Predictive Neuro-fuzzy Controller for Multivariate Process Control

Abstract

A self-learning fuzzy controller with a neural estimator was designed for predictive process control, and it was applied to snack food frying process control. Main features of the designed controller are that it can be applied to plants having nonlinear dynamics, and its structure can be easily extended to multivariable systems. The neural estimator, composed of a time delay multilayer perceptron with output feedback, was structured to model the dynamics of a frying process and to predict the actual plant output affected by controller output after the time lag. The neuro-fuzzy controller was composed of a multilayer feedforward network, and it was trained using the backpropagation algorithm. The neuro-fuzzy controller trained with only two data sets performed the control task successfully, and it showed that the controller had the robustness for the control task starting from the untrained initial conditions by computer simulation.