Research on Element Extraction and Knowledge Graph Construction Method Based on Transmission Network Infrastructure Documents

With the continuous advancement of digitalization in power systems, there is an urgent need for the structured management of the vast amount of engineering elements and relational information embedded in transmission network infrastructure documents. Given the complexity of semantics, diversity of formats, and density of domain-specific terminology in engineering documents, this paper proposes a method tailored for element extraction and knowledge graph construction in transmission network documentation. The proposed approach includes OCR parsing, logical structure recognition, ontology-driven element system construction, and a multi-channel fusion entity extraction mechanism. Additionally, a terminology knowledge enhancement layer is introduced to improve deep models' understanding of domain-specific texts. The integration of SpanBERT and graph neural networks enables the capture of semantic relationships across paragraphs, thereby constructing a semantically coherent engineering knowledge graph. Experimental results show that this method significantly outperforms traditional approaches in element recognition accuracy and relation extraction coverage, effectively enhancing knowledge management and intelligent application capabilities for engineering documents.