Square-Root Cubature Kalman Filters for Training Recurrent Type-2 Fuzzy Neural Networks

This paper proposes a novel learning algorithm benefitting from square-root cubature Kalman filters for training recurrent interval type-2 fuzzy neural networks. The recurrence property in this network is feeding the output of each input to itself. Simulations results show the effectiveness of the proposed network and the proposed learning algorithm.

I. Introduction

In recent years the recurrent type-2 fuzzy neural networks (RT2FNNs) have attracted significant attention for modeling the uncertain dynamics systems. Many structure of the RT2FNNs have been proposed up to now. For instance in [1], a self-structuring RT2FNNs is presented where the rule firing is feeding to itself. In [2], a RT2FNNs similar to [1], is proposed and the feedback weights are optimized by the Kalman filter. In [3], the recurrent fuzzy systems proposed in [3], the rule firing is feeding to itself and the other rules. The effectiveness of this structure is verified by several system identification problems. a In [4], a RT2FNNs is designed for speech detection and the effectiveness of RT2FNNs is shown by simulations. In [5], it is shown that the RT2FNNs are more in noisy environment.