A new technique is presented for the design of evolutionary predictive
controllers in which the optimum parameters of a predictive controlle
r are computed at each time step. Based on a stochastic search techniq
ue using Genetic Algorithms, the technique leads to the design of pred
ictive controllers whose closed-loop performance subject to control co
nstraints is significantly improved over existing predictive control t
echniques. Examples of both unconstrained and constrained control are
presented.