Prediction Of Air Quality Of IMFS-Elman-Hybrid Neural Network Based On Data Decomposition And Reconstruction

With In this era of fast economic development and living standards rise,the quality problem has gradually become the focus of attention at home and abroad,which has a great impact on every enterprise and even the country.Quality management expert juran once said: life is behind the quality dam.It can be seen that the quality problem has a significant impact on social and economic life,and air quality is no exception.And the public has a higher demand for the quality of atmospheric environment.However,with the deepening of urbanization and industrialization,air contamination has become a a hard nut to crack for human beings to face and overcome.On the hand,how to accurately predict the air quality situation and provide a reference model for the optimization of real-time monitoring of air quality and continuous improvement of air quality.On the other hand,with the development of monitoring technology and the improvement of air quality monitoring system,more data can be used.And the rapid development of computer technology has started the trend of information and intelligence,which makes it possible to accurately predict the situation of air quality.The research focus on the air quality data of Shanghai,constructing the IMFs-Elman-Hybird neural network prediction model based on data decomposition,which has practical application value for the research of air quality prediction.Aiming at the problem that Elman neural network is easily affected by data non-stationarity when predicting the Air Pollution Index(AQI),leading to a favourable forecasting trend but low accuracy.Based on complementary ensemble empirical modal decomposition,CEEMD,the author proposes a new hybrid predicting model.Firstly,the CEEMD algorithm is used to decompose the AQI sequence into Intrinsic Mode Function(IMF)and residual components at different time scales,which transformed the prediction study on the non-stationary AQI sequence into the study on the components of multiple stationary sequence functions.Secondly,the partial autocorrelation function and autocorrelation are used to calculate the lag period of each component that the input and output variables of each CEEMD-IMFs-Elman submodel;Thirdly,the comparison of traditional single models(Elman,BP,WNN neural networks)proves that the dynamic Elman neural network is most suitable for air quality index prediction;finally,the empirical mode decomposition family and the Elman neuralnetwork are combined to construct the EMD-IMFs Elman model,EEMD-IMFs-Elman model,CEEMD-IMFs-Elman model,CEEMD-IMFs-BP model,CEEMD-IMFs-WNN respectively.The results show that the mean square error of CEEMD-IMFs-Elman prediction model is 214.46,the average absolute error is 10.85 and the average absolute percentage error is 16.27.The frequency of predicting the correct number of days corresponding to the air quality rating was 78%.As for the problem that the high-noise data has a great impact on the prediction results and the model accuracy needs to be improved.First,calculate the average value of each imfs sequence and z-test,according to the characteristics of each component,those components were divided into three groups: high frequency,low frequency and trend items.Secondly,three groups of components are used to construct the prediction model,IMFs-Elman-Hybris,respectively.Finally,the prediction values of each sub-prediction model were summarized to compared with ceemd-imfs-elman model.According to the error index,the mean square error of the imfs-elman-hybird prediction model is9.86,the average absolute error is 2.40,and the average absolute percentage error is 3.48%,and the frequency of the number of days correctly predicted for the corresponding air quality level is increased to 97%.The model can effectively reduce the influence of non-stationarity on the model prediction results and realize the accurate prediction of air quality grade.It provides an effective basis for further prediction of the trend of AQI in various regions,and also provides a more sufficient reference for government decision-making and management departments to make air pollution control.