Stability and convergence of multipredictor adaptive controllers

The authors deal with global stability and convergence of multipredictor adaptive control algorithms. The algorithms considered are of long-range type, made up by a bank of T separate parallel identifiers, all sharing a common regressor. This feature reduces the computation time and, together with the assumed knowledge of the first T samples of the plant impulse response, makes the analysis possible. Though the first T impulse response samples are assumed known, the numerator and denominator polynomials of the plant transfer function are unknown and are allowed to have marginally stable common factors, modeling the presence of bounded deterministic (recurrent) disturbances. The approach is based on directional forgetting recursive least squares