Share

Reduced-dimension bayesian optimization for calibrating dynamic models of vapor compression cycles.

Number: 2603

Author(s) : MA J., KIM D., BRAUN J. E.

Summary

First-principles dynamic models of vapor compression cycles (VCCs) have demonstrated promising capabilities in capturing complicated thermo-fluid component and system behavior. Development and calibration of these models for achieving reliable predictions is of critical importance for many applications, including control design and fault detection and diagnostics. Nevertheless, the inherent complexity of model descriptions by large systems of differentialalgebraic equations presents significant challenges in model calibration processes that utilize classical gradient-based methods to minimize discrepancies between model predictions and available measurements. This paper presents a reduced-dimension Bayesian optimization (BO) framework for calibrating transient VCC models that employs a lowdimensional probabilistic surrogate model to approximate an expensive-to-evaluate calibration cost function associated with each set of candidate parameters, and optimal search over a feasible parameter space. The proposed approach was implemented for calibrating a large set of parameters including heat transfer coefficients and component geometries for an air-source heat pump based on laboratory measurements. The optimal set of calibrated parameters yielded exceptional prediction accuracy compared against measurements and original models without calibration.

Available documents

Format PDF

Pages: 10 p.

Available

Free

Download

Details

  • Original title: Reduced-dimension bayesian optimization for calibrating dynamic models of vapor compression cycles.
  • Record ID : 30033016
  • Languages: English
  • Subject: Technology
  • Source: 2024 Purdue Conferences. 20th International Refrigeration and Air-Conditioning Conference at Purdue.
  • Publication date: 2024/07/17

Links


See other articles from the proceedings (104)
See the conference proceedings

Indexing