Neural Network Structure for Process Control using Direct and Inverse Process Model

This paper presents a neural network structure that can be used for process control. The proposed structure consists of two feed forward neural networks. One network learns the dynamics of the process to be controlled and the other network learns the inverse dynamics of the neural network process model. The motivation behind this algorithm is that the process gradients which are unknown are learned by the process model estimated by the neural network. The process gradients or some estimates of them are needed for the neural network to learn the inverse dynamics of the process. The output of the inverse dynamic model is the control action that drives the controlled process to desired reference level. The learning methodology that is used for both networks is supervised learning using Davidon´s Least Squares minimization technique. The performance of the proposed neural network control structure is tested on a dynamical second order simulated process.