Volume XLII-3/W9
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3/W9, 23–30, 2019
https://doi.org/10.5194/isprs-archives-XLII-3-W9-23-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3/W9, 23–30, 2019
https://doi.org/10.5194/isprs-archives-XLII-3-W9-23-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

  25 Oct 2019

25 Oct 2019

RESEARCH ON PM2.5 CONCENTRATION COMBINATION FORECASTING MODEL BASED ON COR-SVM

X. Y. Feng, P. Tian, Y. J. Shi, and M. Zhang X. Y. Feng et al.
  • College of Science, Nanjing Agricultural University, Nanjing, China

Keywords: Concentration of PM2.5, Correlation Coefficient, Multivariate Linear Fitting, Support Vector Machine, Combination Forecasting

Abstract. PM2.5 is a pollutant that can enter the lungs, threatening human health and affecting people’s living and traveling. In this paper, we use multivariate linear regression, support vector machine and their combined prediction method to predict the concentration of PM2.5. It is significant for the convenience of healthy life. This paper is based on a series of meteorological data such as O3 concentration, CO concentration, SO2 concentration, PM2.5 concentration and PM10 concentration from 2014 to 2018 in Beijing. By calculating the correlation coefficient between the concentration of PM2.5 and the concentration of the other four components, the multivariate linear regression equation was fitted by using the correlation coefficient with high correlation as the factor of multiple linear regression. Then we use support vector machine regression prediction method to predict the concentration of PM2.5. The combined prediction method is obtained by weighing the two prediction results. It is found that the prediction method of support vector machine is better in dealing with large-scale and small sample data prediction, and the multi-linear fitting method is better in processing short-term prediction. The combined prediction results based on correlation coefficients combine the advantages of the two prediction methods, and the prediction results are more reasonable.