A Method for Multimodal Information Extraction and Knowledge Graph Construction in Substation Secondary System

Area of Science:

• Electrical Engineering
• Computer Science
• Data Science

Background:

• Substation secondary systems generate high-entropy, disordered multi-source heterogeneous data.
• Challenges exist in cross-modal information integration and efficient retrieval.
• Structured processing is needed for effective data management.

Purpose of the Study:

• To propose a method for multimodal information extraction and knowledge graph construction.
• To enable structured processing of heterogeneous data from multiple sources.
• To support information management and decision-making in complex industrial systems.

Main Methods:

• Image data processed using YOLOv8n and Optical Character Recognition (OCR) for positional and semantic extraction.
• Heuristic Circular Stepping Search Algorithm (HCSA) for deterministic path tracing of information flows.
• RoFormer-BiLSTM-CRF model with Rotary Position Embedding (RoPE) for accurate text entity and relationship extraction.
• Domain ontology mapping and string similarity for aligning device entities across modalities.
• Construction of a structured knowledge graph from scattered data.

Main Results:

• Effective extraction of multimodal information from substation secondary systems.
• Successful construction of a structured knowledge graph.
• Demonstrated value in supporting information management and decision-making.
• High-accuracy extraction of entities and relationships from text data.
• Mitigation of effects from uncertain topology and noise interference.

Conclusions:

• The proposed method effectively addresses challenges in processing heterogeneous substation data.
• The knowledge graph construction facilitates structured data utilization.
• The approach provides valuable support for complex industrial system management.