A methodology to improve the predictability of solar energy generation with confirmatory evidence from Germany

Abstract

Using data from Germany's 50Hertz electric power system, this paper begins by observing that the current intraday solar energy forecasts do not fully reflect the day-ahead weather forecasts' information. Solar energy generation also has the property of being highly autoregressive. Based on those properties, an Autoregressive Moving Average with Exogenous Inputs (ARMAX) model is formulated. The model was estimated using 15-minute data from Jan 1, 2014, through Dec 31, 2017. The model is evaluated using out-of-sample data from Jan 1, 2018, to Aug 30, 2020. The 15 minute-ahead out-of-sample predictions have a weighted-mean-absolute-percentage-error (WMAPE) of about 3.84%, substantially less than the WMAPE associated with the intraday solar energy forecasts reported by the system operator over the same period.