Partager

Calibration Bayésienne récursive de modèles énergétiques de bâtiments archétypaux basés sur des données pour le secteur résidentiel : application à une maison expérimentale. 

Recursive bayesian calibration of data-driven archetype building energy models for residential sector: application to a research house.

Numéro : 3465

Auteurs : ABTAHI S. M., ATHIENITIS A. K., DELCROIX B.

Résumé

This paper studies recursive Bayesian calibration of archetype building energy models developed for optimal operation, energy flexibility and resilience in the residential sector. Real-time building performance monitoring has been facilitated recently with low cost faster instrumentations. With an online stream of measured data, recursive model calibration can be employed, where the model is initialized with prior knowledge and is updated with new measurements available to the building automation system. A Markov chain Monte Carlo approach is applied to calibrate two multi-zone archetype models with data from a research house in Québec. Most likely parameter values are estimated with an iterative Metropolis-Hastings algorithm over a 5-day train period and are validated by comparing the measured indoor air temperature in each zone with the model output over another 5-day test period. The time series of the posterior probability distribution shows that heat source activation influences the maximum likelihood values of most parameters and remarkably narrows down their associated credible intervals. Analyzing probability distribution time series helps understand the progression of acquired knowledge with new data and facilitates the comparison of various models.

Documents disponibles

Format PDF

Pages : 10 p.

Disponible

Gratuit

Télécharger

Détails

  • Titre original : Recursive bayesian calibration of data-driven archetype building energy models for residential sector: application to a research house.
  • Identifiant de la fiche : 30030236
  • Langues : Anglais
  • Source : 2022 Purdue Conferences. 7th International High Performance Buildings Conference at Purdue.
  • Date d'édition : 2022

Liens


Voir d'autres communications du même compte rendu (39)
Voir le compte rendu de la conférence

Indexation