24th EANN 2023, 14 - 17 June 2023, León, Spain

Residual Error Learning for Electricity Demand Forecasting

Achilleas Andronikos, Maria Tzelepi, Anastasios Tefas


  Electricity demand forecasting describes the challenging task of predicting the electricity demand by employing historical load data. In this paper, we propose a novel method, named RESidual Error Learning for Forecasting (RESELF) for improving the performance of a deep learning model towards the electricity demand forecasting task. The proposed method proposes to train a model with the actual load values and compute the residual errors. Subsequently, RESELF proposes to train a second model using as targets the computed residual errors. Finally, the prediction of the proposed methodology is defined as the sum of the first model's and second model's predictions. We argue that if the errors are systematic, the proposed method will provide improved results. The experimental evaluation on four datasets validates the effectiveness of the proposed method in improving the forecasting performance.  

*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.