Prediction of PM_2.5 Mass Concentration Based on Complementary Ensemble Empirical Mode Decomposition and Support Vector Regression

To solve the problem of nonlinearity and instability of atmospheric PM_2.5 mass concentration and improve the prediction accuracy, a new hybrid method based on complementary ensemble empirical mode decomposition (CEEMD) and support vector regression (SVR) was established by using the method of "decomposition and integration". First, the original sequence of PM_2.5 mass concentration was decomposed by CEEMD, and several relatively stationary components with different time scales were obtained. Second, each component was predicted by SVR, respectively. Finally, the sum of the predicted values of each component was obtained as the prediction result of the original PM_2.5 mass concentration. The daily average PM_2.5 mass concentration from March 1, 2014 to April 30, 2015 in Wanliu monitoring site of Beijing and the daily average PM_2.5 mass concentration from May 1, 2014 to April 30, 2015 in Huairou monitoring site of Beijing were selected as experimental samples. Results show that CEEMD-SVR model can effectively improve the prediction accuracy of PM_2.5 mass concentration when compared with the EEMD-SVR, EMD-SVR and single SVR models.