The invention relates to a cross-city traffic flow prediction method based on a pre-training model, and belongs to the technical field of information, and the method comprises the steps: carrying out the slicing of a long traffic flow sequence of a data source city, taking the long traffic flow sequence as the input of a pre-training stage, and carrying out the training through employing a mask automatic coding strategy in the pre-training stage; the method comprises the following steps: pre-training a traffic flow sequence of a data source city to generate segment-level representation, and learning a dependency relationship among nodes of the data source city from the segment-level representation to obtain a discrete dependency graph containing road conditions of the data source city; and inputting the traffic flow sequence in the target city into the space-time pre-training model, and obtaining a final prediction result in combination with the segment-level representation obtained by data source city training. According to the method, the parallel computing advantage of Transform can be fully utilized, segment-level representation source cities can be effectively learned and generated from very long-term historical traffic data, and rich spatio-temporal information is provided for cities with scarce downstream data.

    本发明涉及一种基于预训练模型的跨城市交通流预测方法,属于信息技术领域,包括:对数据源城市的长交通流序列进行切片,并将其作为预训练阶段的输入,预训练阶段通过采用掩码自动编码策略进行训练;对数据源城市的交通流序列进行预训练以生成段级表示,并从段级表示中学习数据源城市节点之间的依赖关系,得到一个包含数据源城市道路状况的离散依赖图;将目标城市中的交通流序列输入到时空预训练模型中并结合数据源城市训练得到的段级表示,得到最终的预测结果。本发明能够充分利用Transformer的并行计算优势,有效地从非常长期的历史交通数据中学习和生成段级表示源城市,为下游数据稀缺城市提供丰富的时空信息。


    Access

    Download


    Export, share and cite



    Title :

    Cross-city traffic flow prediction method based on pre-training model


    Additional title:

    一种基于预训练模型的跨城市交通流预测方法


    Contributors:
    XIAN JIAJUN (author) / LI ZEQUAN (author) / YANG DAN (author) / LIN ZHIZHE (author) / ZHOU TENG (author)

    Publication date :

    2024-09-03


    Type of media :

    Patent


    Type of material :

    Electronic Resource


    Language :

    Chinese


    Classification :

    IPC:    G08G Anlagen zur Steuerung, Regelung oder Überwachung des Verkehrs , TRAFFIC CONTROL SYSTEMS / G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen



    Traffic flow prediction model training method, traffic flow prediction method, device and equipment

    GU JUNHUA / CHEN CHEN / LEI WEI et al. | European Patent Office | 2023

    Free access

    Model training method, traffic flow prediction method and device

    BING JUNXIANG | European Patent Office | 2024

    Free access

    Cross-city traffic flow prediction federal transfer learning algorithm based on privacy protection

    YUAN XIAOMING / LUO ZHENYU / ZHANG WENYUAN et al. | European Patent Office | 2024

    Free access

    Federated Transfer Learning for Privacy-Preserved Cross-City Traffic Flow Prediction

    Yuan, Xiaoming / Luo, Zhenyu / Zhang, Ning et al. | IEEE | 2025


    Training method and device of flow prediction model, and traffic flow prediction method and device

    HUANG CHAO / WANG MINGKAIQI / GUO RUNXIA et al. | European Patent Office | 2025

    Free access