Estimation of most optimal azimuthal angles for maximum PV solar efficiency using multivariate adaptive regression splines (MARS)

Abstract

The aim of this study was to build a regression model of solar irradiation in the Kulluk region of Turkey by using the multivariate adaptive regression splines (MARS) technique. Using the well-known data mining algorithm, MARS, this study has explored a convenient prediction model for continuous response variables, i.e., average daily energy production from the given system (Ed), average monthly energy production from given system (Em), average daily sum of global irradiation per square meter (Hd) and average annual sum of global irradiation per square meter (Hm). Four continuous estimators are included to estimate Ed, Em, Hd and Hm: Estimated losses due to temperature and low irradiance (ESLOTEM), estimated loss due to angular reflection effect (ESLOANGREF), combined photovoltaic system loss (COMPVLOSS) and rated power of the photovoltaic system (PPVS). Four prediction models as constructed by implementing the MARS algorithm, have been obtained by applying the smallest generalized cross-validation (GCV) criterion where the means of penalty are defined as 1 and the backward pruning method for the package 'earth' of R software is used. As a result, it can be suggested that the procedure of the MARS algorithm, which achieves the greatest predictive accuracy of 100% or nearly 100%, permits researchers to obtain some remarkable hints for ascertaining predictors that affect solar irradiation parameters. The coefficient of determination denoted as R2 was estimated at the highest predictive accuracy to be nearly 1 for Ed, Em, Hd and Hm while the GCV values were found to be 0.000009, 0.018908, 0.000013 and 0.019021, respectively. The estimated results indicate that four MARS models with the first degree interaction effect have the best predictive performances for verification with the lowest GCV value.