正则化极限学习下细粒度网络入侵检测仿真

Simulation of Fine-Grained Network Intrusion Detection under Regularization Limit Learning

由于网络通信数据产生的速度快,且细粒度网络入侵检测系统通常会产生大量的误报,即误将正常网络活动误认为异常行为,导致对网络的入侵检测精度降低。为了得到高精度和高效率的细粒度网络入侵检测结果,提出一种基于正则化极限学习的细粒度网络入侵检测方法。整个方法划分为两个阶段,第一阶段采用多个尺度不同的滑动窗口将原始网络流量划分为多个观察跨度的子序列,通过小波变换技术对子序列重构获取多层级序列,链式SAE使用特征空间映射形成多层级重构序列。以重构序列误差展开网络流量异常初步判定,并且将各个层级的初步判定结果展开汇总,实现网络流量异常检测。第二阶段通过天牛群优化算法对正则化极限学习的权值和阈值展开联合优化,同时利用优化处理后的正则化极限学习对异常类样本展开细粒度网络入侵检测。实验结果表明,所提方法可以精准检测细粒度网络入侵,检测率保持在95%以上,且检测时间短。

At present,the network communication data is generated quickly.In general,the fine-grained network intrusion detection system produces a large number of false positives,that is,mistakenly identifying normal network activities as abnormal behaviors,resulting in low accuracy of intrusion detection.In order to obtain high-precision and high-efficiency detection results of fine-grained network intrusion,this paper proposed a method of detecting finegrained network intrusions based on regularized extreme learning.The method was divided into two stages.In the first stage,multiple sliding windows with different scales were used to divide the original network traffic into multiple subsequences of observation spans.Then,wavelet transform technology was used to reconstruct the subsequences,thus obtaining multi-level sequences.Moreover,the chain SAE used feature space mapping to form multi-level reconstruction sequences.Based on the reconstruction sequence error,the preliminary determination of network traffic anomalies was carried out,and then the results of each level were summarized to realize the detection of network traffic anomalies.In the second stage,the regularized extreme learning weights and thresholds were jointly optimized by the beetle swarm algorithm(BSA).At the same time,the regularized extreme learning after optimization was used to perform fine-grained network intrusion detection on abnormal samples.Experimental results prove that the proposed method could accurately detect fine-grained network intrusions,with a detection rate of over 95%and a short detection time.