Analysis Of Influencing Factors Of Haze Monitoring In Chengdu-Chongqing Twin Cities Economic Circle Based On Hiersrchical Bayesian Model

In October 2021,The State Council issued the Outline of Construction Planning of Chengdu-Chongqing Twin Cities Economic Circle,raising the construction of Chengdu-Chongqing twin cities economic circle as a national strategy.In February this year,the Ministry of Ecology and Environment pointed out clearly in the Ecological environment Protection Plan of Chengdu-Chongqing twin cities economic circle that regional moderate and mild air pollution is prone to occur during the 14 th Five-year Plan period in Chengdu-Chongqing area,and further improvement is difficult.The analysis of haze influence factors is the key link to solve the problem of air pollution.Although there are many studies on the cause of haze at home and abroad,the researches on the atmospheric environment and haze monitoring of the ChengduChongqing twin cities economic circle as a whole system is relatively lacking.In terms of research methods,mostly focusing on the traditional statistical modeling and conventional machine learning algorithms,and less consideration is given to the nested data structure characteristics between haze monitoring stations and cities,especially the monitoring station data and the economic-social indicators of the city where they are located.Due to the influence of the nested structure,the modeling effect is not ideal,and the hierarchical linear model can just make up for this defect.With the help of Bayesian statistics,it has the advantage of using prior information and learning mechanism,which is helpful to the model fitting optimization.In this paper,the hierarchical linear model of complex data with nested structure is taken as the research object.On the basis of analyzing the deficiency of classical frequentist statistical inference,the bayesian statistical theory is introduced to systematically study the bayesian statistical analysis of hierarchical linear model,including: Parameter estimation,hypothesis testing,model selection and evaluation,etc.Then they were compared with classical model,and the advantages of hierarchical model bayesian analysis are demonstrated.Based on the basic analysis of haze monitoring data and economic-social meteorological indicators in Chengdu-Chongqing twin cities economic circle,a model exploration was carried out,and a three-level Bayesian development model of influencing factors of haze monitoring in ChengduChongqing twin cities economic circle was established according to time-station-region.A comparative analysis of least squares estimation,iterative generalized least squares estimation,empirical bayesian and completely bayesian was carried out.Based on the basic analysis and model construction,problems are found and effective suggestions for haze control was put forward.Details are as follows:Firstly,based on hierarchical linear model,bayesian statistical inference of hierarchical linear model is studied,including bayesian estimation method of hierarchical linear model,bayesian hypothesis testing of fixed effect regression coefficient,random effect regression coefficient and variance component of random effect,Model selection method based on posterior probability and Bayes factor,Model evaluation method based on bayesian information criterion and deviance information criterion.The maximum likelihood estimation,limited maximum likelihood estimation,empirical Bayes and completely Bayes of hierarchical linear models are compared and analyzed to highlight the advantages of completely Bayes.It is found that the thinking essence of hierarchical linear model’s classical the maximum likelihood estimation and limited maximum likelihood estimation is empirical the thought of bayes.Because the parameters are treated as random variables,the construction thinking of hierarchical model is highly consistent with the core idea of Bayes,thus laying the inherent rationality foundation for the fusion of hierarchical model and bayesian analysis.Secondly,the crawler algorithm based on python acquired hourly monitoring data of air quality in Chengdu-Chongqing twin cities economic circle from 2014 to 2021,The proportion of time when air quality is in a state of pollution and the average values of PM2.5 years,quarters and months of PM2.5 at the state control monitoring site are visually analyzed,and the main characteristics and dynamic trends of PM2.5 pollution in Chengdu and Chongqing are extracted.Combined with meteorological,population,regional economic development,industrial structure,agricultural development and other factors related to haze,the linear regression model was explored and found that the modeling effect was not good.Thirdly,by making full use of the nested structure characteristics of haze monitoring data,the hierarchical bayesian development model of influencing factors of haze monitoring in chengdu-Chongqing twin cities economic circle with three levels of time-site-region is constructed.The comparative analysis of classical least squares estimation,iterative generalized least squares estimation,empirical bayes and completely bayes,as well as model selection and evaluation,demonstrates the advantages of completely bayes method.Finally,based on the basic analysis and model construction of haze monitoring data,this paper analyzes and finds out the influencing factors of haze and the main problems existing in air quality in Chengdu-Chongqing twin cities economic circle,and puts forward effective countermeasures for haze control.