Analysis And Multvariate Nonlinear Prediction Model Of Ground-level Ozone Time Series In Shanghai

In recent years, ground-level ozone pollution has become a serious problem at eastern cities of China. With the development of urban, changes in energy structure, and the growth of vehicle number, Shanghai has faced the threat of ozone pollution. Now Shanghai is busy preparing for the 2010 Word Expo, and the air quality gets more attention from the word. The concentrations of ozone is an important indicator as well as the Olympic Games in 2008.Therefore, to analyze the relationships and rules between Ozone concentration and some precursors and meteorological conditions, to carry out the research on ozone concentration prediction are of vital importance for controlling the ozone concentration and avoiding photochemical smog events as well as guaranteed for the 2010 Word Expo.The hourly ozone concentration time series of Luwan and Pudong monitor stations are provided by Shanghai Environmental Monitoring Centre. By analyzing the variation of ozone concentration and its relations with some factors, the method of prediction of mutivariate nonlinear time series was studied. Base on that, an ozone concentration mutivariate nonlinear time series prediction model was constructed, and the prediction results were better than univariate condiction. The main points can be generalized as follows:Firstly, when the statistic methods were used for predicting ozone concentration, the variation of ozone concentration and the relations with some factors were analyzed. The results showed that ozone pollution was very serious in Shanghai, particularly in May and July. Ozone concentrations fluctuated periodically.The reasons of higher ozone concentration in May maybe that it has the proper conditions for ozone formed such as high temperature, little rain, strong solar radiation, low relative humidity. The ozone concentration was higher at weekend than the working day, but the concentration of NO,NO₂ was opposite at Luwan district which lies in the center of Shanghai.This phenomenon was named as "weekend ozone effect", which didn't occur in Pudong district.Secondly, under the framework of filter variable selection method, partial least square regression and stepwise regression were used for analyzing the main affect factors of ozone. The results showed that the main related factors of ozone were different in different seasons. Solar radiation and nitrogen oxides were important all over the year and relative humidity was also mainly to the spring and summer.Thirdly, the nonlinear approximation capability of Support Vector Machine was verified and comparative analyzed the nonlinear neural network model and linear stepwise regression model. The result showed that Support Vector Machines has perfect nonlinear approximation capability, especially for simulating the high-value point.Finally, the ground-level ozone prediction model was constructed based on nonlinear time series theory. Based on the correlation integral C-C method, the time delay and embedding dimension parameters were confirmed and then the multivariate phase space was constructed.After retrained the redundant information and noise and reduced the dimension of model inputs by PLS, Support Vector Machine was used for training.By mean of parameter optimal searching,the optimal model was obtained. The verification results showed that the prediction results of multivariate time series prediction was obtained more satisfactory precision than the prediction results of univariate time series prediction.