Document IIF

Développement de modèles basés sur l'apprentissage pour la régulation centrée sur l'occupant. 

Developing learning-based models for occupant centric control.

Numéro : 3246

Auteurs : KIMBALL R., WEN J., O’NEILL Z., YANG T., LI Y.

Résumé

Use of advanced building control strategies, including model predictive control, is an enabling strategy to achieve Grid-interactive Efficient Buildings (GEB). Many literature-reported control strategies are designed around an ideal building and do not account for the behavior of occupants. Yet research and field studies have shown that occupant behaviors have strong impact on building operation and energy consumption. Occupants who are uncomfortable with the control strategy will often adjust the thermostat, open/close a window, or use a personal fan/heater to better suit their comfort. How to incorporate occupant behaviors into advanced control strategies has been a focus in many of the recent occupant centric control (OCC) studies. Major challenges for OCC development include forecasting occupants’ thermal comfort and behaviors and forecasting building energy with the consideration of occupant behavior. This study explores the feasibility of employing machine learning techniques, including active learning, Artificial Neural Network (ANN), and feature selection, to develop energy forecasting models that incorporate the occupant behaviors into the forecasting. To generate training and testing data needed for the control model formation, a co-simulation virtual building testbed, which utilizes a DOE Prototype residential building model developed in the EnergyPlus environment is developed. The virtual testbed also includes an Occupant Behavior Module (OBM) which is based on a previously reported agent-based-model to simulate occupants’ thermal behavior in the MATLAB SIMULINK environment. Functional Mockup Units (FMU) is used to interface between the EnergyPlus environment and the MATLAB Simulink environment. The virtual testbed is used to generate both training and testing data for typical summer weather. The accuracy and scalability (under different weather and operation conditions) of the ANN-based control models are reported and compared with conventional control models. How to select and evaluate the architecture of the ANN model that is computationally efficient but also can capture the complexity of the interaction between building systems and occupants, is discussed.

Documents disponibles

Format PDF

Pages : 10 p.

Disponible

Gratuit

Détails

  • Titre original : Developing learning-based models for occupant centric control.
  • Identifiant de la fiche : 30030220
  • Langues : Anglais
  • Source : 2022 Purdue Conferences. 7th International High Performance Buildings Conference at Purdue.
  • Date d'édition : N/A
  • Document disponible en consultation à la bibliothèque du siège de l'IIF uniquement.

Liens


Voir d'autres communications du même compte rendu (38)
Voir le compte rendu de la conférence