The invention discloses a traffic flow prediction method and system based on a graph convolution circulation network fused with Transform. The method comprises the following steps: constructing a traffic flow prediction model based on a node adaptive parameter learning (NAPL) module, a data adaptive graph generation (DAGG) module, a GRU module and a Transform module; the output of the NAPL module and the output of the DAGG module are connected with the GRU module, and the output of the GRU module is connected with the Transform module; parameter correction: updating a learning rate and learning rate attenuation in the traffic flow prediction model, the number of hidden layers of the GRU, the number of neurons of the hidden layers, the number of layers of Transformer and a parameter value combination of the number of multi-head attention based on an intelligent optimization algorithm; the optimization degree of the parameter value combination is judged based on prediction accuracy, and the optimal parameter value combination changes towards the optimal parameter value combination; and reconstructing a prediction model based on the optimal parameter value combination and completing training to realize flow prediction. The method can automatically deduce the interdependence relationship between different traffic sequences, achieves the dynamic capture of the whole-course time correlation, and improves the precision of traffic flow prediction.
本发明公开了基于融合Transformer的图卷积循环网络的交通流量预测方法及系统,包括:基于节点自适应参数学习(NAPL)模块、数据自适应图生成(DAGG)模块、GRU模块、Transformer模块构建交通流量预测模型;NAPL模块和DAGG模块的输出连接GRU模块、GRU模块的输出连接Transformer模块;参数修正,基于智能寻优算法更新交通流量预测模型中学习率以及学习率衰减、GRU的隐含层的层数、隐含层的神经元的个数、Transformer的层数和多头注意力的头数的参数取值组合,基于预测准确性判断参数取值组合的优选程度并趋向最佳参数取值组合变化;基于最佳参数取值组合重构预测模型并完成训练实现流量预测;本申请能自动推断不同交通序列之间的相互依赖关系,实现全程时间相关性的动态捕获,提高交通流量预测的精度。
Traffic flow prediction method and system based on image convolution circulation network fusing Transform
基于融合Transformer的图卷积循环网络的交通流量预测方法及系统
2024-05-03
Patent
Electronic Resource
Chinese
Periodic adaptive graph convolution circulation network traffic flow prediction method
European Patent Office | 2023
|Traffic flow prediction method fusing one-dimensional convolution and attribute enhancement unit
European Patent Office | 2023
|Traffic flow prediction method based on adaptive lightweight time convolution network
European Patent Office | 2024
|Traffic flow prediction method based on global diffusion convolution residual network
European Patent Office | 2020
|Traffic flow prediction method based on graph attention convolution network
European Patent Office | 2020
|