“Individual thermal comfort prediction based on non-intrusive sensors”
Architectural Engineering PhD Dissertation Defense by Ati Javid
Starts at
9:00 am
Virtual Location:
Zoom Meeting ID 950 1610 2695
Target Audiences:
Contact:
The Durham School of Architectural Engineering and Construction, durhamschool@unl.edu
Advised by Dr. Iason Konstantzos
Traditional HVAC systems, based on predefined setpoints and the PMV model for comfort evaluation, often fail to account for individual and time-varying thermal preferences, leading to discomfort. Recent advances in machine learning (ML) and the Internet of Things (IoT) offer a transition from the PMV model to personalized comfort models (PCMs). However, PCMs face challenges related to data collection, which can be intrusive and require extensive, time-consuming, and costly data acquisition in real buildings. This dissertation presents a non-intrusive framework for evaluating and predicting occupant thermal comfort in office buildings using low-cost sensors and ML techniques. The proposed framework employs low-cost thermal cameras to measure skin temperature and other comfort-related variables, ensuring enhanced sensor performance through compensation for factors such as target distance. Additionally, leveraging the ASHRAE Global Thermal Comfort Database II, which includes over 109,000 observations, this study aims to improve the generalizability and accuracy of thermal comfort predictions. The framework proposes an approach to identify similar patterns in both ASHRAE and personal datasets, using ASHRAE data to enhance the performance of PCMs by finding similar observations. An experimental analysis was conducted over two months with 20 participants in the HCIBO lab, an innovative center for human comfort analysis. The results demonstrate that integrating skin temperature into comfort prediction models outperforms the traditional PMV model. By utilizing public datasets and advanced data processing techniques, the study improves the performance of PCMs.
This research highlights the potential of low-cost sensors and machine learning algorithms to transform HVAC system operations and enhance indoor environmental quality. By incorporating extensive and diverse data for more accurate thermal comfort predictions, this framework paves the way for more adaptive and personalized building environments that address individual thermal comfort, ultimately improving occupant well-being and productivity.
Traditional HVAC systems, based on predefined setpoints and the PMV model for comfort evaluation, often fail to account for individual and time-varying thermal preferences, leading to discomfort. Recent advances in machine learning (ML) and the Internet of Things (IoT) offer a transition from the PMV model to personalized comfort models (PCMs). However, PCMs face challenges related to data collection, which can be intrusive and require extensive, time-consuming, and costly data acquisition in real buildings. This dissertation presents a non-intrusive framework for evaluating and predicting occupant thermal comfort in office buildings using low-cost sensors and ML techniques. The proposed framework employs low-cost thermal cameras to measure skin temperature and other comfort-related variables, ensuring enhanced sensor performance through compensation for factors such as target distance. Additionally, leveraging the ASHRAE Global Thermal Comfort Database II, which includes over 109,000 observations, this study aims to improve the generalizability and accuracy of thermal comfort predictions. The framework proposes an approach to identify similar patterns in both ASHRAE and personal datasets, using ASHRAE data to enhance the performance of PCMs by finding similar observations. An experimental analysis was conducted over two months with 20 participants in the HCIBO lab, an innovative center for human comfort analysis. The results demonstrate that integrating skin temperature into comfort prediction models outperforms the traditional PMV model. By utilizing public datasets and advanced data processing techniques, the study improves the performance of PCMs.
This research highlights the potential of low-cost sensors and machine learning algorithms to transform HVAC system operations and enhance indoor environmental quality. By incorporating extensive and diverse data for more accurate thermal comfort predictions, this framework paves the way for more adaptive and personalized building environments that address individual thermal comfort, ultimately improving occupant well-being and productivity.