' Carbon Peak, Carbon Neutralization ' accelerates the rapid development of new energy-based power systems. With the access large-scale new energy, it will make the control performance worse. In this paper, a novel EBQ(σ,λ) algorithm based on ensemble learning is proposed to obtain global optimal solution from the perspective of automatic generation control, which can improve the poor control performance standard (CPS) of the grid effectively. The proposed algorithm can not only solve the problem of traditional reinforcement learning estimation bias by reducing the mean squared error of Q value in the next state, but also the introduced sampling parameter σ can weigh between improving the efficiency and better training samples. Meanwhile, the application of eligibility trace can solve the problem of time credit. Ultimately, the proposed algorithm is simulated to be effective in the improved IEEE standard two-area LFC power system model and the Guangdong power grid model. The results show that compared with the traditional algorithm, the proposed method is characterized with exceptional CPS and less carbon emission.