Search optimal fuzzy model using evolutionary algorithm — Application to predict the form function of the composite cylindrical shell

The fuzzy model derived from the input-output data of a given system using a fuzzy clustering algorithm is not generally optimal due to the errors introduced by the projection of the clusters onto the input variables and its approximation by the parametric functions. In the present work, the evolutionary algorithms are used to improve the initial fuzzy model obtained by fuzzy clustering algorithms. The proposed approach is applied to predict the form function of aluminum/polymer composite cylindrical shell. For this application, the performance of four fuzzy clustering algorithms: Gustafson Kessel (GK), fuzzy c-means (FCM), Gath-Geva algorithm (GG) and fuzzy c-regression model (FCRM) are optimized by genetic algorithm (GA) and hybrid particle swarm optimization with constriction factor approach (HPSO with CFA). The search space of evolutionary algorithms is restricted by the constraints in order to avoid loss of initial fuzzy model meaning. The achieved results show that the proposed approach is useful to improve the performance of an initial fuzzy model which is not optimal. The optimal model to predict the considered form function is FCM_GA with a mean square error about 0.2497 and coefficient correlation of 0.9081.