Improving the fuzzy system performance by fuzzy system ensemble

Abstract

This paper proposes a fuzzy system ensemble (FSE) that can improve the system performance in non-linear and complex problems. Each fuzzy system in the FSE is built from two design stages, each stage of which is performed by different genetic algorithms (GAs). The first stage generates a fuzzy rule base that covers as many of the training examples as possible. The second stage builds fine-tuned membership functions that make the system error as small as possible. These two stages are repeated independently upon the different partitions of input-output variables. The system error will be reduced further by invoking the FSE that combines multiple fuzzy systems with an equal system-error weighting method where the weight constant is inversely proportional to the FS's error. Applications of the FSE to both the truck backer-upper control and the Mackey-Glass chaotic time-series prediction are presented. For the truck control problem, control performance of the proposed method is compared with the approach of Wang and Mendel [IEEE Transactions on System, Man, and Cybernetics 22 (1992) 1414] in terms of either the rate of successful controls reaching to the goal or the average traveling distance. For the chaotic time-series prediction problem, prediction accuracy of the proposed method is compared with that of other fuzzy or neural network predictors in terms of non-dimensional error index (NDEI).