Predicting personalized thermal comfort in stratified micro-environments using turbulent jet theories and data-driven models

Stratified micro-environments offer different air distribution stratifications for each occupant and work as a solution for satisfying individual thermal preferences and improving energy efficiency. It is essential to accurately predict the thermal comfort provided in the micro-environments for efficient control. A computational fluid dynamics (CFD) model, validated against experimental measurements, is used to generate data with different conditions of operations systematically. Two acknowledged jet theories, the Abramovich and the Koestel systems, are applied with linear regression models to predict the predicted mean vote (PMV) and the draft rating (DR). Data-driven models include (1) back propagation neural network (BPNN), as a presentative of artificial neural network (ANN), and (2) support vector machine (SVM), as a presentative of non-ANN, are compared with the jet theory models. The demarcation value of the Archimedes number (Ar) between the free-jet condition and the non-free-jet condition is 0.0625. Jet theory models are applied for thermal comfort predictions with free-jet conditions, achieving low mean absolute error (MAE) values of the PMV and the DR, i.e., below 0.2 and 3%, respectively. A sufficient accuracy is also reached for experimental measurements in published studies, with the lowest MAE values of 0.09 for the PMV and 2.10% for the DR. Data-driven models are capable of similarly accurate predictions, advancing their generality for both the free-jet and non-free-jet conditions. The input sets handled by jet theory models perform better than the monitored supply air parameters, showing their value for improving data-driven models with limited data size.