Forecasting and optimizing residential EV flexibility for the Swedish mFRR market using machine learning

The growing adoption of electric vehicles (EVs) presents both opportunities and challenges for electricity systems, particularly in balancing supply and demand through ancillary service markets. In Sweden, plug-in EVs accounted for over 58% of new car registrations in 2024, making grid stability increasingly important. This research investigates the potential of leveraging residential EV flexibility to participate in the Manual Frequency Restoration Reserve (mFRR) market. Using real-world charging data from a sample of 3127 EV chargeboxes from Greenely AB and market data from the Swedish transmission system operator, Svenska kraftnät (SvK), a machine learning (ML) framework is developed to forecast EV availability, defined as the aggregated minimum hourly charging demand. The study compares Long Short-Term Memory (LSTM), eXtreme Gradient Boosting (XGBoost), and Seasonal ARIMA (SARIMA) models, with XGBoost achieving the highest accuracy. These forecasts feed into a Linear Programming Optimization model designed to maximize household revenue by shifting charging to periods with favorable mFRR capacity and mFRR prices, while meeting market, technical, and regulatory constraints. Results show that bi-directional smart charging significantly improves the economic feasibility of mFRR participation, with the most effective scenario increasing net revenue by 45.5% over the baseline. The study also identifies feasible bidding hours, regulatory limitations, and strategies for electricity aggregators, forecasting up to €140,061 in annual gross earnings. By combining ML-based forecasting with the optimization model, this research addresses a key gap in the literature and offers practical insight for EV usage, market bidding strategies, and energy aggregator business models. • XGBoost outperforms other models in predicting EV availability using real-world data. • Smart charging enables mFRR participation by strategically shifting EV loads. • Linear Programming optimizes net revenue by solving for expected costs and earnings. • Smart charging shifting increases net revenue by up to 45.5% vs baseline scenario. • Multi-market strategies boost expected revenue by up to 63.1% vs baseline scenario.