Novel Attention Feature Modeling for Knowledge Fusion in Power Systems Based on Hierarchical Embeddings

Le résumé fourni par la source

As the digitalization and intelligent transformation of power systems continues to advance, power knowledge exhibits characteristics such as multi-source heterogeneity, semantic complexity, and structural coupling. Traditional models struggle to balance accuracy and generalization in knowledge fusion and semantic representation. To address this challenge, this paper proposes a Hierarchical Embedding-Based Attention Framework (HEAF) for power system knowledge fusion. This model constructs a multi-layer semantic embedding framework, categorizing power knowledge into entity, relationship, and scenario layers to achieve hierarchical semantic representation. It introduces a cross-layer attention mechanism to dynamically allocate feature weights, thereby strengthening structural associations and semantic consistency. Experiments conducted on multi-source power knowledge datasets encompassing dispatch, operation and maintenance, and safety supervision data demonstrate that the proposed model outperforms mainstream methods in accuracy, semantic consistency, and robustness, achieving an overall performance improvement of approximately 6%–10%. The findings indicate that integrating hierarchical embedding with attention mechanisms effectively enhances power knowledge understanding and intelligent reasoning capabilities, providing novel technical pathways and theoretical support for operational analysis, risk early warning, and knowledge-driven decision-making in power systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.