Reducing Imbalance Energy by Spatial Balance Area Forecasting

Grid operators must forecast the loads in their spatial balance areas for adequate supply.Standard load profiles (SLPs) facilitate the forecast, but are often inaccurate.Forecasting the deviations between profile-estimate and actual consumption reduces the economic risk of high imbalance energy costs.Forecasting such deviations is challenging because of abrupt changes in the SLPs.The present paper takes on this task and compares several machine learning models, including XGBoost Regressors and a decomposition-based approach to account for the abrupt changes, exemplarily on the differential balancing group data of a German distribution system operator.All proposed methods are capable of producing forecasts that reduce the needed amount of imbalance energy compared to the status quo.Both the XGBoost Regressor and the decomposition-based approach can account for the abrupt changes, with the XGBoost Regressor providing the best overall forecasts and reducing the imbalance energy by 47 % on average in the test period compared to the status quo.