基于优化数据处理的深度信念网络模型的入侵检测方法

发布时间：2018-10-19 06:40

【摘要】：针对目前网络中存在的对已知攻击类型的入侵检测具有较高的检测率,但对新出现的攻击类型难以识别的缺陷问题,提出了一种基于优化数据处理的深度信念网络(DBN)模型的入侵检测方法。该方法在不破坏已学习过的知识和不严重影响检测实时性的基础上,分别对数据处理和方法模型进行改进,以解决上述问题。首先,将经过概率质量函数(PMF)编码和MaxMin归一化处理的数据应用于DBN模型中;然后,通过固定其他参数不变而变化一种参数和交叉验证的方式选择相对最优的DBN结构对未知攻击类型进行检测;最后,在NSL-KDD数据集上进行了验证。实验结果表明,数据的优化处理能够使DBN模型提高分类精度,基于DBN的入侵检测方法具有良好的自适应性,对未知样本具有较高的识别能力。在检测实时性上,所提方法与支持向量机(SVM)算法和反向传播(BP)网络算法相当。
[Abstract]:At present, the intrusion detection of known attack types in the network has a high detection rate, but it is difficult to identify the new attack types. An intrusion detection method based on (DBN) model based on optimized data processing is proposed. In order to solve the above problems, the method improves the data processing and the method model separately on the basis of not destroying the knowledge that has been learned and not seriously affecting the real-time detection. First, the data encoded by the probabilistic quality function (PMF) and normalized by MaxMin are applied to the DBN model. The unknown attack type is detected by fixing other parameters and changing one parameter and cross-validating method by selecting the relatively optimal DBN structure. Finally, the method is verified on the NSL-KDD data set. The experimental results show that the optimal processing of data can improve the classification accuracy of DBN model. The intrusion detection method based on DBN has good adaptability and high recognition ability to unknown samples. In real time detection, the proposed method is comparable to the support vector machine (SVM) (SVM) algorithm and the backpropagation (BP) network algorithm.
【作者单位】：辽宁工程技术大学软件学院;
【基金】：辽宁省教育厅科学技术研究项目(LJYL052)~~
【分类号】：TP393.08

【相似文献】