PM2.5 Prediction Based on Distance Factor

Many cities have set up PM2.5 watching stations for air quality monitoring. When a station does not work, it is necessary to predict the PM2.5 concentration of the fault station by its historical data and the working stations around it. This paper analyses the correlation coefficients between the current and its historical PM2.5 concentration of every station. It also studies the correlation coefficients between the current PM2.5 concentration and the that of the surrounding stations. We find they are both moderately and highly linearly correlated. With linear regression and support vector regression, we build feature vectors to predict PM2.5 concentration by using its historical data and data of its surrounding stations. Besides, this paper combines the historical data of a station with its current surrounding data. It adds auxiliary data called distance factor, and gives different weights according to the distance between the surrounding station and the fault station. That improves the model prediction ability.