Research on Predicting Super-Relational Data Links for Mine Hoists Within Hyper-Relational Knowledge Graphs

Hyper-relational knowledge graphs can enhance the intelligence, efficiency, and reliability of industrial production by enabling equipment collaboration and optimizing supply chains. However, the construction of knowledge graphs in industrial fields faces significant challenges due to the complexity of hyper-relational data, the sparsity of industrial datasets, and limitations in existing link prediction methods, which struggle to capture the nuanced relationships and qualifiers often present in industrial scenarios. This paper proposes the HyLinker model, designed to improve the representation of entities and relations through modular components, including an entity neighbor aggregator, a relation qualifier aggregator, MoE-LSTM (Mixture of Experts Long Short-Term Memory), and a convolutional bidirectional interaction module. Experimental results demonstrate that the proposed method performs well on both public datasets and a self-constructed hoisting machine dataset. In the Mine Hoist Super-Relationship Dataset (MHSD-100), HyLinker outperforms the latest models, with improvements of 0.142 in MRR (Mean Reciprocal Rank) and 0.156 in Hit@1 (Hit Rate at Rank 1), effectively addressing the knowledge graph completion problem for hoisting machines and providing more accurate information for equipment maintenance and fault prediction. These results demonstrate the potential of HyLinker in overcoming current challenges and advancing the application of hyper-relational knowledge graphs in industrial contexts.