Uncertain FlexOffers: a scalable, uncertainty-aware model for energy flexibility

As the usage of Renewable Energy Sources (RES) in electricity grids increases in popularity, energy flexibility has a crucial role. The most common weaknesses of current flexibility models are: i) being hard-coded for specific devices, ii) not scaling for long time horizons and many devices, iii) losing a lot of flexibility if the model is approximated, and iv) not considering the uncertainty affecting flexibility representations, which causes the model to capture too much excess flexibility when imbalance penalties are high. The FlexOffer (FO) model can perform approximations of flexibility with good accuracy across different devices, and scales well to long time horizons and many devices: this work extends FOs to uncertain FOs (UFOs), which keep the good properties while capturing uncertainty. We show that UFOs are very fast by performing optimization in under 5.27 seconds for a 24 hours time horizon, while exact models use more than 29.05 hours for even a 6 hours 15 minutes time horizon, making them totally infeasible in practice. UFOs can capture more flexibility than other uncertain models: UFOs considering energy dependencies can model flexibility without losses for a charging battery, and retain 86.8% of the total flexibility for batteries and 87.5% for EVs when imbalance penalties are high, compared to 79.6% and 74.4% respectively for other models. UFOs allow to aggregate up to 6000 loads for up to 96 time units while retaining 90.5% of the total flexibility: exact models fail already for 330 loads or 21 time units.