Change detection in complex dynamical systems using intrinsic phase and amplitude synchronization

TL;DR

This paper introduces a new method for detecting sharp change points in complex, noisy dynamical systems by analyzing phase and amplitude synchronization across intrinsic time scale decomposition components, improving early detection.

Contribution

The paper presents a novel approach combining ITD-based phase and amplitude analysis with a mutual agreement concept and InSync statistic for high-accuracy change point detection.

Findings

Detects change points 62% earlier than existing methods

Effective in neurophysiological and industrial data

High sensitivity and specificity in identifying transitions

Abstract

We present an approach for the detection of sharp change points (short-lived and persistent) in nonlinear and nonstationary dynamic systems under high levels of noise by tracking the local phase and amplitude synchronization among the components of a univariate time series signal. The signal components are derived via Intrinsic Time scale Decomposition (ITD)--a nonlinear, non-parametric analysis method. We show that the signatures of sharp change points are retained across multiple ITD components with a significantly higher probability as compared to random signal fluctuations. Theoretical results are presented to show that combining the change point information retained across a specific set of ITD components offers the possibility of detecting sharp transitions with high specificity and sensitivity. Subsequently, we introduce a concept of mutual agreement to identify the set of ITD components that are most likely to capture the information about dynamical changes of interest and define an InSync statistic to capture this local information. Extensive numerical, as well as real-world case studies involving benchmark neurophysiological processes and industrial machine sensor data, suggest that the present method can detect sharp change points, on an average 62% earlier (in terms of average run length) as compared to other contemporary methods tested.