Bayesian prediction model of thermally satisfied occupants considering stochasticity due to inter- and intra-individual thermal sensation variations

Abstract

A data-driven probabilistic model is proposed to predict the group-level thermal satisfaction of occupants subjected to a given thermal condition. This model considers the inhomogeneity of inter- and intra-individual variations in thermal sensation votes (TSVs) on the basis that individual variations in TSVs are expected to be undispersed in extreme thermal conditions and scattered in conditions closely matching thermal neutrality. Additionally, unlike conventional deterministic linear regression models, the proposed model adopts an ordinal probit regression model to treat TSVs as ordinal rather than metric variables. Model parameters are estimated by using a Bayesian inference technique to capture the stochastic characteristics of occupants' TSVs. The model's effectiveness is validated against a subset of ASHRAE Global Thermal Comfort Database II. Compared with the conventional model, the proposed model more accurately predicts the variation in TSVs and the thermal conditions in which the occupants are most satisfied.