The invention discloses a traffic flow prediction method and device based on an adaptive hypergraph convolutional neural network. According to the method, dynamic nodes and hyperedge relationships which cannot be captured through a predefined graph or a hypergraph in a traffic flow structure are captured by means of adaptive hypergraph learning; and then a hypergraph convolution method is used to capture the spatial features of the traffic flow in the adaptive hypergraph relationship, the obtained spatial features are used in a recurrent neural network structure to capture time features, and finally the predicted traffic flow is obtained. The implementation method is simple and convenient, the network structure is simple, the model can be obtained by directly training the collected data, and manual modeling of the graph network relationship is not needed; a spatial relationship is learned through a self-adaptive hypergraph, so that the robustness is relatively high, and the interference capability on node position data missing is relatively high; compared with other models of the same type, the method has a better prediction effect, the prediction precision can be improved, and the performance of the model is good.
本发明公开了一种基于自适应超图卷积神经网络的车流量预测方法及装置,该方法借助自适应超图学习捕捉交通流结构中无法通过预定义图或者超图捕捉到的动态节点和超边关系;然后使用超图卷积方法,捕捉车流量在该自适应超图关系中的空间特征,将得到的空间特征使用循环神经网络结构中,用以捕捉时间特征,最终得到预测车流量。本发明的实现方法较为简便,网络结构简单,可以直接通过采集的数据进行训练得到模型,不需要手动对图网络关系进行建模;通过自适应超图学习空间关系,有着较强的鲁棒性,对节点位置数据缺失有着较强的干扰能力;相较于其他同类型模型,有着更好的预测效果,有利于提高预测精度,模型的性能好。
Traffic flow prediction method and device based on adaptive hypergraph convolutional neural network
基于自适应超图卷积神经网络的车流量预测方法及装置
2023-03-21
Patent
Electronic Resource
Chinese
Periodically perceived space-time adaptive hypergraph neural network traffic prediction method
European Patent Office | 2024
|Transform-fused hypergraph neural network-based high-speed station traffic flow prediction method
European Patent Office | 2024
|European Patent Office | 2024
|Traffic prediction method and system based on adaptive hypergraph
European Patent Office | 2023
|