基于Wavelet-LSTM模型的高速公路入口短时交通流预测  

Short-term traffic flow prediction at highway entrances based on Wavelet-LSTM model

作　　者：刘兴国 夏传飞 王秀兰 冯镛 余聪 LIU Xingguo;XIA Chuanfei;WANG Xiulan;FENG Yong;YU Cong(School of Transportation and Logistics Engineering Shandong Jiaotong University,Shandong Jinan 250357 China;Linyi Transportation Law Enforcement Detachment,Shandong Linyi 276000 China;Shandong College of Highway Technician,Shandong Jinan 253020 China)

机构地区：[1]山东交通学院交通与物流工程学院,山东济南250357 [2]山东省临沂市交通运输执法支队,山东临沂276000 [3]山东公路技师学院,山东济南253020

出　　处：《山东交通科技》2023年第5期52-55,共4页

基　　金：国家社会科学基金,项目编号:19BJY173;山东省重点研发计划,项目编号:2021RZA02025;山东省交通科技计划,项目编号:2019B67,2020B50,2022B31。

摘　　要：为提高高速公路短时交通流预测的准确度,基于高速公路实时的收费数据,采用小波分解(wavelet decomposition,Wavelet)和长短时记忆(longshort-term memory,LSTM)相结合的方法,构建Wavelet-LSTM短时交通流组合预测模型,并与单一模型LSTM、随机森林(Randomforest,RF)及组合模型Wavelet-RF进行对比。结果表明,提出的组合预测模型具有更高的预测精度,且能更有效地把握高速公路交通流的变化,该模型预测准确度接近94%,平均绝对百分比误差为6.7%,比LSTM、RF、Wavelet-RF分别提高了11.41%、13.65%、1.73%。借助于更精确的交通流预测模型,可为智慧高速建设提供一定的助力。In order to improve the accuracy of highway shortterm traffic flow prediction,this paper uses a combination of wavelet decomposition(Wavelet)and long short term memory(LSTM)to construct a Wavelet-LSTM short-term traffic flow prediction model based on real-time highway toll data,and compares it with some single model LSTM,random forest(RF)and combined model Wavelet-RF.The results show that the combined prediction model proposed in this paper has higher prediction accuracy and is more effective in grasping the changes of highway traffic flow,with a prediction accuracy of nearly 94%and an average absolute percentage error of 6.7%,which is 11.41%,13.65%and1.73%higher than LSTM,RF and Wavelet-RF respectively,with the help of more accurate traffic flow prediction model,a certain amount of assistance can be provided to the construction of smart highway.

关 键 词：高速公路 短时交通流预测 小波分解 长短时记忆神经网络 

分 类 号：U491.1[交通运输工程—交通运输规划与管理]