基于粗糙Multilayer Perceptron网络的规则提取及生成方法研究

将粗糙集理论同神经网络结合起来,提出了一种基于粗糙集的神经网络体系结构,并以此为基础讨论了神经网络中的规则提取方法.为扩大神经网络的适用范围,利用粗糙集理论中的相关概念来指导神经网络系统的构建,从而使系统参数具有较为明确的物理意义.提出一种对该网络进行训练的学习算法.根据网络中各节点所对应的物理意义以及相应连接权值的大小,给出了一种折衷型算法对粗糙神经网络进行推理规则提取和生成.实验结果表明:粗糙神经网络具有较好的工作性能,并且能给出相应的推理规则.

Approach to rule extraction and generation using rough Multilayer Perceptron networks

A rough neural network is proposed,based on rough sets theory and traditional neural networks.And hence the method of extracting rules from the rough neural network was discussed.To reduce the "black box" nature of neural networks,we use rough sets theory to direct the construction of neural networks,and work out a training algorithm for them.Furthermore,we propose an eclectic method to extract rules from these networks based on the practical meanings of different nodes and the weights between these nodes.Finally we illustrate an experiment which shows the rough neural networks can provide inference rules as well as good working performances.