关键词: 知识图谱, 大语言模型, 决策融合, 人机交互, 航空导航

Abstract: Traditional airborne navigation information fusion methods fail to leverage unstructured knowledge, parse semantics of complex scenarios and support dynamic decision-making. This paper takes large language models as the core of decision reasoning and constructs an intelligent cognitive hub for pilots’ real-time in-flight decision-making. A closed-loop "Knowledge-Perception-Decision" intelligent navigation decision framework is established. Relying on large language models, the system realizes fusion reasoning and intelligent decision-making of multi-source navigation sensor data, overcoming the defect that traditional methods only support offline post-event review. Reasoning is activated only by pilots’ requests without continuous computing overhead. Airborne data is transmitted intermittently, and the end-to-end decision latency is controlled within 1.5 seconds to meet the strict real-time requirements of aviation decision-making.To maximize the reasoning capability of large language models, knowledge graph technology is adopted for structured representation and fast retrieval of navigation knowledge. Graph databases support visual knowledge management and standard interfaces to underpin intelligent decision-making. A multi-dimensional image recognition model is developed to identify weather conditions, geographical environments and flight phases. A multi-sensor data alignment and state perception module is designed to assess sensor availability, classify data precision and assign dynamic weights, ensuring reliable and timely fused data.Experiments demonstrate that the proposed knowledge graph, image recognition and sensor alignment modules enable decision-level fusion of navigation information with outstanding overall performance, significantly boosting cognitive decision accuracy. Deriving interpretable and highly reliable decision strategies from multi-source heterogeneous navigation data, the system improves the robustness and adaptability of navigation decisions in complex airspace, and provides theoretical and engineering references for intelligent airborne navigation.

Key words: Knowledge graph, Large Language Models, Decision Fusion, Human-Computer Interaction, Airborne Navigation