基于加权多图神经网络的蜂窝网络流量预测  

Cellular network traffic prediction based on weighted multi⁃graph neural network

作　　者：李哲辉 宋文飞 王昌海 LI Zhehui;SONG Wenfei;WANG Changhai(The Scientific&Technological Information Center of Henan,Zhengzhou 450003,China;College of Software,Zhengzhou University of Light Industry,Zhengzhou 450000,China)

机构地区：[1]河南省科学技术情报中心,河南郑州450003 [2]郑州轻工业大学软件学院,河南郑州450000

出　　处：《电子设计工程》2025年第9期17-21,26,共6页Electronic Design Engineering

基　　金：河南省科技攻关项目(212102210096,242102210106)。

摘　　要：蜂窝网络流量预测对于优化基站资源配置、降低运营成本具有重要意义。使用图神经网络构建多个邻近基站流量间的非欧时空关联是该领域的主流预测方法。针对当前流量预测方法无法构建基站间多图动态关联的问题,提出一种基于加权多图卷积神经网络的蜂窝网络流量预测方法。使用基站间距离、数据相关性和注意力权重构建基站间的多图关系;基于三种图结构对基站流量数据进行空间特征聚合,并使用多图加权机制融合三种图的嵌入向量;使用LSTM建模时序数据特征得到预测结果。在公开数据集上对方法进行了仿真,实验结果表明,与已有方法相比,该文方法可得到最优预测结果。实验进一步分析了不同图结构和图规模对预测结果的影响,为方法的部署提供了参考。Cellular network traffic prediction is of great significance for optimizing base station resource allocation and reducing operating costs.The graph neural networks which can model non-Euclidean spatiotemporal correlations between multiple neighboring base station traffic are the mainstream prediction methods in this field.In order to solve the problem that the current traffic prediction method based on graph neural network cannot model the dynamic correlation of multiple graphs,a cellular network traffic prediction method based on weighted multi-graph convolutional neural network is proposed.The construction of multi-graph relationships is executed,utilizing the distance between base stations,data correlation,and attention weight.Spatial feature aggregation is carried out based on three graph structures,and a multi-graph weighting mechanism is employed to fuse the embedding vectors of the three graphs.Time series data features are modeled using LSTM to obtain prediction results.The method was simulated on a public data set,and the experimental results show that this paper can obtain optimal prediction results.The experiment further analyzed the impact of different graph structures and graph sizes on the prediction results,providing a reference for the deployment of the method.

关 键 词：蜂窝网络 流量预测 图神经网络 空间聚合 

分 类 号：TN919[电子电信—通信与信息系统]