Transformer substation network disconnection prediction via semantic reasoning with causal modeling

Reliable communication networks are indispensable for the stable operation of smart grids and substations. Currently, WAPI networks have been widely adopted in relevant scenarios. Nevertheless, WAPI networks are confronted with disconnection risks attributed to complex network topologies, dynamic traffic fluctuations, and external environmental disturbances. Most methods rely on correlation analysis and lack causal interpretability, which restricts their effectiveness in rootcause localization and preventive maintenance practices. To address the problem, we propose a disconnection prediction approach that integrates prompt-driven semantic reasoning with structured causal analysis. The approach constructs a causal event graph that models semantic, temporal, and topological dependencies across devices and alarm sequences after extracts heterogeneous information to unified event representation. Based on the established graph, an inference module combines causal path analysis, structural causal models, and counterfactual reasoning to assess the influence of events, predict emerging disconnection risks, and identify plausible root causes with coherent and interpretable justification. By tightly coupling semantic abstraction with causal reasoning, the proposed approach provides a proactive, explainable, and extensible mechanism for anticipating network disruptions and supporting informed maintenance decisions. Experiments demonstrate that the proposed approach improves prediction accuracy and interpretability, verifying its value for smart grid communication networks.