Neural networks in process control: model-based and reinforcement trained controllers

Abstract This article presents several control schemes, based on the use of neural networks, which can be applied to the control of ill-defined non-linear systems as frequently occur in agriculture. The emphasis is on methodology and appropriate software which could later be used to construct control systems in a standard way. As usual when neural networks are used, many experiments are required before desired results are reached. This article presents two software tools which can be used to conduct numerous experiments with neural controllers; one using backpropagation and another genetic algorithms for neural network training. Also, some results of neural control experiments with a bioreactor simulation are presented.