混合前馈型神经网络在入侵检测中的应用研究  

Application of a hybrid feedforward neural network to intrusion detection

作　　者：姚羽[1] 高福祥[1] 邓庆绪[1] 于戈[1] 张守智[2] 

机构地区：[1]东北大学信息科学与工程学院,沈阳110004 [2]辽宁省财政厅信息中心,沈阳110002

出　　处：《控制与决策》2007年第4期432-435,共4页Control and Decision

基　　金：国家自然科学基金项目(60473073);国家863计划项目(2004AA1Z2060);国家973计划项目(2006CB303000);广东省自然科学基金项目(04010589)

摘　　要：提出一种基于混沌神经元的混合前馈型神经网络,用于检测复杂的网络入侵模式.这种神经网络具有混沌神经元的延时、收集、思维和分类的功能,避免了MLP神经网络仅能识别网络中当前的滥用入侵行为的弱点.对混合网络进行训练后,将该网络用于滥用入侵检测.使用DARPA数据集对该方法进行评估,结果表明该方法可有效地提高对具备延时特性的Probe和DOS入侵的检测性能.A hybrid feedforward neural network based on chaotic neuron is proposed to detect complicate network intrusion. The proposed neural network has the capability of time-delay, collection, thinking and classification, based on which the weakness of general neural network is avoided which can only detect current misuse intrusion. The neural network is trained and applied to misuse intrusion detection cases. This approach is evaluated by using DARPA data set. Results show that the system＇s capability of detecting time-delayed Probe and DOS attacks is enhanced effectively by using the proposed approach.

关 键 词：网络安全 入侵检测系统 前馈型神经网络 混沌神经网络 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]