With the continuous evolution of new generation information and communication technologies such as 5G, cloud computing, and artificial intelligence, the world is rapidly entering the fast lane of the digital economy. A dual-layer optimization scheduling method for data centers based on multi-agent proximal strategy network is proposed to address the uncertainty of renewable energy and workload prediction in data centers. Firstly, a dual-layer spatiotemporal optimization scheduling framework for data centers is established, which provides detailed modeling of data center workloads, IT equipment, and air conditioning equipment; On this basis, a dual-layer optimization scheduling model for data centers is proposed. The upper layer schedules the time dimension with the goal of minimizing the total operating cost of IDC operators, while the lower layer schedules the space dimension with the goal of minimizing the operating cost of each IDC. Then, the principle of multi-agent proximal strategy network algorithm is introduced, and the state space, action space, and reward function of the dual-layer optimization scheduling model for data centers are designed. Finally, offline training and online scheduling decisions are conducted for the examples. Simulation results show that the proposed model and method can effectively reduce system costs and energy consumption, achieve optimal workload allocation, and have good economy and robustness.