Takagi-Sugeno Fuzzy System Accuracy Improvement with A Two Stage Tuning

Abstract

In this paper a method for improving accuracy of a Takagi-Sugeno type fuzzy system used in Matlab Fuzzy Logic Toolbox as genfis3 (Sugeno) is proposed. For this fuzzy system, the input space is partitioned using Fuzzy C-means (FCM) clustering algorithm and the consequent parameters are optimized using least square. This improvement is done in a two stage tuning using particle swarm optimization (PSO).In the first stage, PSO is used to optimize input membership functions (mean and variance) and consequent parameters of Takagi-Sugeno fuzzy system. In the second stage, PSO is used to optimize a weighting factor for the rules and a scaling universe of discourse for inputs and outputs variables. To simplify the tuning process and performing it in one optimization stage, a one stage tuning is also discussed to tune same parameters optimized in the two stage tuning. Experimental results with real data applied in data classification problem shows a consistency of getting higher classification accuracy with the proposed tuning methods over the original system.