A state estimation method based on physical information guided dynamic graph attention network pseudo measurement modeling

CN122092400BActive Publication Date: 2026-06-26HOHAI UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HOHAI UNIV
Filing Date
2026-04-23
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing state estimation methods in distribution networks suffer from a lack of data support in sparse measurement scenarios due to factors such as the difficulty in achieving full coverage of measurement equipment and equipment failures. Furthermore, existing models fail to effectively utilize the physical topology of the distribution network, resulting in poor generalization ability in topology-changing scenarios. Additionally, pseudo-measurement results lack physical consistency, making it difficult to support high-precision state estimation.

Method used

A pseudo-measurement modeling method based on physical information-guided dynamic graph attention network is constructed. By adaptively adjusting the aggregation weights between nodes and combining the power-balanced physical constraint embedding loss function, the accuracy and physical consistency of pseudo-measurements are improved by utilizing dynamic graph attention mechanism and Gaussian mixture model.

Benefits of technology

It significantly improves the accuracy and physical consistency of pseudo-measurement generation, enhances the model's generalization ability in scenarios with frequent topology changes, provides high-precision and reliable pseudo-measurement results, and supports state awareness under complex operating conditions of distribution networks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122092400B_ABST
    Figure CN122092400B_ABST
Patent Text Reader

Abstract

The application provides a state estimation method based on physical information guided dynamic graph attention network pseudo measurement modeling, which comprises the following steps: obtaining historical operation data and a topological structure of a power distribution network, constructing a non-directional graph model containing node and branch characteristics, and generating a graph data set; constructing a feature extraction network based on a dynamic graph attention mechanism, adaptively adjusting the aggregation weight of a node and its adjacent nodes, and extracting deep node features of the power distribution network; embedding the power balance physical constraint of the power distribution network into a loss function of the model as a residual vector, and constructing a data and physical hybrid loss function; adopting a training strategy of dynamically adjusting a physical weight, performing offline iterative training on the feature extraction network, and obtaining an optimal pseudo measurement generation model; generating all-network pseudo measurement values based on real-time measurement data and a current topology, and performing power distribution network state estimation after error weight distribution by using a Gaussian mixture model.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Power distribution network topology identification and state estimation method based on message passing neural network and online deep learning

    CN119598655A

  • Power demand dynamic prediction method based on multi-scale hybrid architecture

    CN119990477A