Please use this identifier to cite or link to this item:
|Title:||Smart urban air quality diagnosis using low-cost particulate matter sensors and machine learning||Authors:||Won, Wan-Sik||Keywords:||Engineering::Mechanical engineering||Issue Date:||2022||Publisher:||Nanyang Technological University||Source:||Won, W. (2022). Smart urban air quality diagnosis using low-cost particulate matter sensors and machine learning. Doctoral thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/155250||Abstract:||Particulate matter (PM) concentration is a key parameter for air quality, affecting human health and the environment. The scientific and social interest in PM has been growing as it is revealed that airborne PM has relation to morbidity and premature human mortality associated with numerous adverse effects, such as cardiovascular and respiratory disease. Regulatory authorities around the world have set limit values for 24 hours or annual concentration of the particulates based on accurate measurement of PM concentration. Although the conventional measurement such as filter-based and direct-reading instrument have high-quality and accuracy, they have disadvantage in cost, limiting the quality of spatial distribution. Thus, low-cost sensors are alternatives for expensive conventional instruments, regarding high spatial density PM monitoring. Recently, relatively low-cost and small sensor which works on light scattering principle has become widely available as they have significant advantage in cost, number of deployment, and a distributed manner. However, low-cost PM sensor has inherently low accuracy and lack of reliability dependent on regions because they estimate mass concentration indirectly based on optical measurements of light scattering while PM exhibits hygroscopic properties resulting in bias of light scattering. Accordingly, there is uncertainty associated with meteorological factors, background PM concentration, and aerosol-size distribution, which compromises the reliability of the sensors. Therefore, the low-cost sensors are potentially useful but at the same time require a proper calibration to ensure reliable measurement of PM concentration in diverse field environments. In this research, a regional calibration of low-cost PM sensors by machine learning (ML) using historical weather and air-quality data is proposed to improve the accuracy of ambient-air monitoring. Four-year weather observations and PM concentrations were collected from two regions, Korea and Singapore, to build a visibility-based calibration model. To validate the model, field measurements by a low-cost sensor were conducted over months, revealing that the error after applying the model in the region was significantly reduced. The results show that the regional calibration involving air temperature, relative humidity, and other local-climate parameters can efficiently improve the accuracy of the sensor. The findings suggest that the proposed post-processing by ML using regional weather and air-quality data can be a turning point in outdoor air-quality monitoring, enhancing the applicability of low-cost sensors in various environments.||URI:||https://hdl.handle.net/10356/155250||DOI:||10.32657/10356/155250||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:||MAE Theses|
Updated on May 14, 2022
Updated on May 14, 2022
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.