EXPERT SYSTEM OF MODEL REDUCTION TECHNIQUES BASED ON ARTIFICIAL NEURAL NETWORKS

An expert system of model reduction techniques (ESOMRT) based on artificial neural networks is studied and implemented. This expert ststem is appropriate for both expert users and non-expert users. It can choose proper reduction method based on specific high order model of continuous-time or discrete-time control systems and reduction requirements. It can also evaluate the reduction quality in time domain and/or frequency domain. We have tried to use the knowledge representation of combining procedural method with artificial neural networks(ANN) method and realize semi-automatic knowledge acquisition by using the relearning mechanism of ANN. There are three kinds of working patterns in this system and friendly man-machine interactive interface, its intelligence level is comparatively high and it has a certain practical value. The expert system runs on IBM-PC/XT and 386 machines.