SHORT TERM TRAFFIC FLOW FORECASTING BASED ON TRASNSFORMER
Existing traffic flow prediction models fail to fully obtain the spatial dependence of the road network,ignore the influence of periodicity to traffic data,and lack the ability to model global time dependence.To solve the above problems,a dynamic diffusion convolution and a gated recurrent unit prediction model combined with Transformer is proposed.The dynamic diffusion convolutional networks and gated recurrent unit were used to model the near-term,daily cycle and weekly cycle time of traffic flow.The Transformer layer was used to obtain the global time dependency.The output of each component was weighted and fused to generate the prediction result.The experimental results show that compared with the benchmark model,this method can effectively reduce the prediction error and accurately predict the traffic evolution situation.
万方数据
维普
