METHOD FOR REALIZING BACK OPERATION OF FUZZY NEURAL NETWORK

摘要

PURPOSE: A method for realizing a back operation of a fuzzy neural network is provided to perform a back course of the learning in a fuzzy neural network formed combining a neural network and a fuzzy neuron. CONSTITUTION: A method for realizing a back operation of a fuzzy neural network comprises the steps of: initializing a pure output value of an input layer by the usage of a target fuzzy vector value as an input(102); calculating a presumption inverse of an initial step through activation function from the initializing step(103); forward-feeding an input value transferred to the input layer to calculate an output value(104); examining an ending condition after calculating an error value between the calculated output value and a target output value(106); if the calculated differential value is less than a predetermined error level value, calculating a back propagation value of an error signal(107); calculating a differential value to the pure output value of the input layer by the usage of the error signal(108); after correcting the presumption inverse(109), forward-feeding the input value to return to the step of calculating the output value; and if the calculated differential value is less than the error level value, calculating a new presumption inverse(110).