An integrated approach for evaluation, optimization and prediction of energy flexibility potential in the commercial building sector

• A methodology for assessing and managing energy flexibility in the commercial building sector. • Disaggregation of the electricity load, based on real-time data integrated with information gathered from interviews with technical staff in the building. • A deterministic optimization model that can simulate scenarios involving increases or reductions in flexible loads. • An artificial neural network model that integrates a feasibility classifier to assess whether load adjustments are possible and a regressor to provide quantitative estimates of the resulting load variations. • Integration of deterministic optimization with predictive models can foster the active inclusion of commercial consumers in flexibility services, contributing to the stability and efficiency of smart grids. With increasing reliance on renewable sources, harnessing energy flexibility is a key to managing intermittent generation, ensuring reliability, and optimizing energy use. This work presents the development and implementation of a methodology for assessing and managing energy flexibility in the commercial sector, with a particular focus on a supermarket located in Valencia (Spain). The research involved an initial analysis and disaggregation of the overall electricity load, based on real-time data, and integrated with information gathered from technical literature and interviews with technical staff at the analyzed facilities. An optimization tool, OpTool , was subsequently developed—a deterministic optimization model implemented in Python using linear programming, that can simulate scenarios involving increases or reductions in flexible loads, including low- and medium-temperature refrigeration, heating, and cooling systems, while fully respecting operational constraints and temporal compensation logic. The results showed that the greatest availability of flexibility is concentrated on weekdays and the summer months, with significant variations in the system response. In the case of the supermarket analyzed, flexibility is significantly higher on weekdays than on weekends: approximately 50–52% of scenarios are feasible from Monday to Friday, compared to only 41% between Saturday and Sunday. Seasonally, the best performance is seen in July, with a share of positive scenarios around 74%, while in October the percentage drops to approximately 15%. In addition, to extend the predictive capabilities, PredicTool was developed as an artificial neural network model that integrates a feasibility classifier to assess whether load adjustments are possible and a regressor to provide quantitative estimates of the resulting load variations. Trained on a dataset of 200,000 simulations generated by OpTool , the classifier achieved an accuracy of 87%, while the regressor showed promising potential for refining hourly forecasts. The overall results demonstrate that the integration of deterministic optimization with predictive models can foster the active inclusion of commercial consumers in flexibility services, contributing to the stability, efficiency, and sustainability of smart grids.