Table 4 List of hyperparameter values for passenger flow prediction networks

GEME-Net (Teacher model)

\({C}_{in}\)

Re-parameterized convolutional embedding layer

18

\({C}_{out}\)

Re-parameterized convolutional embedding layer

12

\({C}_{p}\)

Encoder module

32

\({k}_{m}\)

Re-parameterized convolutional embedding layer

3

\({k}_{rc}\)

Re-parameterized convolutional embedding layer

5

\({M}_{e}\)

Event-driven frequency-enhanced module

5

Head number of event-driven frequency-enhanced attention

Event-driven frequency-enhanced module

4

\({k}_{msr}\)

Multi-scale retention rate attention

32

\({H}_{MSR}\)

Multi-scale retention rate attention

6

l

Spatial-temporal adaptive multi-graph convolutional network

3

\({H}_{e}\)

Sparse attention mechanism

4

H

Multi-head self-attention

4

GEME-Net (Student model)

\({d}_{h}\)

-

128

\({r}_{h1}\)

-

16

\({r}_{h2}\)

-

8

Dropout ratio

-

0.1

\({\alpha }_{d}\)

-

0.5