GA-optimized neuro-fuzzy approach for nonlinear system modeling

In order to characterize the behavior of nonlinear dynamic systems many different approaches have been proposed in recent years. One of the best black-box models employed to deal with system nonlinearities is the combination of artificial neural network (ANN) and fuzzy logic system (FLS), which is known as neuro-fuzzy system. However, the gradient-based nature of this combination causes some deficiencies. Therefore, in this paper, an optimization approach which utilizes genetic algorithm (GA) as a derivativefree optimizer is proposed for both designing the structure of neuro-fuzzy model and assessing the model parameters. The whole proposed approach is applied to approximate: first, a nonlinear plant; next, nonlinear dynamic behavior of magneto-rheological (MR) damper, which is widely used in semi-active control of structures and its identification is significantly difficult due to inherent hysteretic and highly nonlinear behavior of the device. Comparisons between the responses of the models and the reference data show high accuracy and feasibility of the proposed approach.