A method for dynamic construction and associated storage of a scarce defect sample library for power distribution networks

CN121501920BActive Publication Date: 2026-05-26ECONOMIC TECH RES INST OF STATE GRID ANHUI ELECTRIC POWER +2
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
ECONOMIC TECH RES INST OF STATE GRID ANHUI ELECTRIC POWER
Filing Date
2026-01-14
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

Existing technologies for identifying defects in power distribution networks suffer from problems such as uneven sample distribution, lack of physical constraints in generative data augmentation, and the risk of model collapse and false confidence in dynamic closed-loop systems, leading to model failure in real-world scenarios.

Method used

A multimodal knowledge graph is constructed, which combines environmental parameters and genealogical labels. The scarcity is dynamically updated through physical consistency verification and the ratio of real to synthetic sample quantity. Augmented samples are generated and real sample collection is driven, realizing the dynamic construction and associated storage of scarcity defects.

Benefits of technology

It effectively improves the physical credibility of the sample library and the robustness of the model in real-world scenarios, prevents model collapse and false confidence caused by over-reliance on synthetic data, and significantly improves the accuracy of defect identification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121501920B_ABST
    Figure CN121501920B_ABST
Patent Text Reader

Abstract

This invention discloses a method for dynamically constructing and associating a scarce defect sample library for power distribution networks, relating to the field of intelligent operation and maintenance technology for power distribution networks. The method includes: constructing holographic samples containing environmental parameters and storing them in a multimodal knowledge graph; configuring genealogical labels for sample nodes to distinguish between real and synthetic sources; calculating defect scarcity values ​​based on the feature distribution scarcity of the holographic samples and the uncertainty of model recognition; generating augmented samples for high-scarcity categories using a generative model, and storing them in the knowledge graph after passing physical consistency verification; dynamically updating the defect scarcity values ​​based on the ratio of real to synthetic samples, wherein when synthetic samples increase while real samples do not, the defect scarcity value is increased based on this ratio to generate a collection strategy for real samples. This invention addresses the problems of existing sample augmentation lacking multidimensional contextual association and physical mechanism constraints, and the collapse and failure of models in real-world scenarios due to over-reliance on synthetic data.
Need to check novelty before this filing date? Find Prior Art