Preprocessing data patterns to simplify the identification of fuzzy models

In this paper we propose both data transformation and modified fuzzy clustering methods to simplify the identification of fuzzy models as well as to improve the system performance. Conventional fuzzy clustering methods contributed the clustering results from the original data to build the fuzzy models. Instead of exploiting the original data to construct the fuzzy models, the given data are transformed into other domains in the hope that the clustering results can provide a positive effect on the system performance. To improve the performance further, both the genetic algorithms and the gradient descent methods succeed in adjusting the projected membership functions from the clustering centroids and the consequent parts in the fuzzy rules. Different types of fuzzy rule are investigated and simulation results are compared. The simulation results verify that the proposed model outperforms conventional models.