Robust Optimization and DRL-Based Control Strategies for Electric Vehicle Aggregators to Provide Frequency Regulation Service

Electric vehicles (EVs) have rapid and precise dynamic response capabilities, massive aggregated EVs can provide valuable frequency regulation ancillary services. This paper first proposes a novel EV feasible domain characterization model to extend sustained regulation capabilities. Next, in the day-ahead optimization phase, an aggregation method for the feasible domain of EV fleet is proposed to enable the treatment of the fleet as an individual entity, and a robust optimization model is developed to maximize the fleet revenue. Then, in the real-time power allocation phase, multiple agents with different structures are designed to handle the uncertainty of EV numbers, and a power allocation model based on twin delayed deep deterministic policy gradient is proposed to efficiently allocate power from the fleet to individual EVs. Experiments show that the optimization model has significant economic benefits, and the power allocation model boasts faster decision-making speeds while maintaining commendable response accuracy compared to other methods.