面向高速公路服务区流量预测的动态时空图神经网络

Dynamic Spatial-temporal Graph Neural Networks for Traffic Prediction in Expressway Service Area

准确地预测驶入高速公路服务区的车流量有助于提升服务区智能化管理的效率。由于流量数据自身的动态性和路网拓扑结构等因素的影响,流量数据中存在复杂的时空相关性。为捕获流量序列中复杂的时空相关性,实现准确地驶入服务区流量预测,提出一种动态时空图神经网络模型。该模型使用感知趋势性变化的多头注意力模块捕获流量数据在时间维度的动态性,通过空间动态图卷积模块捕获驶入服务区的车流量与路网中其他断面流量的动态相关性和空间异质性。基于时空位置嵌入表示,考虑了流量数据在时间维度的顺序性和空间维度的静态属性。模型基于编码器-解码器结构,通过反向传播实现端到端的训练。最后,基于真实高速路网数据的实验说明了提出方法的有效性。

Accurately predicting the traffick flow into the expressway service is helpful to improve the efficiency of intelligent management in the service area.Due to the dynamic nature of traffic data and the topology structure of road network,the traffic series has complex spatio-temporal correlations.In order to capture the complex spatio-temporal correlation in the traffic data and realize accurate traffic prediction,a dynamic spatial-temporal graph attention network model is proposed in this paper.The proposed model leverages a trend-aware self-attention module to capture the dynamics of traffic data in temporal dimension,and uses spatial dynamic graph module to capture the dynamic correlation and spatial heterogeneity be-tween traffic flows entering the service area and other sections in the road network.Based on the spatio-temporal positional embedding,the sequential nature of traffic data in the temporal dimension and the static properties in the spatial dimension are considered.The model is based on the encoder-decoder structure,and can be trained end-to-end via backpropagation.Finally,extensive experiments on real road network data demonstrate the effectiveness of the proposed method.