摘要
提出一种基于注意力机制的时空特征深度学习模型.通过卷积神经网络去学习行程过程中所花费的时间和距离,以及交通拥堵状态信息;然后,通过注意力机制从通道和空间两个角度去捕获影响行程中路段通行时间的异常信息.最后采用双层的长短时记忆网络去学习行程中的路段序列信息,并通过多任务的学习机制从路径和路段两个角度出发去预测路径通行时间.研究结果表明:提出的方法与DEEPTRAVEL模型相比,预测精度的平均绝对误差和平均绝对百分比误差分别提升了8.23%和20.79%.
This paper proposes a spatio-temporal characteristic deep learning model with attention mechanism.The model study travel time anomalies caused by traffic congestion changes during trips through convolutional neural networks,and use the attention mechanism to capture the information that affects the travel time and distance in the journey from two perspectives of channel and space.a two-layer long short time memory network that combined with the above information was used to learn the road sequence information in the trip,and finally a multi-task learning mechanism was used to predict the travel time from the perspective of the path and the road segment.The prediction accuracy of the method proposed in this paper increased by 8.23%and 20.79%in mean absolute error and mean absolute percentage error compared with DEEPTRAVEL model respectively.
作者
杨谊潇
邬群勇
YANG Yixiao;WU Qunyong(Key Lab of Spatial Data Mining and Information Sharing of Ministry of Education,National&Local Joint Engineering Research Center of Satellite Geospatial Information Technology,The Academy of Digital China(Fujian),Fuzhou University,Fuzhou,Fujian 350108,China)
出处
《福州大学学报（自然科学版）》
CAS
北大核心
2023年第3期340-346,共7页
Journal of Fuzhou University(Natural Science Edition)
基金
国家自然科学基金资助项目(42201500)
福建省科技计划引导资助项目(2021H0036)。
关键词
交通信息工程
行程时间预测
注意力机制
网约车订单数据
深度学习
transportation information engineer
estimated time of arrival
attentional mechanism
ride-hailing order data
deep learning
