Control of a non-linear and non-minimum phase multivariable system using a neural predictive controller

Chojeda E.C.; Semino J.O.; Alama W.I. · RIAI - Revista Iberoamericana de Automatica e Informatica Industrial (2023) · DOI: 10.4995/riai.2022.17375

Resumen

In this paper, a Neural Predictive Controller (ANN-MPC) is proposed to control a nonlinear quadruple tank system, which is complex to control due to the nonlincarily of its valves and the interaction between its controlled variables. In addition, the problem is aggravated since the process presents a transient response with inverse dynamics due to being in a non-minimum phase. The ANN-MPC employs a modular artificial neural network framework and the Levcnbcrg-Marquardt training algorithm to more accurately and quickly estimate nonlinear process outputs and avoid model overfitting. Operational data was generated from the plant to train the neural network using Matlab. The performance of the ANN-MPC under reference changes was tested and compared with a linear MPC and a non-linear MPC. The simulation results showed that the ANN-MPC produced a shorter settling time than the linear MPC and generated RMSE values of the outputs similar to those of the NMPC In addition, the computation time required to calculate the optimal control variable was reduced compared to the NMPC.

Instituciones: Universidad de Piura