Prediction of Solar Energy using Time Series Methods

The utilization of solar energy as a source of electricity is increasing day by day, raising interest in prediction of solar irradiation. A successful integration of solar energy sources with existing grid system is the biggest challenge due to volatile and unpredictable behaviour of solar energy. To date, several approaches are proposed to analyse and predict solar irradiation as well as to improve forecast accuracy. The present study concentrates on hourly to monthly forecasting of solar irradiation through various statistical methods, namely AR, MA, ARMA, ARIMA, and Holt Winter’s technique. From the decomposition of time series data, we found that the dataset exhibits seasonality and randomness. The adequacy of the models is assessed from the Root Mean Square Error (RMSE). We note that the model performance improves with the increase of time horizon (from hourly to monthly), probably due to enhanced clarity in seasonality. In case of ARIMA, the RMSE value turns out to be 124.21 in hourly forecasting, whereas this value reduces to 15.66 in monthly forecasting. A similar change has been observed for other models as well.