A method and system for evaluating the innovativeness of power grid technology project requirements
By employing a self-supervised learning and hybrid retrieval mechanism based on a dual-tower structure, the problems of cross-modal semantic bias and single feature in the demand assessment of power grid technology projects are solved. This enables more precise and interpretable multi-dimensional innovative assessments, improving the accuracy and traceability of the assessments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INNOVATION & INNOVATION CENT OF STATE GRID ZHEJIANG ELECTRIC POWER CO LTD
- Filing Date
- 2026-02-10
- Publication Date
- 2026-06-09
AI Technical Summary
Existing technologies for evaluating the innovativeness of power grid technology projects suffer from problems such as isolated single-modal information, cross-style semantic bias, single features, and insufficient interpretability, making it difficult to achieve accurate and comprehensive evaluation.
A self-supervised learning method with a dual-tower structure is adopted. By cross-modal semantic alignment of knowledge graph vector representation with text vectors of standardized documents and demand documents, and a hybrid retrieval mechanism combining knowledge graph subgraph matching and vector library similarity, a multi-dimensional innovative evaluation index is constructed. A hierarchical Bayesian model is used for supervised training to generate interpretable evaluation results.
It achieves accurate matching of cross-modal data, reduces the probability of false detection and false negative detection, provides a reference for structured association and semantic similarity, generates interpretable and innovative evaluation results, and meets the business needs of power grid technology project management.
Smart Images

Figure CN122175400A_ABST