Neural network identification in nonlinear model predictive control for frequent and infrequent operating points using nonlinearity measure

Abstract

In this paper, the problem of optimal system identification in nonlinear model predictive control (NMPC) for highly nonlinear dynamic processes is presented. Due to the short term changes in the operating point, the process may escape from its frequent operating points (FOP) to some infrequent operating points (IOP) for a short period. On the other hand, because the nonlinear model is identified using the operating data, it is mainly accurate for the FOP. Therefore, the NMPC causes tracking error or even instability in the IOP. To handle this problem, in this paper, we present a novel optimal identification algorithm, which is highly depended on the nonlinearity of the understudy plant, to train the nonlinear model of the NMPC. The nonlinear model is selected as a multi-layer perceptron neural network (MLP) which is trained to describe the nonlinear behaviour of the nonlinear dynamic system accurately in the FOP and while it has acceptable performance in the IOP. To validate the proposed algorithm, the well-known nonlinear dynamic pH neutralization process is chosen in both simulation and implementation parts. Finally, the simulation and implementation results prove the effectiveness of the proposed algorithm.

Keywords: Advanced process control; Nonlinear identification; Nonlinear model predictive control; Nonlinearity measure; Volterra series.