Physically-constrained Hybrid Modeling for Vapor Compression Systems

    •  Dong, Y., Qiao, H., Laughman, C.R., "Physically-constrained Hybrid Modeling for Vapor Compression Systems", Thermal and Fluids Engineering Conference, April 2024.
      BibTeX TR2024-038 PDF
      • @inproceedings{Dong2024apr,
      • author = {Dong, Yiyun and Qiao, Hongtao and Laughman, Christopher R.},
      • title = {Physically-constrained Hybrid Modeling for Vapor Compression Systems},
      • booktitle = {Thermal and Fluids Engineering Conference},
      • year = 2024,
      • month = apr,
      • url = {}
      • }
  • MERL Contacts:
  • Research Area:

    Multi-Physical Modeling


Numerical simulations in the HVAC&R industries are crucial for optimizing advanced products, reducing costs, and meeting high-energy efficiency standards. Vapor compression system simulations can be broadly categorized into steady-state or transient. While steady-state evaluations determine system capacity and size, dynamic models offer a more realistic representation of system responses. Traditional dynamic models, based on the conservation laws, often lead to a complicated set of Differential Algebraic Equations (DAEs) that are challenging to solve numerically, especially for large-scale systems like variable refrigerant flow systems. Conversely, black-box models, derived directly from data, offer simplicity and accuracy within a specific operating range but lack flexibility when system architecture changes. In this paper, we propose a physically- constrained hybrid modeling framework for vapor compression systems. This approach adopts a modular based solution scheme so that arbitrary system configurations can be handled, i.e., components can be modeled with flexibility to use either data-driven or physics-based approach. In particular, we train and evaluate the Gated Recurrent Units (GRUs) component models for heat exchangers while use physics-based models for other components. A generic system solver is developed to evaluate the system configuration, formulate, and solve the resulting equations, fulfilling the conservation laws on the system level. A comprehensive comparison between this novel hybrid modeling framework and the traditional physics-based modeling approach is conducted, focusing on the aspects of system dynamics, prediction accuracy, and simulation speed.