Enhanced knowledge graph cascade learning model for cyber–physical systems

Abstract

Recently, the application prospects of knowledge graph technology in cyber–physical systems (CPS) have attracted considerable attention. However, knowledge graph data in various CPS domains are typically collected from sensors or through manual efforts, which inevitably results in incomplete and unreliable data, thereby impacting the performance of downstream task models. This issue is often overlooked in existing studies. This paper proposes an enhanced knowledge graph cascade learning model for CPS. The model performs cascaded and iterative learning of both graph structure and graph representation. By optimizing the graph structure and incorporating hierarchical learning of graph-structured information, the proposed model enhances feature propagation and aggregation during representation learning. Experiments show that our model achieves outstanding results: compared to the baseline models, our approach achieves an average improvement of 2.7% in accuracy on the node classification task and 1.35% in MRR on the link prediction task.