Recurrent fuzzy CMAC in hierarchical form for dynamic system identification

The conventional fuzzy CMAC neural networks perform well in terms of their fast learning speed and local generalization capability for approximating nonlinear functions. However, it requires an enormous memory and the dimension increase exponentially with the input number. In this paper, we use two techniques to overcome these problems: recurrent and hierarchical structures and propose a new CMAC, named Hierarchical Recurrent Fuzzy CMAC (HRFCMAC). Since the structure of HRFCMAC is very complex, the normal training methods are difficult to be applied. A new simple algorithm is given, we can train each sub-block of the hierarchical CMAC independently. A time-varying learning rate assures the learning algorithm is stable.