Construction and application of integrated knowledge graph for mine disasters

In order to achieve a novel disaster early warning mode of “autonomous modeling + integrated early warning + root cause tracing”, and improve the knowledge engineering infrastructure for integrated intelligent disaster warning, this study developed an ontology model for integrated knowledge graphs of disaster based on the “human-machine-environment” system engineering theory. The model was constructed through a top-down approach across four dimensions: temporal, spatial, managerial, and process mechanisms. It encompasses eight core concepts in this field of personnel, equipment, environment, region, process, document, index, and disaster, along with three categories of data attributes (basic information, spatial information, and temporal information). Three types of relationships were established: spatial positioning, numerical correlations, and process mechanism linkages. A hybrid data storage architecture integrating relational, spatial, temporal, and graph databases was built. Data extraction for entities, attributes, and relationships was achieved through a combination of rule-driven workflow engines and manual data supplementation, forming mine-specific disaster-integrated knowledge graphs. By adopting graph database relationship reasoning methods, coupled with anomaly identification criteria for graph objects, root cause analysis of mine disasters was realized. The results show that a “top-down” knowledge graph construction scheme that involves expert modeling and regularized data extraction is suitable in the early stage in the field of integrated analysis of mine disasters; the framework of “eight core concepts, three data attributes, and three relationship types” significantly enriches disaster knowledge systems; the hybrid methodology of rule-driven engines and manual supplementation effectively addresses the needs of mines at varying intelligentization stages; knowledge graph relationship reasoning integrated with object attribute anomaly detection provides a robust technical solution for accident causation analysis.