To prevent a sharp frequency drop due to post-fault power shortage during multi-microgrids islanded operation

it proposes an under-frequency load shedding strategy based on multi-agent deep reinforcement learning.Firstly

it models each microgrid as an agent and formulates a Markov decision process(MDP) using the source-load-storage(SLS) data after islanding.It comprehensively considers the load frequency regulation effect

the importance of the load and the cost of low-frequency load shedding actions.Secondly

the multi-agent system interacts with the environment to learn.The under-frequency load shedding strategy is trained offline and subsequently implemented online through real-time decision-making to execute load shedding actions.Finally

the effectiveness of the proposed strategy is validated using an improved IEEE 37 bus test system.The simulation results demonstrate that the proposed strategy can significantly shorten the decision response time

suppress frequency fluctuations

and accelerate the restoration of the system frequency to the normal operating range while minimizing the total cost of load shedding.