Research into various modeling methods including the artificial neural networks in studying solar radiation for different regions in the Russian Federation

Abstract

The comparison of results obtained using the artificial neural networks (ANN) has been carried out in researching solar radiation for various regions of the Russian Federation with data modeling applying a spline interpolation preserving an interpolant shape or a polynomial (up to tenth degree) in the MATLAB environment obtained when using the MATLAB Basic Fitting-a graphical user interface (GUI). The neural network has been trained by the feed-forward backprop algorithm. The following functions were used: the Bayesian Regularization, the function of gradient descent with regard for moments, the hyperbolic tangent functions. The Mean Square Error (MSE) was chosen as a loss function which was minimized. The 15 input parameters were taken. As a result, the obtained model had high values of correlation coefficients and low values of the Mean Square Error among the target and the output values. The polynomial (up to tenth degree inclusive) coefficients and the norm of the residuals were obtained which were decreasing with the increasing of the polynomial degree. It was shown that the better results are received using the artificial neural networks (ANN), somewhat worse ones employing a spline interpolant and shape-preserving interpolant. At the same time, the proposed models and methods can be used in calculation of solar radiation for various regions in the Russian Federation and other countries.