摘要
本文提出了一种用模糊－神经技术建造专家系统的方法（ＦＮＴ方法）。从领域专家处获取的知识是以模糊规则和隶属函数的形式表示的。根据本文提出的方法，首先将模糊规则和隶属函数用神经网络表示出来（导入）；生成的神经网络用于实现模糊推理，然后利用修改的反传算法训练神经网络，从而提高系统的精度，修改隶属函数，求精模糊规则；最后从神经网络中提取隶属函数和模糊规则（导出），帮助解释神经网络的内部表示和操作。利用本文所提出的方法建造的系统可实现快速的无匹配模糊推理，并具有较强的学习能力。
This paper presents a method of building expert systems with fuzzy-neural technology'(FNT method). The domain knowledge is represented by fuzzy rules and membership functions. WithFNT method, these fuzzy rules and membership functions are derived into a layered neural networkwhichs is then trained by back-propagation algorithm to modify the membership functions and refinefoe fuzzy rules. Finally, the updated fuzzy rules and membership functions are derived out from theneural network to interpret intern31 representation and operations of the neural network. The expertsystems thus built can implement fast inference without rule matching process and have the learningability.
出处
《计算机研究与发展》
EI
CSCD
北大核心
1994年第5期17-23,共7页
Journal of Computer Research and Development
关键词
神经网络
模糊推理
专家系统
back propagation algorithm, neural network, fuzzy inference, data interpretation
