基于改进卷积神经算法的电力物联网视频终端异常检测  

Anomaly detection of power internet of things video terminals based on improved convolutional neural algorithm

作　　者：王文胤 揭英波 王湛雄 朱南海 陈凯明 WANG Wenyin;JIE Yingbo;WANG Zhanxiong;ZHU Nanhai;CHEN Kaiming(Zhanjiang Power Supply Bureau,Guangdong Power Grid Co.,Ltd.,Zhanjiang 524000,Guangdong China)

机构地区：[1]广东电网有限责任公司湛江供电局,广东湛江524000

出　　处：《粘接》2024年第12期142-145,共4页Adhesion

基　　金：广东省2022年第一批省级创新项目(项目编号:030800KK52220017)。

摘　　要：为进一步提高异常流量检测效率,将直觉模糊时间序列模型与卷积神经算法相结合,提出改进卷积神经的电力物联网视频终端异常检测算法。利用多维属性熵值为顶点,利用相似度定义的顶点间熵值变化幅度和边权重构造完整图,建立视频终端流量的直观模糊时间序列模型。实验结果表明,随着终端数据包规模逐渐增大时,F1值接近1,检测效果较好。在卷积神经算法中,75%的异常流量被正确分类,而改进卷积神经算法中,可以使82%的异常流量正确分类。且改进卷积神经算法的FPR比卷积神经算法低78.45%,并具有更短的检测时间。可满足实际电力物联网视频终端异常流量检测应用。In order to further improve the efficiency of abnormal traffic detection,the intuitionistic fuzzy time series model and convolutional neural algorithm were combined to propose an anomaly detection algorithm for power Internet of Things video terminals with improved convolutional nerves.The multi-dimensional attribute entropy value was used as the vertex,and the entropy change amplitude and edge weight between vertices defined by similarity were used to construct a complete graph,and an intuitive fuzzy time series model of video terminal traffic was established.The experimental results indicated that as the size of terminal data packets gradually increased,the F1 value approached 1 and the detection effect was better.In convolutional neural algorithms,75%of abnormal traffic was correctly classified,while in improved convolutional neural algorithms,82%of abnormal traffic could be correctly classified.And the FPR of the improved convolutional neural algorithm was 78.45%lower than that of the convolutional neural algorithm,and it had a shorter detection time.It can meet the practical application of abnormal traffic detection in power IoT video terminals.

关 键 词：卷积神经算法 直觉模糊时间序列 电力物联网 视频终端 异常流量 

分 类 号：TM711[电气工程—电力系统及自动化] TP274[自动化与计算机技术—检测技术与自动化装置]