Evaluation of the Visual Stimuli on Personal Thermal Comfort Perception in Real and Virtual Environments Using Machine Learning Approaches

Abstract

Personal Thermal Comfort models consider personal user feedback as a target value. The growing development of integrated “smart” devices following the concept of the Internet of Things and data-processing algorithms based on Machine Learning techniques allows developing promising frameworks to reach the best level of indoor thermal comfort closest to the real needs of users. The article investigates the potential of a new approach aiming at evaluating the effect of visual stimuli on personal thermal comfort perception through a comparison of 25 participants’ feedback exposed to a real scenario in a test cell and the same environment reproduced in Virtual Reality. The users’ biometric data and feedback about their thermal perception along with environmental parameters are collected in a dataset and managed with different Machine Learning techniques. The most suitable algorithm, among those selected, and the influential variables to predict the Personal Thermal Comfort Perception are identified. The Extra Trees classifier emerged as the most useful algorithm in this specific case. In real and virtual scenarios, the most important variables that allow predicting the target value are identified with an average accuracy higher than 0.99.

Investigating the occupant existence to reduce energy consumption by using a hybrid artificial neural network with metaheuristic algorithms

“…The FFNN-PSO-GSA model is trained using different predictor combinations to be compared against the results reported in the literature. The suggested model yields a classification accuracy ranging between 98.47-98.73% using one predictor (i.e., temperature), higher than the reported value of 67-87% by Salamone et al (2020) and substantially higher than 55-65% reported in Sembroiz et al (2019). As for the temperature and CO2 predictors, the proposed model has an accuracy 2020) and 69-89% in Sembroiz et al (2019).…”

“…Meanwhile, its accuracy ranged between 69% and 89% for temperature and CO2 predictors. Salamone et al (2020) evaluated occupant thermal comfort in real and virtual scenarios by combining the application of the internet of things and machine learning. Different machine learning techniques were fed with users' biometric and feedback data about their thermal comfort along with the environmental parameters.…”

Correlation between Indoor Environmental Data and Biometric Parameters for the Impact Assessment of a Living Wall in a ZEB Lab

“…It is possible to replace the categorical label 22-User following the same approach, considering all environmental and biometric data. In this way, it is possible to verify the importance of individual features to identify the target feature, Users (Figure 9) thus highlighting the interconnection among environmental parameters and biometric data, as discussed in recent studies [54,55]. In the Section 3.2.2, two lists of biometric features are selected.…”

Test rooms to study human comfort in buildings: A review of controlled experiments and facilities

“…It was not possible to assess whether the test rooms are located inside a building or are entirely independent buildings for 10% of the 187 test rooms identified. According to the available information, only 7% of the facilities are independent buildings, external to any other building [34][35][36][37][38][39][40][41][42][43][44][45][46][47]. Five of these independent test rooms are located on a platform that allows the whole structure to rotate [34][35][36][37]41].…”

“…4). Among the 14 experimental facilities built outside, only one does not have windows [46]. At the same time, five include an adjustable envelope to vary the window-to-wall ratio (WWR) [35,37,41,43,45], five have a WWR lower than 0.5 [38,39,42,44,47], and three have a WWR in between 0.6 and 0.8 [34,36,40].…”