Convergence of a multivariable self-tuning controller with integral action (Q1098811)

This note establishes a global convergence for the multivariable self- tuning control algorithm presented recently by the author [Automatica 22, 619-627 (1986)]. The plant is assumed to be described by a CARIMA model, namely, a controlled autoregressive error-integrating moving average which embeds effective integral action on the error vector. The global convergence of the algorithm, presented in the above cited paper, is established by using an interlaced stochastic gradient identification algorithm. It is claimed that this result constitutes the first convergence proof for (multivariable) self-tuning control agorithms based on a (ARIMA representation of the plant.