Feature extraction of time series by using the multifractal signals and its applications

摘要

This report deals with feature extraction of time series based on the multifractal and its applications. We introduce the representation of multifractal by using the wavelet coefficients in contrast to the conventional definition in the time domain. The relation among the multifractal and the moments of wavelet coefficients for the time series is derived. The higher-order moments of wavelet coefficient are easy to calculate, and we can have precise information for each dilation index m. As an application, the basic prescribed waveforms such as the fractal time series and sinusoidal waves are recognized by the system, and also the system is applied to the feature extraction of the time series of stock trends.