基于深度图嵌入的无人机自组网链路预测  

UAV ad hoc network link prediction based on deep graph embedding

作　　者：舒坚[1] 王启宁 刘琳岚[2] SHU Jian;WANG Qining;LIU Linlan(School of Software,Nanchang Hangkong University,Nanchang 330063,China;School of Information Engineering,Nanchang Hangkong University,Nanchang 330063,China)

机构地区：[1]南昌航空大学软件学院,江西南昌330063 [2]南昌航空大学信息工程学院,江西南昌330063

出　　处：《通信学报》2021年第7期137-149,共13页Journal on Communications

基　　金：国家自然科学基金资助项目(No.62062050,No.61762065,No.61962037);江西省自然科学基金资助项目(No.20202BABL202039);江西省研究生创新专项资金资助项目(No.YC2019-S355)。

摘　　要：针对无人机自组网的拓扑时变、节点移动、间歇性连接等特点,提出用时序化图嵌入模型对预处理后的无人机自组网进行表征,基于线性概率计算采样间隔以提高采样效率,将网络结构特征映射为节点间关系,并采用对抗训练提取节点上下文语义特征。利用长短期记忆网络提取无人机自组网的时序特征,预测下一时刻的网络连接情况。采用AUC、MAP、Error Rate作为评价指标。Ns-3仿真实验表明,与Node2vec、DDNE、E-LSTM-D等方法相比,所提方法具有更高的预测准确率。Aiming at the characteristics of the UAV ad hoc network(UAANET),such as topological temporal-varying,node mobility and intermittent connection,a temporal graph embedding model was proposed to present the preprocessed UAANET.To improve the sampling efficiency,the sampling interval was calculated based on linear probability.The network structure features were mapped to the relationship between nodes,and the contextual semantic features of nodes were extracted by adversarial training.With the help of long and short-term memory network,the temporal characteristics of the UAANET were extracted to predict the connection at the next moment.AUC,MAP,and Error Rate were employed as evaluation indexes.The simulation experiments based on NS-3 show that compared with Node2vec,DDNE and E-LSTM-D,the proposed method has a better accuracy.

关 键 词：无人机自组网 图嵌入 链路预测 长短期记忆网络 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]