检验统计量的选取对替代数据方法的影响

替代数据方法是检验时间序列中是否存在非线性的重要统计方法，它可对动力系统实测时间序列数据的随机性或混沌特性进行判定。在应用此方法时，我们发现单一的检验统计量往往不能作出接受或拒绝零假设的正确判断。因此本文引入互信息函数作为检验统计量，同高阶自相关函数和时间反演不对称统计量对时序进行检验的方法进行了比较。实验结果表明，三种统计量对时序的检验各有优势和局限性，交叉检验可获得较好的判断结果。

Effect on Choice of Test Statistics for Surrogate Data Tests

Nonlinear time series,based on the theory of deterministic chaos,has become an important tool for understanding complex dynamics from measurements.It especially offers striking explantions for irregular time series,which nevertheless seems not to dominate a stochastic input. In the analysis of time series, the surrogate data test is often performed to investigate nonlinearity in the data.It is proposed to identify the random or chaotic natures of the data obtained in dynamic analysis. When the method is used, it is found that the test does not always succeed to reject the null hypotheses with a single nonlinear statistic. So the mutual information is used as one of the statistics. Three statistics methods are examined for nonlinear time series, which are the mutual information, the time reversal asymmetry and the three point autocorrelation. Experimental results show that the co-examining gains a good result.