Soft Sensing-Driven CO2 Predictive Models in Educational Buildings
| dc.contributor.author | Meimand, Mostafa | en |
| dc.contributor.committeechair | Ji, Bo | |
| dc.contributor.committeecochair | Jazizadeh, Farrokh | |
| dc.contributor.committeemember | Cho, Jin-Hee | |
| dc.contributor.department | Computer Science | |
| dc.date.accessioned | 2025-01-23T20:33:40Z | |
| dc.date.available | 2025-01-23T20:33:40Z | |
| dc.date.issued | 2024-10-14 | |
| dc.description.abstract | Indoor Air Quality (IAQ) plays a vital role in occupant well-being. Among various factors, CO2 concentration impacts the productivity and cognitive functions of occupants. Different strategies can be utilized to improve IAQ, including context-aware ventilation, air purification technologies, and integration of indoor plants. Existing methods in the literature for reducing CO2 concentrations rely on direct sensing, which requires advanced infrastructure that may prevent scalability. This study investigates a soft sensing approach, utilizing readily accessible features from Building Management System (BMS) to develop predictive models for CO2 concentration, offering a cost-effective alternative to direct sensor-based measurements. We leverage two different datasets to explore the feasibility and accuracy of the soft sensing approach. The first dataset aggregates CO2 data points compiled from existing literature, providing a broad perspective of IAQ variations across various educational settings. The second dataset is a publicly available, high-resolution set of IAQ measurements from several spaces over a month, allowing for detailed model training and testing. By applying machine learning techniques, we developed models that predict CO2 concentrations based on different sets of variables. We observed that the Random Forest model could predict CO2 concentration with a Mean Absolute Error (MAE) of 37.57 by utilizing room temperature, outdoor temperature, and the hour of the day. Moreover, this study assesses the transferability of the predictive models trained on a limited number of data points. We observed that using occupancy percentage results in more transferable models compared to other variable sets. The main contribution of this study to the body of knowledge is the evaluation of the soft sensing approach, which could pave the way for creating more scalable and infrastructure-independent systems to improve indoor air quality in educational facilities. | en |
| dc.description.abstractgeneral | Indoor air quality is crucial for well-being, especially in schools and universities where students and staff spend much of their day. Among different factors, CO2 concentration plays an important role in students’ cognitive function and productivity. Traditional methods use direct sensors to monitor and operate buildings, which can be expensive and cumbersome. This research investigates a cost-effective way to predict indoor carbon dioxide (CO2) level, called soft sensing, using existing, easily accessible data to create models that predict CO2 levels without requiring extensive hardware for all spaces. We tested our models using two types of data: one collected from published studies on indoor air quality and another from a high-quality public dataset of actual air measurements in educational facilities. By applying machine learning techniques, we developed models that can predict CO2 concentrations based on monitored variables, such as room and outdoor temperature and time of day, thereby bypassing the need for extensive new sensor installations. Our finding shows that the created models are accurate and could decrease our need for extensive infrastructure systems. We also explored how well these models can be applied to other spaces, finding that models based on occupancy rates are more generalizable than others. The key finding of this research is that soft sensing can effectively predict CO2 levels in the educational settings of our case study and can be expanded across environments, making it a potentially scalable solution for improving air quality in educational facilities. | |
| dc.description.degree | Master of Science | en |
| dc.format.medium | ETD | en |
| dc.format.mimetype | application/pdf | en |
| dc.identifier.uri | https://hdl.handle.net/10919/124331 | |
| dc.language.iso | en | en |
| dc.publisher | Virginia Tech | en |
| dc.rights | In Copyright | en |
| dc.rights.uri | http://rightsstatements.org/vocab/InC/1.0/ | en |
| dc.subject | Indoor air quality | en |
| dc.subject | Soft sensing | en |
| dc.subject | Educational buildings | en |
| dc.subject | Predictive models | en |
| dc.subject | CO2 Concentration | en |
| dc.title | Soft Sensing-Driven CO<sub>2</sub> Predictive Models in Educational Buildings | en |
| dc.type | Thesis | en |
| dc.type.dcmitype | Text | en |
| thesis.degree.discipline | Computer Science and Application | |
| thesis.degree.grantor | Virginia Polytechnic Institute and State University | en |
| thesis.degree.level | masters | en |
| thesis.degree.name | Master of Science | en |