This article presents different combinations of input parameters based on an intelligent technique, using neural networks to predict daily global solar radiation (GSR) for twenty-five Moroccan cities. The collected measured data are available for 365 days and 25 stations around Morocco. Different input parameters are used, such as clearness index KT, day number, the length of the day, minimal temperature Tmin, maximal temperature Tmax, average temperature Taverage, difference temperature ΔT, ratio temperature T-Ratio, average relative humidity RH, solar radiation at the top outside atmosphere TOA, average wind speed Ws, altitude, longitude, latitude, and solar declination. A different combination was employed to predict daily GSR for the considered locations in order to find the most adequate input parameter that can be used in the prediction procedure. Several statistical metrics are applied to evaluate the performance of the obtained results, such as coefficients of determination (R2), mean absolute percentage error (MAPE), root mean square error (RMSE), normalized root mean square error (NRMSE), mean bias error (MBE), test statistic (TS), linear regression coefficients (the slope “a” and the constant “b”), and standard deviation (σ). It is found that the usage of input parameters gives highly accurate results in the artificial neural network (FFNN-BP) model, obtaining the lowest value of the statistical metrics. The results showed the best input of 25 locations, 12 inputs for Er-Rachidia, Marrakech, Medilt, Taza, Oujda, Nador, Tetouan, Tanger, Al-Auin, Dakhla, Settat, and Safi, seven inputs for Fes, Ifrane, Beni-Mellal, and Meknes, six inputs for Agadir and Rabat, five inputs for Sidi Ifni, Essaouira, Casablanca and Kenitra, four inputs for Ouarzazate, Lareche, and Al-Hoceima. In terms of accuracy, R2 of the selected best inputs parameters varies between 0.9860% and 0.9920%, the range value of MBE (%) being from −0.1076% to −0.5931%, the RMSE between 0.1990 and 0.4580%, the range value of the NRMSE between 0.0355 and 0.8938, and the lowest value MAPE between 0.0019 and 0.0060%. This technique could be used to predict other parameters for locations where measurement instrumentation is unavailable or costly to obtain.
