Congestion recognition is the prerequisite for traffic control and management, vehicle routing, and many other applications in intelligent transportation systems. Different types of roads with traffic facilities provide multi-source heterogeneous field traffic data, which contain the fundamental information and distinct features for congestion recognition. To exploit these traffic big data, in this paper, we propose a machine learning-based framework to tackle the congestion recognition problem. It can be divided in two parts, a digraph-based representation for hybrid urban traffic network and a Dirgraph Convolutional Neural Network (DGCN)-based learning model. At first, the representation incorporates the fundamental traffic variables with the correlation of different traffic flows, and partially decouples the global network topology from local traffic information. And then, to proceed with digraph-based samples, a new type of graph feature extraction method is introduced and the graph Fourier transform is defined accordingly. This distinguishes the proposed model from the conventional graph convolutional networks. Comprehensive experiments are conducted based on real traffic data. The results demonstrate the advantages of the proposed framework over the existing congestion recognition methods.

拥塞识别是交通控制和管理，车辆路线选择以及智能交通系统中许多其他应用程序的先决条件。具有交通设施的不同类型的道路提供了多源异构现场交通数据，其中包含基本信息和用于拥塞识别的独特功能。为了利用这些流量大数据，本文提出了一种基于机器学习的框架来解决拥塞识别问题。它可以分为两个部分，一个用于混合城市交通网络的基于图的表示形式，另一个基于Dirgraph卷积神经网络（DGCN）的学习模型。首先，该表示法将基本交通变量与不同交通流的相关性结合在一起，并将全球网络拓扑与本地流量信息部分分离。然后，为了处理基于图的样本，引入了一种新型的图特征提取方法，并据此定义了图傅里叶变换。这将建议的模型与传统的图卷积网络区分开来。根据实际流量数据进行综合实验。结果证明了所提出的框架优于现有的拥塞识别方法的优点。根据实际流量数据进行综合实验。结果证明了所提出的框架优于现有的拥塞识别方法的优点。根据实际流量数据进行综合实验。结果证明了所提出的框架优于现有的拥塞识别方法的优点。