Figures (11)  Tables (7)

    • Figure 1. 

      Inbound passenger flow variation.

    • Figure 2. 

      Structure of the PSR-CNN-LSTM model.

    • Figure 3. 

      The process of phase space reconstruction.

    • Figure 4. 

      Autocorrelation coefficient.

    • Figure 5. 

      Embedding dimension.

    • Figure 6. 

      Phase space at different delay times.

    • Figure 7. 

      CNN-LSTM structure.

    • Figure 8. 

      LSTM internal structure diagram.

    • Figure 9. 

      Prediction of passenger flow for typical stations of the Shanghai metro.

    • Figure 10. 

      Comparison of error metrics and convergence time for models.

    • Figure 11. 

      Error metrics and convergence time comparison.

    • Card no Date Time Line and station Mode Cost (CNY) Type
      2201252167 04-01-15 19:20:33 Line 7 Changzhong Road Subway 4.0 Full fare
      2702155929 04-01-15 12:52:38 Songjiang Bus 43 Bus 1.0 Full fare
      2201252167 04-01-15 08:55:44 Line 1 Baoshan Highway Subway 3.0 Full fare
      … … … … … … …
      602141128 04-01-15 09:07:57 Songjiang Bus 43 Bus 0 Discount

      Table 1. 

      AFC raw data format.

    • Line Subway station Date Period serial number Station entry person
      1 Baoan Road 04-01-15 37 164
      1 Baoan Road 04-01-15 38 397
      1 … … … …
      1 Baoan Road 04-01-15 55 1,316

      Table 2. 

      Station entry data by 10-min time period.

    • Line Station Number MLES
      1 Bao'an Highway 1 0.025
      1 Caobao Road 2 0.026
      1 Changshu Road 3 0.035
      1 Fujin Road 4 0.022
      1 Gongfu Xincun 5 0.031
      1 Gongkang Road 6 0.030
      1 Shanghai Railway Station 7 0.040

      Table 3. 

      Maximum Lyapunov values of the passenger flow time series.

    • Parameter Value
      Learning rate 0.00283
      Convolutional kernels 35
      Hidden neurons 25
      Dropout rate 0.2
      Kernel size 6 × 6
      Pooling size 4 × 4

      Table 4. 

      Model parameters.

    • Time granularity RMSE R2 MAE
      10 min 19.93 0.99 13.31
      20 min 24.06 0.98 17.91
      30 min 44.09 0.98 28.69

      Table 5. 

      Comparison of prediction results for different time granularities.

    • Model RMSE R2 MAE Convergence
      time (s)      		 Parameters
      CNN 33.07 0.96 21.61 14 1669
      LSTM 39.03 0.94 25.12 20 5520
      CNN-LSTM 27.96 0.97 19.69 34 13084
      ResNet-LSTM-GCN 24.29 0.97 19.63 107 77660
      GCN-GRU 27.07 0.97 16.32 86 17921
      PSR-CNN-LSTM 19.93 0.99 13.31 33 12384

      Table 6. 

      Evaluation of different models.

    • Model RMSE R2 MAE Convergence
      time (s)      		 Parameters
      PSR-CNN 25.90 0.97 18.68 15 7664
      PSR-LSTM 25.47 0.97 15.99 17 6991
      CNN-LSTM (GWO) 24.90 0.97 19.67 38 15049
      PSR-CNN-LSTM GWO) 17.41 0.99 13.16 44 14248
      PSR-CNN-LSTM (SSA) 18.87 0.99 14.79 52 43916
      PSR-CNN-LSTM 19.93 0.99 13.31 33 12384

      Table 7. 

      Ablation study results.