Research on Multi-Service Data Fusion Mechanism and Dynamic Sharing Framework of County-Level Distribution Network Based on Knowledge Graph

To address the challenges of cross-service data integration in power systems and the inefficiency of dispatching decisions caused by evaluation metrics failing to reflect real-time operational status, this study proposes a knowledge graph (KG) construction method tailored for county-level distribution networks. The core steps include: first preprocessing power data to eliminate outliers and noise, then implementing multimodal data feature fusion through attention mechanisms; secondly applying the BERT-BiLSTM-CRF model for entity recognition (ER) and relationship extraction (RE); finally constructing the knowledge graph using Neo4j and achieving real-time updates via an event-driven architecture (EDA). Experimental results demonstrate: entity recognition accuracy reaches $93.2 \%$, query response time is reduced to 65 milliseconds, and inference accuracy remains at $\mathbf{9 3 \%}$.