基于AGRU-Trans融合模型的特长海底隧道交通量预测
Traffic volume prediction of extra-long subsea tunnel based on AGRU-Trans fusion model
黄欣 1谢文红 2陈耀鹏 1李翔 3张素磊4
作者信息
- 1. 青岛理工大学 土木工程学院,青岛 266525
- 2. 台州市杭绍台高速公路有限公司,台州 318000
- 3. 青岛国信建设投资有限公司,青岛 266000
- 4. 青岛理工大学 土木工程学院,青岛 266525;青建集团股份公司,青岛 266071
- 折叠
摘要
为了使海底隧道交通管控部门更好地掌握准确的交通数据,给出行者提供更好的交通引导,通过对循环神经网络和Transformer算法的研究,结合GRU与Transformer模型算法优点并加入自注意力机制,提出一种基于AGRU-Trans融合模型的海底隧道交通量预测模型.选取胶州湾海底隧道市南-黄岛交通运行数据,通过AGRU-Trans融合模型与3种基准模型对比发现,LSTM,Transformer,GRU模型的平均绝对误差(MAE)相比AGRU-Trans分别大了31.48%,67.54%,20.57%;均方根误差(RMSE)分别增长了35.63%,38.45%,32.02%.结果表明:AGRU-Trans融合模型的预测结果与真实数据贴合性最好,预测精度均高于基准模型,因此,基于此方法可为胶州湾海底隧道管理部门对车辆进行精细化诱导及管控提供理论参考.
Abstract
In order to help the subsea tunned traffic control department of master accurate traffic data and provide better traffic guidance to travelers,this paper studies recurrent neu-ral network(RNN)and Transformer algorithm,combines the advantages of GRU and Transformer model algorithm and adds self-attention mechanism.It finally proposes a traffic volume prediction model of subsea tunnel based on AGRU-Trans fusion model.It then se-lects the Shinan-Huangdao traffic operation data of Jiaozhou Bay subsea tunnel and compares the AGRU-Trans fusion model with three benchmark models.It is found that the mean ab-solute error(MAE)values of LSTM,Transformer and GRU models are respectively 31.48%,67.54%and 20.57%larger than that of AGRU-Trans.Root mean square error(RMSE)increased by 35.63%,38.45%and 32.02%respectively.The results show that the prediction results of AGRU-Trans fusion model fit the real data best,and the prediction accuracy is higher than that of the benchmark models.Therefore,it can be concluded that this model can provide theoretical reference for the management department of Jiaozhou Bay subsea tunnel to conduct refined vehicle guidance and control.
关键词
海底隧道/交通量预测/AGRU-Trans融合模型/循环神经网络/Transformer模型Key words
subsea tunnel/traffic volume prediction/AGRU-Trans fusion model/recurrent neural network(RNN)/Transformer model引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据