Abstract: To improve the reality，useful and whole performance of network Intrusion Detection System（IDS），the solution to get independent characters used to（or evidence） IDS is presented in the paper.The approach of auto-recognition of inter-relativity of the characters is developed，which is classification method for characters.The characters picked up by the method are used as the input of back propagate neural network.The characters are inter-independent，or weak relative.On the base of the chosen characters，an IDS is built.Tests show that the approach and IDS developed in the article is useful and available.It is conclusion that a set of inter-independent characters should be provided for IDS.The inter-relative degree of characters can be required by the method developed in the article.

Key words: Intrusion Detection System（IDS）, inter-relativity of character, back propagate neural network

摘要： 为了提高网络入侵检测系统（IDS）的实时性、可用性以及整体性能，提出了一种自动识别特征相关性的方法（特征分类法）。用该方法提取出的互相独立（或相关性很小）的特征作为反向传播神经元网络的输入，以此为基础建立了一个IDS。实验证明该方法以及所建立的IDS效果较好。结论表明通过分类可以求得一组特征互相之间的相关程度，进一步可求得互相独立（或相关性很小）的特征。

关键词: 入侵检测系统, 特征相关性, 反向传播神经元网络