Enlarged Scaling Ranges and Smooth Boundary for Detection of Chaos
Zhenya HE,Wenjiang Pei,Qingqing Sun,Luxi Yang
DOI: https://doi.org/10.3969/j.issn.1004-9037.2000.04.001
2000-01-01
Abstract:Correlative dimension is one of the most popular measures to the analysis of nonlinear time series. However, one may obtain suspicious results when applying it to a short and noisy data set. Our intention in this. paper is to concentrate on the estimation of correlative dimension from a short and noisy data set. Compared with the basic G-P algorithm, two developments are achieved in the proposed algorithm. Firstly, in order to enlarge scaling ranges from a short data set, we take into account the local structure of the attractor for the calculation of the correlative dimension. In detail, we apply a special vector quantization technique, the Neural Gas algorithm, to find reference vectors that represent the structure of the data manifold. Secondly, we modify the Heavside function to fuzzy probabilities in order to handle the noise present in the data. From a theoretical point of view the invariance of the probability measure is essential and must not be violated. From an experimental point of view, however, we are forced to calculate relative frequencies serving as estimates for the requested probabilities. We expect the fuzziness of the new. functions that enter the probability calculations to be ideal to compensate for this noise influence. Finally, The improvement is demonstrated for short and noisy time series generated by Lorenz system.