A Mamdani type multistage fuzzy neural network model

In this paper, a new multistage fuzzy neural network model is proposed to overcome the dimensionality problem of single-stage fuzzy neural networks. The model arranges single-stage reasoning stages in a multistage manner, where the consequence of one stage can be passed to the next stage as a fact. The network structure in each individual stage is developed based on Lin and Lee´s (1991) fuzzy neural network model in which Mamdani´s fuzzy reasoning is adopted. Given the stipulated input-output data pairs, an appropriate fuzzy rule set can be created through a hybrid learning process. Simulation Results show that the new model uses less resources than its single-stage counterpart to achieve favourable performance. Some interesting results have also been found in convergence and robustness