Abstract: In response to the multi-objective collaborative optimization challenges faced by data center distribution systems in the context of high proportion access of distributed power sources, such as operational economy, renewable energy consumption, and safety and stability, research is conducted on intelligent scheduling based on deep deterministic policy gradients. By constructing an optimization objective function that integrates operating costs, energy abandonment penalties, and safety constraints, and designing a deep reinforcement learning intelligent agent that combines priority experience replay, autonomous collaboration and real-time decision-making of gas turbines, energy storage systems, flexible loads, and external power grids in the continuous action space within the data center can be achieved. Through comparative analysis of optimization on typical days, it is shown that this method can effectively reduce the overall operating cost of the system, significantly improve the photovoltaic consumption rate, and ensure safety constraints such as voltage, providing an effective intelligent solution for the transformation of high energy consuming data centers towards green, economic, and elastic operation.