Nonlinear adaptive predictive control

Nonlinear model predictive control algorithms that explore the specific features of distributed collector solar field dynamics are considered. The key point explored in this chapter is to show that, by making a change in the manipulated variable, together with a change in the timescale, it is possible to obtain exactly linear plant models, either of input-output or state-space type. In practice, the change in the timescale is implemented in discrete time by use of a variable sampling interval. Using this transformed linear model, the controller computes then the manipulated variable by minimizing a quadratic cost in a receding horizon sense. The use of state-space models requires a state-estimator, whose design is described. Adaptation is embedded by estimating the parameters of the linearized model using recursive least squares. Two adaptive control algorithms, named WARTIC-i/o and WARTIC-state that embody this approach are described. Experimental results with a distributed collector solar field, that show the influence of the choice of the optimization horizon of the quadratic cost on the resulting performance, are described. The method described in this chapter allows to perform closed-loop step responses of large amplitude with almost no overshoot and a fast raising time.