A data-driven framework for estimating residential energy flexibility for aggregated demand-side management
Ioannis Papias, Vasilis Michalakopoulos, Elissaios Sarmas, Vangelis Marinakis, Gabriel Antonesi, Tudor Cioara, Ionuț Anghel
Abstract
The increasing penetration of renewable energy sources (RES) into power grids necessitates innovative strategies to manage their inherent variability. Demand-side energy flexibility has emerged as a solution to ensure grid stability and maximize RES integration. However, existing methods for flexibility quantification lack generalizability across different building types, making them costly to replicate and often struggle with accuracy. This study proposes an artificial intelligence (AI) and data-driven methodology to forecast the aggregated energy flexibility in residential buildings. The methodology integrates dynamic baseline calculations, statistical feature engineering, and bidirectional long short-term memory (BiLSTM) models to predict flexibility metrics and provide insights through energy flexibility indicators (EFIs). Two distinct datasets, a residential multi-apartment building in Austria and a single residential building in Ireland were used to validate the accuracy, scalability, and adaptability of the proposed method across diverse building typologies. The flexibility predictions were further classified into peak and off-peak intervals, enabling an understanding of the energy dynamics and supporting targeted demand response (DR) strategies. Results indicate a strong predictive performance for the energy flexibility of buildings, achieving values exceeding 0.85 for energy consumption and up to 0.97 for photovoltaic (PV) production, with classification accuracy reaching 92%.