Real-time recurrent interval type-2 fuzzy-neural system identification using uncertainty bounds

In this paper, a novel real-time recurrent interval type-2 fuzzy neural system identification is presented using intelligent algorithm. Interval type-2 fuzzy neural network (FNN) by adding feedback connection on the input layer is introduced to handle uncertainties which arise from the noisy training data, noisy measurements used to activate the fuzzy logic system (FLS) and linguistic uncertainties. In order to overcome the iterative type-reduction overhead, the intelligent algorithms are proposed using uncertainty bounds, inner- and outer-bound sets, which provide estimates of the uncertainties contained in the output of an interval type-2 FLS without having to perform the costly computations of type-reduction. Duffing forced oscillation system is fully illustrated to be identified and simulation results show that not only similar identification performance to one that use type-reduction can be achieved but also significantly faster real-time identification can be performed.