Hybrid retrieval-augmented generation method for emergency decision-making in carrier-based aircraft launch and recovery

Objectives Carrier-based aircraft launch and recovery operations are central to the combat capability of aircraft carriers. Their complexity and high risk make emergency decision-making a critical factor in ensuring flight safety. Methods To address the scarcity of special-condition cases and the difficulty of accumulating decision-making experience in carrier-based aircraft launch and recovery operations, this paper proposes an intelligent decision-support method based on hybrid retrieval-augmented generation. First, a domain knowledge graph for carrier aviation is constructed based on authoritative professional books and relevant Wikipedia entries, forming a structured knowledge representation containing 1,284 entities, 1,879 relations, and 57 thematic communities. Second, vector retrieval and graph retrieval are integrated, and reciprocal rank fusion is used to combine dual-channel evidence. Evidence deduplication, filtering, and structured presentation are further employed to improve retrieval quality. Finally, an evaluation benchmark containing 23 simulated scenarios is constructed, and domain experts are invited to conduct scoring-based evaluation. Results Experimental results show that the proposed method outperforms baseline methods across all evaluation dimensions, achieving an 8% improvement over the base model and validating the effectiveness of hybrid retrieval-augmented generation for decision support in the specialized field of carrier aviation. Conclusions This study provides a generalizable technical approach for intelligent decision support in knowledge-intensive and high-risk scenarios.