Dynamic fuzzy learning rate in a self-evolving interval type-2 TSK fuzzy neural network

In every self-evolving interval type-2 fuzzy neural network (IT2FNN), there are several pre-given parameters which are to be adjusted before runtime and their changes may drastically affect the system´s performance. Mostly, the pre-given parameters are adjusted by trial-and-error which is a time-consuming procedure. One of these pre-given parameters is the learning rate. In this paper, an Interval Type-2 Takagi-Sugeno-Kang Fuzzy Neural Network is investigated. In structure learning phase, fuzzy clustering is used to generate a new rule. Then parameters of the new rule are adjusted in parameter learning phase which benefits from Gradient Descent Algorithm. In this case, in order to reduce complexity and adjust parameters more precisely and to enhance the self-organizing property, adjusting of the learning rate online by using simple fuzzy rules is proposed. Then the proposed IT2FNN is used for identification of a nonlinear system. Experimental results indicate that the IT2FNN with the proposed idea achieves good results without the need to tune learning rate manually.