was found that the new method has a better accuracy within 2% deviation and a faster simulation speed in less than 1 second than the traditional physics-based model. The proposed method, which does not have a convergence problem, is user-friendly for non-experts. This approach also provides a way for existing systems to adjust operation parameters and detect faults. Future work can be making the current model more robust and reliable. In addition, how to combine the strengths of the knowledge-based method and physics-based methods needs to be further studied.

Available documents

Format PDF

Pages: 7

Available

  • Public price

    20 €

  • Member price*

    15 €

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

Details

  • Original title: Development of dynamic modeling framework using convolution neuron network for variable refrigerant flow systems.
  • Record ID : 30028504
  • Languages: English
  • Subject: Technology
  • Source: 2021 Purdue Conferences. 18th International Refrigeration and Air-Conditioning Conference at Purdue.
  • Publication date: 2021/05
  • Document available for consultation in the library of the IIR headquarters only.

Links


See other articles from the proceedings (184)
See the conference proceedings