New hypothesis testing-based rapid change detection for power grid system monitoring

The vulnerability of power grid systems to malicious attacks is one of the most pressing problems faced concerning power grid systems. Based on the dynamics of the generators, we show that the time evolution of the power grid system can be modelled by a discrete-time linear state-space model. We employ approaches based on hypothesis testing for failure and intrusion detection of the monitored power grid system. We develop a new locally optimum unknown direction (LOUD) test to detect changes in matrices or vectors and apply this approach to power grid failure and intrusion detection problems. We provide numerical results which show that, unlike the standard generalised likelihood ratio-based approach, the LOUD test is able to produce decisions right after the change has occurred without waiting to collect additional data while it performs nearly as good, within a few percent in the case considered, as the optimum but unachievable likelihood ratio test for the known change. We employ realistic simulations of the IEEE 14 bus system to more fully evaluate the LOUD test.