不平衡数据下基于SVM增量学习的指挥信息系统状态监控方法  

C4ISR state monitoring method based on SVM incremental learning of imbalanced data

作　　者：焦志强[1,2] 易侃 张杰勇 姚佩阳[1] JIAO Zhiqiang;YI Kan;ZHANG Jieyong;YAO Peiyang(College Information and Navigation,Air Force Engineering University,Xi’an 710077,China;Unit 95910 of the PLA,Jiuquan 735018,China;Science and Technology on Information Systems Engineering Laboratory,Nanjing 210007,China)

机构地区：[1]空军工程大学信息与导航学院,陕西西安710077 [2]中国人民解放军95910部队,甘肃酒泉735018 [3]信息系统工程重点实验室,江苏南京210007

出　　处：《系统工程与电子技术》2024年第3期992-1003,共12页Systems Engineering and Electronics

摘　　要：针对指挥信息系统历史状态样本有限的特点,基于支持向量机(support vector machines,SVM)设计了一种面向不平衡数据的SVM增量学习方法。针对系统正常/异常状态样本不平衡的情况,首先利用支持向量生成一部分新样本,然后通过分带的思想逐带产生分布更加均匀的新样本以调节原样本集的不平衡比。针对系统监控实时性要求高且在运行过程中会有新样本不断加入的特点,采用增量学习的方式对分类模型进行持续更新,在放松KKT(Karush-Kuhn-Tucker)更新触发条件的基础上,通过定义样本重要度并引入保留率和遗忘率的方式减少了增量学习过程中所需训练的样本数量。为了验证算法的有效性和优越性,实验部分在真实系统中获得的数据集以及UCI数据集中3类6组不平衡数据集中与现有的算法进行了对比。结果表明,所提算法能够有效实现对不平衡数据的增量学习,从而满足指挥信息系统状态监控的需求。To the characteristic of limited historical sample of command,control,communication,and computer,intelligence,surveillance and reconnaissance(C4ISR),an incremental learning method based on support vector machines(SVM)is designed for imbalanced data.To the imbalance of normal/abnormal state samples of the system,first use the support vector to generate a part of new samples,and then use the idea of banding to generate new samples with a more uniform distribution to adjust the imbalance ratio of the original sample set.In view of the high requirements for real-time monitoring of the system and the continuous addition of new samples during operation,the classification model is continuously updated by incremental learning.On the basis of relaxing the KKT(Karush-Kuhn-Tucker)update triggering conditions,by defining the sample importance and the introduction of retention rate/forgetting rate to reduce the number of training samples required in the incremental learning process.In order to verify the effectiveness and superiority of the algorithm,the experimental part compared the existing algorithms in the real system data set and the UCI data set with 3 types and 6 groups of imbalanced data sets.The results show that the proposed algorithm can effectively realize the incremental learning of imbalanced data,so as to meet the requirements of the C4ISR state monitoring.

关 键 词：指挥信息系统 系统监控 支持向量机 不平衡数据 增量学习 

分 类 号：TP315[自动化与计算机技术—计算机软件与理论]