Attention Mechanism with Spatial-Temporal Joint Deep Learning Model for the Forecasting of Short-Term Passenger Flow Distribution at the Railway Station

Table 3

Comparison of prediction performances obtained using different TGs in different baseline models.

Time granularity	10 s			20 s			30 s			60 s
Indicators	RMSE	MAE	MAPE (%)	RMSE	MAE	MAPE (%)	RMSE	MAE	MAPE	RMSE	MAE	MAPE (%)
ARIMA	16.282	10.281	55.1	43.069	23.602	58.7	63.227	35.636	63.2%	120.975	63.829	88.1
BPNN	3.973	2.975	14.1	8.843	5.877	18.5	14.824	8.627	23.9%	44.443	28.725	26.6
CNN	5.961	3.239	18.4	10.604	8.685	22.3	31.238	14.611	27.4	47.961	25.614	28.6
RNN	5.626	5.879	13.8	9.365	6.174	18.3	30.962	12.962	23.9%	46.716	22.606	23.6
LSTM	5.203	5.808	13.7	8.722	5.114	17.5	28.806	10.589	23.9%	31.320	17.701	22.4
ST-Bi-LSTM (no graph)	2.913	2.214	12.7	7.047	4.961	17.2	14.878	8.534	23.3%	29.499	17.782	21.7
ST-Bi-LSTM (1A)	2.927	2.023	12.9	6.744	4.297	16.3	13.696	8.759	20.9%	34.015	22.683	31.4
ST-Bi-LSTM (no A)	2.849	2.094	12.6	6.614	4.291	15.9	13.247	8.191	20.1%	25.461	14.874	19.4
ST-Bi-LSTM (no T)	2.931	2.145	13.1	6.761	4.306	16.2	13.600	8.35	20.8%	28.103	16.912	20.7
ST-Bi-LSTM (no T&A)	2.986	2.966	13.5	6.651	4.395	16.5	13.855	8.527	20.2%	30.216	17.411	22.1
Adam ST-Bi-LSTM	2.889	2.018	13.0	6.640	4.309	16.2	13.377	8.346	19.7%	26.513	14.821	19.4
ST-Bi-LSTM	2.716	1.921	12.1	6.605	4.282	15.7	13.215	8.126	18.4%	25.133	15.036	18.9