This paper proposes a rigorous anomaly detection
scheme, developed to spot power system operational changes
which are inconsistent with the models used by operators. This
novel technique relies on a state observer, with guaranteed
estimation error convergence, suitable to be implemented in real
time, and it has been developed to fully address this important
issue in power systems. The proposed method is fitted to the
highly nonlinear characteristics of the network, with the states
of the nonlinear generator model being estimated by means
of a linear time-varying estimation scheme. Given the reliance
of the existing dynamic security assessment tools in industry
on nominal power system models, the suggested methodology
addresses cases when there is deviation from assumed system
dynamics, enhancing operators’ awareness of system operation.
It is based on a decision scheme relying on analytical computation
of thresholds, not involving empirical criteria which are likely to
introduce inaccurate outcomes. Since false-alarms are guaranteed
to be absent, the proposed technique turns out to be very useful
for system monitoring and control. The effectiveness of the
anomaly detection algorithm is shown through detailed realistic
case studies in two power system models.