This study aims to optimize the multi-factory remanufacturing process by integrating product allocation, U-shaped disassembly scheduling, and drone delivery. Traditional methods struggle with high complexity and concurrency, prompting us to propose an improved Dueling DQN algorithm. This algorithm enhances learning efficiency through decoupled value-advantage estimation. The problem is divided into three distinct stages: (1) product-factory allocation, (2) disassembly scheduling under precedence constraints, and (3) drone route optimization, formulated as a Traveling Salesman Problem (TSP) with subtour elimination. A profit-maximizing model is developed to balance the revenue from recovered components against the costs of disassembly, operations, and transport. Experiments across six product complexity cases demonstrate that the Dueling DQN algorithm achieves near-optimal solutions, with superior convergence, profit, and stability compared to DQN. This framework addresses multi-factory collaboration, dynamic scheduling, and drone logistics, offering scalable solutions for industrial systems.