The invention discloses a hierarchical spatio-temporal feature enhancement framework for traffic flow prediction. The hierarchical spatio-temporal feature enhancement framework comprises a periodic embedding module (PEM), a normalization layer module (NormLayer), an adaptive parallel channel attention fusion module (AP-CAF), a hierarchical serial feature fusion enhancer (HS-FFE) and a multi-layer perceptron module (MLPLayer). According to the method, lightweight embedding processing can be carried out by adopting time periodicity, a time relationship is captured, a parallel channel attention fusion component captures different types of feature information, and a serial feature enhancement component improves understanding of a model on a complex relationship between features, so that the model can better capture an internal mode and correlation between the features; the multi-layer sensing module processes the features on different levels, and it is ensured that the model obtains the overall space-time features on different levels more comprehensively.
本发明公开了用于交通流量预测的分层时空特征增强框架,包括周期性嵌入模块(PEM)、归一化层模块(NormLayer)、自适应并行通道注意力融合模块(AP‑CAF)、层次化串行特征融合增强器(HS‑FFE)和多层感知机模块(MLPLayer)。本发明能够采用时间周期性进行轻量化嵌入处理,捕获时间关系,并行通道注意力融合组件捕捉不同类型的特征信息,串行特征增强组件提高模型对特征之间复杂关系的理解,使模型能够更好地捕捉特征之间的内在模式和相关性,多层感知模块在不同层次上对特征进行处理,确保模型在不同层次上更全面获取时空总体特征。
Hierarchical spatiotemporal feature enhancement framework for traffic flow prediction
用于交通流量预测的分层时空特征增强框架
2024-08-02
Patent
Electronic Resource
Chinese
TRAFFIC FLOW PREDICTION SYSTEM USING SPATIOTEMPORAL STOCHASTIC MODEL
European Patent Office | 2015
|Air Traffic Flow Prediction with Spatiotemporal Knowledge Distillation Network
DOAJ | 2024
|