Real time Multiple Generalized Predictive Control based on a Neuro-fuzzy model

This paper presents an implementation of Multiple Generalized Predictive Control (MGPC) based a Neuro-fuzzy model to control a real time gaseous nonlinear process. Firstly, a Neuro-fuzzy model, which describes the dynamic characteristics of the process is identified and optimized. Because the consequent part of the identified Neuro-fuzzy model is represented as a form of several Auto-Regressive Exogenous (ARX) sub models, MGPC are then designed based on the ARX sub models to work collaboratively to control the process. To validate the improvement of the proposed control strategy, two real time closed loop tests, referred as set-point tracking and disturbance rejection are investigated. In both tests, the performances of the proposed MGPC are benchmarked to the performances of traditional single GPC, which is designed based on a single ARX model. Results from the comparison show that the MGPC based Neuro-fuzzy model is able to improve the performance of nonlinear control system.