Please use this identifier to cite or link to this item: https://hdl.handle.net/10356/141578
Title: Transforming thermal comfort model and control in the tropics : a machine-learning approach
Authors: Hu, Weizheng
Keywords: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
Engineering::Computer science and engineering::Computing methodologies::Simulation and modeling
Issue Date: 2020
Publisher: Nanyang Technological University
Source: Hu, W. (2020). Transforming thermal comfort model and control in the tropics : a machine-learning approach. Doctoral thesis, Nanyang Technological University, Singapore.
Project: Building and Construction Authority (BCA) - Green Buildings Innovation Cluster: NRF2015ENC-GBICRD001-012
Behavioural Studies in Energy, Water, Waste and Transportation Sectors: BSEWWT2017_2_06
Data Science & Artificial Intelligence Research Centre @ NTU: DSAIR@NTU
Abstract: The building sector is one of the primary energy consumers in Singapore. According to the report of Building and Construction Authority (BCA), Singapore's building sector consumes up to 38% of the nation's electricity, and the Heating, Ventilation and Air Conditioning (HVAC) systems account for 40~50% of the total energy consumed by buildings. As a result, the reduction of HVAC systems energy expenditure would save a lot of energy and lower the electricity bill. However, the HVAC systems directly impact the occupants' indoor thermal comfort, which influences people's productivity and health. Existing studies suggested that office tenants in comfortable thermal environments are more productive. In contrast, Sick Building Syndrome (SBS) symptoms have been attributed to thermal discomfort, potentially leading to long-term health hazards. It follows that it is a challenge to reduce the energy consumption of the HVAC systems while ensuring occupants’ thermal comfort. In this thesis, we investigate the challenge mentioned above to strike the trade-off between occupants' thermal comfort and energy saving of HVAC systems. First, we validate the canonical Fanger's Predicted Mean Vote (PMV) model, survey the related literature and find that it suffers from three major issues: 1) Inadequate modeling parameters; 2) Insufficient modeling dataset; and 3) Inaccurate comfort prediction. For the first issue, the PMV model only considers six thermal comfort parameters: air temperature, mean radiant temperature, relative humidity, air velocity, metabolic rate, and clothing insulation. However, we survey the existing literature and find that other than the six PMV parameters, some other environmental and personal factors, especially the human vital signs (e.g., skin temperature, heart rate, etc.), also link to human thermal comfort. To address this issue, we adopt the off-the-shelf wearable device (e.g., Microsoft Band 2) for human vital sign tracking and the mature wireless sensor network (WSN) product is used for environmental data monitoring. For the second issue, the PMV model was developed in 1970 by using the paper-based questionnaires to collect data from test subjects manually. However, from the view of current big-data standard, the obtained modeling dataset is relatively small, which potentially affect the model accuracy. To address this issue, we develop an Internet of Things (IoT) based data acquisition system, including a back-end server, iOS & Android mobile Apps and a management website, and then use it to support our in-situ measurements and conduct one-year data collection experiments to obtain our own dataset. For the last issue, existing research works indicate that the PMV is inaccurate to predict human thermal comfort. Moreover, the PMV model is a static model which makes it difficult to adapt to the different thermal environment with various occupants dynamically. To address this issue, we propose to use the emerging deep learning techniques to develop next-generation thermal comfort models for improving the prediction accuracy. Our developed Heterogeneous Transfer Learning (HTL) based Intelligent Thermal Comfort Neural Network (HTL-ITCNN) outperforms the PMV model by on average 73.9%. In addition, for learning-based approaches, it is flexible to introduce new parameters to an existing model or update the developed models by training with newly obtained training data, to further improve the performance. Then, we aim to reduce the HVAC energy consumption based on our developed learning-based thermal comfort models. Since PMV model has been adopted by the American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE) standard 55, it significantly affects the building design and operation, especially the thermal control (e.g., HVAC control). Currently, based on the PMV model, a fixed set-point temperature (e.g., 23°C or 24°C) is set for the whole building. However, a building has multiple zones, and occupants in different zones may have heterogeneous thermal comfort demands. Considering the inaccurate comfort prediction issue of the PMV model, we argue that the PMV model may not provide the proper guidance on the thermal control of a multi-zone building. To address this issue, we use the deep reinforcement learning technique to develop the smart thermal comfort mechanism for multi-zone buildings. Our Multi-Agent Deep Reinforcement Learning (MADRL) based approach can achieve from 2.8% up to 15.4% energy saving under different thermal comfort demands. Eventually, we integrate all our solutions introduced above for striking the trade-off between occupants’ thermal comfort and energy saving of HVAC systems to build an intelligent thermal comfort management (iTCM) system. With iTCM system, we design and implement a novel human behavioral study with a gamification approach to encourage human behavioral changes towards greener, smarter and healthier buildings. It is well-known that there is no perfect HVAC control mechanism which can fulfill every occupant's thermal comfort demand. To mitigate those occupants who may be in discomfort, we aim to derive the economic model and use it to judiciously balance between the HVAC electricity cost saving and the incentive expenditure, maximizing the total cost saving while maintaining occupants' satisfaction levels over 84%. Moreover, we develop an interesting demo website by using Digital-Twin technology. Digital-Twin technology is used to create the digital replica of various entities (e.g., processes, systems, people) in the real world, and then use them to build the related virtual world to simulate the dynamics of the real world. The aforementioned system, solutions and novel studies can efficiently reduce the HVAC energy consumption while maintaining occupants' indoor satisfaction levels. Also, the collected dataset will benefit the human thermal comfort research community. Furthermore, the proposed approaches and obtained results may provide the guidelines to future research on indoor thermal comfort model and control.
URI: https://hdl.handle.net/10356/141578
DOI: 10.32657/10356/141578
Rights: This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).
Fulltext Permission: open
Fulltext Availability: With Fulltext
Appears in Collections:SCSE Theses

Files in This Item:
File Description SizeFormat 
Final_Thesis_HU_Weizheng.pdf6.12 MBAdobe PDFView/Open

Page view(s) 20

592
Updated on Feb 5, 2023

Download(s) 10

306
Updated on Feb 5, 2023

Google ScholarTM

Check

Altmetric


Plumx

Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.