Short-Term Daily Univariate Streamflow Forecasting Using Deep Learning Models

Table 5

Performance comparison of the proposed models for different input time lags in Gummera station.

Gummera	T + 1					T + 2					T + 3					T + 4
	RMSE	MAE	MAPE	R^2	TTPE (sec)	RMSE	MAE	MAPE	R^2	TTPE (sec)	RMSE	MAE	MAPE	R^2	TTPE (sec)	RMSE	MAE	MAPE	R^2	TTPE (sec)
MLP	17.67	7.84	0.18	0.90	0.88	17.68	7.87	0.19	0.90	0.75	17.43	7.76	0.18	0.90	0.61	17.56	9.73	0.78	0.90	0.95
GRU	17.66	8.39	0.41	0.90	0.51	17.76	7.82	0.16	0.90	1.43	17.47	8.29	0.35	0.90	1.24	17.99	10.54	1.03	0.89	0.53
S-LSTM	17.66	8.29	0.39	0.90	1.54	17.71	8.04	0.22	0.90	0.75	17.69	8.51	0.32	0.90	2.99	17.69	8.41	0.35	0.90	3.31
Bi-LSTM	17.63	7.98	0.27	0.90	0.92	17.93	8.39	0.29	0.89	2.34	17.83	9.18	0.56	0.89	2.77	17.56	8.56	0.39	0.90	1.84

TTPE(training time per epochs).The bold values indicate the best performance score for each time lag.