Fuzzy inference device using neural network

A fuzzy inference device for determining an inference operational quantity from system input variables and output variables in accordance with inference rules each constituted by an antecedent and a consequent, comprising:

• an inference rule determiner having means for receiving the system input variables and including
• a a variable reducing processor which receives data of the input variables and reduces a quantity of the input variables so as to determine a number of the input variables,
• b an evaluator for evaluating the determined number of the input variables performed by said variable reducing processor,
• c a cluster generator which receives data of the input variables determined by said variable reducing processor and formulates a plurality of data groups so as to determine a number of fuzzy inference rules by determining representative data of each of the data groups,
• d a cluster evaluator for determining the fuzzy inference rules in said cluster generator;
• an antecedent neural element,, operatively coupled to said inference rule determiner, including
• e a standard membership generator for generating learning data to determine a membership function of an antecedent of fuzzy inference by using input data belonging to each of the fuzzy inference rules determined by said cluster generator,
• f a neural network learner which has a learning function of a neural network which is set according to the learning data generated in said standard membership generator,
• g an antecedent membership function generator for determining a shape of the membership function by using the learning data in said standard membership generator and the learning function in said neural network learner;
• an output data divider, operatively coupled to said inference rule determiner, for dividing data of the output variables into output data portions such that each of the output data portions corresponds to each of the fuzzy inference rules determined by said cluster generator of said inference rule determiner;
• a situational change processor, operatively coupled to said inference rule determiner, including
• h an initialization determiner for determining an initial value of a search parameter to determine a real number of a consequent of fuzzy inference by using the input data and output data in said output data divider,
• i a gradient vector determiner for calculating a difference of inference accuracy for the search parameter by starting from the initial value in said initialization determiner,
• j a consequent calculator for calculating a real number of the consequent on the basis of the difference calculated by said gradient vector determiner;
• an inference operational quantity determiner, operatively coupled to said antecedent neural element and to said situational change processor, which performs fuzzy inference by using the membership function in said antecedent membership function generator and the real number in said consequent calculator to determine the inference operational quantity;
• and
• an overall evaluator, operatively coupled to said inference operational quantity determiner, for evaluating, in accordance with an evaluation criterion, the inference operational quantity outputted by said inference operational quantity determiner.