A smooth gain scheduling generalized predictive control

Nonlinear model predictive control is an emerging control technique dealing with high nonlinearities of industrial plants. However, it suffers from many hurdles such as the numerical problems related to the resulting no convex nonlinear optimization problem, time consuming, and difficulties in analyzing properties such as stability. In this paper, to side-step these difficulties, an infinite gain scheduling generalized predictive control is designed to control a benchmark high nonlinear plant instead of nonlinear predictive control. A neural model of the plant is identified and used as an internal model of the generalized predictive control scheme. The neural model is linearized successively and a filtering process is used to smooth the adaptation of the linearized model every sample time. The results show good performance in tracking the reference and rejecting abrupt changes in measured disturbances. The filtering process improved the results in terms of rapidity, overshoots damping, and smoothing the control signal.