A distribution network fault inspection method, system, device and storage medium

Through multi -weight sharing construction technology and model parameter self -optimization methods, the problem of large calculation and slow optimization iteration in the process of intelligent recognition model construction and optimization of the power distribution network, realizing the independent update and rapid optimization of the model weight, improvement of model training Efficiency and accuracy.

CN117251764BActive Publication Date: 2025-01-24STATE GRID INTELLIGENCE TECHNOLOGY CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202311226603.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-09-21
Publication Date
2025-01-24
Estimated Expiration
2043-09-21

AI Technical Summary

Technical Problem

In the construction and optimization process of the existing distribution network intelligent identification model, there are problems such as large amount of calculation, slow optimization iteration, inability to iterate independently during the model use, and inability to match different network architectures.

Method used

Adopt a multi -weight sharing construction technology and model parameter self -optimization method. Through the initialized inspection analysis of the weight status and branch structure of the model, selectively delete or share according to the number of weights of weights, and the design vector calculation has a model parameter to optimize the model parameters, and and also optimize the model parameters. Parameter updates are performed in different different phase architectures.

Benefits of technology

The independent update and rapid optimization of model weights are realized, the number of network structures and weights is reduced, the efficiency and accuracy of model training are improved, and the independent iteration optimization of the model in the process of use.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117251764B_ABST
    Figure CN117251764B_ABST
Patent Text Reader

Abstract

The present invention discloses a distribution network fault inspection method, system, equipment and storage medium, comprising: obtaining distribution network inspection data samples; performing network structure training on a constructed inspection analysis model, comprising: initializing the weight state and the branch structure of the weight connection in the inspection analysis model, and selectively deleting or sharing the corresponding weight and branch structure according to the number of times the weight is used during the training process; defining a vector calculation directed graph, forming the inspection analysis model after the network structure training into a corresponding network architecture in the form of a vector calculation directed graph, self-optimizing model parameters by using a differentiable continuous gradient descent search method on the network architecture, and performing model parameter training after the optimized model parameters are secondary-connected with the original model parameters, thereby obtaining an optimal inspection analysis model; and using the optimal inspection analysis model to perform abnormal diagnosis on the distribution network inspection data.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Neural network unit structure searching method and system based on attention mechanism

    CN114626506A

  • Power transmission and transformation inspection model network structure searching method and system and storage medium

    CN114881225A