Evolutionary identification of a recurrent fuzzy neural network with enhanced memory capabilities

An enhanced memory TSK-type recurrent fuzzy network (EM-TRFN) is proposed in this paper, for dynamic control of nonlinear systems. The network employs feedback connections in the rule layer, with their synaptic links being implemented through finite impulse response (FIR) filters. Thus, the network structure is enriched in terms of past information processing capabilities. Both structure and parameter learning are performed through a hybrid evolutionary algorithm, with its representation scheme employing variable-length mixed-type chromosomes. Comparative results in a control problem of a dynamic system prove the EM-TRFN's structural merits, as well as the proposed learning algorithm's ability in dealing with complex search spaces.