Chaos-based two-stage framework for abnormal ship behavior identification using AIS data

Abstract

Abnormal ship navigation behavior is a critical precursor to maritime incidents. Conventional detectors based on rules often fail to characterize the nonlinear and irregular dynamics that arise during maneuvering. This study proposes a ship behavior analytics framework with two stages informed by chaos theory that integrates multivariate chaos screening with Cross-Track Distance (XTD) control monitoring for robust and operationally interpretable abnormality assessment from Automatic Identification System (AIS) data. AIS time series are resampled to a uniform time step, and missing observations are handled using segmentation controlled by data gaps to preserve consistent sliding window evaluation while mitigating discontinuity artifacts. The first stage performs sliding window chaos screening for Speed Over Ground (SOG), Course Over Ground (COG), and Drift Angle (DRIFT) using the 0-1 test, Largest Lyapunov Exponent (LLE), and delay coordinate phase space reconstruction. Using multiple indicators improves robustness to noise and parameter sensitivity and provides complementary evidence of dynamical instability. The second stage analyzes XTD dynamics to reveal incipient control instability even when lateral deviation remains moderate. Derivative and statistical descriptors computed within each window quantify intensified lateral corrections and elevated activity at short horizon frequencies. Experimental results show that combining multivariate chaos metrics with control descriptors derived from XTD improves sensitivity to subtle yet operationally meaningful instabilities and supports interpretable abnormal ship behavior identification. The proposed framework provides an extensible basis for navigational safety monitoring and maritime surveillance and can support maritime traffic management through targeted screening and diagnostics.