Ji Anning, Ma Xintao
Department of Traffic Management, Jilin Police College, Changchun, China.
School of Management Science and Information Engineering, Jilin University of Finance and Economics, Changchun, China.
PLoS One. 2025 May 28;20(5):e0323787. doi: 10.1371/journal.pone.0323787. eCollection 2025.
Recently, accurate traffic flow prediction has become a significant part of intelligent transportation systems, which can not only satisfy citizens' travel need and life satisfaction, but also benefit urban traffic management and control. However, traffic forecasting remains highly challenging because of its complexity in both topology structure and time transformation. Inspired by the propagation idea of graph convolutional networks, we propose ripple-propagation-based attentive graph neural networks for traffic flow prediction (T-RippleGNN). Firstly, we adopt Ripple propagation to capture the topology structure of the traffic spatial model. Then, a GRU-based model is used to explore the traffic model through the timeline. Lastly, those two factors are combined and attention scores are assigned to differentiate their influences on the traffic flow prediction. Furthermore, we evaluate our approach with three real-world traffic datasets. The results show that our approach reduces the prediction errors by approximately 2.24%-62,93% compared with state-of-the-art baselines, and the effectiveness of T-RippleGNN in traffic forecasting is demonstrated.
近年来，精确的交通流预测已成为智能交通系统的重要组成部分，它不仅能满足市民的出行需求并提高生活满意度，还能有利于城市交通管理与控制。然而，交通流预测仍然极具挑战性，因为其在拓扑结构和时间变化方面都很复杂。受图卷积网络传播思想的启发，我们提出了基于涟漪传播的注意力图神经网络用于交通流预测（T-RippleGNN）。首先，我们采用涟漪传播来捕捉交通空间模型的拓扑结构。然后，使用基于门控循环单元（GRU）的模型通过时间线来探索交通模型。最后，将这两个因素结合起来并分配注意力分数，以区分它们对交通流预测的影响。此外，我们用三个真实世界的交通数据集对我们的方法进行了评估。结果表明，与最先进的基线方法相比，我们的方法将预测误差降低了约2.24% - 62.93%，证明了T-RippleGNN在交通流预测中的有效性。
