Prediction Of PM2.5 Concentration Based On Mixed Model Method

The haze weather occurs frequently,and for a long time seriously affects people's health as well as life and work.Among the air pollutants causing haze,PM2.5 is the most serious one.Many studies have shown that PM2.5 can increase the risk of lung infection,cardiovascular and cerebrovascular diseases.And PM2.5 concentration is affected by many factors,PM2.5 concentration varies and changes rapidly even in different urban areas of the same city.The estimation of PM2.5 concentration in a certain period of time in the future is extremely important for people to reasonably arrange their daily life.In this paper,Lanzhou city and Wuhan city which have distinct natural and social environments,are selected as the research objects.Taking the average daily PM2.5 concentration of January 1,2015 and March 15,2018 of the two regions as the original data,a new mixed prediction model is proposed.First,Singular Spectrum Analysis(SSA)was used to decompose and reconstruct the average daily concentration of PM2.5.For the reconstructed time series individually predicting with BP neural network optimized by Mind Evolutionary Algorithm(MEA)and Support Vector Regression(SVR)optimized by Grey Wolf Optimization(GWO).Secondly,add and summarize the predicted results of the two methods,the another Support Vector Regression(SVR)optimized by Grey Wolf Optimization(GWO)hybrid model was used to predict the new time series.A hybrid model SSA-BP-SVR-SVR is proposed.In order to verify the effectiveness of the new mixed model proposed above,this paper compares this model with other mixed models and single models,such as EEMDBP-SVR method,PSO-SVR method,SSA-GRNN-SVR method and RBFN method.The final results show that the new hybrid model method proposed in this paper has a relative advantage in predicting PM2.5 concentration in a short time.