Construction And Empirical Study Of Air Pollution Forecast Model For Typical Cities In Beijing-Tianjin-Hebei Region Based On Data Analysis

In recent years,with the continuous development of Chinese industry and economy,the air pollution issues have become more and more serious.Especially,air pollution in the Beijing-Tianjin-Hebei region was particularly significant,which was mainly reflected in the high air quality index（AQI）and the serious over-standard concentration of atmospheric pollutants,such as PM_2.5 and O₃.At present,there are many air pollution monitoring datas in China,but they are not fully utilized.Therefore,the research on the construction of air pollution prediction models has attracted more and more attention,which could provide a strong reference for the traceability and control of air pollution.Among the existing models,statistical prediction models（such as linear prediction models and BP artificial neural network models）are the most widely used.Compared with numerical prediction models,statistical prediction models have the advangtages of simple,economical,and easy to implement.However,there are still some challenges for the statistical prediction models,such as poor fitting of non-linear data and interpretability.In this thesis,the air pollution prediction models based on data analysis was proposed,and visualization and sensitivity analysis of proposed models were carried out.The obtained models presented good prediction performance,certain interpretability and application value.（1）In this thesis,the air quality data and related meteorological data of typical cities（Beijing,Tianjin,Shijiazhuang,Tangshan,Xingtai）in Beijing-Tianjin-Hebei from 2015 to 2018 were obtained,and the air pollution status of each city was studied.The research showed that although the air quality of each city had obviously improved from 2015 to 2018,several typical pollutants such as PM_2.5 and O₃ still exceeded the standard.Based on the status,the thesis analyzed the primary pollutants and their time distribution characteristics in each city.The results showed that PM_2.5 and O₃ were the primary pollutants in winter and summer respectively,and PM_2.5and O₃ were finally determined to be the key atmospheric conventional pollutants in each city.At the same time,in order to improve the accuracy of the prediction models,the correlation analysis of the meteorological indicators of PM_2.5 and O₃ in each city was carried out,and the meteorological indicators that should be added to the prediction model of each city were determined.（2）Based on the above data analysis,three kinds of PM_2.5 and O₃ prediction models were constructed for each city respectively.The results showed that the prediction performance of the BP neural network model was significantly better than the linear prediction model.Subsequently,by adjusting and optimizing the parameters of the BP neural network model,Beijing,Tianjin,Shijiazhuang,Tangshan,and Xingtai PM_2.5-optimal parameter BP neural network models were finally obtained.The coefficients of determination（R²）were:0.93,0.93,0.97,and 0.95;O₃-optimal parameter BP neural network models,the determination coefficients（R²）were:0.90,0.88,0.90,0.93 and 0.90 respectively.（3）The inversitagtion results showed that although the prediction accuracy of the above models was high,there was still a defect,i.e.poor prediction ability for high concentrations of PM_2.5 and O₃.Therefore,based on the significant seasonal characteristics of PM_2.5 and O₃,Beijing was used as an example to construct winter PM_2.5 and summer O₃-BP neural network models in 2018 respectively.The optimization results indicated that the fitting ability of winter PM_2.5 and summer O₃ was significantly improved,and the determination coefficients（R²）increased to 0.97 and0.90 respectively.It indicated that the optimization effect of the model based on the seasonal characteristics of the city was significant,and had certain enlightenment for the optimization of the future prediction model.At the same time,in order to facilitate the application of the model in engineering practice,Python software was used to realize the visualization of the model,and constructed the Beijing PM_2.5 and O₃prediction program based on the prediction models.The program had presented a promising application value after verification.（4）After completing the construction of PM_2.5 and O₃-BP prediction models for various cities,the detail sensitivity analysis on the models was also conducted to enhance the interpretability of the model and clarify the sensitivity of the model output to the changes of each model variable.The results showed that the sensitivity of PM_2.5and O₃-BP prediction models in each city to the changes of various model variables was much lower than 100%,indicating that the performance was relatively stable when responding to changes in variables.At the same time,by comparing the sensitivity of each model variable in different cities,it is found that when BP neural network model was applied to different cities,the sensitivity of each model variable had certain difference.The reason might be related to the difference of geographical location,energy and industrial structure and enterprise production characteristics of each city.Therefore,it is necessary to evaluate the sensitivity of each model variable before its application in different regions.