An enhanced memory TSK-type recurrent fuzzy network for real-time classification

Abstract

An enhanced memory TSK-type recurrent fuzzy network (EM-TRFN) is proposed in this paper, suitable for modeling complex dynamic systems. Feedback connections, formulated using finite impulse response (FIR) synaptic filters, are employed in the network architecture, serving as internal memories of multiple past firing values, used to determine the current rule firings. Thus, high-order temporal capabilities are embedded in the network, rendering it capable of modeling highly complex nonlinear temporal processes. The structure of the EM-TRFN is evolved in an on-line fashion, with concurrent structure and parameter learning. The proposed network is combined with the predictive modular fuzzy system (PREMOFS), leading to an efficient system for on-line time-series classification. Simulations on a gait identification problem indicate the efficiency of the proposed system. © 2007 EUCA.