engleski

Application of Fuzzy Neural Networks in Predictive Control of Batch Distillation

Identification and control of nonlinear processes is a difficult task. For this purpose we used an adaptive predictive control based on a Fuzzy Neural Network (FNN) model of the process. Standard on-line and on-line learning algorithms for FNN are described. Further on it is investigated how the FNN model can be linked to the generalized predictive controller (GPC). In original GPC design purely linear transfer function model is used for long-range prediction. In this paper instantaneous linearization of FNN model is used to extract the best current linear model for the GPC controller. In this way shortcomings of the classical approach in case when process has strong nonlinearities and/or time dependence, can be surpassed. Applicability of this approach is tested on an example of binary batch distillation process with high nonlinearity and time dependence.

Predictive control; fuzzy-neural networks; distillation column

nije evidentirano