市政技术2026，Vol.44Issue(9)：132-137,6.DOI:10.19922/j.1009-7767.2026.09.132
基于非对称时空注意力网络的交通拥堵预测与疏散决策
Traffic Congestion Prediction and Evacuation Decision-making Based on Asymmetric Spatiotemporal Attention Network
阮春喜 1甘锐 2邓志鹏2
作者信息
- 1. 广州市道路事务中心,广东 广州 510635
- 2. 广州信息投资有限公司,广东 广州 510627
- 折叠
摘要
Abstract
Aiming at the problem of lag and misjudgment in traditional prediction models caused by the asymmetric evolution characteristics of urban traffic congestion,namely rapid formation and slow dissipation,this paper proposes an asymmetric spatiotemporal collaborative attention architecture(AST-Former)that integrates physical laws and deep learning.Breaking the paradigm of partial improvement within existing symmetric architectures,the architecture formalizes the characteristic of"rapid formation and slow dissipation"as a direction-aware factor,and designs an asymmetric modulation function to dynamically adjust spatiotemporal attention weights.This endows the model with the inherent capability to rapidly focus on congestion formation signals and maintain persistent memory of dissipation states.As a high-precision perception engine,the architecture is further deeply integrated with an expert rule system to form a closed-loop intelligent decision support system with"perception-decision"integration functions.Experimen-tal results based on large-scale real traffic data demonstrate that the AST-Former model significantly outperforms mainstream baseline models in prediction accuracy and state classification accuracy.The research findings can pro-vide innovative solutions for agile management and active diversion of urban traffic.关键词
非对称注意力/交通拥堵预测/时空协同/深度学习/智能决策系统Key words
asymmetric attention/traffic congestion prediction/spatiotemporal collaboration/deep learning/intel-ligent decision system分类
交通工程引用本文复制引用
阮春喜,甘锐,邓志鹏..基于非对称时空注意力网络的交通拥堵预测与疏散决策[J].市政技术,2026,44(9):132-137,6.
