基于孪生神经网络的物联网通信异常数据捕获  

Abnormal Data Capture of IoT Communication Based on Twin Neural Network

作　　者：佟冬 张珠玲 TONG Dong;ZHANG Zhu-ling(School of Computer Science and Engineering,Jilin University of architecture and Technology,Changchun Jilin 130000,China;College of Communication Engineering,Jilin University,Changchun Jilin 130000,China)

机构地区：[1]吉林建筑科技学院计算机科学与工程学院,吉林长春130000 [2]吉林大学通信工程学院,吉林长春130000

出　　处：《计算机仿真》2021年第12期304-307,434,共5页Computer Simulation

摘　　要：针对传统物联网通信异常数据捕获精度较低、捕获耗时长等问题,提出基于孪生神经网络的物联网通信异常数据捕获方法。通过构建特征提取单元与区域推荐网络单元的孪生神经网络结构,去除网络Alex Net框架中两个卷积层与全连接层,在各卷积层后添加SE-Network,架构用于捕获异常数据的孪生神经网络模型;利用该网络模型降维处理异常数据特征,利用粒子群算法中标签约束策略,将二维数据转化到三维恒定空间中,完成异常数据种类划分;结合待捕获数据与候选数据特征,确定数据之间相似度,基于交叉相关层度量相似度,实现物联网通信异常数据捕获。仿真结果表明:采用所提方法可有效提升物联网通信异常数据捕获的精度,且捕获速度较快。Generally, the traditional method of capturing abnormal data in Internet of Things has thedisadvantages of low precision and long time-consuming. In this regard, we report a method of capturing abnormaldata in Internet of Things communication based on twin neural network. The twin neural network structure of featureextraction unit and regional recommendation network unit was established to eliminate the two convolution layers andfull connection layer in the Alex net framework. SE-Network was introduced into each convolution layer for construc-ting a twin neural network model for capturing abnormal data. The network model was applied to reduce the dimensionof abnormal data features. Label constraint strategy was also used to transform two-dimensional data into three-di-mensional constant space, thus completing the classification of abnormal data. The similarity between the data wasdetermined via combining the features of the data to be captured and the candidate data. Based on this similarity, theabnormal data of Internet of things communication was captured. Simulation results show that the method has high ac-quisition accuracy and fast acquisition speed.

关 键 词：孪生神经网络 物联网 通信数据 异常数据 特征图 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]