一种用于异常检测的实值否定选择算法  

Improved real-value negative selection algorithm for anomaly detection

作　　者：李亚伦[1] 柴争义[2] 陈国强[3] Li Yalun;Chai Zhengyi;Chen Guoqiang(School of Electronic and Information Engineering,Tianjin Polytechnic University,Tianjin 300387,China;School of Computer Science and Software Engineering,Tianjin Polytechnic University,Tianjin 300387,China;School of Computer and Information Engineering,Kaifeng 475004,China)

机构地区：[1]天津工业大学电子与信息工程学院,天津3000387 [2]天津工业大学计算机科学与软件学院,天津3000387 [3]河南大学计算机与信息工程学院,河南开封475004

出　　处：《南京理工大学学报》2019年第5期586-591,共6页Journal of Nanjing University of Science and Technology

基　　金：国家自然科学基金(61972456);河南省高等学校重点科研项目(14A520079);河南省科技攻关计划(162102210168)

摘　　要：为了提高实值否定选择算法的效率,通过分析检测器的更新过程,针对已有实值否定选择算法检测器生成冗余的问题,提出一种减少冗余检测器生成的优化算法。所提算法检测器采用二次否定选择,并通过改变接受和拒绝零假设的条件来快速更新检测器集,进而减少无效检测器的生成。采用该算法对合成数据集2DSyntheticData和实际的Iris数据集进行了实验,结果表明,该算法误警率低,所需检测器的数量明显减少,适合实时异常检测。To solve the problem that existing real-valued negative selection algorithms generate redundancy detectors,an improved detector generation algorithm is proposed by analyzing the updating process of the detector set.The proposed algorithm uses dual negative selection and quickly updates the detector set by changing the conditions of accepting and rejecting the null hypothesis,thereby it reduces the generation of invalid detectors.The 2DSyntheticData and the actual Iris data set are used to test the algorithm.Experimental results show that the false positive rate of the algorithm is lower,especially the number of detectors is significantly reduced,and it is suitable for the real-time anomaly detection.

关 键 词：否定选择算法 异常检测 零假设 检测器生成 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP309[自动化与计算机技术—控制科学与工程]