Abstract: Flexible manufacturing cells, as an important organizational form for realizing multi-variety, small-batch, mixed-line production, have been widely applied in the manufacturing of high-end components. However, in actual operation, the scheduling process faces challenges in multi-resource coordinated decision-making and dynamic response due to factors such as re-entrant processes, machine blocking, and the coupling of external buffering with cross-regional transportation. To address these issues, a multi-resource coordinated scheduling method considering external buffers and transportation is investigated. Firstly, the production system is modeled as a re-entrant blocking flow shop scheduling problem with a central buffer, and a unified disjunctive graph is constructed to integrate processing, buffering, and transportation processes. On this basis, the scheduling problem is formulated as a Markov Decision Process, and a deep reinforcement learning-based dynamic scheduling framework is proposed. A hybrid feature extraction model combining graph neural networks and long short-term memory (GNN-LSTM) is employed to capture the spatiotemporal characteristics of the production system. Finally, extensive experiments based on a representative industrial scenario are conducted. The results demonstrate that the proposed method consistently outperforms traditional heuristic rules under various operating conditions, effectively reducing the makespan while exhibiting strong robustness and real-time performance.