Detecting dynamical change in nonlinear time series

We present a robust, model-independent technique for measuring changes in the dynamics underlying nonlinear time-serial data. After constructing discrete density distributions of phase-space points on the attractor for time-windowed data sets, we measure the dissimilarity between density distributions via L-1-distance and chi(2) statistics. The discriminating power of the new measures is first tested on the Lorenz model and then applied to EEG data to detect the transition between non-seizure and epileptic activity. We find a clear superiority of the new measures in comparison to traditional nonlinear measures as discriminators of changing dynamics. (C) 1999 Published by Elsevier Science B.V. All rights reserved.