人工神经网络与遗传算法相结合的入侵检测模型的研究

发布时间：2018-06-21 22:49

本文选题：入侵检测 + 粗糙集理论　；参考：《江苏科技大学》2015年硕士论文

【摘要】：伴随着互联网络和计算机技术的高速发展,互联网通信已经渗透到政治、经济、文化和生活以及科学的各个领域,极大地影响着人类社会各个方面的进步和发展。同时,它也极大地影响和改变着人们的生活、工作和学习。但是,随着互联网络的发展,各种安全问题也随之出现。有关统计数据显示,在全球范围内,每隔20秒就会发生一起网络入侵事件。网络上的黑客能够轻松的盗取你的私密文件,盗取你的银行存款信息,破坏你的个人账目信息,将你的私密信函公之于众,随意改正、扰乱和破坏你的数据库中的信息,甚至直接破坏你的磁盘和计算机等硬件设施,导致你的网络处于瘫痪或者崩溃状态。所以,研究一些切实有效的互联网安全技术来保障计算机系统和互联网系统的安全,已经成为学术界和商界研究的热点问题。入侵检测是一种主动防御入侵的安全方式,是维护网络安全的主要模块。因此该技术成为人们研究的热点。目前入侵检测技术正在朝着自动化和智能化方向发展。所谓入侵检测的智能化是将遗传算法和人工神经网络算法等智能化算法用于入侵检测。传统的入侵检测系统误报率和漏报率都很高,不能够很好的识别新兴的攻击类型。随着网络数据量的急剧增加,传统的入侵检测技术不能胜任实时性的要求。此外,传统的遗传算法和神经网络算法还具有收敛速度慢和容易陷入局部极小值等缺点。本文通过深入分析研究入侵检测的特点和遗传算法以及神经网络算法的结构。为了适应实时性要求,首先基于粗糙集理论,对传统遗传算法的选择交叉变异等环节做了一系列改进。利用了粗糙集理论的知识,重新设置了适应度函数,使其能够快速删除冗余的属性,约简大型知识系统。然后对神经网络算法添加自己的想法,做了一系列改进,将约简后的大型知识系统输入到改进后神经网络中,使其能够正确高效的识别入侵数据。最后在Windows环境下利用Rosetta和MATLAB软件编写仿真程序,采用KDDCUP99数据集,设置了两组试验以证明本文提出的算法的有效性。
[Abstract]:With the rapid development of Internet and computer technology, Internet communication has penetrated into the fields of politics, economy, culture, life and science, which has greatly affected the progress and development of all aspects of human society. At the same time, it also greatly affects and changes people's life, work and study. However, with the development of Internet, all kinds of security problems appear. Statistics show that globally, a cyber intrusion occurs every 20 seconds. Hackers on the Internet can easily steal your private documents, steal your bank deposit information, destroy your personal accounts, publish your private letters to the public, correct them at will, disrupt and destroy the information in your database. Even directly damage your disk and computer hardware facilities, resulting in your network in a state of paralysis or crash. Therefore, the research of some effective Internet security technologies to protect the security of computer systems and Internet systems has become a hot issue in academia and business circles. Intrusion detection is a kind of active defense against intrusion, and it is the main module to maintain network security. Therefore, this technology has become the focus of research. At present, intrusion detection technology is developing towards automation and intelligence. Intelligent intrusion detection is based on genetic algorithm (GA) and artificial neural network (Ann). The traditional intrusion detection system has high false alarm rate and false alarm rate, so it can not identify the emerging attack types. With the rapid increase of network data, the traditional intrusion detection technology can not meet the requirements of real-time. In addition, the traditional genetic algorithm and neural network algorithm also have the disadvantages of slow convergence speed and easy to fall into local minimum. In this paper, the characteristics of intrusion detection and the structure of genetic algorithm and neural network algorithm are studied. In order to meet the real-time requirements, a series of improvements are made to the selection of crossover mutation in traditional genetic algorithm based on rough set theory. By using the knowledge of rough set theory, the fitness function is set up so that the redundant attributes can be quickly deleted and the large knowledge system can be reduced. Then we add our own ideas to the neural network algorithm and make a series of improvements to input the reduced large-scale knowledge system into the improved neural network so that it can recognize intrusion data correctly and efficiently. At last, the simulation program is compiled by Rosetta and MATLAB in Windows environment, and the KDDCUP99 data set is used to set up two groups of experiments to prove the validity of the proposed algorithm.
【学位授予单位】：江苏科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP18;TP393.08

【参考文献】