Publication Type

Conference Proceeding Article

Version

acceptedVersion

Publication Date

4-2020

Abstract

Dense deployments of commodity air quality sensors have proven effective to provide spatially-resolved information on urban air pollution in real-time. However, long-term operation of a dense sensor deployment incurs enormous maintenance expenses and efforts. A cost-effective alternative is to first collect measurements with an initial dense deployment and then rely on a small subset of sensors for air quality map generation. To avoid dramatic accuracy degradation in air quality maps generated using the downscaled sparse deployment, we design MapTransfer, an air quality map generation scheme which augments the current sensor measurements from the downscaled sparse deployment with appropriate historical data from the initial dense deployment. Due to the spatiotemporal complexity of air pollution, it is challenging to select the best historical data and fuse them with measurements from the downscaled deployment to accurate map generation. To overcome this challenge, MapTransfer adopts a learning-based data selection scheme and integrates the best historical data with the current measurements via a multi-output Gaussian process model at sub-region levels. Evaluations on a large-scale PM2.5 sensor deployment show that MapTransfer reduces the overall mean absolute error of air quality maps by 45.9%, compared with using data from the downscaled deployment alone.

Discipline

Software Engineering

Research Areas

Software and Cyber-Physical Systems

Publication

Proceedings of 2020 IEEE/ACM Fifth International Conference on Internet-of-Things Design and Implementation, Sydney, Australia, April 21-24

First Page

1

Last Page

13

Identifier

10.1109/IoTDI49375.2020.00010

Publisher

IEEE

City or Country

Sydney, Australia

Additional URL

https://doi.org/10.1109/IoTDI49375.2020.00010

Download

Share

COinS