Studies On The Dynamics Statistical Model Of SO₂ Pollution Forecast In Xi'an City

This paper makes use of SO₂ concentration value monitored daily from January 2002 to December 2005 of Xi'an city and the ground meteorological data in the corresponding period, starting from air pollution diffusion equation, by analysis draws Xi'an SO₂ air pollution forecast dynamics statistical models. The model is considered the role of weather conditions on the concentration of pollution, and pollution emissions into the role of pollution concentrations, in comparison with the previous air pollution forecast statistical models, is based on the more credible physical base.Before Modelling,tranquilization test of source release and SO₂ concentration time distribution assumptions test is conducted, the choices for modeling modules and mathematical models ensure the mathematical direction of modelling.While on the basis of analyse on spatial and temporal characters of SO₂ Concentration in Xi'an city, the reasons for the trend of changes in the initial exploration are carried out . The choice to forecasting factors upgrades the degree of attention to several indirect and qualitative factors, the partial correlation analysis between forecast factors and SO₂ concentration and the test results after modeling of the significant level show that the choice to forecasting factors is suitable relatively.In the Statistical modeling phase, aimed at the phenomenon of serious relatively multicollinearity, by comparison to the diverse regression projects, the respective applications with Ridge regression and Partial Least Squares regression establishes SO₂ concentration forecasting equation, by compared with common statistical regression methods, the greatest advantages of Partial Least Squares regression method are its withdrawal to the biggest variables information components, elimination of harms from muticollinearity, and the most explanatory to variables.The test results after modeling to Ridge regression model and Partial Least Squares regression model indicate that the quality of Partial Least Squares regression dynamicsstatistical model as recommended model is relatively better, its fitting error rate is 18.2%, the scores of forecast class veracity and veracity achieves 96.1 and 79.8 respectively;the test result of the basic assumptions of the linear regression to Partial Least Squares regression model shows that it meet the three basic assumptions of linear regression model ,and the all statistics dealing in the stages of statistical modeling is meaningful;the test result of a number of indicators indicates that the quality of dynamics statistical model built in the paper attain the expected request.