Abstract: The large-scale access to new energy with strong random characteristics will bring strong random disturbance to the power grid. The traditional control methods can not effectively solve the problems of frequency instability and worse control performance standards caused by a strong random disturbance in the distributed power grid mode. From the point of secondary frequency modulation, this paper proposes a multi-agent cooperative control algorithm for distributed multi-area interconnected power grid, i.e. over-relaxation double Q learning algorithm to obtain multi-area cooperation control. The proposed algorithm introduces an over-relaxation factor based on fast Q-learning ω. To accelerate the calculation of the optimal value function, at the same time, the double Q learning strategy is introduced to solve the problem of overestimation of the active exploration value in the reinforcement learning of the Q algorithm system, so as to improve the update efficiency and convergence performance of the algorithm. Through the simulation of the improved IEEE standard two-area load frequency control model and Yunnan interconnected power grid model, the proposed algorithm shows better control performance and convergence speed.