Fine Tuning of Fuzzy Rule-Base System and Rule Set Reduction Using Statistical Analysis

Learning and tuning of fuzzy rule-based systems is the core issue for linguistic fuzzy modeling. To achieve an accurate linguistic fuzzy model genetic learning of initial rule base is introduced and evolutionary simultaneous tuning of nonlinear scaling factors and fuzzy membership functions (MFs) are employed. Novel evolutionary algorithm is applied for simultaneous optimization process due to its computational efficiency and reliability. To preserve the interpretability issue, linguistic hedges are utilized, which slightly modify the MFs. Interpretability issue is further improved by introducing the statistical based fuzzy rule reduction technique. In this technique, most appropriate rules are selected by computing the activation tendency of each rule. Further, focusing on granularity of partition, linguistic terms for input and output variables are modified and new reduced rule base system is developed. The proposed techniques are applied to nonlinear electrohydraulic servo system. Extensive simulation and experiment results indicate that proposed schemes not only improve the accuracy but also ensure interpretability preservation. Further, controller developed based on proposed schemes sustains the performance under parametric uncertainties and disturbances.