Toward data-driven identification of essential factors causing seasonal change in daily electricity demand curves

The forecast of medium-/long-term electricity demand helps the system operators and energy suppliers to plan appropriate facilities and supply. In particular, information of daily electricity demand curve is necessary for several problems. Traditionally, a regression approach focusing on a few variables has been studied widely from the viewpoint of deriving possible scenarios. However, the prediction of demand had becoming difficult with these traditional frameworks, so that further detailed regression approaches considering seasonal factors among numerous variables have been studied. In this study, the authors propose an approach to identify the essential factors causing seasonal change in daily demand curve using seasonal models constructed based on machine learning techniques; in this scheme, the consistency of selected variables in seasonal models plays a key role for deriving interpretable results. This study introduces an approach to derive the minimal number of important variables for identification of essential factors causing seasonal change in demand.