Harmonic analysis based hourly solar radiation forecasting model

Abstract

In this study, a novel two-step hourly solar radiation modelling procedure is developed. In the first step, hourly solar radiation data of a single day is considered and its periodic Fourier series coefficients are calculated. The first nine Fourier series coefficients are considered to accurately render the data. Among these nine coefficients, the yearly variations of the first two coefficients exhibit relatively large magnitudes and therefore are mathematically modelled. In the second step, various NN models are built to predict the solar behaviour using the Fourier model and the past data. The network inputs are the solar radiation data belonging to early sunrise hours of each day, and the model generated values of first two Fourier series coefficients. The NN outputs are the differences between actual and model values of first two Fourier series coefficients, together with the actual values of the following seven Fourier series coefficients. The NN models are trained and tested using different simulation parameters. Comparative results of the models are discussed in detail on hourly solar radiation data obtained from Izmir region of Turkey. It was observed that, once the model is constructed, it is possible to explain the behaviour of the solar radiation data in any day just by using the solar radiation data obtained early in the morning.