基于集成学习技术的业务系统入侵数据篡改攻击识别  

Identification of business system intrusion and data tampering attacks based on integrated learning technology

作　　者：张文明 ZHANG Wenming(State Grid Zhejiang Electric Power Co.,Ltd.,Pujiang County Power Supply Company,Pujiang 322299,Zhejiang China)

机构地区：[1]国网浙江省电力有限公司浦江县供电公司,浙江浦江322299

出　　处：《粘接》2024年第11期143-146,共4页Adhesion

摘　　要：提出了一种基于集成学习技术的入侵数据检测方法,并使用焦点损失函数处理数据不平衡,提高数据篡改分类能力,并利用随机森林、深度学习、支持向量机与本文所提出方法进行性能比较。结果表明,当迭代次数大于80时,4种模型的收敛速度开始增加,并最终在迭代次数为140,趋于收敛。其中模型收敛的速度分别为深度学习>集成学习技术>随机森林>支持向量机。集成学习技术方法实现了95.83%的准确率,92.46%的精度,97.47%的召回率和94.90%的F1得分。相对于随机森林模型,集成学习技术方法在F1分数方面提高了约1.63%。集成学习技术的训练时间及检测时间分别为12、26 ms,均小于其他方法。In this paper,an intrusion data detection method based on ensemble learning technology was proposed,and the focus loss function was used to deal with the data imbalance,improve the data tampering classification ability,and the performance of random forest,deep learning and support vector machine was compared with the proposed method.The experimental results showed that when the number of iterations was greater than 80,the convergence speed of the four models began to increase,and eventually tend to converge at an iteration number of 140.The speed of model convergence was as follows:deep learning>ensemble learning technology>random forest>support vector machine.The integrated learning technology method achieved an accuracy of 95.83%,an accuracy of 92.46%,a recall rate of 97.47%,and an F1 score of 94.90%.Compared to the random forest model,the ensemble learning technology method improved F1 scores by approximately 1.63%.The training time and detection time of integrated learning technology were 12 ms and 26 ms respectively,which were smaller than those of other methods.

关 键 词：电力系统 数据篡改 检测 识别技术 

分 类 号：TP274.4[自动化与计算机技术—检测技术与自动化装置]