A recurrent self-evolving fuzzy neural network with local feedbacks and its application to dynamic system processing

This paper proposes a recurrent self-evolving fuzzy neural network with local feedbacks (RSEFNN-LF) for dynamic system processing. A RSEFNN-LF is composed of zero-order or first-order Takagi–Sugeno–Kang (TSK)-type recurrent fuzzy if–then rules. The recurrent structure in a RSEFNN-LF comes from locally feeding the firing strength of a fuzzy rule back to itself. A RSEFNN-LF is constructed on-line via simultaneous structure and parameter learning. In structure learning, an efficient rule and fuzzy set generation algorithm is proposed to generate fuzzy rules on-line and reduce the number of fuzzy sets in each dimension. In parameter learning, the consequent part parameters are learned through a varying-dimensional Kalman filter algorithm whose input dimension varies with structure learning. The antecedent part and feedback loop parameters are learned using a gradient descent algorithm. The RSEFNN-LF is applied to dynamic system identification, chaotic sequence prediction, and speech recognition problems. This paper also compares the performance of the RSEFNN-LF with other recurrent fuzzy neural networks.