基于生成对抗网络的异常行为模拟算法研究  

Research on Network Abnormal Behavior Simulation Algorithm Based on Generative Adversarial Network

作　　者：赵维[1] ZHAO Wei(Jilin Police College,Changchun 130117)

机构地区：[1]吉林警察学院

出　　处：《长春理工大学学报（自然科学版）》2019年第6期138-142,共5页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：吉林省教育厅项目(吉教科合字[2016]第558号)

摘　　要：信息技术飞速发展导致了网络上的信息日益增加,随之而来的网络攻击日渐频繁,其频率和破坏力都在不断上升,攻击的隐匿性也越来越高。隐藏在大量信息下的网络攻击和异常行为,亟需有效的检测方法。训练机器学习检测算法时,对异常样本的数量要求较高。当异常样本在训练数据集中比例较小时,获得的模型检测效果较差。本文提出一种基于生成式对抗网络(Generative Adversarial Network,GAN)的异常数据模拟算法,用于提高训练样本中异常数据集的比例,解决了训练样本数据不均衡的问题,并利用K-means算法验证了生成样本数据的质量。The rapid development of information technology has led to the increasing amount of information on the net-work,which is then followed by increasingly frequent network attacks.These attacks are characterized by heightened frequency,intensified destructive power and enhanced concealment.Under such background,there is an urgent need to develop an effective detection method to discover the network attacks and abnormal behaviors hidden in a large amount of information.However,when training detection algorithms using machine learning,there is a high requirement for the amount of abnormal samples.This is to say,when the abnormal samples take up a small proportion of the whole train-ing data set,the detection performance of the obtained model will be relatively poor.To address this problem,a simu-lation algorithm for abnormal data based on Generative Adversarial Network(GAN)is proposed in this paper,so as to increase the proportion of abnormal data in the whole training samples.In this way,the imbalance problem existing in the training samples can be solved.Furthermore,K-means algorithm is used to verify the quality of the generated sample data.

关 键 词：生成对抗网络 异常行为模拟 网络攻击 K-MEANS算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]