Function approximation by neurofuzzy GMDH with error backpropagation learning empirical comparisons of approximation accuracy with multilayered neural networks

摘要

An adaptive learning network has been proposed, in which the Radial Basis Functions networks were applied to the partial descriptions of network type GMDH. In this paper, we derive a learning ruleby formulating on optimization problem, and report on results of simulation studies about its ability to approximate nonlinear mappings. The first two examples use synthetic data generated by nonlinear functions and chaotic time series from the Mackey-Glass differential delay equation respectively.