基于车牌数据和LSTM的动态交通分配研究  

Research on Dynamic Traffic Assignment Based on License Plate Data and LSTM

作　　者：张南[1] 谢薇 ZHANG Nan;XIE Wei(School of Transportation and Logistics,Southwest Jiaotong University,Chengdu 611756,China)

机构地区：[1]西南交通大学,交通运输与物流学院,四川成都611756

出　　处：《综合运输》2022年第8期92-99,共8页China Transportation Review

基　　金：国家自然科学基金项目(61873216)。

摘　　要：动态交通分配是交通控制与诱导的重要理论基础,传统的数学解析模型和交通仿真模型存在求解困难和参数标定繁琐等问题,在大规模实例路网上应用具有一定局限性。车牌数据具有样本量大、覆盖面广、精度高等优势,本文选取15min作为统计时间间隔从车牌数据中获取动态OD和路段转弯流量,在确定OD矩阵在路网的最大影响时段数后,作为LSTM的输入输出数据建立动态交通分配模型。以某市核心区域为例,选取历史数据构造训练集,并用测试集评估训练好的模型。结果表明模型分配速度较快,分配结果与实际路段流量相比,不论是同时段还是同路段,分配值和观测值的拟合度均较高,且整体误差较小,说明了模型的可行性和有效性。Dynamic traffic assignment is an important theoretical basis for traffic control and guidance.Traditional mathematical analytical models and traffic simulation models have problems such as solution difficulty and cumbersome parameter calibration,which have certain limitations in the application of largescale example road networks.License plate data has the advantages of large sample size,wide coverage and high accuracy.This paper selects 15min as the statistical time interval to obtain dynamic OD and road section turning flow from the license plate data.After determining the maximum influencing period of OD matrix in the road network,it is used as the input and output data of LSTM to establish the dynamic traffic allocation model.Taking the core area of a city as an example,the historical data is selected to construct the training set,and the test set is used to evaluate the trained model.The results show that the model has a faster assignment speed.Compared with the actual section flow value,the fitting degree of both the test set results and the real data is higher in the case of same time intervals or same sections.And the overall error is small,which shows the feasibility and effectiveness of the model.

关 键 词：城市交通 动态交通分配 长短时记忆网络 车牌数据 改进的Dijkstra算法 

分 类 号：U491.1[交通运输工程—交通运输规划与管理]