PDF(2972 KB)
PDF(2972 KB)
PDF(2972 KB)
基于自适应时空同步建模的交通流预测
Adaptive Spatio⁃temporal Synchronous Modeling Based Traffic Flow Prediction
为准确捕获路网中不同交通节点之间的时空关联关系,提出一种基于自适应时空同步建模的交通流预测方法。首先,构建全局节点嵌入和不同子图的偏置生成多个既相互关联又有一定区别的时空子图,将不同的时空子图进行拼接生成静态自适应时空图,从不同的维度上描述路网中不同节点间的时空关联关系。其次,为了更好地建模不同交通节点间动态变化的时空关联关系,设计了一种新的动态自适应时空图构建方法,在有效描述不同交通节点间动态时空关联关系的同时,降低了动态时空图的计算复杂度。最后,在3个来自真实路网的公开数据集上进行测试,测试结果表明:与LSTM、DCRNN、STGCN、ASTGCN、GWN、STSGCN、STFGNN、STGODE、S2TAT等9种基线方法相比,所提方法具有更高的预测精度。在数据集PEMS08上,与最优基线方法S2TAT相比,该方法的平均绝对误差e MAE、平均绝对百分比误差e MAPE和均方根误差e RMSE分别降低了8.65% 、9.25%和6.04%。
In order to accurately capture the spatio-temporal correlation between different traffic nodes in the road network, a traffic flow prediction method based on adaptive spatio-temporal synchronization modeling was proposed. Firstly, the global node embedding and the bias of different sub-graphs were constructed to generate multiple sub-spatiotemporal graphs that are both related and different, and the different sub-spatiotemporal graphs were spliced to generate a static adaptive spatio-temporal graph, which describes the spatio-temporal correlation between different nodes in the road network from different dimensions. In addition, in order to better model the dynamic spatio-temporal relationship between different nodes, a new dynamic adaptive spatio-temporal synchronization graph construction method was designed, which can effectively describe the dynamic spatio-temporal relationship between different traffic nodes and reduce the computational complexity of the dynamic spatio-temporal graph. Finally, three public datasets derived from a real-world road network, were utilized for testing. The results of the experiments demonstrated that: when compared with nine baseline methods, including LSTM, DCRNN, STGCN, ASTGCN, GWN, STSGCN, STFGNN, STGODE, and S2TAT, the proposed method exhibited superior prediction accuracy. Specifically, on the PEMS08 dataset, in comparison with the optimal baseline method S2TAT, the proposed method achieved reductions of 8.65%, 9.25% and 6.04% in mean absolute error(MAE) e MAE, mean absolute percentage error(MAPE) e MAPE and root mean square error(RMSE) e RMSE, respectively.
智能交通系统 / 交通流量预测 / 图神经网络 / 自适应时空图 / 时空同步建模 / 深度学习
intelligent transportation systems / traffic flow prediction / graph neural network / adaptive spatio-temporal graph / spatio-temporal synchronous modeling / deep learning
| 1 |
|
| 2 |
|
| 3 |
吴鹏,叶宝林,吴维敏, 等. 基于改进混沌粒子群算法的交通信号控制[J]. 计量学报, 2024, 45(12): 1876-1884.
|
| 4 |
|
| 5 |
|
| 6 |
|
| 7 |
荣斌, 武志昊, 刘晓辉,等. 基于时空多图卷积网络的交通站点流量预测[J]. 计算机工程, 2020, 46(5): 26-33.
|
| 8 |
叶宝林, 戴本岙, 张鸣剑, 等. 基于图卷积网络的交通流预测方法综述[J]. 南京信息工程大学学报, 2024, 16(3): 291-310.
|
| 9 |
闫旭, 范晓亮, 郑传潘, 等. 基于图卷积神经网络的城市交通态势预测算法[J]. 浙江大学学报 (工学版), 2020, 54(6): 1147-1155.
|
| 10 |
|
| 11 |
|
| 12 |
|
| 13 |
|
| 14 |
|
| 15 |
|
| 16 |
|
| 17 |
|
| 18 |
|
| 19 |
|
| 20 |
|
/
| 〈 |
|
〉 |