AN IMPROVED TRANSFORMER RAIL TRANSIT SHORT-TERM PASSENGER FLOW FORECASTING METHOD CONSIDERING FUSION ATTENTION MECHANISM

In view of the high non-linearity and dynamic space-time dependence of short-term passenger flow, an improved Transformer rail transit short-term passenger flow forecasting method based on attention mechanism is proposed, which takes advantage of the attention mechanism and the advantages of encoders and decoders. The Transformer model was improved, and the inbound and outbound passenger flow was modeled separately, so that the decoder could integrate the inbound and outbound passenger flow. The correlation features between the spatiotemporal sequence data were further captured. This paper took the North railway station of Chengdu Rail Transit as an example. The experimental results show that compared with the other four prediction methods, the improved Transformer model has the best prediction effect.