Real-Time Estimation of the Urban Air Quality with Mobile Sensor System

Recently, real-time air quality estimation has attracted more and more attention from all over the world, which is close to our daily life. With the prevalence of mobile sensors, there is an emerging way to monitor the air quality with mobile sensors on vehicles. Compared with traditional expensive monitor stations, mobile sensors are cheaper and more abundant, but observations from these sensors have unstable spatial and temporal distributions, which results in the existing model could not work very well on this type of data. In this article, taking advantage of air quality data from mobile sensors, we propose an real-time urban air quality estimation method based on the Gaussian Process Regression for air pollution of the unmonitored areas, pivoting on the diffusion effect and the accumulation effect of air pollution. In order to meet the real-time demands, we propose a two-layer ensemble learning framework and a self-adaptivity mechanism to improve computational efficiency and adaptivity. We evaluate our model with real data from mobile sensor system located in Beijing, China. And the experiments show that our proposed model is superior to the state-of-the-art spatial regression methods in both precision and time performances.