State-space self-tuning controllers for linear stochastic multivariable systems

This dissertation presents state-space self-tuning controllers for a class of linear stochastic multivariable systems. The relationship between the state-space model and the auto-regressive moving average model is discussed. A joint algorithm for system parameters identification and states estimation is then developed, which provides the basis of several self-tuning schemes. Using state-feedback techniques, we develop three explicit self-tuning control algorithms based on optimal control, on pole-assignment and on generalized minimum variance control strategies respectively. Both optimal control and pole-assignment self-tuning controllers can be applied to any unstable and/or non-minimum phase systems commonly encountered in practical applications. A self-tuner derived based on the low-order estimated model instead of the original high-order system is also presented. This is of practical significance since it requires less computational efforts. Simulation examples are provided to illuminate the design idea and to demonstrate the potential of the developed self-tuning algorithms.