Partager

Approche de modélisation prédictive des performances par apprentissage statistique pour un compresseur de pompe à chaleur au dioxyde de carbone. 

Statistical learning performance prediction modelling approach for a carbon dioxide heat pump compressor.

Numéro : 1163

Auteurs : VENTER P., VAN ELDIK M., COETZER R.

Résumé

Accurate compressor performance predictions are an integral part of any heat pump cycle simulation, and furthermore, the operation thereof. Heat pumps can be incorporated to increase the temperature of a medium, typically water, at various flow rates and inlet conditions. To operate a heat pump at the most efficient or economical condition, while transferring sufficient thermal energy to the heated medium, the performance prediction of the compressor is required. Even though analytical models concerned with the simulation of compressor performance can be found in literature, dimensional knowledge of the unit is typically required. This is further complicated if the compressor operates at various speeds, under a range of mass flow rates, temperatures, and pressures.
However, if sufficient operational data are available either from the manufacturer or acquired via an experimental setup, a compressor’s performance can be characterised through data analysis. The need therefore exists to formulate a statistical modelling approach, based on compressor data, that can be used to predict the unknown operational conditions required to help quantify the working of a heat pump cycle.
This paper demonstrates how statistical learning techniques can be incorporated to accurately predict the outlet conditions of a compressor that operates under various conditions. A semi-hermetic CO2 (carbon dioxide) compressor is used that operates with a VFD (variable frequency drive). The most accurate of four statistical learning techniques under investigation was combined polynomial and logistic regression. Compressor discharge temperature is less accurately predicted than the discharge pressure; however, within an error of 1.35⁰C or 1.21%.
Although the numerical predictions are specific to the chosen unit, the technique of applying statistical learning models can be extended to any compressor, provided sufficient operational data are available.

Documents disponibles

Format PDF

Pages : 9 p.

Disponible

Gratuit

Télécharger

Détails

  • Titre original : Statistical learning performance prediction modelling approach for a carbon dioxide heat pump compressor.
  • Identifiant de la fiche : 30030283
  • Langues : Anglais
  • Sujet : Technologie
  • Source : 2022 Purdue Conferences. 26th International Compressor Engineering Conference at Purdue.
  • Date d'édition : 15/07/2022

Liens


Voir d'autres communications du même compte rendu (83)
Voir le compte rendu de la conférence

Indexation