Utilizing metaheuristic algorithms to optimize the rule base in fuzzy systems

Fuzzy systems are a useful means that are applied to various problems, including decision making, taxonomy, modeling, prediction, and control. The major challenge in using such systems is designing a fuzzy rule base with optimized parameters to maintain a desirable system performance. In this paper, a hybrid particle swarm optimization and opposition-based differential evolution training method is proposed and used to optimize the Gaussian membership function parameters of the rule base in a fuzzy system of type Takagi-Sugeno-Kang (TSK). In this dissertation, the effect of soft computing methods, e.g. evolution computing, on a zero-order TSK fuzzy system is investigated to control two non-linear plants. This paper considers a hybrid computing approach consisting of: opposition-based differential evolution (ODE) and particle swarm optimization (PSO). Results of training a zero-level TSK fuzzy system used to control two non-linear plants indicate that the proposed hybrid algorithm has a better classification accuracy in comparison to other training approaches. Moreover, this study uses heuristic opposition-based differential evolution (ODE) and particle swarm optimization (PSO) algorithms (HODEPSO) and applies them to two accuracy-oriented fuzzy system (FS) design problems. For these two models, all free parameters of a first-level Takagi-Sugeno-Kang (TSK) system are also optimized using the HODEPSO algorithm. The models used in our experiments are the Mackey – Glass chaos time series and a real-world economic problem whose future values are predicted using the proposed algorithm. Finally, results of these experiments also show that HODEPSO has the minimum average training and test error in comparison to other training methods.