Self-organising interval type-2 fuzzy neural network with asymmetric membership functions and its application

Abstract

For the identification and modelling problems of a nonlinear system with complex uncertainties, a self-organising interval type-2 fuzzy neural network structure with asymmetric membership functions (SIT2FNN-AMF) is developed. First, a fuzzy c-means algorithm with four fuzzifier parameters is used to partition the input data to obtain the uncertainty means and widths of the fuzzy rule antecedent; then, according to the cluster validity criterion, the number of fuzzy rules is determined. Thus, identifications of the structure and rule antecedent parameters are automatically completed. The consequent part uses the Mamdani model, and the initial value of the consequent parameter is an interval random number. The fuzzy rule parameters are tuned by the gradient descent method. Finally, the proposed SIT2FNN-AMF is applied to simulations of nonlinear system identification and soft-sensing model for ethylene cracking furnace yield. The comparison of simulation results obtained with a conventional fuzzy neural network and interval type-2 fuzzy neural network verifies the performance of the proposed SIT2FNN-AMF.