Machine learning based forecasting of EV charging load in a parking lot for optimal participation in frequency services

This work assesses the potential of electric vehicles participating in frequency services in the Nordics. For this, data from a workspace parking lot is used to create artificial load profiles to take the perspective from an aggregator. The study is then divided into two parts: Firstly, a machine learning model is developed to forecast the parking lot load. In a second step, the predictions are given to a rolling-horizon mixed integer linear program that optimally allocates the capacities to Frequency Containment Reserve services. It is found that the machine learning approach almost doubles the profitability compared to offering bids just based on historical values. Finally, a hypothetical market structure is considered, where the FCR-D late auction is moved to an hour-ahead intra-day auction. The analysis shows that the opportunity to correct bids intra-day improves participation in frequency services and triples profits compared to the day-ahead auction.