With the rapid advancement of the industrial Internet and intelligent manufacturing technologies, cloud manufacturing has become increasingly significant as an advanced paradigm for achieving globally optimized resource allocation. However, previous scheduling models frequently fail to comprehensively address unexpected scenarios such as task conflicts, equipment maintenance interruptions, and dynamic resource contention, resulting in delayed system responses and imbalanced multi-objective optimization in practical applications. To overcome these challenges, this paper proposes an intelligent scheduling model based on a two-layer deep reinforcement learning approach. The first layer architecture autonomously selects optimization objective weights, such as prioritizing time efficiency or cost control. The second layer dynamically employs nine composite scheduling rules generated by combining task-priority strategies and resource allocation strategies and innovatively incorporates a multi-dimensional reward mechanism to coordinate the optimization process. Simulation experiments validate the effectiveness of the proposed method in handling multi-objective scheduling in complex environments.