|
| Ref. | Methodology | Forecast horizon | Findings |
|
| Method of EMD |
| [10] | EMD-ELM, ELM | 5min, 30 min | (i) The EMD-SCA-ELM with 15 min time interval data forecasting method provides an enhancement in the performance of the ELM model concerning short-term PV power forecasting |
| [11] | EMD-CNN | 1–6–12 h | (i) The results show that the proposed model has less error than other conventional forecasting models, and as the prediction time scale becomes larger, the prediction accuracy of EMD-CNN does not decrease too much, and the complexity of the EMD-CNN model does not increase in the case of too many decomposition components |
| [12] | EMD-SCA-ELM,ELM, EMD-ELM, and SCA-ELM | 5 min, 30 min, 60 min | (i) The results signify that the recommended technique performs in an outstanding manner than the conventional ones while addressing short-term PV power |
| [13] | EMD-BPNN | 1–12–24 h | (i) Decomposition-based BPNN model performs better as compared to the BPNN method due to EMD; the input space was expanded due to the increased number of data points created by the IMF |
|
| Method of VMD |
| [15] | ARMA, DBN, EMD-ARMA-DBN, EEMD-ARMA-DBN, and VMD-ARMA-DBN | 1 d | (i) The short-term prediction accuracy of the nonlinear PV power time series in this work proposes a multifrequency combined prediction model based on VMD mode decomposition |
| [14] | VMD-CNN | 8 h | (i) The proposed hybrid model improves short-term PV power forecasting precision and can meet the needs of practical projects. With the rapid development in the field of deep learning, the model will have advantages in computation efficiency and become more practical in the near future. |
| [16] | VMD | 1 d | (i) Determines the candidate feature set of each component using the incremental search method and sorts the features in the candidate feature set in descending mRMR value order |
| [17] | OVMD-IPSO-LSTM | 1 d | (i) The advanced decomposition method is developed to decompose the PV power into the different fluctuation components more effectively |
|
| Method of WD |
| [18] | LS-SVM + WD | 1–24 h | (i) Analysis of three forecast models concludes that the LS-SVM with WD also permits reaching the greatest revenue with lower costs for unbalancing penalty with respect to the ANN and the LS-SVM |
| [19] | DWT-CNN-LSTM | 1 d ahead | (i) The proposed DL technique-based day-ahead solar irradiance forecasting model has a high potential for future practical applications |
| [20] | WD + ANN | 5 d | (i) The proposed model is validated by experimental data that predict the output power PV systems accurately, which is useful to enhance the safety and stability of the electrical grid |
| [21] | Wavelet-ANFIS | 10 min, 30 min, 60 min | (i) Demonstrates that the connective forecast with discrete wavelet decomposition and ANFIS could be an outstanding tool for the short-term forecasting of PV output power |
|
| Method of CEEMDAN |
| [5] | CEEMDAN–CNN–LSTM | 1 h | (i) An average RMSE of 38.49 W/m2 indicates that CEEMDAN-CNN-LSTM model has a relatively stable prediction performance in different climatic conditions |
| [22] | CEN-SCA-BiLSTM | 1 h | (i) The CEN-SCA-BiLSTM model obtains the smallest |
| [23] | CEEMDAN-AE-GRU | 1 d | (i) A desirable method for accurate short-term PV power forecasting |
| [24] | CEEMDAN-FIG-ILSTM-ARIMA | 7 h | (i) Decomposition and reconstruction of historical PV output power |
|
