基于改进的动态神经网络的网络入侵检测模型

Network Intrusion Detection Models Based on Improved Dynamic Neural Network

提出了一种改进型的动态神经网络,并成功地将其应用于网络入侵检测系统中。对于给定的全连接的动态神经网络,在通过学习以后可以成为部分连接的神经网络系统,从而降低了计算的成本。针对目前常见的4种不同类型的网络攻击行为(即DoS,Probe,R2L,和U2R),利用给定的改进型的动态神经网络分别构建相对应的检测系统。然后使用改进的遗传算法对给定的动态神经网络的权值和开关参数进行调节,以适应不同类型的入侵检测。最后利用KDD’99网络入侵检测数据对所提出的网络入侵检测模型进行训练和测试,初步试验结果表明,所提出的入侵检测系统具有较高的检测率。

This paper extends these works by applying an improved dynamic neural network （IDNN） to network intrusion detection system. The weights and parameters of the improved dynamic neural network are tuned by improved genetic algorithm （IGA）. A given fully connected feed-forward dynamic neural network may become a partially connected network after learning, which implies that the cost of implementing the IDNN may be reduced to very low. Four network intrusion detectors （corresponding to Dos, Probe, R2L and U2R, respectively） based on the improved dynamic neural network are developed, which can give balance detection accuracy between different types of intrusion. The proposed intrusion detectors are evaluated by KDD＇99 network intrusion detection data sets. Experimental results demonstrate that the proposed intrusion detectors have enough accuracy for network intrusion detection.