Experiments on Charge-Station demand stationΒΆ

  • Experiment Setting

    • Dataset

|        Attributes        | **Beijing** |
| :----------------------: | :---------: |
|        Time span         |             |
| Number of riding records |             |
|    Number of stations    |     629     |

Following shows a map-visualization of bike stations in NYC, the latest built stations have deeper color.

In the data preprocessing stage, we removed the stations with average daily traffic flow smaller than 1, since predictions for these stations are not significant in real life. The remaining station number are 717, 444 and 378 for NYC, Chicago and DC, respectively.

  • Network parameter setting (STMeta model)

    Following shows the parameters we used and a short explanation of the parameter meaning. To know more about the parameter, please refer to the API reference.

| Parameter  | Value |                  Meaning                   |
| :--------: | :---: | :----------------------------------------: |
|    GLL     |   1   |          number of GCLSTM layers           |
| LSTMUnits  |  64   |       number of hidden units in LSTM       |
|  GALUnits  |  64   |       number of units used in GAtteL       |
|  GALHeads  |   2   |       number of multi-head in GAtteL       |
| DenseUnits |  32   | number of units in penultimate dense layer |
|     TC     |   0   |           correlation threshold            |
|     TD     | 1000  |             distance threshold             |
|     TI     |  500  |           interaction threshold            |
  • Experiment Results

    • CG-GCLSTM Only use correlation graph in the model

    • DG-GCLSTM Only use distance graph in the model

    • IG-GCLSTM Only use interaction graph in the model

Beijing
HM 1.13594
ARIMA 5.60477
HMM 5.42030
XGBoost 5.32069
LSTM 5.13307
CG-GCLSTM 4.64375
DG-GCLSTM 4.67169
IG-GCLSTM 4.77809
ST_MGCN (Multi-Graph) 4.41732
STMeta 4.22640

Add trend and period into feature:

(C6-P7-T4) means the length of closeness, period and trend are 6, 7, and 4 respective.

Default C6-P7-T4

NYC Chicago DC
HM (C6-P7-T4) 4.55474 3.28585 2.74502
HM (C0-P7-T4) 4.27844 3.18290 2.68013
XGBoost 4.14909 3.02530 2.73286
GBRT 3.94348 2.85847 2.63935
LSTM 3.92746 2.92663 2.65197
DG-GCLSTM 3.88572 3.00055 2.60095
IG-GCLSTM 3.79187 2.97707 2.58739
CG-GCLSTM 3.77422 2.98797 2.59339
STMeta 3.73464 2.79475 2.47565
  • Model training records

    Following data was collected on a Windows PC with CPU : Interl i7 8700K, Memory: 32 GB, GPU: Nvidia GTX 1080Ti.

NYC City SingleGraph-GCLSTM(Average) STMeta ST_MGCN
Number of trainable variables 19749 61993 249245
Converaged training time 2 hours 6 hours 51 hours
  • Source Code

Use the ./Experiment/STMeta_Master_Bike.py to train the model or view evaluation results.

Back To HomePage