Table 1 Summary of recent literature on solar radiation forecast/prediction.
Author/References | Case study | Research objective | Models used | Performance of models |
|---|---|---|---|---|
Sun et al.16 | Beijing China | Improvement of the performance of solar radiation forecasting and comparison with other models | Decomposition-clustering-ensemble learning | NRSME = 2.96% MAPE = 2.83% Directional forcast = 88.24% |
Belmahdi et al.17 | Tetouan city Morocco | Building models that can forecast monthly mean daily global radiation | Time series (ARMA and ARIMA) | ARIMA (0.2,1) gave a better performance than ARMA (2,1) with 64.05% and 24.32% improvement respectively |
Blal et al.18 | Adrar Algeria | Statistically comparing the predictive models used for daily average global radiation estimation and hourly global solar radiation study on the horizontal surface under different weather conditions (Studying solar radiation under various conditions of climate) | Six Ambient temperature models | Model (M4) gave R2 of 0.8753 being best M1 = 0.7099 M5 = 0.8193 |
Heng et al. 19 | United States | The model used for forecasting with accuracy and stability objective for global monthly average radiation | nondominated sorting-based multiobjective bat algorithm (NSMOBA) | Gave satisfactory accuracy and stability |
Kisi et al. 20 | Turkey | Connectionist system evolution for daily scale prediction of solar radiation | Dynamic evolving neural-fuzzy inference system (DENFIS) | Provided better accuracy in monthly SR prediction than the benchmark models |
Ghimire et al. 21 | Australia | Integration of CNN and LSTM for short-term GSR prediction | hybrid model based on a convolution network CLSTM | Performed better than other DL models and the benchmark models |
Rodríguez-Benítez et al. 22 | Spain | Extension of a temporal horizon of ASI-based nowcast to match the satellite-based prediction. Increasing the temporal latency and resolution of the satellite-based nowcasting to match that of ASI-based prediction | all-sky imager (ASI) model | ASIs are preferable to other models since it overcomes most challenges that other models encounter |
Peng et al. 23 | Alabama USA | Construction and evaluation of the performance of DL models based on biLSTM, SCA, and CEEMDAN for hourly solar radiation prediction over multi-step horizons | deep learning model based on Bi-directional long short-term memory (BiLSTM), sine cosine algorithm (SCA), and complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) which can be called CEN-SCA-BiLSTM model | CEN-SCA-BiLSTMmodel gave the smallest RMSE, MAE, MASE, and largest R when compared with other competitors |
Campo-Ávila et al. 24 | Spain | Prediction of one day ahead hourly global solar radiation | A model that combines clustering, regression, and classification | RMSE less than 20% |
Lai et al. 25 | Brazil | Hourly solar forecasting with Feature Attention-based Deep Forecasting (FADF) | A deep learning-based hybrid method | RMSE 11.88% on Itupiranga dataset and 12.65% on Ocala dataset when compared with smart persistence |
Guermoui et al. 26 | Algeria | multi-step ahead forecasting of daily global and direct horizontal solar radiation components in the Saharan climate | Weighted Gaussian Process Regression (WGPR), | RMSE = 3.18 and R2 = 85.85% for 10th daily global horizontal radiation and RMSE = 5.23 and R2 |
Gürel et al. 15 | Turkey | Using four different models to predict monthly average daily global SR data | ML algorithm-based models | R2 = 0.952 ~ 0.993 RMSE and MAPE less than 10% |
Zhuo et al. 27 | China | To simultaneously predict the multi-time scale (daily and monthly mean daily) and multi-component (global and diffuse) solar radiation | combined multi-task learning and Gaussian process regression (MTGPR) model | Average R2 ranges 0.19 ~ 0.48%, RMSE improved 0.57 ~ 0.65% and rRMSE improved 0.51% ~ 0.52% for daily prediction. For monthly prediction the range is 2.62 ~ 2.65%, 5.50 ~ 12.07% and 5.21 ~ 12.08% respectively for R2, RMSE and rRMSE |
Makade et al. 28 | India | Developing a comprehensive review of the works done by Indian researchers in solar radiation modeling and carrying out a statistical analysis of the developed solar radiation model | GSR Model M-78 | MPE varies between -8.1186% and 6.9383% and the coefficient of determination between 0.6345 and 0.9616 |
Prasad et al. 29 | Australia | Development of a hybrid model that handles issues with nonstationarity in multiple predictor inputs utilizing a self-adaptive approach while giving a good accuracy of the forecast of short-term | multivariate empirical mode decomposition method (MEMD) – Singular Value Decomposition (SVD)- Random Forest (RF) model (hybrid MEMD-SVD-RF model) | Generated a better and more reliable forecast Average R2 of 0.98 and RMSE of 1.05 |
Z. Pung et al. 30 | Alabama US | To study the performances of DL algorithms for the prediction of solar radiation | An ANN model and a recurrent neural network (RNN) model | RNN model improved by 47% in NMBE and 26% in RMSE |
Puah et al. 31 | Malaysia | Producing a comparable forecast performance in relation with the Supervised Learning | Regression Enhanced Incremental Self-organising Neural Network (RE-SOINN) | Achieved higher accuracy when compared to others MASE = 0.65755 RMSE = 73.945 |
Narvaez et al. 32 | Colombia | Develo[ping accurate site-adaptation as well as solar radiation model using ML and DL | ML-based model | 38% better performance than the traditional methods |
Karaman et al. 33 | Karaman Turkey | Using different activation functions to obtain the best response from ELM and ANN after their performance has been compared | extreme learning machines (ELM) and Artificial Neural Network (ANN) | ELM has better performance with RMSE = 0.0297 and Performance of 95% |
A˘gbulut et al. 34 | Turkey | Prediction of daily global solar radiation from 4 different provinces having diverse solar radiation distribution | support vector machine (SVM), artificial neural network (ANN), kernel and nearest-neighbor (k-NN), and deep learning (DL) models | R2 ranges from 85.5%—93.6% MAPE 15.92%—30.24% rRMSE 14.10%—25.19% |
Al-Rousan et al. 35 | Jordan | Reviewing different prediction methods employed in predicting solar radiation | Multi-layer perceptron (MLP), Support Vector Machine Regression (SVMR), and Linear regression (LR) | R2 = 0.9513, 0.8477 and 0.8477 respectively for MLP, SVMR and LR while MAPE = 0.0001, 0.0418 and 0.0434 |
Sunhra Das 36 | India | To carry out short term solar forecasting for different days of the year | A model for prediction of solar radiation on tilted surface | RMSE = 8.9, 6.7, and 8.3 for Jan 29th, Apr 1st, and Oct 6th respectively |
Bounoua et al. 37 | Morocco | Evaluation of the potential of three ensemble methods based on regression trees (Bagging, Boosting, and RandomForest) in estimating the daily GHI | empirical and machine-learning methods | Random Forest method with the following result R: 87.53–96.20%; nMAE: 5.84–11.81%; nRMSE: 7.85–15.33% outperformed others |
Shadab et al. 38 | India | extending the ARIMA models for spatial forecasting of monthly average insolation as well as finding the most suitable location for solar power projects based on the forecasts | Seasonal ARIMA (SARIMA) model | R2 = 0.9293, Root Mean Square Error = 0.3529, Mean Absolute Error = 0.2659 and Mean Absolute Percentage Error = 6.556 |
Srivastava et al. 39 | India | forecasting of the 1-day-ahead to 6-day-ahead solar radiation levels using four ML models | MARS, CART, M5 and random forest models | Random Forest provided the best result while the Cart has the worst result. From best to worst we have Random Forest > M5 > MARS > CART |
