Tracking control using self-organizing neural network

An identification method and a tracking controller for nonlinear discrete-time systems using the "neural-gas network" are proposed. The neural-gas network is a kind of self-organizing network, and was developed by Martinet and Schulten (1991). The system is identified by estimating a hypersurface in the space of input and output sequences using the neural-gas network. The metric of the space of the synapse weight is modified to increase efficiency of learning. The hypersurface is expressed with a method by means of rational Bezier surface or direct interpolation. An inverse model of the system is derived from the surface, which is applied to a tracking control problem.