Criterial convergence of the least squares method in an adaptive control system (Q1593858)

The least squares method is a standard approach to the estimation of the parameters of linear regression. In adaptive control problems the least squares method is applied to the linear-regression equation in which the regressors depend on the history of noises. In the standard problem of adaptive minimal variance for a linear scalar minimum-phase plant, the condition of a uniformly nondegenerated information matrix is not fulfilled. In this paper it is proved that the least squares method ensures criterion convergence, i.e., the cost functional is minimized, although the estimates of the parameter may diverge. Special specifying actions were introduced that guarantee that the information matrix is nondegenerate to a certain degree. It is proved that the estimates of the coefficients are consistent and the closed control system is optimal. The proof is based on the Lyapunov quadratic function method.