Construction and Application of Power Industry Knowledge Graphs

The Artificial Intelligence Application Institute, established in 2018 within the China Electric Power Research Institute, focuses on five research directions—intelligent perception, big data, platform technology, intelligent cognition, and AI applications—supporting AI research and deployment across the power sector.

Traditional power knowledge engineering relied on expert systems, but the growing complexity and openness of power data demand a shift toward knowledge graphs that can structurally represent concepts, entities, events, and their relationships, enabling more effective semantic integration.

Power domain knowledge representation distinguishes between generic knowledge (e.g., equipment names, voltage levels) and specialized knowledge (e.g., customer billing details), with a business layer capturing actions such as maintenance procedures, allowing the mapping of physical grid topology to graph structures.

Ontology construction combines top‑down expert‑driven modeling with bottom‑up data‑driven extraction, often using a hybrid approach to incorporate both structured database schemas and unstructured textual sources, resulting in comprehensive domain ontologies.

The proposed NoDKG architecture comprises four layers: data acquisition (handling structured, semi‑structured, and unstructured data), graph construction (leveraging graph databases like Neo4j and relational databases for multimedia), knowledge computation (embedding, reasoning, and path analysis), and graph application (intelligent search, QA, recommendation, and decision support).

Practical implementations include power dispatch fault handling, operation‑inspection work‑order processing, and intelligent customer service QA, each integrating domain-specific ontologies with NLP models such as Bi‑LSTM‑CRF for entity and relation extraction.

Future challenges involve extracting knowledge from heterogeneous noisy data, efficient reasoning over grid topology, establishing robust quality evaluation metrics for domain graphs, and developing scalable application frameworks to support emerging digital service providers in the power ecosystem.

In summary, power sector knowledge graphs aim to structurally capture the complex entities and relationships of the electric grid, providing a foundation for semantic search, reasoning, and AI‑driven decision making, while acknowledging that they are a complementary tool rather than a universal solution.