基于自适应门控图神经网络的交通流预测

交通流预测是智能交通系统中的重要组成部分，由于交通数据的复杂性，长期而又准确的交通流预测一直是时间序列预测中最具挑战性的任务之一。近年来，研究人员将基于图神经网络的时空图建模方法应用于交通流预测任务，并取得了良好的预测性能。然而，现有的图建模方法仅通过预定义的邻接结构反映道路网络中的空间依赖关系，忽略了各节点之间的序列关联关系对预测的重要性。针对这一局限性，提出了一种自适应门控图神经网络（Ada-GGNN），其核心为通过空间传递模块同时捕获道路网络的空间结构及自适应的时序相关性，并通过门控机制学习节点上的时间序列特征。在两个真实交通网络数据集PeMSD7和Los-loop上的实验结果证明了该模型具有更优越的性能。

Traffic flow prediction based on adaptive gated graph neural network

Traffic flow prediction is an important part of intelligent transportation system. Due to the complexity of traffic data, long-term and accurate traffic flow prediction has always been one of the most challenging tasks in time series forecasting. In recent years, researchers have applied spatial-temporal graph modeling methods based on graph neural networks to traffic flow prediction tasks which achieved good prediction performance. However, existing graph modeling methods only reflect the spatial dependence in road networks through predefined adjacency structures, ignoring the importance of time-series correlation between nodes for prediction. Aiming at this limitation, this paper proposed an adaptive gated graph neural network(Ada-GGNN), the core of which was to simultaneously capture the spatial structure of the road network and adaptive time series correlation through the spatial passing module, and learned time-series features on nodes through the gating mechanism. The experimental results on two real-world traffic network datasets PeMSD7 and Los-loop show that the model has better performance.