Abstract:
Current PM2.5 model forecasting data greatly deviate from the measured concentration. In order to solve this problem, support vector machine (SVM) was applied to set up a dynamic model. The data of PM2.5 model forecasting (WRF-CHEM) concentration and the five main model forecasting meteorological factors were used as training data of SVM. The data were provided by Shanghai Meteorological Bureau in Pudong New Area (from November in 2012 to November in 2013). The dynamic model was used to improve the forecasting accuracy of PM2.5 concentration one hour in advance. SVM model was compared with radical basis function neural network (RBFNN), Multi-variable Linear Regression (MLR) and WRF-CHEM. Experimental results show that the proposed algorithm greatly improves the forecasting accuracy of PM2.5 concentration one hour in advance. SVM model performs better than RBFNN, MLR and WRF-CHEM, and has better forecasting ability for the condition with concentration dramatic changing.