A novel soft computing model (Gaussian process regression with K-fold cross validation) for daily and monthly solar radiation forecasting (Part: I)

The main objective of this paper is to present Gaussian Process Regression (GPR) as a new accurate soft computing model to predict daily and monthly solar radiation at Mashhad city, Iran. For this purpose, metrological data was collected from Iranian Meteorological Organization for Mashhad city located at the North-East for the period of 2009-2014. All the collected data include of maximum, minimum and average daily outdoor temperature (Tmax, Tmin and Tave), daily relative outdoor humidity (Rh), daily sea level pressure (p), day of a year (N), sunshine hours (Ns), daily extraterrestrial radiation on horizontal surface (H0) and daily global solar radiation on horizontal surface (H). Results of sensitivity analysis showed that (N/Ns, Tave, Rh, H0) is the best data set group for evaluation of daily global solar radiation at this region. For the GPR model, MAPE, RMSE and EF were 1.97%, 0.16 and 0.99, respectively. Monthly evaluation showed that the main model is not suitable for every month, so for every month, perfect model was trained and tested. Generalizability and stability of the GPR model was evaluated by different sizes of training data with 5-fold analysis. The results showed that GPR model can use with small size of data groups.