The invention relates to a traffic flow prediction method based on a space-time bidirectional attention mechanism, and the method comprises the following steps: dividing a whole city into grids with a plurality of regions, and carrying out the definition of a flow variable; a traffic flow prediction network model with a Transform model as a basic framework is constructed, the traffic flow prediction network model comprises a convolution encoder used for extracting spatial-temporal characteristics and a spatial-temporal decoder used for traffic flow prediction, and the traffic flow Pout of the urban area in the l time periods after the T moment is predicted according to k time steps Din before the T moment extracted from the historical observation set; the convolutional encoder for extracting the spatial-temporal characteristics comprises a local CNN network, a multi-head local position attention layer, a full-connection feedforward network layer FC-FF layer and a residual connection Norm layer with layer normalization, and the encoder takes Din as input and outputs an encoding state Den.
本发明涉及一种基于时空双向注意力机制的交通流预测方法,包括下列步骤:将整个城市划分成成具有多个区域的网格,进行流量变量的定义;构建以Transformer模型作为基础架构的交通流预测网络模型,包括,用于提取时空特征的卷积编码器和用于交通流量预测的时空解码器,根据从历史观测集中提取T时刻前的k个时间步骤Din来预测T时刻后l个时间段的城市区域交通流流量Pout;用于提取时空特征的卷积编码器包括局部CNN网络、多头局部位置注意力层、全连接的前馈网络层FC‑FF层和具有层归一化的残差连接Norm层组成,编码器以Din作为输入,并输出编码状态Den。
Traffic flow prediction method based on space-time bidirectional attention mechanism
一种基于时空双向注意力机制的交通流预测方法
2024-09-13
Patent
Electronic Resource
Chinese
Traffic flow prediction method based on space-time attention mechanism
European Patent Office | 2024
|Traffic flow prediction method based on cyclic space-time attention mechanism
European Patent Office | 2024
|Traffic flow prediction method based on space-time decomposition and attention mechanism
European Patent Office | 2025
|Fusion attention mechanism bidirectional LSTM for short-term traffic flow prediction
Taylor & Francis Verlag | 2024
|European Patent Office | 2023
|