The large-scale integration of renewable energy sources and electric vehicle clusters has introduced strong stochastic disturbances

which amplify frequency oscillations and challenge the performance of automatic generation control(AGC). However

existing reinforcement learning approaches exhibit low efficiency in utilizing experience samples

which hinders their ability to accurately reflect the dynamic state changes of the AGC system

thereby leading to suboptimal control strategies. To overcome these limitations

this paper proposes a twin delayed deep deterministic policy gradient algorithm for AGC

incorporating brain-inspired memory cooperative experience replay. By integrating short-term

long-term

and expert experience samples

it cooperatively utilizes and interactively samples multi-source experiences

enabling the agent to consider real-time feedback

historical data

and external guidance for more accurate state reflection and optimal control. The simulation results of both the two-area load frequency control model with electric vehicle cluster access and the four-area model of the Central China power grid indicate that the proposed algorithm can effectively suppress frequency oscillations and improve the performance of AGC.