Please use this identifier to cite or link to this item: https://scholarbank.nus.edu.sg/handle/10635/226228
Title: SPATIAL-TEMPORAL INDOOR ENVIRONMENTAL SATISFACTION PREDICTION USING GRAPH NEURAL NETWORKS
Authors: MAHMOUD MOHAMED MOHAMED ALI ABDELRAHMAN
ORCID iD:   orcid.org/0000-0001-8027-6936
Keywords: spatial-temporal, data analysis, graph neural networks, indoor environmental satisfaction, IEQ, GNN, build2Vec
Issue Date: 2-Nov-2021
Citation: MAHMOUD MOHAMED MOHAMED ALI ABDELRAHMAN (2021-11-02). SPATIAL-TEMPORAL INDOOR ENVIRONMENTAL SATISFACTION PREDICTION USING GRAPH NEURAL NETWORKS. ScholarBank@NUS Repository.
Abstract: The quality of a building's indoors (Building Indoor Environmental Quality (IEQ)) and how occupants perceive it ( Indoor Environmental Satisfaction (IES)) have gained considerable attention recently. IES has a severe impact on occupants' life, health, productivity, and well-being. An occupant level of satisfaction is linked to personal factors (e.g., gender, age, height, weight, metabolic rate, acclimatization, and other physio-psychological factors); environmental factors (e.g., indoor temperature, humidity, lighting, noise, and CO2 levels); and building spatial design factors (e.g., amount of space, location of the desk, furniture, cleanness of space, colors and textures). The personal and environmental factors have been extensively adopted in literature to predict occupants' satisfaction through statistical regression and Machine Learning (ML). However, the building's spatial data (e.g., windows, doors, furniture, walls, fans, HVAC components, and equipment) and the occupant location in relation to them have not been sufficiently investigated, even though many researchers have reported their significance. Therefore, this research aims to develop a framework and model for predicting spatial-temporal occupants' indoor environmental satisfaction called Build2Vec. Build2Vec employs the building's spatial data from the Building Information Model (BIM) in a real-world setting. The Build2Vec framework also collects momentary data about the occupant's location and satisfaction level using indoor localization and Ecological Momentary Assessment (EMA), respectively. The aggregation of these data is combined into a Graph network structure (i.e., objects and relations) and fed into a Graph Neural Network (GNN) model to predict the spatial-temporal occupants' satisfaction. The research is conducted in two sets of longitudinal survey experiments on thermal comfort. These experiments took place over four months and a total of 47 participants. Then, a prediction model was developed using spatial data, thermal comfort votes, sensors data, and graph neural networks. The results of a test implementation show 86% prediction accuracy which outperforms the baseline models by 14-28%. Also, a framework for adaptive spatial-temporal sampling has been developed using Build2Vec. This model could potentially reduce the number of sensors deployed in buildings as well as the sample size by leveraging the building's spatial data.
URI: https://scholarbank.nus.edu.sg/handle/10635/226228
Appears in Collections:Ph.D Theses (Open)

Show full item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
AbelrahmanMA.pdf19.73 MBAdobe PDF

OPEN

NoneView/Download

Page view(s)

24
checked on Sep 22, 2022

Download(s)

2
checked on Sep 22, 2022

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.