基于拟牛顿算法神经网络的入侵检测系统的研究

近年来，神经网络技术在入侵检测中得到了广泛应用，其中最具代表性的是BP神经网络，但其本身所具有的局部极小性质限制了检测性能的提高。利用人工神经网络可以解决当前其他入侵检测方法中所遇到许多问题，有望成为异常检测中统计方法的替代品，是研制具有学习和适应能力的入侵检测系统重要手段之一。通过抽取部分混合实例以及典型攻击实例进行模式训练、测试后，在BP神经网络优化算法进行对比研究的基础上，利用拟牛顿算法对传统BP算法进行改进，从而提高入侵检测系统的收敛度，检测率。实验分析可得，在一定的训练方法基础上，基于拟牛顿算法优化神经网络和其他几种算法相比，在针对多种攻击类型上检测率有不同程度的提高。

Research of Intrusion Detection System based on the Quasi-Newton Algorithm in Neural Networks

In recent years, the neural network technology obtained the widespread application in the Intrusion, what most has represents is the BP neural network, but the local minimum nature of itself has limited the detection performance enhancement. The use of artificial neural network can solve the current intrusion detection methods other encountered many problems, is expected to be a statistical anomaly detection method alternative is to develop the ability to learn and adapt to the intrusion detection system as one important means. In this paper, part of the mixed samples, as well as the typical mode of attack examples of training, testing, the BP neural network optimization algorithm on the basis of comparison, the use of quasi-Newton algorithm to improve the traditional BP algorithm, intrusion detection system so as to enhance the degree of convergence, detection rate. Experimental analysis of the availability of certain training methods based on quasi Newton algorithm based on neural networks and several other algorithms, in a variety of attack types for detection rate improved to different extents.