基于改进多因子优化蝙蝠算法的网络入侵检测方法  

etwork Intrusion Detection Method Based on Improved Multi-factorial Optimization Bat Algorithm

作　　者：张震[1] 张思源 田鸿朋 ZHANG Zhen;ZHANG Siyuan;TIAN Hongpeng(School of Electrical and Information Engineering,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]郑州大学电气与信息工程学院,河南郑州450001

出　　处：《郑州大学学报（工学版）》2024年第5期52-60,94,共10页Journal of Zhengzhou University（Engineering Science）

基　　金：国家重点研发计划重点专项(2018XXXXXXXXXX);河南省重大公益专项(201300311200);河南省重点研发专项(231111211600)。

摘　　要：针对高维网络数据存在大量冗余和不相关的特征导致入侵检测准确率低的问题,提出了一种改进的多因子优化蝙蝠算法(IMFBA)用于数据特征选择,筛选出具有最大信息量的特征子集,提高网络入侵检测精度。首先,在多因子优化框架下设计全局特征选择任务和局部特征选择任务,并通过基于蝙蝠算法所设计的选型交配和垂直文化传播算子实现不同任务间的信息共享,从而帮助全局特征选择任务更快锁定最优解空间,提高算法收敛速度和稳定性。其次,通过将反向学习策略和差分进化引入蝙蝠算法,重新设计算法初始解选择阶段及个体更新过程,弥补其缺少突变机制的不足,增强解的多样性,帮助算法摆脱局部最优。最后,提出一种自适应参数调整策略,根据潜在最优解质量决定其指导个体更新的权重,避免在多任务特征选择过程中出现知识负迁移现象,实现全局搜索与局部开发之间的平衡。实验结果表明:IMFBA所选特征子集对网络入侵数据集KDD CUP 99和NSL-KDD分类结果的准确率分别为95.37%和85.14%,相较于完整特征集提升了3.01百分点和9.78百分点。IMFBA算法能选择更高质量特征子集并提升网络入侵检测准确率。In addressing the challenge of diminished intrusion detection accuracy resulting from the abundance of redundant and irrelevant features in high-dimensional network data,an improved multi-factorial optimization bat algorithm(IMFBA)was introduced for precise data feature selection,with the ultimate goal of improving network intrusion detection accuracy.Within the multi-factorial optimization framework,global and local feature selection tasks were formulated.Information exchange between these tasks was facilitated by selection and vertical cultural transmission operators,strategically designed based on the bat algorithm.The global feature selection task was accelerated in identifying optimal solution spaces,thereby enhancing the algorithm′s convergence speed and stability.By incorporating the reverse learning strategy and differential evolution into the bat algorithm,the initial solution selection stage and individual updating process were refined to address the absence of a mutation mechanism,fostering solution diversity and aiding the algorithm in escaping local optima.An adaptive parameter adjustment strategy was introduced,determining weightings for guiding individual updates based on potential optimal solution quality.This could mitigate the risk of knowledge negative transfer during multi-task feature selection,achieving a balance between global exploration and local exploitation.The feature subsets selected by IMFBA demonstrate classification accuracy of 95.37%and 85.14%on the KDD CUP 99 and NSL-KDD intrusion detection datasets,respectively.This reflected increased by 3.01 percentage points and 9.78 percentage points compared to the complete dataset.Experiment results confirm the efficacy of EMFBA in selecting higher-quality feature subsets and,consequently,enhancing network intrusion detection accuracy.

关 键 词：入侵检测 网络安全 特征选择 蝙蝠算法 多因子优化 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]