基于BP神经网络的入侵检测技术的研究

发布时间：2018-11-13 19:31

【摘要】：随着计算机技术的迅速发展及网络的广泛应用，网络攻击事件频繁发生，并且入侵的手段层出不穷，使得防火墙越来越难以单独地保障网络安全。因此，在增强网络的安全性方面，许多组织和专家都倾向于更强有力的主动安全防护策略，其中，入侵检测技术作为一种有效的解决途径，已成为了一个研究的焦点。本文首先对入侵检测技术及其发展趋势进行了相关介绍。分析和总结了传统的入侵检测技术所存在的一些缺陷，比如误报率、漏报率高，对未知攻击的识别能力差，实时检测攻击的能力低等。近些年，，人工神经网络技术开始运用到了入侵检测领域，为入侵检测技术开辟了新的研究途径。因此，本文引入了人工神经网络的相关理论，并重点研究了在神经网络领域被广泛使用的BP神经网络(Back Propagation NeuralNetwork)。本文首先详细论述了BP神经网络的工作原理、基本思想及算法的推导过程，同时对其所存在的问题和几种目前较常使用的改进方法进行了分析和介绍。然后，针对入侵检测，本文在自适应学习率法的基础上进行了改进，提出了一种新的改进算法：每次迭代时，根据误差和局部梯度两者的变化信息，采用不同的学习率分别对各个节点之间的连接权重和阈值进行调整，使权值的修改更有针对性。同时，也引入了带有可调因子的转移函数，克服BP网络在误差平坦区收敛速度慢的缺点，以提高BP网络在入侵检测中的检测能力和收敛速度。最后，本文采用KDDCUP99数据集中的部分数据在Matlab7.0下对本文所提出的改进算法、自适应学习率法和标准BP算法分别进行了仿真实验，并对实验结果进行分析，结果证明该算法能够达到预期的效果。
[Abstract]:With the rapid development of computer technology and the wide application of network, network attacks occur frequently, and the means of intrusion emerge endlessly, which makes it more and more difficult for firewall to protect network security alone. Therefore, in enhancing the security of network, many organizations and experts tend to more powerful active security protection strategy, among which, intrusion detection technology as an effective solution has become a focus of research. This paper firstly introduces the intrusion detection technology and its development trend. Some shortcomings of traditional intrusion detection techniques are analyzed and summarized, such as false alarm rate, high false alarm rate, poor ability to identify unknown attacks, and low ability to detect attacks in real time. In recent years, artificial neural network technology has been applied to the field of intrusion detection, which has opened up a new research approach for intrusion detection technology. Therefore, this paper introduces the theory of artificial neural network, and focuses on the research of BP neural network (Back Propagation NeuralNetwork)., which is widely used in the field of neural network. In this paper, the working principle, basic idea and derivation process of BP neural network are discussed in detail. At the same time, the existing problems and several improved methods are analyzed and introduced. Then, based on the adaptive learning rate method, an improved algorithm is proposed for intrusion detection, which is based on the information of error and local gradient in each iteration. The weight and threshold of each node are adjusted with different learning rates, so that the modification of weights is more targeted. At the same time, the transfer function with adjustable factor is introduced to overcome the shortcoming of the slow convergence rate of BP network in the error flat area, so as to improve the detection ability and convergence speed of BP network in intrusion detection. Finally, the improved algorithm, the adaptive learning rate method and the standard BP algorithm are simulated by using part of the KDDCUP99 data set under Matlab7.0, and the experimental results are analyzed. The results show that the algorithm can achieve the desired results.
【学位授予单位】：河北工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.08;TP18

【相似文献】