基于改进深度卷积生成对抗网络的入侵检测方法

Intrusion Detection Method Based on Improved Deep Convolutional Generative Adversarial Network

针对入侵检测系统因采用的网络攻击样本具有不平衡性而导致检测结果出现较大偏差的问题,提出一种将改进后的深度卷积生成对抗网络(deep convolution generation adversarial network,DCGAN)与深度神经网络(deep neural network,DNN)相结合的入侵检测模型(DCGAN-DNN),深度卷积生成对抗网络能够通过学习已知攻击样本数据的内在特征分布生成新的攻击样本,并对深度卷积生成对抗网络中生成网络所用的线性整流(rectified linear unit,ReLU)激活函数作出改进,改善了均值偏移和神经元坏死的问题,提升了训练稳定性。使用CIC-IDS-2017数据集作为实验样本对模型进行评估,与传统的过采样方法相比DCGAN-DNN入侵检测模型对于未知攻击和少数攻击类型具有较高检测率。

An intrusion detection model(DCGAN-DNN)that combines an improved deep convolutional generative adversarial network(DCGAN)with a deep neural network(DNN)was proposed to address the problem of large bias in the detection results of intrusion detection systems due to the unbalanced nature of the network attack samples used.The intrinsic feature distribution of the known attack sample data was learned by the deep convolutional generative adversarial network to generate new attack samples.Rectified linear unit(ReLU)activation function used for generating networks in deep convolutional generative adversarial networks was improved to ameliorate the problems of mean shift and neuron necrosis and to enhanced training stability.The CIC-IDS-2017 dataset was used as an experimental sample to evaluate the model,the DCGAN-DNN intrusion detection model was used to compare with traditional oversampling methods and was found to have a high detection rate for unknown attacks and a few attack types.