[1]孙大盟,欧阳安杰,何立明.基于改进STGCN深度学习框架的交通速度预测[J].计算机技术与发展,2024,34(11):133-139.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0225]
　SUN Da-meng,OUYANG An-jie,HE Li-ming.Traffic Speed Prediction Based on Improved STGCN Deep Learning Framework[J].,2024,34(11):133-139.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0225]
点击复制

基于改进STGCN深度学习框架的交通速度预测()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
34
期数:
2024年11期
页码:
133-139
栏目:
人工智能
出版日期:
2024-11-10

文章信息/Info

Title:
Traffic Speed Prediction Based on Improved STGCN Deep Learning Framework
文章编号:
1673-629X(2024)11-0133-07
作者:
孙大盟; 欧阳安杰; 何立明
长安大学 信息工程学院,陕西 西安 710064
Author(s):
SUN Da-meng; OUYANG An-jie; HE Li-ming
School of Information Engineering,Chang’an University,Xi’an 710064,China
关键词:
智慧交通系统; 交通速度预测; 图卷积网络; 位置注意机制; 时空相关性
Keywords:
smart transportation system;  traffic speed prediction;  graph convolutional network;  position attention mechanism;  spatial -temporal correlation
分类号:
TP31;U491.2
DOI:
10.20165/j.cnki.ISSN1673-629X.2024.0225
摘要:
实时准确的交通速度预测对于加快智慧交通建设和推动智能交通系统发展至关重要。 然而交通网络具有复杂的空间结构和动态随机的时变特征,致使现有预测方法无法准确捕捉其隐藏的时空相关性。 为了充分挖掘数据中隐藏的动态时空特性,并提高预测准确性,该文提出了一种基于 STGCN 框架的交通速度预测改进算法,即时空注意力图神经网络(STA-GNN)。 该算法采用可学习的位置注意力机制,有效聚合邻近节点信息,从而获取道路网络中的空间相关性。 同时,引入带有门控机制的以一维因果卷积网络为内核的时序卷积网络,来捕获时间序列中的时间相关性,并通过残差块连接来提高模型的泛化能力。 所提方法在 PeMSD7 数据集上进行了 15 分钟、30 分钟和 45 分钟的交通速度预测实验。 实验结果显示,该模型在 45 分钟预测任务中,均方根误差相较于 STGCN 模型降低了约 10. 2% 。 表明 STA-GNN 模型在中长期交通速度预测任务中表现更为出色。
Abstract:
Real- time and accurate traffic speed prediction is crucial for accelerating smart transportation development and advancing intelligent traffic systems. However,the complex spatial structure and dynamic stochastic temporal characteristics of traffic networks make it challenging for existing prediction methods to accurately capture their hidden spatiotemporal correlations. To fully exploit the dynamic spatiotemporal features hidden in data and enhance prediction accuracy,we propose an improved traffic speed prediction algorithm based on the STGCN framework, namely the Spatiotemporal Attention Graph Neural Network ( STA - GNN). This algorithm employs a learnable position attention mechanism to effectively aggregate information from neighboring nodes,thereby capturing spatial correlations within road networks. Simultaneously,it introduces a one-dimensional causal convolutional network with gate mechanisms as the kernel of the temporal convolutional network to capture time correlations in time series data. This is further enhanced through residual block connections to improve model generalization capability. The proposed approach was evaluated on the PeMSD7 dataset for traffic speed prediction experiments at 15 -minute,30 -minute,and 45 -minute intervals. Experimental results demonstrate that the proposed model achieves approximately 10. 2% reduction in root mean square error compared to the STGCN model in the 45-minute prediction task,indi-cating superior performance of the STA-GNN model in mid- to long-term traffic speed prediction tasks.

相似文献/References:

[1]赵嘉雨,段亚茹,何立明.基于 GRU-GCN-RDrop 模型的交通速度预测[J].计算机技术与发展,2023,33(04):120.[doi:10. 3969 / j. issn. 1673-629X. 2023. 04. 018]
　ZHAO Jia-yu,DUAN Ya-ru,HE Li-ming.Traffic Speed Prediction Based on GRU-GCN-RDrop Model[J].,2023,33(11):120.[doi:10. 3969 / j. issn. 1673-629X. 2023. 04. 018]

更新日期/Last Update: 2024-11-10