Traffic flow forecasting

This is a fork project from guoshnBJTU/ASTGCN-r-pytorch.

Run

Notebook: Google Colab

• on PEMS04 dataset

  python prepareData.py --config configurations/PEMS04_[MODEL].conf

  python train_[MODEL]_r.py --config configurations/PEMS04_[MODEL].conf

Prediction on sensor 1 in the first 100 step

Contributing

• We modified the ASTGCN model with full three information from last hour, yesterday and previous week.
• We compare ASTGCN with GCN, LSTM, HA in various settings.

Result

Methods	HA (h/hdw)	LSTM (h/hdw)	LSTMGC (h/hdw)	ASTGCN (h/hdw)
MAE@1h	38.75/28.15	29.46/23.91	37.1/22.52	22.53/21.84