Partager

Application de réseaux neuronaux prédictifs à la simulation de systèmes de conditionnement d'air de véhicules électriques à batterie.

Application of feedforward neural networks to simulate battery electric vehicle air conditioning systems.

Numéro : 2166

Auteurs : JAIN A., FRAMKE N., TIWARI A., SPASOV M.

Résumé

Machine learning techniques have garnered extensive interest recently to reduce complicated physics-based models. This reduction enables a general speedup of simulations and allows the creation of plant models that can run substantially faster than the physics-based models while maintaining the same accuracy. One promising approach has been using feedforward neural networks (NNs) as a substitute for some of the physics-based sub-systems that make up the system model of a battery and/or fuel cell-powered electric vehicle. The thermal management systems in these vehicles are of utmost importance for component conditioning and obtaining higher efficiencies that translate into greater vehicle range. The two-phase circuit of a thermal management system is generally the bottleneck in the simulation speed of such a system due to complicated phase transitions. We start here with a two-phase circuit in a representative vehicle thermal management system and replace it with feedforward neural nets. We then assess the capabilities of neural nets to reproduce important physical quantities. A commercially available software GT-SUITE is used for model generation and neural-net training. We train the neural nets based on the compressor speed, component flow rates, and temperatures, and use the trained network as a proxy for the two-phase circuit. We found that good and valid assumptions are necessary at the interface between the NN metamodel and the remaining physical circuit to utilize NN. We also observe that static NNs generally perform well for circuits having relatively low thermal inertia. These findings are crucial for a user in assessing the applicability of these metamodels. One key application is the study of battery degradation where the degradation timescale is in making the traditional physics-based techniques infeasible. Another potential application of NNs is in controls law development in electric vehicles.

Documents disponibles

Format PDF

Pages : 10 p.

Disponible

Gratuit

Télécharger

Détails

  • Titre original : Application of feedforward neural networks to simulate battery electric vehicle air conditioning systems.
  • Identifiant de la fiche : 30030502
  • Langues : Anglais
  • Sujet : Technologie
  • Source : 2022 Purdue Conferences. 19th International Refrigeration and Air-Conditioning Conference at Purdue.
  • Date d'édition : 10/07/2022

Liens


Voir d'autres communications du même compte rendu (224)
Voir le compte rendu de la conférence

Indexation