Median-Linear model for solar radiation estimation by comparison with ANFIS and Angström-Prescott methods

This paper aims to propose a novel approach to estimating solar radiation in relation to sunshine duration. A hybrid Median-Linear (MedLin) model is presented including both the median value and the linear solar radiation equation. First stage is concerned with two-stage alternative of the proposed model with inputs and outputs exhibiting a seasonality structure and it is seasonally adjusted using median values. In the second stage, the linear equation for the difference values is determined with estimations the difference in solar radiation. Finally, the median value and the estimated difference value are added. Monthly total solar radiation values are estimated using monthly total sunshine duration data from Mardin, Adana, Bozova, Hakkari, Şırnak, and Van stations in Southeast of Türkiye for the application. The predictions of the MedLin approach are compared with the results of the Angström-Prescott, ANFIS, and Mean-Linear (MeanLin) models. As for the prediction of monthly total solar radiation data at Mardin and Hakkari stations, the proposed MedLin approach yields the best results compared to other methods with mean absolute percentage error 3.508% and 8.951%, respectively. Angström-Prescott model for Adana, MeanLin method for Bozova, ANFIS model for Şırnak, and MeanLin model for Van provided the best predictions with mean absolute percentage error 3.721%, 3.285%, 4.278%, and 13.101%, respectively. The results show that the MedLin technique proposed in this paper is useful for solar radiation estimation.