Day-ahead solar forecasting using time series stationarization and feed-forward neural network

Abstract

Solar forecasting is a pivotal factor in a viable solar energy deployment to support reliable and cost-effective grid operation and control. This paper proposes a new approach to overcome one of the most significant challenges in solar generation forecasting, i.e., the limited availability of the stationary data sets. This challenge is addressed by converting the non-stationary historical solar irradiance data into a stationary set, which will be further validated using an ADF test. This conversion will be followed by a neural network-based forecasting and proper post-processing steps. Numerical simulations exhibit the performance of the proposed method, which has achieved a mean absolute percentage error (MAPE) of less than 1% under different weather conditions.