考虑纵向时空特性的双向交通流组合预测研究  

Research on Bidirectional Traffic Flow Combination Prediction Considering Longitudinal Spatiotemporal Characteristics

作　　者：武月 侯越[1] WU Yue;HOU Yue(School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州交通大学电子与信息工程学院,兰州1730070

出　　处：《兰州交通大学学报》2024年第3期39-48,共10页Journal of Lanzhou Jiaotong University

基　　金：国家自然科学基金(62063014);甘肃省自然科学基金(22JR5RA365)。

摘　　要：针对现有研究方法未充分考虑现实路网双向交通流的纵向时空特性,道路方向辅助信息特征再组织性不足的问题,提出一种考虑纵向时空特性的双向交通流组合预测模型。采用卷积深度信念网络预训练机制细粒化提取上下游纵向路段间的空间特性,利用双向长短时记忆网络提取双向交通流时序特性,通过二者组合的方式完成纵向时空特性的深层次挖掘,并在此基础上,结合改进的双向注意力机制对模型提取的特征进行方向性自组织规划,从而实现针对不同方向性类别的自适应权重分配过程。实验结果表明:所提模型的平均绝对百分比误差和均方根误差相较于未考虑上下游路段影响的模型的分别降低了3.01%、5.57%,相较于考虑上下游路段影响但未改进注意力机制的模型的分别降低了0.29%、4.87%。A model for combined prediction of bidirectional traffic flow is proposed to address the issues of insufficient consideration of the longitudinal spatiotemporal characteristics of real road networks and inadequate reorganization of directional auxiliary information features in existing research methods.This model adpots a pre-training mechanism based on convolutional deep belief networks to finely extract spatial characteristics between upstream and down-stream longitudinal segments,and employs bidirectional long short-term memory networks to capture the temporal characteristics of bidirectional traffic flow.By combining the two,a deep-level exploration of the longitudinal spatio-temporal characteristics is achieved.Furthermore,an improved bidirectional attention mechanism is integrated into the model to perform directional self-organized feature extraction,enabling adaptive weight allocation for different directional cat-egories.The experimental results indicate that compared to the model that did not consider the influence of upstream and downstream sections,the average absolute percentage error and root mean square error of the proposed model decreased by 3.01%and 5.57%,respectively.Compared to the model that considered the influence of upstream and downstream sections but did not improve the attention mechanism,the respective reductions were 0.29%and 4.87%.

关 键 词：智能交通 双向交通流预测 深度学习 纵向时空特性 双向注意力机制 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]