Forecasting day-ahead electric load demand on Greek Energy Market

Abstract

The task of Electric Load Demand Forecasting (ELDF) is a pivotal one for power systems, accompanied by many applications, e.g., power systems operations and planning. In this work, we deal with the problem of ELDF on Greek Energy Market. The objective of this work is two-fold. First, we aim to provide an evaluation study for selecting the optimal input features for training a day-ahead load forecasting model (24 hours of the next day), as well as an effective model architecture. Second, we aim to improve the baseline forecasting performance, proposing a regularization methodology. The experimental evaluation indicates the optimal input features and model for the ELDF task, while the effectiveness of the proposed regularization method is also validated.