Asymptotically efficient adaptive control in stochastic regression models (Q1821070)

The paper is concerned with the issue of global convergence in self- tuning control. In adaptive control, there is a conflicting requirement on the control input to provide on the one hand good output regulation and on the other hand sufficient excitation so as to provide reliable estimates. This suggests that the parameter inaccuracies be taken into account in the control design. The problem is difficult to solve, both computationally and analytically. A more practical approach is used in self-tuning control, where the contrl input is generated using the parameter estimates as if they were the true ones. The issue of convergence of such a scheme is then important. The paper discusses how the dilemma between the need for information and the objective of efficient control can be resolved in the construction of asymptoticaly efficient self-tuning regulators for a general stochastic regression model. The basic idea is to introduce white-noise probing inputs whenever the data suggests inadequate information for parameter estimation.