Table 5 Summary of performance metrics and adjustable training parameters of deep learning models for forecasting the axial strain response creep datasets at various confining stress levels (best highlighted in bold).

Stages (MPa)	DNN forecast models	Epochs	Training loss	Input chunk length	Output chunk length	SMAPE (%)	MAPE (%)	RMSE	MAE
5	N-BEATS	100	0.0454	24	12	1.46	1.45	0.332	0.287
	TCN	400	2.260	24	12	6.03	6.28	1.352	1.177
	RNN	400	0.892	24	12	25.08	22.16	4.437	4.283
	TF	400	0.450	24	12	13.04	11.09	1.856	1.21
15	N-BEATS	100	0.0099	24	12	3.98	4.06	0.465	0.381
	TCN	400	0.838	24	12	5.46	5.49	0.645	0.535
	RNN	400	0.188	24	12	28.12	25.68	4.491	2.39
	TF	400	0.327	24	12	25.52	23.43	3.954	2.98
25	N-BEATS	100	0.0291	100	20	4.62	4.54	0.957	0.961
	TCN	400	0.575	24	12	2.88	2.85	0.704	0.592
	RNN	400	0.872	24	12	12.33	10.05	1.950	1.21
	TF	400	0.398	24	12	10.05	11.73	8.232	2.45
35	N-BEATS	100	0.0277	50	10	4.26	4.34	0.640	0.473
	TCN	400	0.348	24	12	4.46	4.34	0.540	0.472
	RNN	400	0.233	24	12	22.18	24.98	2.983	1.79
	TF	400	0.568	24	12	15.08	13.06	2.221	2.13

Quick links

Search