Closed-loop persistent identification of linear systems with unmodeled dynamics and stochastic disturbances (Q1614415)

The authors study the essential issues of time complexity and probing signal selection for persistent identification of a linear time-invariant system in a closed-loop setting. Certain basic properties of the identification mapping are established by setting up both upper and lower bounds on identification accuracy as functions of the length of observation, size of unmodeled dynamics, and stochastic disturbances. The inherent impact of unmodeled dynamics on identification accuracy and the benefits of reducing time complexity by stochastic averaging of disturbances are demonstrated. Periodic and full rank inputs are shown to process generic feedback-invariant features that are desirable for persistent closed-loop identification. It is shown that the mixed formulation, in which deterministic uncertainty of system dynamics is blended with random disturbances, is beneficial to the reduction of identification complexity.