时间序列非平稳检测方法的对比分析

对比分析了检测时间序列是否平稳的两种方法: 一种是基于简单的统计方法检测时间序列非平稳性的run-test (简称RT)方法(考虑了3种典型方法: 均值RT、方差RT和标准差RT), 另一种是利用动力非平稳检测的图示方法(space time-index, STI)。 通过分析人工生成的已知平稳性的时间序列和实际湍流时间序列, 对比了两种方法在时间序列非平稳检测上各自的优劣: RT方法简单、高效, 但精确度低; STI方法精确度高, 但复杂、低效。同时, 分析了3种不同的RT方法对不同类型时间序列的平稳性检测能力, 表明方差RT方法最适用于湍流时间序列非平稳性检测。

Comparative Analysis of Detection Methods of Non-stationarity in Time Series

Two methods are applied to detect non-stationarity in time series. One is the run-test (RT) method, which depends on statistic theory; the other is a graphical method called space time-index (STI), which is a dynamical method. Three different RT methods and STI method are given to detect the non-stationarity in artificial time series with known non-stationarity and experimental turbulence time series measured in atmospheric boundary layer. The results show that the RT method which has lower accuracy is simpler and more efficient than STI. Besides comparing the merits of the STI and RT methods, the results also show that the variance RT method is the most effective to detect the non-stationarity in experimental turbulence time series among the three ones of RT methods.