Personalized visual satisfaction profiles from comparative preferences using Bayesian inference

This study presents the development of personalized models of occupant satisfaction with the visual environment in private perimeter offices. A set of experiments was designed and conducted to collect comparative preference data from four human test-subjects. A probit model structure, with assumed latent satisfaction utility model in the form of multivariate Gaussian function, was adopted for developing the preference model. Four distinctive satisfaction and preference models were trained with a variational inference algorithm using experimental data. The posterior estimations of model parameters, visual preference probabilities and inferred satisfaction utility functions were investigated and compared, with results reflecting the different characteristics of the subjects. The developed visual satisfaction utility function was designed for use in personalized control, where occupants could balance their own satisfaction expectation and energy considerations.