Air Traffic Flow Prediction with Spatiotemporal Knowledge Distillation Network
Table 1
The experimental results of the proposed ST-KDN and other six comparison methods in the next 10 minutes, 20 minutes, 30 minutes, 40 minutes, 50 minutes, and 60 minutes.
Metric
Method
10 min
20 min
30 min
40 min
50 min
60 min
MAE
SVR
1.898696
2.244607
2.484253
2.767143
3.052133
3.309817
AGCRN
1.794790
1.869949
1.940945
2.007488
2.073016
2.138081
ASTGCN
1.795160
1.890413
1.962362
2.056936
2.060439
2.158814
GMAN
1.805774
1.798546
1.801262
1.819169
1.848141
1.886582
STSSL
1.638903
1.804966
1.980573
2.128636
2.288099
2.457236
TESTAM
1.841422
1.857893
1.883490
1.924575
1.977991
1.989194
ST-KDN
1.763453
1.770182
1.771767
1.778287
1.779059
1.781851
RMSE
SVR
2.565081
2.983550
3.265532
3.604007
3.952299
4.265697
AGCRN
2.519696
2.634144
2.735415
2.832605
2.927571
3.023671
ASTGCN
2.495914
2.630218
2.737751
2.866813
2.889380
3.003969
GMAN
2.518119
2.509231
2.513937
2.539927
2.584160
2.645740
STSSL
2.375745
2.645554
2.916242
3.149616
3.423993
3.716680
TESTAM
2.563212
2.595453
2.638512
2.712400
2.807796
2.912532
ST-KDN
2.500712
2.509924
2.506504
2.512559
2.511632
2.514018
MAPE
SVR
0.387429
0.459490
0.513001
0.576568
0.640230
0.697416
AGCRN
0.368877
0.385325
0.404844
0.423151
0.441527
0.460654
ASTGCN
0.370427
0.385325
0.404844
0.423151
0.441527
0.460654
GMAN
0.384672
0.379004
0.376588
0.377863
0.382285
0.389442
STSSL
0.318166
0.333431
0.377861
0.381922
0.399758
0.408216
TESTAM
0.379114
0.380991
0.388400
0.391331
0.397276
0.401991
ST-KDN
0.372582
0.373806
0.374326
0.376572
0.375481
0.376291
The optimal results for each index within each prediction interval are indicated by bold values. Table shows prediction performance of seven different methods on the real data set in the next 10 minutes (Q = 1), 20 minutes (Q = 2), 30 minutes (Q = 3), 40 minutes (Q = 4), 50 minutes (Q = 5), and 60 minutes (Q = 6).
