Research On The Construction Of Different Integrated Prediction Models And Their Applications In Air Pollution Prediction

Air pollution has become an important environmental problem that affects public health and restricts economic development.Scientific prediction theory and prediction information are of great theoretical value and practical significance in promoting air pollution prevention and alleviating the economic losses caused by air pollution.In recent years,decomposition integrated model is one of the complex time series methods which play an important role in the field of air pollution prediction.Specially,decomposition integrated model is an effective and widely used method for air pollution prediction.In previous researches,it can be found that decomposition integrated models still have the following problems in predicting air pollution:Firstly,the decomposition modes are generally subjectively refactored in the existing studies,which obviously lacks theoretical support.Secondly,most of the decomposition models are linearly integrated and do not consider the influence of meteorological factors on air pollution.Thirdly,there is uncertainty in optimal model selection for predicting air pollution with decomposition integrated models,which leads to less robust predictions.Based on the above problems,the following studies are carried out.For the first problem,this research introduces sample entropy and random forest to refactor the modes and constructs the CEEMD-MR-Hybrid prediction model.The proposed CEEMD-MR-Hybrid model is used to predict the daily SO₂concentrations in five cities,including Baoshan and Shenzhen.The prediction results show that the prediction accuracy and fitting goodness of CEEMD-MR-Hybrid are higher than those of the subjective refactored integrated model.For example,compared with random refactor integrated models,Imapes of all CEEMD-MR-Hybrid models in Shenzhen are above 4%.The research results suggest that the modes refactor based on feature selection can effectively improve the accuracy of the decomposition integrated model in predicting SO₂.For the second problem,this research builds VMD-GAM-Correction integrated model by employing the generalized additive model to integrate all modes’predictions and integrating meteorological factors for error correction.The proposed VMD-GAM-Correction model is used to predict daily CO concentrations in Lanzhou and Shenzhen.The prediction results reveal that the proposed model has higher prediction accuracy than the linear integrated model without meteorological factors.For example,compared with linear integrated models,Imapes of all VMD-GAM-Correction models in Lanzhou are above 9%.The research results indicate that the GAM non-linear integrated model with meteorological factors for error correction can effectively improve accuracy in predicting CO.For the third problem,this research introduces Bas Elman method to construct VMD-GAM-Bas Elman nonlinear combined model for avoiding less robust forecasts caused by the uncertainty of optimal model selection.The proposed VMD-GAM-Correction model is used to predict daily average air quality indexes in Lhasa and Lijiang.The prediction accuracy of VMD-GAM-Bas Elman is higher than that of the optimal individual integrated models and linear combined model.For example,compared with the optimal individual integrated model and linear combined model,MAPEs of VMD-GAM-Bas Elman models for AQI prediction of Lhasa improve by8.89%and 1.35%respectively.The research results show that the non-linear combined model based on Bas Elman is more robust for predicting daily air quality indexes.For the shortcomings of the integrated prediction of air pollution prediction,this research proposes an integrated prediction model based on feature refactor,designs a nonlinear integrated method integrating meteorological factors,and constructs a robust nonlinear combined prediction model,which provides important prediction theoretical support and data support for air pollution prevention and control.