The traditional rough logic neural network can do research in information systems and decision-making and reveal the substance of rough set theory, but cannot get good results when dealing with the problem of non-single-value input The rough neuron with upper and lower boundary can deal with the above problem, and with the development of rough set,the concept of upper and lower boundaries has been widely used. Comprehening the above advantages, this paper propounded the construction and studying of a kind of rough logic neural network. It is made up of rough logic neural network and rough neurons (each variable in this pattern has both upper and lower bounds), which is called boundary rough logic neural network. First the paper gave the basic knowledge about rough neuron,rough logic and decision-making, and then propounded the structure of boundary rough logic neural network and learning methods,then gave the two models about it and compared the advantages and disadvantages between them. It indicated that this type of neural network,compared with traditional rough logic neural network, can be more efficiency when dealing with the problem non-single-valued and continuous approximation function. At last it proposed the optimized directioa%传统的粗逻辑神经网络可以对信息系统及粗推理决策进行研究,能够深入地揭示粗糙集理论实质,但对于处理非单值输入问题不能取得良好的效果.粗糙神经元的上边界和下边界恰好能解决这一方面的问题,且随着粗集理论的不断发展,上下边界的概念得到了广泛的应用.综合两个方面的优点,提出了一种粗逻辑神经网络的构造与学习方法.它主要由传统粗逻辑神经网络和粗糙神经元的思想(模式中每一个特征变量都包含上界和下界两个边界)构成:边界粗逻辑神经网络.首先给出了粗糙神经元和粗逻辑及决策的基本知识,然后提出了边界粗逻辑神经网络的结构和学习方法及两种模型并比较了模型间的优缺点.与传统粗逻辑神经网络相比,这类神经网络能更有效地处理非单值和连续近似域函数问题.最后提出可以进一步优化的方向.
展开▼
