Analysis Of Influencing Factors And Prediction Of PM2.5 Under CEEMDAN’s Perspective

In recent years,people’s awareness of environmental protection continues to improve,China’s atmospheric environment has been significantly improved,but there are still some areas such as Beijing,Tianjin and Hebei affected by air pollution,and China’s atmospheric standards are significantly different from European and American countries,and most of the heavy air pollution is PM2.5 plays a leading role.The fluctuation of the daily concentration of PM2.5 largely reflects China’s atmospheric conditions.Therefore,the study of PM2.5 will always exist in China’s modern and green development.Empirical Mode Decomposition(EMD)and its extension method are data adaptive and self-driven decomposition methods,which can effectively improve the prediction ability of PM2.5 concentration.However,the practical significance of its decomposed eigenmode function(IMFs)has always been a key and difficult problem.If the signal meaning of IMFs can be reasonably and effectively interpreted,It will greatly improve the scientificity and effectiveness of the later analysis and prediction work,and make up for the fuzzy problem of mathematical theory.Taking the daily concentration of PM2.5 in Hefei City in recent 10 years as an example,this paper reconstructs several IMFs processed by Complete Ensemble EMD with Adaptive Noise(CEEMDAN)into low-frequency reconstructed signals,residual signals and high-frequency reconstructed signals,and makes two studies on them.On the one hand,it combines literature and historical data and gives a reasonable interpretation of practical significance based on the wave characteristics of reconstructed signals;on the other hand,it establishes multiple combined models based on reconstructed signals and predicts the daily PM2.5 concentration in Hefei City through dimensionality reduction.Reconstruction of signal interpretation.The annual periodicity of low-frequency reconstructed signal fluctuates and corresponds to the seasonal characteristics of PM2.5 concentration in Hefei,which is high in winter and low in summer.The short term turning point of low frequency reconstructed signal in late May corresponded with the burning of agricultural straw.The long-term decline trend of residual signal is dominated by the environmental policy requirements of Hefei City.The short-term extremely high site of high-frequency reconstructed signal in each cycle corresponds to the Spring Festival at the beginning of the year.The high-frequency reconstructed signal of PM2.5 is consistent with the fluctuation characteristics of the high-frequency information reconstructed by the principal components of the comprehensive air pollution factors,indicating that the high-frequency reconstructed signal is affected by SO2 and other polluting gases.PM2.5 prediction.Each reconstructed signal was fitted and optimized by LSTM model,PROPHET model and SVM model respectively,and 27 complete PM2.5predicted concentration data were obtained by corresponding addition of the predicted data.Then dimension reduction and optimization were performed by LASSO model,SCAD model and MCP model.At the same time,LSTM model,PROPHET model and SVM model were used to construct simulation and optimize the prediction of the original data,and the results were used for analysis and comparison.The results show that the prediction performance of multiple dimensionality reduction models under CEEMDAN method is significantly higher than that of direct simulation of raw data.The dimensionality reduction models under CEEMDAN method have higher prediction performance,and the SCAD model and MCP model under CEEMDAN method are slightly better than the LASSO model.