State Estimation with ARMarkov Models

The ARMarkov models were originally developed for adaptive neural control, and later for predictive control, and state-space identification. Recently, an interaction matrix formulation has been developed that explains the internal structure of the ARMarkov models and their connection to the state-space representation. Using the interaction matrix formulation, we show in this paper how a state estimator can be identified directly from input-output data. The conventional approach is to design such a state estimator from knowledge of the plant, and the difficult-to-obtain process and measurement noise statistics. A numerical example compares the identified state estimator with an optimal Kalman filter derived with perfect knowledge of the plant and noise statistics. 1 Graduate Student, Department of Mechanical Engineering. 2 Assistant Professor, Department of Mechanical and Aerospace Engineering, Dissertation Advisor. 3 Professor, Department of Mechanical Engineering, Dissertation Co-Advisor.