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

Table 2

Traffic flow 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 (%)
HA	38.03	59.24	27.88
ARIMA(2003)	33.73	48.80	24.18
STGCN(2018)	21.16	34.89	13.83
DCRNN(2018)	21.22	33.44	14.17
ASTGCN(r)(2019)	22.93	35.22	16.56
GWN(2019)	24.89	39.66	17.29
LSGCN(2020)	21.53	33.86	13.18
STSGCN(2020)	21.19	33.65	13.90
STFGNN(2021)	20.48	32.51	16.77
Z-GCNETs(2021)	19.50	31.61	12.78
STG-NCDE(2022)	19.21	31.09	12.76
ST-AGRNN(ours)	18.97	30.003	12.81
Model	PeMSD8
	MAE	RMSE	MAPE (%)
HA	34.86	59.24	27.88
ARIMA(2003)	31.09	44.32	22.73
STGCN(2018)	17.50	27.09	11.29
DCRNN(2018)	16.82	26.36	10.92
ASTGCN(r) (2019)	18.25	28.06	11.64
GWN(2019)	18.28	30.05	12.15
LSGCN(2020)	17.73	26.76	11.20
STSGCN(2020)	17.13	26.80	10.96
STFGNN(2021)	16.94	26.25	10.60
Z-GCNETs(2021)	15.75	25.11	10.01
STG-NCDE(2022)	15.45	24.81	9.92
ST-AGRNN(ours)	14.95	23.15	9.21