Anomaly Detection in Complex Systems

摘要

This paper presents a novel concept of anomaly detection in complex systems using tools of Symbolic Dynamics and Pattern Discovery. Anomaly is defined as a deviation from the nominal behaviour and can be associated with parametric changes that may gradually evolve. The proposed methodology is based on two-time-scale analysis of observed asymptotic behaviour of the dynamical system. This concept of small change detection in dynamical systems is elucidated through an example of the forced Duffing equation with a slowly varying dissipation parameter.