Computer design of a new predictive adaptive controller coupling neural networks and kalman filter applied to siso and mimo control

摘要

This work presents a predictive control algorithm based on constraint neural networks as internal non-linear model with a tuning algorithm based on the Kalman filter. The algorithm utilises a sequential quadratic programming algorithm to compute the next action of the manipulated process variables. The predictive control parameter, the suppression factor, is optimised on-line by a standard Kalman filter. The suppression factor is identified by a method based on the relative gain. The algorithm was tested on distinct chemical processes, a penicillin fermentation process (SISO) and a fixed bed catalytic reactor (MIMO). It shows that the suppression factor can be identified on-line, but a scaling factor has to be introduced because the process derivatives can become large. The proposed procedure still reduces the number of parameters to be adjusted in case of MIMO systems.