Nonlinear system control based on multi-resolution radial-basis competitive and cooperative networks

초록

Most mapping neural networks to date can learn only one-to-one and many-to-one mapping. However, the inverse control of certain nonlinear systems requires to handle one-to-many inverse mapping, which most conventional mapping neural networks are unable to cope with. This paper presents a novel Multi-resolution Radial-basis Competitive and Cooperative Network (MRCCN) capable of dealing with one-to-many inverse mapping for control. MRCCN self-organizes a collection of local clusters of various locations, shapes and sizes to represent an arbitrary many-to-many mapping under uniform mapping accuracy. MRCCN is able to retrieve multiple outputs based on the cooperative decision of local clusters relevant to the given input. The highlight of this paper is the demonstration that MRCCN is able to control those nonlinear systems that can not be handled by conventional mapping networks due to the one-to-many mapping involved. Simulation results are shown. © 1995.

키워드