北京交通仿真建模的新技术探索与实践

Exploration and Practice of New Technologies for Transportation Simulation Modeling in Beijing

交通仿真是城市交通规划和综合治理不可或缺的技术支撑手段。面向超大城市交通发展需求,北京市建立了数据驱动和AI赋能的交通仿真平台技术架构,对交通仿真关键技术进行了重点攻关。基于深度学习技术构建北京交通计算图,解决了多源时空数据融合的难题,提高了城市交通网络运行状态判别精度;依据交通需求场景库提出了基于通用地理神经网络的出行分布预测模型,预测精度较传统方法大幅提高;创建了中国特色的微观交通行为库,构建了微观驾驶行为模型,更加贴近本地驾驶行为特征和规律;在动态交通分配模型中构建了瓶颈点排队模型,更加真实地模拟了过饱和交通流的形成与消散过程;研发了面向大规模网络交通分配的并行算法,大幅提高仿真运算效率。最后以北京城市副中心为仿真平台实证场景,预测并评估交通规划方案的实施效果。

Transportation simulation represents an indispensable technical support for urban transportation planning and comprehensive governance.To address the transportation development demand in a megacity,a data-driven and AI-enabled technical framework for transportation simulation platform has been established in Beijing,with a research focus on critical transportation simulation technologies.Based on deep learning technology,this paper presents the development of a big data driven Transportation Computational Graph,alternatively Beijing Computational Graph(BTCG),addressing the difficulty in multi-source spatiotemporal data fusion and improving the assessment accuracy of the operating state of urban transportation network.Using a Travel Pattern database,the paper proposes a travel distribution prediction model based on the Universal Geographic Neural Network(UGNN),substantially improving the prediction precision compared with traditional methods.In addition,a micro traffic behavior database with distinctive Chinese characteristics and a micro driving behavior model are developed,which are closer to the characteristics and rules of local driving behavior.Moreover,the paper presents a bottleneck queuing model associated with the dynamic traffic assignment model,more accurately simulating the formation and dissipation process of over-saturated traffic flow.A parallel algorithm is also developed for large-scale network traffic assignment,greatly improving simulation operation efficiency.Finally,Beijing Municipal Administrative Center is used as an empirical scenario for the simulation platform to predict and evaluate the implementation effect of the transportation planning scheme.