Energy-Efficient Short-Term Load Forecasting for Responsive Demand-Side Analytics for the European Electricity Market

Maintaining a balance between electricity generation and consumption is essential for reliable power grids.On the European energy market, Balance Responsible Parties (BRPs) rely heavily on accurate Short-Term Load Forecasting (STLF) to manage grid stability and determine daily positions on the day-ahead and intraday energy markets.This paper addresses the need for a lightweight, energy-efficient STLF solution that can be deployed on resourceconstrained devices, such as analysts' laptops, to ensure ease of use and minimal deployment complexity.We propose the Dynamic Multi-Domain (DMD) forecasting framework, and integrate it with a range of widely used machine learning models, covering treebased ensembles and Deep Learning models.The proposed DMD framework provides accurate forecasts across multiple, correlated domains, selecting the domain to forecast during inference rather than training.To improve speed and reduce resource consumption, we propose a SHAP-Based Forward Feature Selection (SFFS) algorithm based on Shapley values, which also supports explainable forecast models.We perform an extensive experimental evaluation on several real-world load datasets which shows that our DMD-XGBoost model outperforms alternative DMD models, with up to 78% and 79%, 76%, 69.3% MAPE improvements, respectively, while completing training 3.7-15x faster than its closest competitor in terms of accuracy on multi-domain datasets.Our SFFS algorithm speeds up training by a further 1.6-8.2x,making DMD-XGBoost even more efficient and furthermore makes the model explainable.