Artificial Neural Networks for On-Line Trained Controllers

This paper deals with the use of artificial neural networks employed as an on-line trained controller for a real process and simulation model control. Well-known back-propagation method is used as a learning algorithm intended to minimize the difference between the plant’s actual response and the desired reference signal. The influence of neural network’s parameters on a controlled plant output is discussed. We also attempted to find the rules of these parameters adjustment in view of the type of a transfer function in Laplace transform and tested the robustness of our controller burdened with the error signal. Some simulation and real process control results are also presented to evaluate the proposed design. Discussed in the last chapter are the possibilities of creating an adaptive neural controller. Key-Words: back-propagation, artificial neural nets, neural controller, adaptive neural controller