Characteristic decomposition and the multivariable generalisation of predictive self-tuning control

Presents a framework for the multivariable generalisation of predictive self-tuning control. The proposed algorithm relies on a ´characteristic subsystem´ description of the system, which reduces the multivariable control problem to a set of scalar problems so that currently existing single-input/single-output algorithms can be directly applied. This provides the user with the same flexibility in controlling the loops (individual input/output pairs) of the multivariable system as that associated with standard scalar implementations. The result is a multivariable self-tuning algorithm that achieves loop-by-loop control in a true, generalised Nyquist framework.