A stable one-step-ahead predictive control of nonlinear systems (Q1975542)

A one-step control of a single-input single-output nonlinear analytic plant is considered. The control at the current step is chosen such that the squared difference between the setpoint and output at the next step is minimal. To avoid overshooting, a quadratic penalty function is added to the cost function. Also, to ensure equality between the model and the plant output, an uncertainty parameter is added to the general nonlinear model. Neural networks models, radial basis functions, NARMAX models, and other general nonlinear models can be equally used. It is shown that the unconstrained closed-loop system is asymptotically stable in the large, and asymptotically stable under the input-output constraints. The effectiveness of the proposed control architecture is demonstrated on a recognized benchmark problem and on the continuous-stirred tank reactor.