Nonlinear Model Predictive Controller Design for Identified Nonlinear Parameter Varying Model

Abstract

In this paper, a novel nonlinear model predictive controller (MPC) is proposed based on an identified nonlinear parameter varying (NPV) model. First, an NPV model scheme is present for process identification, which is featured by its nonlinear hybrid Hammerstein model structure and varying model parameters. The hybrid Hammerstein model combines a normalized static artificial neural network with a linear transfer function to identify general nonlinear systems at each fixed working point. Meanwhile, a model interpolating philosophy is utilized to obtain the global model across the whole operation domain. The NPV model considers both the nonlinearity of transition dynamics due to the variation of the working-point and the nonlinear mapping from the input to the output at fixed working points. Moreover, under the new NPV framework, the control action is computed via a multistep linearization method aimed for nonlinear optimization problems. In the proposed scheme, only low cost tests are needed for system identification and the controller can achieve better output performance than MPC methods based on linear parameter varying (LPV) models. Numerical examples validate the effectiveness of the proposed approach.