考虑交通事件影响的高速公路短时行程时间预测  

Freeway Short-term Travel Time Prediction Considering the Impact of Traffic Events

作　　者：潘杰 石京[1] PAN Jie;SHI Jing(Department of Civil Engineering,Tsinghua University,Beijing 100084,China)

机构地区：[1]清华大学土木工程系,北京100084

出　　处：《交通工程》2024年第8期45-53,74,共10页Journal of Transportation Engineering

基　　金：国家自然科学基金项目(51578319和51778340)资助。

摘　　要：本研究目的是基于历史行程数据并考虑交通事件,构建可提高高速公路短时行程时间预测精度的方法。基于深层机器学习理论,设计加权均方根相似(Weighted-RMSS)模型,利用经纬度将行程分段,考虑高速公路车辆时空流动性的时间传递,计算当前行程时间和历史案例行程时间的相似性,提高了行程时间预测精确度。在此基础上,结合交通事件数据建立交通事件影响矩阵,建立LGBM模型(Light Gradient Boosting Machine)用于短时行程时间预测,并利用广州高速公路平沙至机场南路段实测数据进行验证。研究结果表明,开发2个模型效果均优于传统KNN模型,且考虑了交通事件影响的LGBM模型的预测精度高于Weighted-RMSS模型,达到95.68%,比较不同未来预测时间得出预测5 min效果最佳,精度可达96.18%。本研究在短时行程时间预测上有显著的优越性,有助于为驾驶人提供准确的出行时间,有利于高速公路的交通管理。The purpose of this study is to build a method to improve the prediction accuracy of short-time travel time on highways considering traffic events based on historical travel data.Based on the deep machine learning theory,a Weighted-RMSS model is designed.The travel time is segmented by longitude and latitude,and the similarity between the current travel time and the historical case travel time is calculated by considering the time transfer of the spatial and temporal mobility of highway vehicles.On this basis,combined with the traffic event data,the influence matrix of traffic events was established,and the algorithm was improved to establish the LGBM model(Light Gradient Boosting Machine)for short time prediction.The measured data from Pingsha to Airport South section of Guangzhou Expressway are used.The average percentage error(M APE)and root mean square error(RMSE)are used as evaluation index of model prediction precision for validation.The results show that the two models developed in this study are better than the traditional KNN model,and the prediction accuracy of the LGBM model considering the influence of traffic events is higher than that of the Weighted-RMSS model,reaching 95.68%.Comparing different future prediction times,it is found that the prediction effect of 5 minutes is the best,and the accuracy can reach 96.18%.This study has significant advantages in short travel time prediction,which is helpful to provide accurate travel time for drivers and is beneficial to highway traffic management.

关 键 词：高速公路 行程时间预测 时空流动性 加权均方根相似模型 交通事件影响 LGBM模型 

分 类 号：U491[交通运输工程—交通运输规划与管理]