Evolving Generic Feedforward Path Nonlinear Controllers from Open-loop Plant Response Data

In this paper a generic functional approximator in the form of a neural network is enabled as a generic nonlinear controller. This means such a nonlinear feedforward path controller structure can be used for any plants, in a similar way to a conventional linear controller. An evolutionary learning mechanism for training such controllers directly from open-loop plant response data is also presented. Furthermore it is able to achieve competing goal such as short rise-time and no overshoots. Finally, this controller is applied to a linear second-order servo-system and a nonlinear water-tanksystem.