Frequency Regulation Provision by EV Aggregator With Dynamic Balance Between User Charging Anxiety and Aggregate Flexibility

Aggregating a large number of electric vehicles (EVs) can provide considerable capacity for frequency regulation (FR) services. However, disaggregating real-time FR commands to individual EVs results in both immediate impacts to EVs and cascading impacts on future FR bidding. It is crucial to manage the trade-off between these two impacts during disaggregation. This paper proposes a deep reinforcement learning (DRL)-based FR provision method that balances user charging anxiety and aggregate flexibility in real-time disaggregation. The immediate and cascading impacts of disaggregation are characterized in two objectives: 1) user compensation costs related to charging anxiety, and 2) the preservation of EV aggregate flexibility. Incorporating these two objectives, a multi-objective disaggregation optimization problem is formulated with a learnable soft switching factor to balance these objectives. To determine the optimal soft switching factor and achieve coordination between bidding and disaggregation, the DRL algorithm is employed to jointly optimize market bidding decisions and the soft switching factor. Additionally, a consistency-based solution is derived from the disaggregation problem to alleviate computational burdens and protect user privacy. Numerical simulations verify that the proposed method achieves at least 21.2% higher profits for EV aggregators compared to single-objective methods while maintaining low computational complexity.