Summary
With the rapid evolution of new-type power systems, engineers struggle to retrieve scenariospecific specifications from massive unstructured standard documents due to vocabulary gaps and limited semantic understanding.
Read more Read lessThis study proposes an intelligent recommendation method for intelligent distribution grid standards based on standard modularization and multimodal fusion. A total of 186 core power standards are structurally decomposed to build a high-precision knowledge base. A three-path framework is designed: a structural path using Word2Vec and Graph Attention Networks
(GAT) to capture relationships among modules, a semantic path using Sentence-BERT
(SBERT) for text representation, and a keyword path using the Best Matching 25 (BM25) algorithm for accurate term recall. These heterogeneous features are fused through adaptive weighting.
Experimental results show a Top-20 recall of 0.904, outperforming single-model baselines.
The approach effectively reduces semantic drift and provides robust support for standards recommendation and the digitalization of intelligent distribution grids.
Additional informations
| Publication type | Session Materials |
|---|---|
| Reference | D2_11573_2026 |
| Publication year | |
| Publisher | CIGRE |
| Country | China, People's Republic of |
| Study committees | |
| File size | 2 MB |
| Price for non member | 30 € |
| Price for member | 30 € |
Authors
WANG Sining - State Grid INFO&TELECOM GROUP CO.,LTD, Beijing.; XIA Baobing - State Grid INFO&TELECOM GROUP CO.,LTD, Beijing.; HAN Yuxin - State Grid INFO&TELECOM GROUP CO.,LTD, Beijing.
Keywords
Intelligent distribution grid, Standards recommendation; Multimodal fusion; BM25; GAT
