Table 6.

A summary of the studies comparing different models using the same dataset.

Location	Model	Input Parameters	Output Parameter	Data Scale	Statistical Benchmarks	Key Findings	References
China	ANN, and Empirical regression models (Model 1, Model 2)	Sunshine percentage, and clearness index	Monthly mean daily diffuse solar radiation	1995 to 2004	RMSE, MPE, and MBE	ANN is superior to empirical models. ANN estimated the actual values of Zhengzhou with 94.81% accuracy.	[113]
Türkiye (73 different locations)	ANN and MLR	Months of the year, latitude satellite-estimated LST, longitude, and altitude	Solar radiation forecasting	2000 to 2002	R2, RMSE, and MBE	ANN achieved high accuracy compared to MLR.	[114]
Iran	Five empirical models, WR, GEP, and ANN	Daylight hours, extraterrestrial global solar radiation, daily mean clearness index, and daily temperature	Daily global solar radiation	1982 to 2016	GPI, MAE, RMSRE, MBE, RMSE, RRMSE, U95, MARE, R2, erMAX, and t-stat,	The statistical metric results gave that the best prediction performance was exhibited by the ANN method.	[115]
Paris, France	ARMA, SIM, SVM, and NN	Global solar radiation	Hourly solar radiation	January 1, 2004, to December 31, 2015	nRMSE	NN model gave better performance than other models.	[116]
Kerman, Iran	3rd degree empirical model, ANN, SVM–RBF, SVM–WT	Daily clearness index	Diffuse solar radiation	2006 to 2012	r, RMSE, and MABE	The SVM–WT method has better estimation accuracy than its competitors with 0.9631 of r, 0.6940 MJ/m2 of RMSE, and 0.5757 MJ/m2 of MABE.	[117]
Tamil Nadu (India)	SVM, ANN, and empirical models	Relative humidity, longitude, day length, month, latitude, maximum and minimum temperature, and bright sunshine hours	Monthly mean daily global solar radiation	2003 to 2012	MBE, MAPE, RMSE, t-stat, and r-value	SVM algorithms gave better results than both those of ANN and empirical models.	[118]
Iran	Empirical models, ordinary and coupled ANN models	Sunshine duration, minimum and maximum air temperatures, and daily global solar radiation	Daily global solar radiation	1992 to 2015	R2, RMSE, and MBE	The prediction performance of the ordinary ANN models was enhanced considerably after being coupled with a genetic algorithm.	[119]
Abu Musa Island, Iran	SVR, MLFFNN, FIS RBFNN, and ANFIS	Inputs (N1): Wind speed, temperature, relative humidity, pressure, and local time Input (N2): Solar radiation	Hourly solar radiation	2010 to 2016	r, RMSE, and MSE	The results of N1 give that, MLFFNN and SVR methods exhibited the best prediction performance with r = 0.9999 and 0.9795, respectively. Furthermore, ANFIS, MLFFNN, and SVR methods obtained a correlation coefficient of over 0.95 in the test data for N2.	[75]
Four climatic zones of China	12 ML models, and 12 versions of the Ångström–Prescott model	Daily historical data	Daily global solar radiation	1966 to 2015	R2, RMSE, U95 MBE, t-stat, and NRMSE	Each prediction method used the same dataset and ML methods gave lower error values than empirical models. Among the ML methods, four models come to the fore: ANFIS, ELM, LSSVM, and MARS.	[120]
Four provinces (Şırnak, Kilis Ankara, and Karaman) in Türkiye	Angstrom type-empirical models, RSM, Holt-Winters, and ANN	Wind speed, pressure, relative humidity, ambient temperature, and sunshine duration	Monthly average daily global solar radiation	2008 to 2018	MAPE, RMSE, MBE, t-stat, and R2	Each model used the same dataset, and ANN exhibited the best results for global solar radiation data with R2, MAPE, RMSE, MBE, and t-stat of 0.9911, 4.93%, 0.78 MJ/m2, 0.1323 MJ/m2, and 0.58, respectively.	[112]
Five locations, Morocco	22 empirical models, RF, MLP, Boost, and Bag	Relative humidity, ambient temperature, wind speed, and solar radiation	Daily global solar radiation	2011 to 2015	r, nMAE, and nRMSE	RF method gave the best performance. r, nMAE and nRMSE are 81.73–95.14%, 5.88–13.86%, and 8.22–18%, respectively. Among the empirical models, the TG1 model was recommended. r, nMAE and nRMSE are 72.38–93.46%, 6.96–17.94%, and 9.89–22.39%, respectively.	[41]
Alabama, United States	KNNR, ANN, SVM, and BILSTM	Global solar radiation	Hourly solar radiation	May 1, 2011, to February 18, 2013	RMSE, MAE, and R2	The BILSTM model outperformed KNNR, ANN, and SVR methods in terms of RMSE, MAE, and R2 evaluation benchmarks.	[121]
North Carolina, and Southern Spain	MLP, ELM, GRNN, SVM, RF, and XGBoost	Temperature-based variables	Daily extraterrestrial solar radiation	2000 to 2018	MBE, RMSE, RRMSE, NSE, R2, and GPI	MLP and SVM are recommended for arid and semi-arid areas, while RF and XGBOOST are recommended for semi-humid and humid areas.	[122]
Tetouan in Morocco	ARIMA, FFNN, and k-NN	Top of atmosphere radiation, clearness index, maximum, average, delta, and ratio temperature	Daily global solar radiation	January 1, 2013, to December 31, 2015	MAPE, RMSE, MBE, NRMSE, Ts and σ	FFNN (6 × 10 × 1) gave better results than those of time series, and k-NN model with very low error magnitudes.	[123]