车联网大数据驱动下融合深度学习与交通流模型的快速路多目标协调预测控制方法

City expressway carries majority of the daily urban traffic demands. The traffic flow prediction based expressway proactive control strategy has been proven to be an efficient way of improving the efficiency, safety and pollution emissions performance of the urban expressway. However, most current expressway proactive control strategy was designed based on traditional macroscopic traffic flow data collecting and analysis, to which their control performance is limited. This project planned to: take advantage of the macro and microscopic traffic flow data enriched by the connected vehicle environment, and reveal the impact mechanism of the expressway control method (ramp metering, variable speed limit) to the traffic flow. Establish the connected vehicle enabled big data driven traffic flow model with deep learning methodology to enhance the accuracy of predicting real-time expressway traffic flow dynamics; base on the model prediction result and the online collected traffic condition variables, design a novel multi-objective proactive expressway control strategy with prediction-optimization-control-feedback framework to coordinate the ramp metering and variable speed limit control for optimized expressway operation efficiency, safety and pollution emissions; a field data based microscopic traffic simulation will be conducted to validate the proposed prediction model and control strategy. This project aims at the basics of traffic flow theory, the application of proposed control strategy will help to further improve the efficiency of expressway network, mitigate traffic congestion, and reduce potential traffic accidents and pollution emissions.

快速路是城市日常交通需求的主要载体，而基于交通流状态预测的快速路预测控制方法可有效提高快速路的运行效率与安全水平。然而现有快速路控制方法局限于宏观交通流数据的采集与分析，难以全面揭示交通流状态变化机理，导致控制效果受限。本项目计划：利用人工智能深度学习算法充分发掘车联网技术带来的宏、微观交通流大数据，融合传统宏观交通流模型，深入分析控制影响下快速路交通流状态变化机理；设计以所建立交通流预测模型及路网状态在线分析模型为基础，利用预测-优化-控制-反馈架构协同入口匝道控制与可变限速控制实现对路网运行效率、安全水平、能耗排放等多目标实时同步定量预测优化的快速路预测控制方法；基于路网实测数据建立微观交通仿真平台，检验所建立预测模型与控制方法的可行性与有效性。项目关键科学问题瞄准交通流理论这一学科基础研究领域，其应用可进一步提高快速路运行效率，缓解交通拥堵，减少交通安全隐患与能耗排放。

作为城市发展主动脉的快速路无疑承担着巨大的交通压力，因此智能化交通主动管理措施的研究，对于缓解交通拥堵、提升生态效益与降低事故风险具有至关重要的意义。然而，现有的快速路主动控制方法研究中存在交通流预测模型精确度不足与优化目标不够全面等问题。本项目结合车联网大数据信息与深度学习卷积神经网络算法，展开交通流模型优化与快速路多目标协调预测控制方法研究。本项目取得的主要研究成果包括： 1）完成了车联网微观大数据驱动下，融合深度学习算法的交通流控制状态预测机理研究与交通流模型建模,该模型可以依据数据潜在的时空交叉联系补齐路段车辆轨迹与交通流状态信息，强化城市路网的微观交通态势感知； 2）完善了基于交通流变化机理的快速路网络多目标主动控制方法设计，利用宏、微观交通流状态参数，进一步优化多手段网联反馈控制；3）针对所设计快速路主动控制方法进行仿真实现与有效性评估，所设计的控制方法对快速路路网的交通流效率、路网安全性能及路网能耗排放效率皆有大幅优化。本项目的研究成果进一步提升了交通流模型的精确性，为交通管理者提供多目标优化下的高性能交通控制解决方案。