Artificial Neural Network (ANN) and it's application for autopilot have been systemically investigated in this paper. A new method for training ANN on-line has been studied in detail, and so a new ANN-based control method, so-called ANN-forward backward control structure, has been proposed here, in which there is an ANN-based dynamical nonlinear model of the controlled process. It is just this ANN model that gives the Neuro-controller (NC) the priority to be trained sufficiently in every sample time, which makes it easy to utilize an almost no-ready NC immediately, without taking long time to search for initial value and long learning period for NC to be trained to conduct satisfactorily.
|