Combination of Breeding Swarm Optimization and Backpropagation Algorithm for Training Recurrent Fuzzy Neural Network

The usage of recurrent fuzzy neural network has been increased recently. These networks can approximate the behaviour of the dynamical systems because of their feedback structure. The Backpropagation of error has usually been used for training this network. In this paper, a novel approach for learning the parameters of RFNN is proposed using combination of the backpropagation and breeding particle swarm optimization. A comparison of this approach with previous methods is also made to demonstrate the effectiveness of this algorithm. Particle swarm is a derivative free, globally optimizing approach that makes the training of the network easier. These can solve the problems of gradient based method, which are instability, local minima and complexity of differentiating.