An adaptive learning algorithm for a neo fuzzy neuron.

Abstract

In the paper, a new optimal learning algorithm for a neo-fuzzy neuron (NFN) is proposed. The algorithm is characteristic in that it provides online tuning of not only the synaptic weights, but also the membership functions parameters. The proposed algorithm has both the tracking and filtering properties, so the NFN can be effectively used for prediction, filtering, and restoration of non-stationary noisy stochastic and chaotic signals. A special feature of the proposed algorithm is its computational simplicity in comparison with the other learning procedures for neuro-fuzzy systems. is large. Certain problems may arise in the processing of non-stationary signals, since the second-order procedures with exponential forgetting can be numerically instable. To overcome these difficulties, a neuro-x 1 x 2 NS NS