This study addresses the challenges of insufficient data privacy protection and inefficient computational resource scheduling in multi-party collaboration scenarios. We propose a blockchain-integrated trusted computing sandbox task scheduling framework that incorporates deep reinforcement learning. By synergizing the immutability of blockchain with the privacy-preserving capabilities of trusted computing sandboxes, we establish a multi-layer collaborative architecture comprising user layer, blockchain layer, scheduling decision layer, and computational execution layer. A novel deep reinforcement learning algorithm (PPDAC) is developed using Proximal Policy Optimization with a dual-critic mechanism, integrating a multi-objective reward function that simultaneously considers task priority, resource compatibility, and load balancing. The framework achieves dynamic optimization of task-to-sandbox resource matching through adaptive policy iteration. Simulation experiments demonstrate that compared to conventional genetic algorithms and Deep Q-Network approaches, the PPDAC algorithm proposed in this paper has better performance in indicators such as average task completion time, resource utilization, and load balancing, verifying the feasibility of this framework in effectively improving scheduling efficiency while ensuring data privacy and security. The proposed provides a novel technical pathway for privacy-preserving intelligent scheduling in multi-party collaborative applications, particularly in energy data sharing and medical collaborative computing scenarios.