基于高速门架数据和深度学习的通行时间分布预测研究  

Study on travel time distribution prediction based on high-speed gantry data and deep learning

作　　者：蒋俊杰 JIANG Junjie(Guizhou Communications Polytechnic University,Guiyang 551400,China)

机构地区：[1]贵州交通职业大学,贵州贵阳551400

出　　处：《无线互联科技》2025年第5期81-84,89,共5页Wireless Internet Science and Technology

基　　金：贵州交通职业大学院级科研项目;项目名称:面向高速道路通行时间分布预测模型(2024ZR01);项目编号:2024YB03ZK。

摘　　要：高速公路交通流量的复杂性和波动性对通行效率与安全性提出了更高的要求。为精确预测高速公路通行时间分布,文章结合贵阳至黄果树高速公路段连续门架系统采集的交通流数据,提出了一种基于注意力机制的长短期记忆网络(Long Short-Term Memory-Attention, LSTM-Attention)组合模型。该模型通过LSTM捕捉时间序列中的长期与短期依赖关系,利用注意力机制动态聚焦于关键时间步,显著提升了通行时间分布的预测精度。研究结果表明,LSTM-Attention模型在节假日与非节假日2种场景下均表现出较高的预测精度,能够有效捕捉交通流量的时变特性和波动模式,尤其在节假日期间复杂交通场景下表现出显著优势。相比传统模型,LSTM-Attention模型在平均绝对误差(Mean Absolute Error, MAE)和均方误差(Mean Squared Error, MSE)上均实现了显著优化,为动态路径规划、交通管理优化和应急响应提供了重要的技术支持。文章研究成果不仅验证了深度学习技术在交通预测中的有效性,也为复杂交通场景的通行时间分布预测提供了新的研究思路。The complexity and volatility of highway traffic flow impose higher demands on traffic efficiency and safety.To accurately predict the distribution of highway travel times,this paper proposes a hybrid model based on an Attention Mechanism and Long Short-Term Memory Network(LSTM-Attention),utilizing traffic flow data collected from the continuous gantry system on the Guiyang-Huangguoshu highway segment.The model leverages LSTM to capture long-term and short-term dependencies in time series data,while the attention mechanism dynamically focuses on critical time steps,significantly enhancing the accuracy of travel time distribution predictions.The results demonstrate that the LSTM-Attention model achieves high prediction accuracy under both holiday and non-holiday scenarios,effectively capturing the time-varying characteristics and fluctuation patterns of traffic flow.Notably,it exhibits significant advantages in complex traffic conditions during holidays.Compared with traditional models,the LSTM-Attention model achieves substantial improvements in Mean Absolute Error(MAE)and Mean Squared Error(MSE),providing essential technical support for dynamic route planning,traffic management optimization,and emergency response.The findings not only validate the effectiveness of deep learning techniques in traffic prediction but also offer a novel research approach for predicting travel time distributions in complex traffic scenarios.

关 键 词：高速公路通行时间 深度学习 注意力机制 LSTM 交通流量预测 

分 类 号：U491.1[交通运输工程—交通运输规划与管理]