Dynamic model has always been the focus of data science research. The results of dynamic models with different physical backgrounds vary greatly. Considering the time-varying characteristics and heteroskedasticity of long time exchange rate series, the GARCH (Generalized Autoregressive Conditional Heteroskedasticity) model can be used to depict the pattern of variance fluctuation and clustering over time, and the research on variance can be extended to time-varying conditional variance. In this paper, the EGARCH (Exponential GARCH) model is further adopted to analyze the high-frequency exchange rate data from 2005 to 2017, which can reduce the parameter restrictions of the GARCH model, and at the same time, the characteristics of the exponential ensure the non-negative nature of the estimated variance, so as to predict the volatility more accurately. Moreover, the different effects of positive and negative random shocks on volatility can be analyzed to fully describe the data asymmetry. Finally, we use the clustering algorithm of the EGARCH model for data mining, evaluating the influence of endogenous variables and exogenous events on the RMB exchange rate.