Türkiye's solar radiation forecasting with different machine learning approaches

SR is an important parameter for studies related to energy conversion, green building concept, meteorology, global climate change, agriculture and animal husbandry. Since the receivers required to determine solar radiation are not available for all points, this parameter needs to be estimated by various methods. In this study, the annual average solar radiation values (kWh/m2) of T & uuml;rkiye's 3600 grid points for the years 2004-2021 were used to estimate solar radiation for different test points across the country. SR values were estimated using two different machine learning techniques, namely MARS and LSSVR on MATLAB platform. Inverse distance weighting interpolation technique was used for solar radiation maps. The estimated data were mapped in ArcMap environment. SR was estimated using the location information of neighboring measurement grid points and the periodicity component of year values. The data used in the models are the data from HELIOSAT, obtained from the MGM. SR estimates obtained from the test points using different combinations were compared with the observed data. In these comparisons, root mean square error, mean absolute error, mean absolute relative error, Nash-Sutcliffe model efficiency coefficient and coefficient of determination methods were used. Grid-based variation, scatter graphs, Taylor and Violin diagrams of the estimated SR values were created. In addition, Kruskal-Wallis test and Wilcoxon test were applied. LSSVR, one of the machine learning methods, gave very successful prediction results. Thus, it is shown that machine learning algorithms can be an easier and alternative method compared to the traditional methods accepted in the literature.