摘要
提出了描述变量间包括线性和非线性的相关程度的一种以互信息为基础的广义相关系数指标Rg ,给出三种算法 :等间距法、等概率法和等概率递推法。等概率法计算出的结果与序列的总体概率分布无关 ,分组较为客观、合理 ,计算简便 ,较其他算法更优越。在等概率法基础上 ,通过统计实验途径 。
The correlation between variables is of importance in the analysis of nonlinear dynamic system. However traditional approaches are usually centered on the analysis of autocorrelation coefficient and mutual correlation coefficient or other indexes that only reflect the linear correlation between variables. Obviously it is not sufficient for the real system that often has nonlinear correlation structure. Mutual information is used in such analysis. A general correlation coefficient is proposed and three different algorithms: equidistance algorithm, equal probability algorithm and successively equal probability algorithm, are discussed. The equal probability algorithm, which has no relationship with the studied population distribution, is the best and simplest. A quantile table of general correlation coefficient based on the equal probability algorithm is provided for statistic test purpose.
出处
《四川大学学报（工程科学版）》
EI
CAS
CSCD
2002年第3期1-5,共5页
Journal of Sichuan University (Engineering Science Edition)
基金
国家自然科学基金重大资助项目 (5 0 0 9962 0 )
