Forecasting andOptimizing Residential EVFlexibility for the SwedishmFRR Market UsingMachine Learning

Sammanfattning

The growing adoption of electric vehicles (EVs) presents both opportunities and challengesfor modern electricity systems, particularly in balancing supply and demand through ancillaryservice markets. In Sweden, where plug-in EVs accounted for over 58% of new car registrationsin 2024 (EAFO, 2025), ensuring grid stability is increasingly critical. This thesis investigatesthe potential of leveraging residential EV flexibility to participate in the Manual FrequencyRestoration Reserve (mFRR) market. By utilizing historical charging data from Greenely, an energy-tech company, and market datafrom Svenska kraftnät (SvK), a machine learning framework was developed to forecast EVavailability. The study compares the performance of LSTM, XGBoost, and SARIMA models inpredicting aggregated residential EV load, with XGBoost achieving the highest accuracy. Theseforecasts inform a linear programming optimization model designed to maximize householdrevenue by engaging in ”EV Flexibility”, which is the process of shifting charging (usingSmart charging controls) to periods with favorable mFRR capacity and activation prices, whileadhering to technical and market regulatory constraints. Results indicate that partial and bi-directional smart charging significantly improves theeconomic feasibility of mFRR participation for households, even under limited flexibilityassumptions, compared to a baseline case. The study further identifies feasible bidding hours,highlights challenges posed by regulatory thresholds, and suggests strategies for electricityaggregators like Greenely to optimize participation. By combining machine learning with a linear programming optimization model, this researchfills a critical gap in the literature on participation of residential EVs in frequency markets andprovides actionable insights for grid planning, market design, and energy-aggregator businessmodels.