Table 3.

Comparison of SMI-TCN-BILSTM with other neural network models.

Model	MSE	RMSE	MAE	MAPE	R2
BP34	0.1005	0.317	0.2487	2.85%	88.00%
SVR35	0.0331	0.1820	0.1475	1.74%	96.05%
LSTM9	0.0286	0.1692	0.1215	1.38%	96.58%
TCN-LSTM33	0.0433	0.2081	0.1446	1.69%	94.83%
CNN-LSTM36	0.0293	0.1711	0.1316	1.52%	96.50%
CNN-GRU-Attention14	0.0302	0.1738	0.1339	1.53%	96.39%
BiLSTM11	0.0247	0.1571	0.1180	1.35%	97.05%
TCN-BiLSTM	0.0245	0.1566	0.1148	1.32%	97.07%
SSA-TCN-BiLSTM	0.0225	0.1501	0.1082	1.23%	97.31%
SMI-TCN-BiLSTM(Ours)	0.0132	0.1147	0.0851	0.96%	98.43%

The model input data in the above tables are processed with Savitzky-Golay filter and MIC to ensure fairness of the prediction results.