Intelligent Control Using Cubic Neural Network with Multi-Levels of Information Abstraction

Abstract

Neural network control has been applied to various objects, and showed its validity. From the viewpoint of intelligent control, however, it is considered that most of utilizations of neural network are no more than a technique in which a part of information processing is replaced with that using a neural network and they are not able to cope with abnormal situation. This study introduces a new concept of hierarchical information abstraction. This study presents a new technique of intelligent control using the neural network, called "Cubic Neural Network" (CNN) which possesses multi-levels of information abstraction, and furthermore presents a systematic synthesis method of CNN. Therefore, qualitative knowledge or information can be used in addition to ordinary feedback control using only quantitative measured values. And a robust control which has large applicable regions is realized. And, as an example, the control of inverted pendulum including supposed or not supposed large parameter change is dealt with, and the usefulness of the present method is shown.