Multi large language model collaboration framework for few-shot link prediction in evolutionary fault diagnosis event graphs

Fault-tolerant control is crucial for ensuring flight safety in aircraft. However, existing methods for fault diagnosis in nonlinear systems face challenges such as data sparsity, limited generalization, and lack of explainability. To address these challenges, this paper proposes a multi-large language model (LLM) collaboration framework for few-shot link prediction in evolutionary fault diagnosis event graphs. The framework consists of two modules: the Clustering Language Model (LMc) and the Prediction Language Model (LMP). LMc utilizes the semantic understanding capabilities of LLMs to cluster entities and decompose large-scale graph data into smaller subgraphs, mitigating the impact of data sparsity on link prediction. LMP leverages the reasoning capabilities of LLMs to perform link prediction within each subgraph and fuses the prediction results to enhance accuracy and generalization. The completion of the link serves as a means to an end, which is to conduct fault diagnosis reasoning on a more detailed knowledge graph, thereby significantly improving the accuracy of fault diagnosis. Experimental results demonstrate that the proposed framework outperforms traditional embedding models and existing meta-learning methods on multiple datasets, particularly for sparse and background-rich datasets. This approach offers a novel solution for fault diagnosis in nonlinear systems, with significant theoretical and practical value.