Online Self-Distillation for Electric Load Demand Forecasting on Greek Energy Market

Even though Knowledge Distillation (KD) has been extensively studied during the recent years considering classification tasks, the research considering forecasting problems is extremely limited, despite the fact that the underlying reasons for the need of KD, linked with requirement for effective and fast models, are also apparent in such problems. In this work, we propose an online distillation method, named Online Self-Distillation for Forecasting (OSDF) for ameliorating the baseline forecasting performance considering the Electric Load Demand Forecasting (ELDF) problem on Greek energy market. The experimental evaluation, considering a typical and a fully realistic setup validates the effectiveness of the proposed OSDF method.