Document IIF

Hybrid Predictive Modeling Framework for R290 Heat Pump System Performance in Electric Vehicles.

Numéro : 0077

Auteurs : SHAO W., REN Z., YANG T., DAI J., ZOU H.

Résumé

This paper presents a hybrid predictive modeling framework for R290 heat pump systems in electric vehicles, integrating a virtual sensor for refrigerant mass flow rate with heat exchanger performance prediction via machine learning. A systematic evaluation of eighteen algorithms across eight categories is conducted for three tasks: mass flow rate prediction, condenser HTC and FPD prediction, and evaporator HTC and FPD prediction. The EXT algorithm consistently outperforms all other methods, achieving an R² of 0.976 for mass flow rate and 0.995 for FPD prediction in both heat exchangers. A two-step feature selection strategy, combining Pearson correlation clustering with SHAP analysis, is employed to identify key dimensionless parameters and reveal the distinct physical mechanisms governing condensation and evaporation. By eliminating the need for costly flow measurement instrumentation without compromising predictive accuracy, the framework offers a cost-effective solution for thermal management optimization in electric vehicles. This paper presents a hybrid predictive modeling framework for R290 heat pump systems in electric vehicles, integrating virtual refrigerant mass flow sensing with heat exchanger performance prediction via machine learning. The framework comprises a compressor-based virtual flow sensor, a two-step feature selection strategy combining Pearson correlation clustering with SHAP analysis for physical interpretability, and a systematic evaluation of 16 algorithms across eight categories for predicting condenser and evaporator HTC and FPD. The Extra Trees algorithm consistently outperforms all others, achieving R² of 0.976 for mass flow rate, 0.995/0.992 for condenser/evaporator HTC, and 0.913/0.940 for condenser/evaporator FPD. Feature importance analysis reveals distinct physical sensitivities: condensation HTC is most influenced by Reeq and Xtt, whereas evaporation HTC is governed by ρf/ρg, Reeq, Xm, and Fr. The framework eliminates costly flow meters while maintaining predictive accuracy, offering a cost-effective solution for real-time EV thermal management.

Documents disponibles

Format PDF

Pages : 16 p.

Disponible

  • Prix public

    20 €

  • Prix membre*

    15 €

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

Détails

  • Titre original : Hybrid Predictive Modeling Framework for R290 Heat Pump System Performance in Electric Vehicles.
  • Identifiant de la fiche : 30035448
  • Langues : Anglais
  • Sujet : Technologie
  • Source : 17th IIR-Gustav Lorentzen Conference on Natural Refrigerants. Proceedings. Hamilton, New Zealand, August 23-26 2026.
  • Date d'édition : 08/2026
  • DOI : http://dx.doi.org/10.18462/iir.gl2026. 0077

Liens


Voir d'autres communications du même compte rendu (91)
Voir le compte rendu de la conférence

Indexation

  • Thèmes : N/A
  • Mots-clés : N/A