A multi-layer intelligent control strategy for multi-regional power system with electric vehicles: A deep reinforcement learning approach

The operating status, control resources, and accidental events within the power system exhibit significant uncertainty, and the integration of cluster electric vehicles (EVs), characterized by high permeability. These introduce both opportunities and challenges to power system regulation. This paper proposes a multi-layer intelligent control strategy for multi-regional power generation that incorporates the participation of EVs. Initially, a multi-regional interconnected control structure is designed, predicated on the frequency modulation ancillary service market. The upper-level model predictive controller rapidly issues total power generation commands based on regional control deviations, while a lower-level deep reinforcement learning controller executes comprehensive control. This control considers system frequency modulation mileage, regional control deviation, and the demand loss of EV users as objectives. Subsequently, an autonomous decision-making model for EV owners' charging modes is established. Based on this, charging and discharging participation coefficients for EVs are formulated, and the corresponding classification of EVs is conducted, delineating the regulatory margin boundary for individual EVs. Utilizing Minkowski sum theory, the stochastic boundary of the regulation margin for an EV charging station (CS) cluster is calculated. Consequently, the adjustable capacity range for future CSs can be determined in the current stage. Finally, the multi-experience replay pool theory is applied to enhance the MATD3 algorithm. Simulation results demonstrate that the proposed strategy can effectively reduce regional control deviations and control costs, and achieve more intelligent CSs management.