Skip to main content
. 2021 Mar 18;226:120403. doi: 10.1016/j.energy.2021.120403

Table 5.

Performance comparison of different techniques.

MAPE RMSE MAE IR(MAPE)
BPNN 1 46.78% 11.1041 8.4084
2 9.56% 2.2285 1.6749 79.56%
3 28.51% 6.6743 5.3789 39.06%
4 7.17% 1.9194 1.5775 84.67%
MLR 1 76.27% 14.5121 12.8406
2 76.19% 14.6423 12.7107 0.10%
3 64.45% 15.4351 12.0035 15.50%
4 56.28% 12.9765 10.4790 26.21%
SVM 1 36.17% 8.9761 7.4284
2 26.97% 7.1991 6.4495 19.20%
3 46.59% 14.4929 10.1242 −28.81%
4 23.60% 5.2307 4.0115 34.75%
LSTM 1 54.43% 14.5938 11.1827
2 32.53% 9.7595 8.1046 40.24%
3 44.40% 8.2702 6.9034 18.43%
4 31.88% 6.7175 5.2260 41.43%
RNN 1 58.92% 12.3870 11.2355
2 22.53% 4.7991 4.4857 61.76%
3 50.54% 15.0736 11.4218 14.22%
4 17.02% 4.7706 3.3586 71.11%

Note: “1” means that historical data are used to predict. “2” means that historical data and text features are used to predict. “3” means that historical data and financial data are used to predict. “4” means that historical data, financial data, and text features data are both employed to predict. IR(MAPE) means improving the rate of MAPE from “1” to “2 (3 or 4)”. The grid search method is used to determine the parameters of adopted algorithms [37]. Appendix C lists the final parameter values of these forecasting models in all examples.