A multivariable neural predictive control algorithm

A model-based predictive control algorithm for multivariable nonlinear plants with constraints is the subject of the paper. The predictive control strategy, which is very efficient for reasonably wide class of dynamic processes, is extended to deal with nonlinear systems. For prediction and optimisation a feedforward neural model is used. Since. no analytical solution to the problem exists a nonlinear constrained optimisation algorithm must be used. Effective analytical formulae needed to apply a gradient-based algorithm, the choice of which is also discussed, are thoroughly derived in the paper. An illustrative simulation example is presented and discussed. Copyright (C) 2001 IFAC.