Integrated Underfrequency Load Shedding Strategy for Islanded Microgrids Integrating Multiclass Load-Related Factors

Reducing the decision response time of load shedding while considering the comprehensive value of load shedding is one of the main challenges faced in emergency control of islanded microgrids. However, the existing underfrequency load shedding strategies do not fully consider the multiple factors associated with the load, and load assessment and load shedding decision-making are separated; this results in a long response time for underfrequency load shedding decisions for islanded microgrids. Therefore, in this paper, an integrated underfrequency load shedding strategy for islanded microgrids is proposed, which integrates multiclass load-related factors. This strategy first constructs an integrated underfrequency load shedding model for islanded microgrids on the basis of multiclass load-related factors such as the load frequency regulation effect, load shedding cost, and three-phase system power unbalance degree. Then, the load shedding model is described as a Markov decision process (MDP), and the environment, action space, and reward function are defined considering the load shedding objectives and constraints of islanded microgrids. Finally, a novel twin delay deep deterministic policy gradient method with softmax and dual buffer replay (DBR-SD3) is developed to determine the optimal integrated underfrequency load shedding strategy. This approach integrates softmax and the dual buffer replay mechanism into twin delay deep deterministic policy gradient (TD3), which greatly improves the ability of the agent to learn the optimal load shedding strategy in a complex microgrid operating environment. The simulation results based on the improved IEEE 37-bus microgrid and IEEE 118-bus microgrid verify that the proposed integrated load shedding strategy can greatly reduce the decision response time, correct the three-phase power unbalance of the system while minimizing the load shedding cost, and restore the system frequency to a normal level more quickly. Moreover, even under strong noise interference, the proposed strategy can produce stable load shedding decisions and has strong robustness and adaptability.