Equality index and learning in recurrent fuzzy neural networks

Abstract

A novel learning algorithm for recurrent neurofuzzy networks is introduced in this paper. The core of the learning algorithm uses equality index as the performance measure to be optimized. Equality index is especially important because its properties reflect the fuzzy set-based structure of the neural network and nature of learning. Equality indexes are strongly tied with the properties of the fuzzy set theory and logic-based techniques. The neural network recurrent topology is built with fuzzy neuron units and performs neural processing consistent with fuzzy system methodology. Therefore neural processing and learning are fully embodied within fuzzy set theory. The performance recurrent neurofuzzy network is verified via examples of nonlinear systems modeling. Computational experiments show that the recurrent fuzzy neural models developed are simpler and that learning is faster than both, static neural and neural fuzzy networks and alternative recurrent fuzzy neural networks.