基于神经网络模型分割的入侵检测方法

摘要

To reduce resource consumption, rise speed of training and testing, a new method oflDS based on division of NN is proposed. The NN is divided into several small-scale neural networks according to the characteristic of the intrusion to perform learning and detecting this intrusion. In order to rise the speed of training and testing and to reduce the number of weights, every sub-neural network is divided into several smaller models. The detecting algorithm is developed and simulation is performed. The experimental results demonstrate that this method has the merit of fast learning and lower consumption of system resources, this method also improve the detection rate.%针对神经网络在入侵检测应用中存在资源消耗大、学习效率低等不足,提出一种基于神经网络模型分割的入侵检测方法.该方法根据当前典型攻击的特征,为每类攻击分别建立独立的子神经网络,对该类攻击进行学习和检测.然后再将每个子神经网络分割成多个更小的子模型,来降低学习时间和减少神经网络各层之间的连接权数目.设计了相应算法并进行仿真实验.实验结果表明,提出的方法提高了入侵检测的速度,降低了系统资源的消耗,提高了检测率.