PGDRT: Prediction Demand Based on Graph Convolutional Network for Regional Demand-Responsive Transport

Algorithm 1

Training procedure of the proposed method.

Require:

1: Past demands

2: Future true demand

3: Adjacency matrix , degree matrix , hop:

Ensure: future prediction demand

4: While training do

5:   For all do

6:     (1) Spatial dependency: apply Chebyshev to each adjacency matrix

7:     rescale (normalize )

8:     For alldo

9:        Chebyshev

10:     End for

11:     (2) Temporal dependency: apply with contextual gating (CG) and ConvLSTM

12:     

13:   End for

14: (3) Spatial-temporal dependency: apply with FC (fully connected) and GCN (graph convolution network)

15:   

16:   Compute loss:

17: End while