密度估计下异构网络数据异常辨识算法设计  

Design of Anomaly Identification Algorithm for Heterogeneous Network Data under Density Estimation

作　　者：尚建贞 王欣欣 SHANG Jian-zhen;WANG Xin-xin(Henan University of Animal Husbandry and Economy School of Information Engineering(School of Software),Henan Zhengzhou 450000,China;North China University of Water Resources and Electric Power School of Mechanical Engineering,Henan Zhengzhou 450000,China)

机构地区：[1]河南牧业经济学院信息工程学院(软件学院),河南郑州450000 [2]华北水利水电大学机械学院,河南郑州450000

出　　处：《计算机仿真》2024年第12期477-481,共5页Computer Simulation

基　　金：河南省高校重点科研项目(21A520044)。

摘　　要：在海量数据中辨识异常数据是确保异构网络安全运行的前提条件,为提高数据异常辨识精度、敏感性和效率,提出一种基于密度估计的异构网络数据异常辨识算法。采用稀疏去噪自编码网络消除异构网络中存在的噪声数据,避免噪声数据对辨识过程产生干扰,提高辨识结果的稳定性;通过密度估计获取异构网络数据的特征曲线,对其展开加权叠加处理,获得异构网络数据特征,并采用MISE最小准则优化窗宽,提高特征提取精度;引入欧几里得距离计算网络数据特征之间的相似度,设定异常辨识阈值,完成异构网络数据的异常辨识。实验结果表明,所提算法具有较高的辨识精度、敏感性和辨识效率。Identifying abnormal data in massive data is a prerequisite for ensuring the safe operation of heterogeneous network.To improve the accuracy,sensitivity,and efficiency of anomaly identification,this paper proposed an algorithm of identifying the anomaly of heterogeneous network data based on density estimation.Sparse denoising selfcoding network was used to eliminate noisy data in heterogeneous network,thus avoiding the interference from noisy data and improving the stability of identification results.Moreover,the feature curves of heterogeneous network data were obtained by density estimation.After the weighted superposition,the features of heterogeneous network data were obtained.Meanwhile,the MISE minimum criterion was adopted to optimize the window width,and thus to improve the accuracy of feature extraction.Finally,Euclidean distance was introduced to calculate the similarity between network data features.After setting the anomaly identification thresholds,we completed the anomaly identification of heterogeneous network data.The experimental results show that the proposed algorithm has high identification accuracy,sensitivity,and identification efficiency.

关 键 词：密度估计 稀疏去噪自编码网络 异构网络 密度估计 数据异常辨识 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术]