AUTONOMOUSLY CONSTITUTING METHOD FOR FUZZY NEURAL NETWORK AND CONTROLLING METHOD USING FUZZY NEUTRAL NETWORK CONSTITUTED BY AUTONOMOUSLY CONSTITUTING METHOD

摘要

PROBLEM TO BE SOLVED: To autonomously generate a fuzzy rule even for a system to be controlled having high nonlinearity by autonomously adding and deleting a fuzzy rule in a learning process and obtaining an optimum fuzzy rule for the system to be controlled. SOLUTION: It is judged whether or not a means error E(n) after (n)-time learning is larger than a threshold value Th2 multiplied by the coefficientα(α<1) of a mean error E(n-1) after (n-1)-time learning (step 2). Here, when E(n) is larger than Th2, it is judged that the learning is stagnant and a rule adding process is performed (step 3), but when small, it is judged that the learning is proceeding and a learning process is performed (step 4). When all consequent combination loads wf that a membership function is concerned in have linear relation with the coefficient wf of at least other two membership functions in the same direction, it is judged that the membership function as an object to be evaluated can be complemented with other membership functions in the same direction which have the linear relation with the coefficient wf and the membership function is deleted (steps 6, 7).