Dynamic system modeling using a recurrent interval-valued fuzzy neural network and its hardware implementation

Abstract This paper first proposes a new recurrent interval-valued fuzzy neural network (RIFNN) for dynamic system modeling. A new hardware implementation technique for the RIFNN using a field-programmable gate array (FPGA) chip is then proposed. The antecedent and consequent parts in an RIFNN use interval-valued fuzzy sets in order to increase the network noise resistance ability. A new recurrent structure is proposed in RIFNN, with the recurrent loops enabling it to handle dynamic system processing problems. An RIFNN is constructed from structure and parameter learning. For hardware implementation of the RIFNN, the pipeline technique and a new circuit for type-reduction operation are proposed to improve the chip performance. Simulations and comparisons with various feedforward and recurrent fuzzy neural networks verify the performance of the RIFNN under noisy conditions.