Invariant Process Control Using Neural Networks

Using neural networks in direct inverse control is a promising approach mainly because of the simplicity of implementation in which the inverse model has immediate utility for control. This paper presents a study of the behaviour of such a controller. Specifically in the control of a water temperature control system it was discovered that if a NN is trained with its input being the rate of change of the plant output but instead, as a controller, the input of the NN is the plant error, the control loop is reduced to a first order system. However this is subject to the plant parameters being closely matched.by the network Thus the plant behaves as a first order system and whilst doing so becomes invariant to disturbances.