Fuzzy control using neural network techniques

A method of fuzzy control using a multilayer neural network which learns fuzzy rules using the error backpropagation algorithm is proposed. To demonstrate the method, a motor servo control was simulated to confirm that tracking could be conducted. The authors also investigated the relationship between fuzzy rule number or effect of learning rules and output using a three-dimensional output expression. The more the network learns, the clearer the undulation is, but the number of rules which were learned does not affect the input-output relationship seriously if rules express a similar relationship. This system was compared with an ordinary fuzzy control method presented by E.H. Mamdani (1976). In this case, the input-output relationship is rugged. It is pointed out that one of the advantages of using a neural network is insensitivity to damage. It was found that if certain important connections are severed, the effects are critical