融合双注意力机制的GNN多维时间序列预测  

Multivariate time series forecasting with a graph neural network and dual attention mechanism

作　　者：范航舟 梅红岩[1] 赵勤 张兴[1] 程耐 FAN Hangzhou;MEI Hongyan;ZHAO Qin;ZHANG Xing;CHENG Nai(School of Electronics&Information Engineering,Liaoning University of Technology,Jinzhou 121000,China)

机构地区：[1]辽宁工业大学电子与信息工程学院,辽宁锦州121000

出　　处：《智能系统学报》2024年第5期1277-1286,共10页CAAI Transactions on Intelligent Systems

基　　金：辽宁省教育厅科学研究项目(JZL202015404,LJKZ0625,JYTMS20230869).

摘　　要：针对现有多维时间序列数据(multivariate time series,MTS)预测中变量间依赖关系捕获能力不足和时间序列数据多通道信息利用不充分的问题,提出一种融合双注意力机制的多维时间序列预测模型(feature fusion and dual attention mechanism based GNN,FFDA-GNN)。该模型将图神经网络与空间注意力机制融合,用于增强多变量之间依赖关系捕获能力;利用并行的多层膨胀卷积和通道注意力机制,对时间序列数据进行多通道的特征提取,实现对时间序列数据多通道信息的充分利用,从而提升预测性能。在经济、电力、交通3个领域数据集上与基准模型进行对比实验,该模型预测精度优于其他基准方法,有良好的可行性。To address the issues of insufficient capture of dependency relationships between variables and inadequate utilization of multichannel temporal information in multivariate time series,a forecasting model fused with a dual attention mechanism(FFDA-GNN)is proposed.This model integrates a graph neural network with a spatial attention mechanism to enhance the capture of dependencies among multiple variables.Additionally,parallel multilayer inflation convolution and channel attention mechanisms are used for multi-channel feature extraction from temporal data.This approach fully utilizes multi-channel information and enhances forecasting performance.Comparative experiments conducted on datasets from the fields of economics,electric power,and traffic show that the proposed model achieves better prediction accuracy than other benchmark methods and demonstrates superior feasibility.

关 键 词：多维时序预测 图神经网络 注意力机制 特征融合 时间卷积网络 深度学习 卷积神经网络 时空特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]