无线电通信技术2025，Vol.51Issue(3)：530-537,8.DOI:10.3969/j.issn.1003-3114.2025.03.012
基于注意力机制的动态时空感知网络
Dynamic Spatio-Temporal Perception Network Based on Attention Mechanism
摘要
Abstract
Traffic flow prediction is an important task in multivariate time series prediction,and accurate traffic flow prediction can be used as auxiliary information for decision-making of traffic management department.Due to complex spatio-temporal and non-linear characteristics of traffic flow,it is still challenging to perform real-time efficient traffic flow prediction.Existing methods use parameter matrix as adjacency graph to learn spatial features,lacking intuitive explanations.A Dynamic Spatio-Temporal Perception Network(DST-PN)based on the attention mechanism is proposed in this paper to learn complex patterns of traffic flow.DSTPN first constructs a spatio-temporal perception graph using the Breccitis distance,and then models dynamic spatial correlation and temporal correlation between nodes using the attention mechanism with multi-scale convolutions.Experimental results on real datasets show that DSTPN outperforms state-of-the-art methods in terms of Root Mean Square Error(RMSE),Mean Absolute Error(MAE),and Mean Absolute Percentage Er-ror(MAPE).It demonstrates the effectiveness of the proposed DSTPN model on the traffic flow prediction task.关键词
时空数据库/注意力机制/机器学习/交通预测/人工智能Key words
spatio-temporal database/attention mechanism/machine learning/traffic prediction/artificial intelligence分类
信息技术与安全科学引用本文复制引用
张艺婷,蒲咏秋,刘钊勇..基于注意力机制的动态时空感知网络[J].无线电通信技术,2025,51(3):530-537,8.基金项目
国家自然科学基金(62272066) National Natural Science Foundation of China(62272066) （62272066）
