基于LSTM-Attention的高速公路短时速度预测方法研究  

Short-Term Prediction Method of Expressway Speed Based on LSTM-Attention

作　　者：姚进强 孙超 孟少寅 陈天怡 张永捷 张子彦 YAO Jinqiang;SUN Chao;MENG Shaoyin;CHEN Tianyi;ZHANG Yongjie;ZHANG Ziyan(Intelligent Transportation Research Branch of Zhejiang Communications Investment Group Co.,Ltd.,Hangzhou 310030,China;Shenzhen Urban Transport Planning Center Co.,Ltd.,Shenzhen 518057,China)

机构地区：[1]浙江省交通投资集团有限公司智慧交通研究分公司,杭州310030 [2]深圳市城市交通规划设计研究中心股份有限公司,深圳518057

出　　处：《交通与运输》2025年第1期55-60,共6页Traffic & Transportation

摘　　要：作为智能交通系统的重要组成部分,高速公路短时速度预测可为交通管理者提供动态调控依据,以提高城市交通系统运行效率。针对G92杭州湾环线高速公路ETC门架数据,进行分路段细化拆分与10 min级颗粒度归集;为解决统计学模型对波动性数据特征提取不足与RNN等模型存在的梯度消失问题,提出一种基于长短期记忆网络(LSTM)算法模型与Attention机制相结合的高速公路短时速度预测模型,其中,LSTM模型具有在捕获时间序列数据长期依赖关系捕捉和数据特征提取的优势,Attention机制的引入又可使时间序列在数据处理时更加注重关键信息,从而实现高速公路常规环境下短时通行状况的有效预测。As an important part of intelligent transportation system,highway section speed prediction can provide dynamic regulation basis for traffic managers and improve the operational efficiency of urban transportation system.In this paper,for ETC gantrydata of G92 Hangzhou Bay Ring Highway,sub-section refinement splitting and 10-minute granularity attribution is carried out.In order to solve the problem of insufficient feature extraction of fluctuating data by statistical model and gradient disappearance of RNN and other models,this paper proposes a highway short-term speed prediction model based on the combination of Long Short-Term Memory Network(LSTM)algorithmic model and Attention mechanism.The LSTM model has the advantage of capturing long-term dependency relationship capture and data feature extraction in capturing time series data,and the introduction of Attention mechanism can also be used to predict the speed of the highway.The introduction of LSTM-Attention model can make the prediction process more focused on key information when dealing with long time series data,so as to achieving effective prediction of traffic conditions in the conventional environment of highways.

关 键 词：高速公路 ETC门架数据 路段速度短时预测 LSTM-Attention模型 

分 类 号：U491[交通运输工程—交通运输规划与管理]