数据驱动的城市路网短时交通流预测

Data-driven Short-term Traffic Flow Prediction in Urban Road Network

文中立足于大数据时代的城市交通背景,总结现有短时交通流预测的研究现状,内容涵盖统计学模型、机器学习模型、传统深度学习模型及新颖的图神经网络等预测方法.根据预测模式将现有研究划分为单节点交通流预测及网络级交通流预测两大类别.将前者进一步细分为考虑交通流时变特征的预测方法及考虑空间相关性的预测方法,将后者按照所使用的预测模型归纳为基于卷积神经网络的预测方法及基于图神经网络的预测方法,论述了图神经网络中涉及到的拓扑网络构建方法.总结了现阶段预测方法中的不足,指出了包括融合多维交通特征、考虑多源数据时空特征的协同预测,以及融合时空复杂网络与交通预测等七点未来研究的重点方向.

Based on the urban traffic background in the era of big data, this paper summarized the current research status of short-term traffic flow forecasting, including statistical models, machine learning models, traditional deep learning models and novel graph neural networks. According to the prediction model, the existing research was divided into two categories: single-node traffic flow prediction and network-level traffic flow prediction. The single-node traffic flow prediction was further subdivided into the prediction method considering the time-varying characteristics of traffic flow and the prediction method considering the spatial correlation. According to the used prediction models, the network-level traffic flow prediction was summarized into the prediction method based on convolution neural network and the prediction method based on graph neural network, and the topological network construction method involved in graph neural network was discussed. The shortcomings of forecasting methods are further summarized, and seven key directions of future research, such as integrating multidimensional traffic characteristics, collaborative forecasting considering spatio-temporal characteristics of multi-source data, and integrating spatio-temporal complex network with traffic forecasting, are pointed out.