On Learning of Fuzzy Contoller with Neural Network

The fuzzy controller (FC) consists of two parts. First one is the control rule part which is referred to as linguistic rules. And it is written in the form of 'If∿THEN∿'. Second one is fuzzy reasoning which derives the reasoning value from the control rules. So it is easy for us to understand FC. However, there are two problems. First one is how to make the control rules. We usually take the rules from an expert. But it is not easy. Because all rules which the extpert has are not linguistic rules. The neural network has ability of learning. So many people expect that it is useful to get the control rules identified from the expert. And several researchers study with respect to it. Horikawa, one of them, presents a fuzzy controller using a neural network. The controller can automaticaly identify the control rules and tune the membership functions utilizing control data of the expert. The controller uses the rules written in the form of 'IF∿THEN y=f', where 'f' is constant. However, when FC uses this rules, it needs many rules. In this paper, we present a new fuzzy controller with a neural network which uses the rules written 'IF∿THEN y=f(・)', where f(・) is a linear function. Because when FC uses this rule, it need not many rules. In order to demonstrate the effectiveness of this FC, we simulate by using a computer to control a first order delayed system with a dead time. And we compare this controller with Horikawa's controller