面向流量预测的交通智脑关键技术研究

In order to improve the quality of traffic large-scale data forecast and analysis in wide application of traffic congestion and road safety, public information services and other areas, this project takes the multi-source heterogeneous data in the transportation field as the research object, takes the in-depth learning and the knowledge map as the main technical means, and studies the construction technology of traffic knowledge map based on multi-source data fusion for the characteristics of traffic data: complexity, diversity and fragmentation. In order to solve the problems that the current traffic flow forecasting model relies heavily on experts’ experience, lack of utilization of large data and limited ability of deep causal reasoning, this project studies data-driven, no-intervention (end-to-end), humanized intelligence models and short- time traffic flow forecasting model based on the in-depth learning. In addition, this project studies the traffic brain coordination model possessing the sensing capability of real-time scene and environment based on multi-source data fusion, for the deficient responding capability of current flow forecasting method in the application of real-time, dynamic and unexpected events. Finally, for the typical multi - source and heterogeneous traffic large data, traffic intelligent brain prototype system has been studied based on the large data calculation framework. The implementation of the project will meet the calculation and storage requirements where massive multi-source traffic data are used in applications, and improve the scalability of system forecast model, the efficiency and accuracy of traffic flow forecasting.

为提高广泛应用于交通拥堵、道路安全以及公共信息服务等领域交通大数据预测分析的质量，本项目以交通领域多源异构大数据为研究对象，以深度学习和知识图谱为主要技术手段，开展面向流量预测的交通智脑关键技术研究。针对交通数据复杂性、多样性、碎片化等特征，研究基于多源数据融合的交通知识图谱构建技术；针对目前交通流量预测模型严重依赖专家经验、交通大数据利用不足、深层因果推理能力有限等问题，研究数据驱动、无人干预（端到端）、类人化智能模型以及基于深度学习的短时交通流量预测模型；针对当前流量预测方法在应对实时、动态、突发事件响应能力的不足，研究基于多源多模态融合的具备实时场景和环境感知能力的交通智脑协同模型；针对典型的多源异构交通大数据研究基于大数据计算框架的交通智脑原型系统。项目的实施将满足大规模多源交通数据在实际应用中的计算与存储需求，提高系统预测模型的可扩展性以及交通流量预测的高效性与精准性。

交通拥堵已经成为制约城市经济和社会发展的瓶颈，在城市发展过程中的短板效应日益明显。目前仅仅增加交通供给只能在短期内缓解交通拥堵问题，而更多的需要通过有效合理的技术手段来实现。因此，开展面向流量预测的交通智脑关键技术研究尤为必要。.本项目以交通领域多源异构大数据为研究对象，以深度学习为主要技术手段，开展了面向流量预测的交通智脑关键技术研究。针对交通数据复杂性、多样性、碎片化等特征，研究了基于多源数据融合的交通知识图谱构建技术；针对目前交通流量预测模型严重依赖专家经验、交通大数据利用不足等问题，构建了数据驱动的基于深度学习的短时交通流量预测模型；针对当前流量预测方法在应对实时、动态、突发事件响应能力的不足，构建了基于多源多模态融合的具备实时场景和环境感知能力的交通智脑协同模型；针对典型的多源异构交通大数据实现了基于大数据计算框架的交通智脑原型系统。.本项目的实现能满足大规模多源交通数据在实际应用中的计算与存储需求，提高系统预测模型的可扩展性以及交通流量预测的高效性与精准性，为交通拥堵、道路安全以及公共信息服务等领域交通大数据预测分析提供有效的服务。项目已按既定计划执行完成。