Artificial neural networks for global and direct solar irradiance forecasting: a case study

The current study treats the prediction of the direct normal and the global horizontal irradiances (DNI and GHI) for five years measurements in Oujda city, Morocco. The prediction is performed by the mean of proposed and developed models based on the Artificial Neural Network method, which is the most common machine-learning algorithm today. The purpose of this research is to achieve more accurate daily and monthly GHI and DNI by using many combinations of a new calculated input; cloudy sky index (C-s) with temperature (T), wind speed (WS), wind direction (WD), relative humidity (RH), diffuse, and global horizontal irradiances (DHI and GHI). Several models are developed by forming combinations according to the variation of the input number. These models are evaluated by different statistical tools: Mean Bias Error (MBE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and correlation coefficient (R-2). For the DNI estimation, the most appropriate inputs were GHI, C-s, T, RH, WS, WD, and BP. Thus, the best performance is found at the monthly time scale to be -0.35%, 3.33%, 1.26%, and 0.98 for MBE, r-RMSE, MAPE and R-2 respectively. The overall errors obtained for the best GHI prediction at the daily time scale are -0.0004%, 0.0099%, 0.0067% and 1 for MBE, r-RMSE, MAPE, and R-2, respectively. The cloudy sky index has shown good accuracies for the estimation of GHI and DNI. Moreover, literature about the prediction of different solar irradiance components under Moroccan climate is reviewed in this work.

Automated summary

This study predicts the direct normal and global horizontal irradiances using the Artificial Neural Network method with new calculated input variables for more accurate daily and monthly forecasts. The cloudy sky index shows good accuracy in estimating GHI and DNI, with the best performance observed at the monthly time scale.