Robust segmentation of switching dynamics in time series

A method to identify nonstationary dynamics in time series, based on Annealed Competition of Experts algorithm (ACE), has been developed by J. Kohlmorgen, et al. [2000]. Incorrect selection of embedding dimension and time delay of the signal significantly affects the performance of the ACE method. In this paper, we utilize systematic approaches based on mutual information and false nearest neighbor to determine appropriate embedding dimension and time delay. Using this ameliorated approach, we have enhanced the performance of the ACE algorithm in determining the location of the switching dynamics in time series. Furthermore, application of the ACE algorithm to time series containing multiple time-varying frequency components reveals that the method achieves better time resolution than does the short-time Fourier transform.