Metro Passenger Flow Prediction Model Based on WT Transformer Algorithm

In this paper, we propose a deep learning neural network model WTransformer for multi-feature fusion input, which combines the Transformer model with multi-feature fusion input to capture the spatial and temporal characteristics of traffic data. The model couples weather, temperature and other relevant external factors to improve the overall prediction performance and enhance the stability of the model. In this paper, the model performance is validated using Nanjing Metro Line 1 and total metro passenger traffic data combined with Valid Loss, MSE, MAE, etc. The comparison results show that compared with typical forecasting models (LSTM, TransAD, etc.), the improved model WTransformer has higher prediction accuracy and stronger ability to predict the peak of metro passenger traffic.