Short-term imbalance forecasting with machine learning for proactive mFRR controlling

The main topic of the work is responding to the system operation challenges caused by the intensive spread of renewable energy sources through the development of frequency controlling; essentially examining the possibility of a change of approach, transforming the reactive mFRR activation into proactive, which allows for economic and other system optimization considerations in addition to the existing activation principles (low available revolving reserve, power plant outage). This proactive behavior requires an imbalance forecast, the exploratory analysis of which forms the content of the document.A test environment is presented in which the possible mFRR services are simulated using the product definitions according to the MARI platform. The test is based on Light Gradient Boosting quantile regression imbalance forecasting using different confidence levels, which are then converted into standard products based on an activation protocol. The key performance indices are the proactively activated mFRR energy during the test period, its ratio to total imbalance, and the proportion of counter-activation caused. These results are presented in the study with different confidence levels and activation criteria.