Abstract
Characterizing complex time series observed in the natural world is generally very difficult. Fourier analyses or wavelet analyses are often used for it but satisfactory results are hard to gain. One of the reasons of this is treating a compound signal as a single signal and separation of signals from the original will be needed in such cases. In this study, a short-term prediction method based on chaos theory is applied to feature extraction of complex time series. The prediction method is developed for predictive control method by the authors and the optimal sampling period is defined in it as one of important features of time series. The efficiency of the optimal sampling period as a feature for separation of time series is discussed fundamentally through numerical examples for a forced damped pendulum.
