Linear predictive control based on approximate input-output feedback linearisation

The computational burden related to model-based predictive control (MBPC) of constrained nonlinear systems hampers its real-time application. To avoid this, input-output feedback linearisation (IOFL) techniques are used to linearise the process model over a wide operating range. The resulting linear model is then integrated in a linear MBPC scheme allowing for standard linear control techniques to be applied. However, the process input constraints become nonlinearly related with the optimisation variable due to the state-dependent nonlinear feedback law. In this paper a new method to IOFL of general multivariable discrete-time systems is proposed. By adopting an approximate IOFL based on a suitable linear model approximation, a new linear and state dependent input mapping is obtained which further enables the MBPC solution to be found through a single quadratic programming optimisation. The performance of this new technique is compared with other well-known schemes for the control of a Van der Vusse chemical reaction taking place in a CSTR