摘要
为了进一步满足零信任电力物联网络的安全需求,文章提出了一种基于混合注意力机制的深度学习异常检测方案。首先,针对网络全要素数据进行用户与实体行为分析,实现多源异构数据中主体、行为与属性的深度关联挖掘。其次,以卷积神经网络为基础,融合多头注意力与空间注意力机制,基于多种特征通道对数据特征进行强化学习,实现对模型权重参数的优化调整,以提高异常检测的准确率。基于公开数据集对所提方案进行了验证,实验结果表明,该方案具有较好的异常检测性能,能够更好地满足零信任电力物联网的应用需求。
A hybrid attention mechanism-based deep-learning anomaly detection model is proposed in this paper to meet the security and efficiency challenge of zero-trust power IoT network.First,user and entity behavior analysis is conducted for the whole element data of the network to mine the association among subjects,behaviors,and attributes in multi-source heterogeneous data with deep learning.This model takes the convolutional neural network(CNN)as its backbone and combines the multi-attention and spatial attention mechanisms.Reinforcement learning of data features based on multiple feature channels optimizes model weight parameters to improve the accuracy of anomaly detection.The test and validation of the proposed method have been carried out on the public database.The experimental results show that the proposed method has better performance,which meets the application requirements of zero-trust power IoT.
作者
陈石
夏飞
张颂
刘行
王正琦
CHEN Shi;XIA Fei;ZHANG Song;LIU Xing;WANG Zhengqi(Information&Telecommunication Branch,State Grid Jiangsu Electric Power Co.,Ltd.,Nanjing 210000,Jiangsu Province,China;Nanjing NARI Information&Communication Technology Co.,Ltd.,Nanjing 210003,Jiangsu Province,China)
出处
《电力信息与通信技术》
2025年第9期73-79,共7页
Electric Power Information and Communication Technology
基金
国网江苏省电力有限公司科技项目资助“基于流量的零信任身份标识和行为分析关键技术研究”(J2023178)。
关键词
零信任
电力物联网
异常检测
用户与实体行为分析
混合注意力机制
zero trust
power Internet of Things
anomaly detection
user and entity behavior analytics
hybrid attention mechanism
