A NEURAL FUZZY INFERENCE SYSTEM  

作　　者：Lu Jing 

机构地区：[1]Computer Science Department, Oklahoma State University

出　　处：《Journal of Electronics(China)》2013年第4期401-410,共10页电子科学学刊（英文版）

摘　　要：This paper proposes a new neural fuzzy inference system that mainly consists of four parts. The first part is about how to use neural network to express the relation within a fuzzy rule. The second part is the simplification of the first part, and experiments show that these simplifications work. On the contrary to the second part, the third part is the enhancement of the first part and it can be used when the first part cannot work very well in the fuzzy inference algorithm, which would be introduced in the fourth part. Finally, the fourth part "neural fuzzy inference algorithm" is been introduced. It can inference the new membership function of the output based on previous fuzzy rules. The accuracy of the fuzzy inference algorithm is dependent on neural network generalization ability. Even if the generalization ability of the neural network we used is good, we still get inaccurate results since the new coming rule may not be related to any of the previous rules. Experiments show this algorithm is successful in situations which satisfy these conditions.This paper proposes a new neural fuzzy inference system that mainly consists of four parts. The first part is about how to use neural network to express the relation within a fuzzy rule. The second part is the simplification of the first part, and experiments show that these simplifications work. On the contrary to the second part, the third part is the enhancement of the first part and it can be used when the first part cannot work very well in the fuzzy inference algorithm, which would be introduced in the fourth part. Finally, the fourth part ＂neural fuzzy inference algorithm＂ is been introduced. It can inference the new membership function of the output based on previous fuzzy rules. The accuracy of the fuzzy inference algorithm is dependent on neural network generalization ability. Even if the generali- zation ability of the neural network we used is good, we still get inaccurate results since the new coming rule may not be related to any of the previous rules. Experiments show this algorithm is successful in situations which satisfy these conditions.

关 键 词：Fuzzy logic Neural network Relation within fuzzy rule . Graph solution 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置] U270.1[自动化与计算机技术—控制科学与工程]