Adaptive Non-linear Predictive Control

Most predictive control algorithms, including the Generalized Predictive Control (GPC) (Clarke et al. 1987) are based on linear dynamics. Many processes are severely non-linear and would require high order linear approximations. Another approach, which is presented here, is to extend the basic adaptive GPC algorithm to a non-linear form. This provides a non-linear predictive controller which is shown to be very effective in the control of processes with non-linearities that can be suitably modelled using general Volterra, Hammerstein and bilinear models. In developing this algorithm, the process dynamics are not restricted to a particular order as is the case with the current non-linear adaptive algorithms. Simulations are presented using a number of examples and the steady state properties are discussed.;The Non-Linear Generalized Predictive Control (NLGPC) algorithm is tested on a non-linear batch reactor system by carrying out a number of experiments and comparing its performance with other control strategies. The NLGPC is shown to outperform the constrained Self-Tuning PID (STPID) controller by Katende and Jutan (1993) and the Generalized Minimum Variance (GMV) controller by Clarke and Gawthrop (1975). It is also shown to have better performance than the well known GPC algorithm by Clarke et al. (1987). The advantage of th NLGPC over the other controllers is attributed to its adaptive nature and use of non-linear process models in its design