Kalman Filter Based Secure State Estimation and Individual Attacked Sensor Detection in Cyber-Physical Systems

In this paper we propose two real-time attack detection and secure state estimation algorithms, namely Rolling Window Detector (RWD) and Novel Residual Detector (NRD). These algorithms are basically developed based on Kalman state estimation. In the former, we present a statistical testing approach which is handled over a finite time horizon $T$ to detect individual attacked sensors. The latter extends the X2-detector to be able to detect individual compromised sensors. Both methods then will be employed together with a modified version of Kalman filter to perform a secure state estimation with a relatively low estimation error. Efficiency of the algorithms will be assessed in both unstealthy and stealthy scenarios. Productivity of the methods will be underlined in the stealthy case, which is of much more significance among cyber-security challenges. Simulation results on an IEEE 14-bus power grid test system along with a comprehensive comparison between the performance of RWD and NRD with a recently introduced tool, which is the only other method that tries to detect individual attacked sensors, proves the effectiveness of the algorithms.