Objective To address the challenges in multi-sortie carrier-based aircraft support scheduling, such as complex multi-type object relationships, frequent resource competition, and tightly coupled task dependencies, this study aims to develop a scheduling approach that balances cross-sortie resource coordination with local task optimization, while enhancing rapid adaptability and generalization capability in dynamic combat environments.

Method A meta-learning-enhanced heterogeneous graph scheduling method (Meta-HGS) is proposed. A heterogeneous tripartite graph is constructed, comprising carrier-based aircraft, support tasks, and support stations. A heterogeneous graph attention network is employed to model node types and relation types differentially. Features are aggregated across sortie-level, task-level, and resource-level granularities, enabling unified optimization of cross-sortie resource competition and task temporal constraints. To further enhance adaptability under dynamic task distributions, a meta-learning mechanism is incorporated. This mechanism consists of a Meta-Critic network for cross-task value evaluation and a Task−Actor encoding network for extracting task-specific policy representations. Through inner- and outer-loop parameter updates, the Meta-HGS framework achieves rapid policy adaptation and fast convergence across different tasks. Additionally, the HGAT explicitly models heterogeneous node interactions through meta-path-based neighbor aggregation, thereby preserving the semantic information of task-task dependencies and task-station assignments. This integrated approach allows the model to handle complex multi-type object relationships, frequent resource competition, and tightly coupled task dependencies, ensuring stable and efficient scheduling across diverse and dynamic operational scenarios.

Results Across three simulation scenarios of different scales, Meta-HGS reduced the average completion time by approximately 5.4% compared to GA, PSO, DQN, and G-A2C, while maintaining advantages in responsiveness and solution accuracy. The average completion times for small-, medium-, and large-scale scenarios were 42.1, 68.4, and 95.3, respectively, outperforming other methods by 1.6%–12.3%. Average response time (ART) and scheduling time (AST) remained the lowest, with ART at 45.3, 78.5, 156.2, and AST at 128.5, 256.3, 512.8, exceeding learning-based methods. The performance gap relative to the optimal solutions obtained using OR-Tools was only 1.5%–3.1%, demonstrating high precision and stability. Although memory consumption was slightly higher than that of heuristic methods, it remained comparable to DQN and G-A2C, supporting the overall performance gains. These results indicate that Meta-HGS balances efficiency, real-time performance, and accuracy, providing an effective and robust solution for multi-sortie carrier-based aircraft support scheduling.

Conclusion The Meta-HGS-based scheduling method effectively captures multi-granularity heterogeneous relationships, significantly improves the efficiency and timeliness of multi-sortie carrier-based aircraft support scheduling, and demonstrates strong task transferability and environmental adaptability. This approach provides a generalizable technical framework for intelligent scheduling in highly dynamic and strongly coupled support scenarios.