Partager

Document IIF

Physics-informed neural network for deep learning of finned-tube evaporator performance: From the perspective of system modeling.

Auteurs : LU S-Y., CHEN E-Q., XU T-Y-H., LIANG X. Y., ZHANG C. L.

Type d'article : Article de la RIF

Résumé

The finned-tube evaporator is widely used in refrigeration and heat pump systems. Accurate and efficient prediction of its performance is crucial for achieving digital twin and intelligent operation of complex systems, such as the multi-split variable refrigerant flow system. From the perspective of system modeling, neural networks for evaporator performance modeling should be able to predict the performance of energy (heat transfer rate), momentum (pressure drop), and mass (refrigerant inventory). However, existing neural networks generally fail
to fully satisfy these requirements. To address this limitation, this paper proposes a physics-informed neural network (PINN) for finned-tube evaporators from the perspective of system modeling. Distinct from conventional neural networks, the proposed PINN incorporates refrigerant inventory as an output parameter and the energy balance is integrated into the loss function as a physical constraint. Thus enhanced, the proposed PINN achieves excellent overall accuracy with low local error. The mean absolute percentage errors (MAPE) for predicting the total heat transfer rate, sensible heat transfer rate, refrigerant pressure drop, air pressure drop, and refrigerant inventory are 0.15%, 0.48%, 0.38%, 0.36%, and 0.30%, respectively. The root mean square errors (RMSE) are reduced by 42.4%, 11.2%, 18.4%, 34.1%, and 48.2%, respectively, compared to the previous neural network.

Documents disponibles

Format PDF

Pages : 14 p.

Disponible

  • Prix public

    20 €

  • Prix membre*

    Gratuit

* meilleur tarif applicable selon le type d'adhésion (voir le détail des avantages des adhésions individuelles et collectives)

Détails

  • Titre original : Physics-informed neural network for deep learning of finned-tube evaporator performance: From the perspective of system modeling.
  • Identifiant de la fiche : 30034783
  • Langues : Anglais
  • Sujet : Technologie
  • Source : International Journal of Refrigeration - Revue Internationale du Froid - vol. 185
  • Date d'édition : 05/2026
  • DOI : http://dx.doi.org/10.1016/j.ijrefrig.2026.02.021

Liens


Voir d'autres articles du même numéro (27)
Voir la source

Indexation