Improving the design of clustered neural fuzzy models for optimization

Abstract

This article discusses the application of hierarchical clustering techniques in the design of clustered neural fuzzy approximations for optimization problems. Different clustering techniques are investigated for addressing the function approximation problem from data samples. The hierarchical clustering process has the advantage of providing a rational manner to determine the adequate number of clusters. We present a methodology for the iterative design of the neural fuzzy network model using the clustering scheme. Experiments with some analytical functions are performed to confirm the methodology discussed. Finally, we investigate the design of a superconducting magnetic energy storage device with two solenoids and three design parameters. The results indicate that the employment of hierarchical clustering techniques for generating neural fuzzy approximations is a valuable technique to solve and to reduce the computational cost of practical optimization problems.