Abstract: The integration of large-scale wind and solar renewable energy generation into the power grid introduces significant uncertainty, which makes the global optimization decision of the system face challenges. This paper proposes a design method of cooperative optimal scheduling strategy of source-grid-load-storage based on generative adversarial network (GAN) modification. Firstly, considering the operation characteristics of various adjustable resources in the new power system, a cooperative optimal scheduling model of source-grid-load-storage based on proximal policy optimization (PPO) algorithm is constructed. Secondly, the GAN is introduced to modify the advantage function of the PPO algorithm, which reduced the variance of the value function and improved the efficiency of agent exploration. Then, the discriminator in the GAN is combined with the expert strategy to guide the generator to generate the scheduling strategy. Finally, discriminator constantly confronted the generator to find the Nash equilibrium point, and got the optimal scheduling strategy. The analysis of an example shows that the intra-day optimal scheduling strategy designed in this paper adopts the proximal policy optimization based on correction of the generative adversarial networks algorithm, which shortens the convergence time of the training process compared with the traditional PPO algorithm, and improves the absorptive capacity of renewable energy through online control.