|
| Reference | Model type | Location of data used | Performance |
|
| Lee et al. [42] | Angstrom-type equations | South Korea (Seoul) | nRMSE (%) = 43.09
nMBE (%) = −12.24 |
| Brabec et al. [43] | Heteroscedastic model (HR) | Romania (Timisoara) | nRMSE (%) = 45.80 |
| Voyant et al. [44] | MLP | France (Ajaccio, Corsica) | nRMSE (%) = 40.55 |
| Voyant et al. [44] | ARMA | France (Ajaccio, Corsica) | nRMSE (%) = 40.32 |
| Voyant et al. [44] | Hybrid MLP-ARMA | France (Ajaccio, Corsica) | nRMSE (%) = 36.59 |
| Trapero et al. [45] | ARIMA | Spain (Castilla-La Mancha) | nRMSE (%) = 37.34
nMBE (%) = 1.89 |
| Akarslan and Hocaoglu [46] | Adaptive approach | Turkey (Çanakkale) | nRMSE (%) = 34.86
RMSE = 68.4132 W/m2 |
| Zhao et al. [47] | 3D-CNN with AR | NREL database | nRMSE (%) = 34.97
nMBE (%) = 0.65
FS = 5.89 |
| Bae [48] | ANN | South Korea (Yuseong-gu, Daejeon) | RMSE = 71.41 W/m2
R = 0.9264 |
| Bae [48] | NAR | South Korea (Yuseong-gu, Daejeon) | RMSE = 111.41 W/m2
R = 0.7912 |
| Bae [48] | SVM | South Korea (Yuseong-gu, Daejeon) | RMSE = 58.72 W/m2
R = 0.9562 |
| Qing and Niu [30] | LSTM | Cape Verde (Santiago) | RMSE = 76.245 W/m2 |
| Yu [29] | LSTM | USA (Hawaii) | RMSE = 66.69 W/m2
MAE = 46.04 W/m2
R = 0.95 |
| Yu [29] | SVR | USA (Hawaii) | RMSE = 144.43 W/m2
MAE = 96.25 W/m2
R = 0.75 |
| This study | Model II-BD based on LSTM-MLP | USA (Denver, Colorado) | nRMSE (%) = 32.27
RMSE = 62.16 W/m2
MAE = 26.65 W/m2
MBE = −0.4547 W/m2
R = 0.9737 |
|
