| Summary: | 碩士 === 淡江大學 === 水資源及環境工程學系 === 92 === Various methods of transforming non-Gaussian time series to Gaussian time series have been developed. However, most of the methods are still not convincing. In this study, we would like to look into a simpler method of transforming non-Gaussian time series.
The bias of autocorrelation function is first analyzed by varying the distribution characteristics of non-Gaussian time series. Second, the autocorrelation function of original time series and synthesized time series are compared. The synthesized time series are generated with different order series and different marginal distributions. It is found that the bias of auto correlation function is influenced by the skewness instead of the mean and variance of the same marginal distribution. Moreover, the bias will be identical when the same marginal distribution and the coefficient of skewness are used for an AR model of a specific order.
It is also shown that the GOS generation method can not only model the correlation characteristics of a time series but also the mean, variance and coefficient of skewness. Therefore, it is a feasible method to analyze non-Gaussian time series.
|
|---|
