Solar Irradiation Forecasting Uses Time Series Analysis

Abstract

In recent years, the ever-increasing demand on electricity cannot be followed by the amount of energy supply mainly produced from fossil. In order to mitigate this issue, many researches have been conducted in the field of renewable energy (RE), which aims at reducing the dependence on fossil based energy. In Indonesia, one form of RE that has a promising potential is solar power. Unfortunately, solar power has intermittent nature and thus, the output of RE plants (solar power plants) tends to be unpredictable. It has been found that there is a linear relationship between the amount of solar irradiation and the power produced by the solar power plants. This linear relationship can be exploited to predict the power produced by the solar power plant by performing forecasting on the amount of solar irradiation. This paper discusses three methods for forecasting solar irradiation. The first forecasting method discussed in this paper is multivariate linear regression. In this method, the forecasting is performed by using data on temperature, air pressure, humidity, and wind speed. The other two methods are support vector regression (SVR) and hybrid fast Fourier transform-autoregressive (FFT-AR). In these two methods, the forecasting process is performed based on the historical solar irradiation data sets. All the aforementioned three methods are applied to the solar irradiation data obtained from Cirata solar plant and the automatic weather station, the meteorology, climatology, and geophysics agency. Based on the study discussed in this paper, it can be concluded that the hybrid FFT-AR method offers a better performance than the other two methods. The hybrid FFT-AR method produces a normalized root mean square error (NRMSE) value of 7.18% for all scenarios. The smallest NRMSE is 4.5%, which is produced in FFT (99) – AR (48) hybrid model and which is used for short-term forecasting (7-days forecast period).