UC Riverside

This dissertation studies two topics related to the problem of estimation in networked systems, namely, the problem of secure state estimation against possible stealthy false data injection attacks in electric power network, and the problem of distributed dynamic state estimation using networked local multi-agents. The two topics are, respectively, presented in Part I and Part II.

The study in Part I is motivated by the importance of the state estimation in electric power systems as many applications such as Optimal Power Flow, Automatic Generation Control and Contingency Analysis are highly dependent on the state estimation. Therefore, it is critical to have an accurate and reliable estimate of the states as bad estimates will offer the system operator with inaccurate information about the system, which may cause wrong decisions and finally, cause malfunction or even power blackout in the network. The stealthy attack, as a strategically designed false data injection attack against the power system state estimation mechanism, is able to let the corrupted measurements bypass residue-based bad data detectors with the same probability as that of uncorrupted measurements, and to fool the system operator with the deviated estimates. While most of the existing articles assume the network topology to be fixed, the effects of switching topologies on such an attack is shown in this work. In Part I, a new mechanism is proposed to eliminate the possibility of such an attack via strategically shutting down some preselected transmission lines by turns and therefore switching the network topologies. The necessary and sufficient condition to achieve such elimination, as well as the general form of possible attacks when the elimination is impossible, are both formulated. The case where the attack is only stealthy in a subset of the preselected is also studied. The general form of the possible estimate deviations caused by this "partially" stealthy attack is derived. Simulations and case studies are provided using different IEEE bus systems to show the efficiency of the proposed strategy, to discuss the countermeasures in the case when there always exist possible stealthy attacks, and to show how the possible deviations introduced by a "partially" stealthy attack could be affected by the decisions made by both the attacker and the system operator.

Part II studies the problem of distributed dynamic state estimation using networked local agents with sensing and communication abilities. This problem has become a popular research area in recent years due to its wide range of applications such as target tracking, region monitoring and area surveillance. Specifically, Part II considers the scenario where the local agents take local measurements and communicate with only their nearby neighbors to estimate the state of interest in a cooperative and fully distributed manner. A distributed hybrid information fusion (DHIF) algorithm is proposed in the scenario where the process model of the target and the sensing models of the local agents are linear and time varying. The proposed DHIF algorithm is shown to be fully distributed and hence scalable, to be run in an automated manner and hence adaptive to locally unknown changes in the network, to have agents communicate for only once during each sampling time interval and hence inexpensive in communication, and to be able to track the interested state with uniformly upper bounded estimate error covariance. It is also explored the very mild conditions on general directed time-varying graphs and joint network observability/detectability to guarantee the stochastic stability of the proposed algorithm. Then the DHIF algorithm is extended to two more general scenarios, namely, the scenario with nonlinearities involved in both the process and the sensing models, and the scenario with uncertain process models. In the former scenario, a nonlinear DHIF algorithm is proposed by adopting the unscented transformation approach. In the latter one, two algorithms are proposed by following the two well-known multiple model (MM) paradigms, namely, the first order generalized pseudo Bayesian and the interacting MM approaches. The extended algorithms in both scenarios inherit the aforementioned advantages of the original DHIF algorithm. The stability is also rigorously analyzed in each extended case with sufficient conditions formulated.