Based On The Spatial-temporal Data Model Of PM2.5 Concentration Prediction

With the rapid development of social economy,industry,agriculture and constructionproduce a lot of PM2.5.A large number of studies have shown that PM2.5 will damage human health,and increase the incidence of human disease;at the same time,PM2.5 has strong destructive effect on the environment,which is one of the main reasons for the formation of haze.There is a lot of research on the formation,influence and emergency measures of PM2.5.In order to provide a reliable basis to control PM2.5 pollution,this paper has carried on the study for statistical modeling of PM2.5,which is driven by spatial-temporal data.With the social attention to a series of issues brought about by environmental pollution,many sites provide real-time data on the level of pollution in domestic urban areas.In order to build the database used for modeling,a real-time online data acquisition system based on the network API interface is developed,the acquisition system using JAVA programming software and POST way of parameters passing to achieve data collection of hourly PM2.5concentration value at the monitoring site of a domestic city.The current modeling of PM2.5 concentration is mostly focused on the selection of linear model predictors.Based on the spatial characteristics of PM2.5 concentration,the relationship between the PM2.5 concentration in the monitoring site of urban area is studied,and the regions with great influence on the target monitoring areas are selected as the spatial variables of the model;the PM2.5 concentration is also affected by other external variables,the existing results show that the influence of temperature,humidity,wind speed and traffic volume on PM2.5 concentration is very larg,therefore,the relationship about PM2.5 concentration between weather variable and traffic volume is studied,and the variables with strong correlation with PM2.5 concentration are selected as external variables;PM2.5 concentration has the dynamic characteristics of time,that is the current concentration value is affected by the past,therefore,the PM2.5 concentration in the past time is chosen as the time variables.On the basis of spatial variables,time variables and external variables,the spatial-temporal data model is proposed.Finally,the optimal model structure of the spatial-temporal data model is selected by the information criterion AIC,and validate the spatial-temporal data model with the mean interpolated data set.Due to network,equipment and other reasons,the data may be missing.In this paper,Kalman filter based on AR model,EM algorithm and regression model are used to interpolatethe missing data,and the method with the highest interpolation precision is selected.Based on the complete data set and the spatial-temporal data model,the RFF algorithm is used to predict the PM2.5 concentration.The results show that the RFF algorithm has high online calculation accuracy.