A data driven approach to constrained control

This thesis presents a data-driven approach to constrained control in the form of a subspace-based state-space system identification algorithm integrated into a model predictive controller. Generally this approach has been termed model-free predictive control in the literature.

Previous research into this area focused on the system identification aspects resulting in an omission of many of the features that would make such a control strategy attractive to industry. These features include constraint handling, zero-offset setpoint tracking and non-stationary disturbance rejection.

The link between non-stationary disturbance rejection in subspace-based state-space system identification and integral action in state-space based model predictive control was shown.

Parameterization with Laguerre orthonormal functions was proposed for the reduction in computational load of the controller.

Simulation studies were performed using three real-world systems demonstrating: identification capabilities in the presence of white noise and non-stationary disturbances; unconstrained and constrained control; and the benefits and costs of parameterization with Laguerre polynomials.