利用哈里斯鹰算法优化卷积神经网络的入侵检测研究

Research on Intrusion Detection Using Harris Hawk Algorithm to Optimize Convolutional Neural Network

以往利用卷积神经网络(CNN)搭建入侵检测模型时,需用人工经验设定网络结构,导致其网络性能很难发挥最优。为此,提出利用哈里斯鹰算法(HHO)对CNN的网络结构进行自适应优化,构建入侵检测模型。首先针对传统CNN全连接层易发生过拟合的问题,采用全局池化层(GAP)对参数进行缩减;然后采用哈里斯鹰算法选取CNN最佳网络结构,避免人工干预引起的检测不确定性,从而缩短参数选择时间,提升入侵检测模型的适用性和入侵检测性能。在NSL-KDD数据集的实验结果表明:所提哈里斯鹰算法优化改进后的卷积神经网络构建的入侵检测模型,检测准确率93.68%,误报率1.65%,检测性能优于SVM、AdaBoost、BP入侵检测模型。

In the past,having convolutional neural networks(CNN)adopted to build intrusion detection models asks for artificial experience to set the network structure which resulting in the difficulty in playing its network performance better.To this end,applying Harris Hawk algorithm to adaptive optimization of CNN structure to build an intrusion detection model was proposed.Firstly,aiming at the overfitting in the fully-connected layer of traditional CNN,having the global pooling GAP layer used to reduce parameters;then having the Harris Hawk algorithm adopted to select CNN’s optimal structure to avoid uncertainty of detection incurred by the manual intervention so as to shorten parameters selection time and improve applicability and intrusion detection performance of the intrusion detection model.The experimental results on the NSL-KDD dataset show that,the intrusion detection model established with Harris Hawk algorithm-improved CNN has a detection accuracy of 93.68%and a misinforming rate of 1.65%and its detection performance outperforms that of SVM,AdaBoost and BP intrusion detection models.