Optimal State Estimation of Cyber-Physical Systems

Abstract

This thesis focuses on state estimation problems of cyber-physical systems to overcome the challenges brought by their features, e.g, resource limitations in sensor networks and model uncertainties due to the changes in interconnections of components. To achieve smart allocation of the scarce resources of cyber-physical systems, two event-based state estimation problems are formulated and solved for systems described by hidden Markov models utilizing a new reference measure approach with the change of probability measure. For a linear Gaussian system with an energy harvesting sensor, the joint conditional probability distribution of the state and energy is obtained based on the event-triggered information received at the remote estimator under the energy-dependent measurement transmission policy. The robust state estimation problems are investigated for linear Gaussian systems with event-triggered scheduling and systems with unknown exogenous inputs utilizing the risk-sensitive approach, where closed-form risk-sensitive state estimates are derived. A fully distributed robust consensus-based filtering algorithm for systems measured by a sensor network is proposed with stability analysis on the local estimators. Based on the proposed results, state estimates that are optimal in a certain sense can be calculated in a simple recursive way, which are potentially applicable to industrial processes. The effectiveness of the proposed methods is validated by simulation examples, showing performance improvements in certain scenarios in the sense of mean estimation errors.