中文摘要：

传统多元多尺度熵算法在处理有限长时间序列时,会使均值曲线产生较大的波动,并且阈值的选取也会对结果产生较大的影响。因此,在传统多元多尺度熵的基础上首先对传统粗粒化方式进行了改进,改进后的算法采用滑动均值滤波使粗粒化后各尺度上的时间序列与原始时间序列长度一致,减小了所计算多元多尺度熵的离散性。此外,本文算法在保持多元样本熵硬阈值优点的同时,通过定义模糊隶属度函数来统计两复合延迟矢量距离略大于阈值的情况,既降低了传统方法对阈值的依赖性,也很好的解决了传统阈值所导致的不稳定现象。最后用仿真数据对该算法进行了验证,并将其应用于不同人体步态加速度信号的复杂度评价和分类,结果表明改进算法的识别效果明显优于传统多元多尺度熵。

英文摘要：

Traditional Multiple multiscale entropy algorithm at moment of dealing with time series of limited length,would led to curve fluctuations larger and threshold selection will also have a greater impact on the results. There-fore,on the basis of traditional Multiple multiscale entropy,firstly this paper improved the way of traditional coarsegrained process,the method improved coarse-grained way of traditional multiple multiscale sample entropy by slid-ing mean filter so that coarse-grained time series equal to the length of original time series on each scale,reduce thecompute discreteness of multivariate multiscale entropy. In addition,algorithm both maintain the advantage of hardthreshold of multiple multiscale sample entroy and count the distance of two composite delay vector slightly greaterthan the threshold value by defining fuzzy membership function,not only reducing the dependence of the thresholdof multiple multiscale sample entropy,but also solving the instability caused by the traditional threshold. Finally,the algorithm was validated in the emulated data,and applied it to different human gait acceleration signal complexi-ty evaluation and classification.The results show that improved multiple multiscale entropy recognition is betterthan traditional multivariate multiscale entropy.