神经网络的模糊规则提取

AN APPROACH TO FUZZY RULES EXTRACTION OF NEURAL NETWORKS

神经网络的“黑箱问题”为该技术的广泛应用带来了一定限制,由于神经网络在一定条件下可与模糊系统相互转换,从神经网络中提取模糊规则为“黑箱问题”的解决提供了有效手段。本文在阐述基本概念的同时,分析了把连续值网络转化为二值网络和从神经网络到模糊系统的转换进行模糊规则提取的两类方法,通过解决Iris问题的实验结果比较了两类方法的性能。

Owing to the shortcoming of being black boxes , the extension of the applications of neural networks into a wide range of areas is limited . Because a neural network can be approximated to any degree of accuracyly using a fuzzy system under certain assumptions, extracting fuzzy rules from neural networks is an effective method which can be provided to solve this problem. In this paper we discuss the fundamental concepts and in the meantime analyse two classes of fuzzy rules extraction techniques: the one converting continuous - valued networks into binary - valued networks and the other directly through functional equivalence between neural networks and fuzzy systems. Using the resolution of the Iris problem, the performances of these techniques have been compared.