Driving Force Extraction And Prediction Model Applied In Atmospheric Composition Analysis

Based on the theory of Slow Feature Analysis, we put forward a method which extracted drving force from nonlinear timeseries itself. The new method is different from the pass one according to timeseries features and physical impact to set ideal drving force, but extracted from time series itself.Then we embedding drving force in the reconstruction of nonlinear time series prediction model, to examine the rationality of the real drving force of extraction.Starting from ideal nonlinear time series produced by the Logistic mapping.Extracting the external driving forces by slow feature analysis, then embedding drving force into the "ideal" nonlinear time series forecasting model, and to predict the experiment. Results show that the prediction model can effectively improve the accuracy of prediction.Using the method in actual time series, get the following conclusion:(1) To select the embedding dimension while extracting the factor of external driving forces in a certain range has no significant influence to the factor of external focing collected in the real life.Different embedding dimension has the same trend of the factor of external focing,the only difference is that a very little displacement in fine degree of the data.(2) Based on slow feature analysis, embedding the external factors in the building of the prediction model can improve the accuracy of the forecasting effictively.The correlation coefficient in the first step of the forecasting increased from0.6982to0.7390for the time series of atmospheric aerosol number concentration.To the time series of the ozone in winter,the correlation coefficient in the first step of the forecasting increased from0.6015to0.7828.And the correlation coefficient in the first step of the forecasting increased from0.7391to0.9160for the forecasting test of ozone time series in the past20years.(3) According to the theory of slow feature analysis,pull the slower output signal in the building of the prediction model as a second external forcing factors,to estabalish a double external forcing mode. The correlation coefficient in the first step of the forecasting increased from0.6982to0.7475for the time series of atmospheric aerosol number concentration.To the time series of the ozone in winter,the correlation coefficient in the first step of the forecasting increased from0.6015to0.9079.This indicates that the second external forcing factors has been embedded in can also improve the accuracy of forecasting in a small scale.