ST-AGRNN: A Spatio-Temporal Attention-Gated Recurrent Neural Network for Traffic State Forecasting

Table 1

Traffic speed prediction performance under different benchmark methods in the PeMSD4 and PeMSD8 datasets (bold is the best; underline is the second best.).

Model	PeMSD4 (MAE/RMSE/MAPE(%))
	15 min	30 min	60 min
HA	2.54/4.96/5.56	2.54/4.96/5.56	2.54/4.96/5.56
ARIMA(2003)	2.51/5.72/5.32	2.75/6.34/5.69	3.21/7.36/6.56
DCRNN(2018)	1.35/2.94/2.68	1.77/4.06/3.71	2.26/5.28/5.10
STGCN(2018)	1.47/3.01/2.92	1.93/4.21/3.98	2.55/5.65/5.39
ASTGCN(2019)	2.12/3.96/4.16	2.42/4.59/4.80	2.73/5.21/5.46
GWN(2019)	1.30/2.68/2.67	1.70/3.82/3.73	2.03/4.65/4.60
LSGCN(2020)	1.45/2.93/2.90	1.82/3.92/3.84	2.22/4.83/4.85
USTGCN(2021)	1.40/2.69/2.81/	1.64/3.19/3.23	2.03/4.25/4.32
ST-AGRNN	1.19/2.36/2.17	1.45/2.98/2.69	1.76/3.63/3.24
Model	PeMSD8 (MAE/RMSE/MAPE(%))
	15 min	30 min	60 min
HA	1.98/4.11/3.94	1.98/4.11/3.94	1.98/4.11/3.94
ARIMA(2003)	1.90/4.87/5.11	2.12/5.24/5.21	2.79/6.22/5.62
DCRNN(2018)	1.17/2.59/2.32	1.49/3.56/3.21	1.87/4.50/4.28
STGCN(2018)	1.19/2.62/2.34	1.59/3.61/3.24	2.25/4.68/4.54
ASTGCN(2019)	1.49/3.18/3.16	1.67/3.69/3.59	1.89/4.13/4.22
LSGCN(2020)	1.16/2.45/2.24	1.46/3.28/3.02	1.81/4.11/3.89
USTGCN(2021)	1.14/2.15/2.07	1.25/2.58/2.35	1.70/3.27/3.22
ST-AGRNN	1.015/2.07/1.82	1.24/2.63/2.21	1.53/3.33/2.71