基于集成降噪自编码的网络入侵多模式匹配算法设计

Multi-pattern matching algorithm design of network intrusion based on integrated noise reduction own coding

针对高维网络入侵受到层次限制,入侵识别易出现局部极值化,导致入侵匹配时间延长的问题,提出并设计基于集成降噪自编码的网络入侵多模式匹配算法。引入降噪自编码网络,将深度学习法运用于在网络入侵检测中,构建深层网络模型,根据逐层贪婪编码模式实现预训练,完成高维深层网络入侵特征的自适应性提取及挖掘,通过监督式微调解决机器学习中局部极值问题,增强网络入侵检测效果与泛化性能,实现网络入侵初步检测。利用多模式匹配实现初步识别结果的最终匹配,完成网络入侵模式诊断。搭建实验平台对所提算法进行验证,将网络中用户行为日志数据当作实验数据样本,实验结果表明,负载均衡方差保持在0.85~1.0,匹配时间最短,访问文本串中字符的数目增长幅度很小,证明了算法鲁棒性和可实践性很强。

With the high-dimensional network intrusion limited by the level,the local extreme value easily occurring in the intrusion identification,and the intrusion matching time extending,this paper proposes multi-pattern matching algorithm design of network intrusion based on integrated noise reduction own coding.Noise reduction own coding network is introduced,deep approach of learning is employed in network intrusion detection,and deep network model is constructed.Extraction and mining of high-dimensional network intrusion signature adaptivity is achieved,which is based on the implement of pre-training in accordance with layered greedy coding.By solving the problem of the local extreme value by supervisory microposition,detective effectiveness and generalization performance is improved and initial surveys of network invasion is accomplished.Network intrusion pattern is diagnosed,which is on the basis of the final match of the preliminary results according to multi-pattern matching.The proposed algorithm is verified by establishing the experimental platform.Network user behaviors log data is analyzed and the results indicate that when the load balance variance is kept 0.85 and 1.0,matching time is the shortest and the number of characters of access text string slightly increases,which shows that the robustness and practicalness of the algorithm is powerful.