Observer Based Restricted Structure Generalized Predictive Control for quasi-LPV Nonlinear Systems

Abstract

An observer based Restricted Structure Generalized Predictive Control (RS-GPC) algorithm is proposed. The novel feature is to assume the state-observer within the feedback loop is of reduced order. The aim is to inherit the natural robustness of low-order controllers and to provide a solution that may be easily simplified for real-time implementation. The nonlinear discrete-time, multivariable plant model is represented by a state-space system that may be in Linear Parameter Varying or State-Dependent forms. The controller gains are computed to minimize the type of cost-function that is found in traditional model predictive control but with some additional terms that enable gain magnitudes and the rate of change of control gains to be minimized. The cost-function also includes dynamically weighted tracking-error and control signal costing terms. The optimal controller includes a reduced order observer and a time-varying control gain matrix within the loop and background processing for the gain computations. Hard constraints may be imposed on the gain and rate of change of gain and on the control and output signals. Copyright (C) 2020 The Authors.