Intelligent control method using cubic neural network with multi-levels of information abstraction

This study presents a new architecture of neural networks named “cubic neural network (CNN)”, which possesses multi-levels-of-information processing and has the ability of parallel distributed signal processing as an intelligent control method. Each level of this CNN processes different degree of abstracted signals and it enables adaptation for abnormal state. In this paper, the fundamental construction, the learning method and the abstraction method of CNN are described. The control ability of this method was verified by the experimental results of an inverted pendulum with large change of parameters