基于时空依赖图神经网络的交通流量预测

Traffic Flow Prediction Based on Spatio-Temporal Dependent Graph Neural Network

摘要: 准确、实时的交通预测是智能交通系统的重要组成部分，对交通规划、交通管理、交通控制具有重要意义。然而，由于交通路网中存在复杂的时空依赖性，实现高度准确的交通流量预测是一项具有挑战性的任务。本文采用时空依赖图神经网络进行交通预测。首先，探究了交通流量、速度和密度之间的关系，依据格林希尔兹提出的关系式，通过速度数据得到了交通流量数据，为现有交通流量数据的稀缺性提供了解决方案。其次，本文采用基于图卷积网络和门控递归单元的模型，建模交通路网的时空依赖性，具体而言，图卷积网络可捕获道路网络的拓扑结构，建模空间依赖性；门控递归单元可捕获道路上交通数据的动态变化，建模时间依赖性。最后，本文采用自有数据集来评估本文方法，实验结果表明，与传统方法相比，本文采用的方法将预测误差降低了约6.3%~97.73%，并证明了模型在交通预测方面的优越性。

Abstract: Accurate and real-time traffic prediction is an important part of intelligent transportation system, which is of great significance for traffic planning, traffic management and traffic control. However, achieving highly accurate traffic flow forecasts is a challenging task due to the complex spatio-temporal dependencies in the transportation road network. In this paper, spatio-temporal dependent graph neural network is used for traffic prediction. First, the relationship between traffic flow, speed and density is explored, and based on the relational equation proposed by B. D. Greenshields, the traffic flow data is obtained from the speed data, which provides a solution to the scarcity of the existing traffic flow data. Second, this paper uses a model based on graph convolutional networks and gated recurrence units to model the spatio-temporal dependence of the traffic road network. Specifically, graph convolutional networks capture the topology of the road network and model the spatial dependence, while gated recurrence units capture the dynamic changes of traffic data on the road and are used to model the temporal dependence. Finally, this paper uses its own dataset to evaluate the proposed method. The experimental results show that the method used in this paper reduces the prediction error by about 6.3%~97.73% compared to the traditional method and demonstrates the superiority of the model in traffic prediction.