PM2.5 Quality Concentration Prediction Based on Local Average Decomposition and Support Vector Regression

In view of the nonlinear and nonstationary characteristics of atmospheric PM2.5 mass concentration, in order to improve the prediction accuracy of PM2.5 mass concentration. Herein, we use the "decomposition and integration" prediction method, established a mixed prediction model of local average decomposition (LOCAL Mean Decomposition, LMD) and minimum daily support vector machines (LSSVM). Firstly, LMD was used to decompose the original time series of PM2.5 mass concentration, and several relatively stationary components with different time scales are obtained, then the SVR algorithm is used to predict each component separately, at last, obtaining the sum of the predictive values of each component as the prediction result of the original PM2.5 quality concentration. We select the PM2.5 daily average mass concentration from March 1, 2014 to April 30, 2015 from the Wanliu Monitoring Station in Haidian District, Beijing. The PM2.5 daily the average mass concentration is used as an experimental sample set. The results of the research were compared with EEMD-LSSVM, EMD-LSSVM and a single LSSVM model, indicating that the LMD-LSSVM model effectively improves the predictive accuracy of PM2.5 quality concentration.