
第 卷第 期
哈尔滨工程大学学报
Vol
年 月
Journal of Harbin Engineering University
Apr
确 定 相 空 间 重 构 嵌 入 维 数 的 研 究
刘树勇 朱石坚 俞翔
海军工程大学 振动与噪声研究所湖北 武汉
摘要由于相空间重构是船舶机械系统混沌振动识别的重要基础研究了确定相空间重构嵌入维数的新方法首
先应用相空间局部流形变化最小原理和互信息方法计算重构延迟时间在此基础上提出无阈值计算嵌入维数方法
在嵌入维数过低的情况下吸引子被压缩在低维空间中产生了伪邻点随着维数的增加邻近点对之间的距离逐渐
增大伪邻点数逐步减小 研究发现吸引子上邻近点之间距离的变化速度随嵌入维数增加并不是无限增大而是受
一定条件的限制通过定义指标 G 对邻近点对的演化行为进行定量刻画当嵌入维数达到某一特定值时G 曲线趋向
于饱和吸引子结构最为有序因而所对应的嵌入维数为最小嵌入维数该方法在计算过程中不需要选择阈值因而
更具有客观性 通过对 Lorenz 以及 Henon 系统产生的混沌数据进行分析证实了该方法的有效性同时研究了数据
长度噪声以及延迟时间对该方法的影响
关键词混沌识别吸引子相空间重构嵌入维数
中图分类号TN文献标识码A文章编号
Determinating the embedding dimension in phase space reconstruction
LIU Shuyong ZHU Shijian YU Xiang
Institute of Noise and Vibration Naval University of Engineering Wuhan China
AbstractThis paper presents a new method for determining embedding dimensions in phase space reconstruction
The method is the basis of chaotically vibratory signal identification for shipsmechanical systemsThe principles of
local manifold distortion minimization and the mutual information method were applied to determine delay timeOn
this basis we proposed a new method to calculate embedding dimensions without the use of a thresholdSimulation
results showed that if the embedding dimension is lower the attractors are compressed and pseudo adjacent points
are formedHowever with increasing dimensions the distance between adjacent points increases while the num
ber of pseudo adjacent points decreasesWe found that the rate of change in distance between adjacent points of the
attractor increases with increasing dimensions but not infinitely as it is limited by certain conditionsTo charac
terize the evolutionary behavior of the adjacent point pair an index G is defined as a function of the embedding di
mension dWhen parameter d approaches d
the Gd curve tends to be saturated and the corresponding d
is
the optimal dimension for reconstructionCompared with the FNN false nearest neighbor method the method
presented is more objective because it is not necessary to choose the thresholdThe effectiveness of this method was
verified with chaotic signals generated from Lorenz and Henon systemsFinally the influences of noise data length
and delay time on this method were studied as well
Keywo rdschaotic identification attractor phase space reconstruction embedding dimension
收稿日期
基金项目国家自然科学基金资助项目
作者简介刘树勇 男 博士研究生Email lsydhsinacom
船舶机械系统在一定的参数条件下可能产生混
沌振动 无论是为了尽量避免还是为了充分利用混
沌
其首要问题是需要对混沌振动信号进行准确
识别通常采用的方法是对信号进行定性和定量分
析如分析实测信号的频谱图相空间图计算关联
维数
Lyapunov 指数
和 K
熵等这些计算都是
基于对实测数据进行相空间重构而实现的根据