基于机器学习的交通流预测方法综述

A Review of Machine Learning-based Traffic Flow Prediction Methods

摘要：通过文献梳理、专家访谈和试验场景构建等方法，分析了道路指定断面和区域路网宏观交通流预测的国内外研究现状和发展趋势，归纳了局部断面交通流预测方法，包括传统机器学习、递归神经网络和混合模型，分析了卷积神经网络、图神经网络和融合多因素网络的特点，阐述了方法的原理、优势、局限性和应用场景，总结了现有场景交通数据集类别，从采样周期与采集方式角度归纳了国内外主流交通数据集。分析结果表明：递归神经网络可以有效获取交通数据的历史规律，但存在梯度爆炸、计算复杂度高、长时预测准确度不佳等问题；图神经络针对路网拓扑连接关系引入了图结构，在考虑路网和交通流数据的时空相关性上具有明显优势；融合多因素网络充分考虑天气、道路、事故等内外部因素的影响，有效提升了交通流预测的实时性和鲁棒性；由于交通数据采集困难、外部因素影响难以量化、机器学习方法可解释性差等原因，交通流预测方法的改进受到了限制；未来应从交通信息有效挖掘和图卷积方法完善两方面入手，拓宽图结构在交通领域的应用和考虑非常态交通场景，进一步揭示交通数据的内在规律，开发更准确、高效的交通流预测方法，推动交通流预测在工业界的落地应用。

Abstract: Through literature review, expert interviews, experimental scenario construction, etc, this paper analyzes the current research status and development trends of macroscopic traffic flow prediction of specific road sections and regional road networks at home and abroad. Also, traffic flow prediction methods of local sections are summarized, including traditional machine learning, recurrent neural networks, and hybrid models. The characteristics of convolutional neural network, graph neural network, and multi-factor network are analyzed, and the principles, advantages, limitations, and application scenarios of these methods are expounded. Furthermore, the paper concludes the existing categories of traffic datasets and the mainstream traffic datasets at home and abroad on a basis of sampling period and acquisition methods. The findings illuminate that recurrent neural network can effectively capture operation rules of traffic data but is accompanied with limitation such as gradient explosion, high computational complexity, and poor long-term prediction accuracy. Graph neural network, by considering the topological connectivity of road networks, exhibits significant merits in capturing the spatiotemporal correlations between road networks and traffic flow data. As for multi-factor network，it takes into account the impacts of external factors such as weather, road conditions, and accidents, thereby enhancing the real-time performance and robustness of traffic flow prediction. However, the improvement of traffic flow prediction methods is constrained owing to challenges in data acquisition, quantifying external influences, and poor interpretability of machine learning methods. Therefore, future research should conducted in terms of effective acquisition of traffic information and improvement of graph convolution methods. Expanding the application of graph structure in transportation and exploring non-stationary traffic scenarios are necessary to further reveal the intrinsic laws of traffic data. Meanwhile, traffic flow prediction methods should be more accurate and efficient to promote its practical applications in the industry.

关键词：智能交通系统；交通流预测；机器学习；图卷积网络；混合模型；交通数据集

Keywords: intelligent transportation systems (ITS); traffic flow prediction; machine learning; graph convolutional networks; hybrid models; traffic datasets.

来源：Journal of Traffic and Transportation Engineering

发表时间：2023-05-03

检索：徐小焱

翻译：蒲婷

一审：翟雅婷

二审：彭莉

三审： 罗玲娟

上传发布： 姜浩