Low-frequency sampling non-intrusive industrial load decomposition method based on improved C-MTL-GRU

By improving the two-stage fuzzy C-MTL-GRU method, and combining fast Fourier decomposition and multi-task learning GRU neural network, the accuracy problem of decomposing industrial user load data is solved, and efficient power equipment sensing and energy-saving scheme formulation are realized.

CN116522071BActive Publication Date: 2026-07-10STATE GRID JIANGSU ELECTRIC POWER CO LTD +1
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Patents(China)
Current Assignee / Owner
STATE GRID JIANGSU ELECTRIC POWER CO LTD
Filing Date
2023-03-24
Publication Date
2026-07-10

AI Technical Summary

Technical Problem

Existing technologies struggle to efficiently decompose load data from industrial users, especially given the diverse range of equipment and insufficient data granularity, making it difficult to effectively monitor electrical equipment and develop energy-saving solutions.

Method used

An improved two-stage fuzzy C-MTL-GRU method is adopted, which combines fast Fourier decomposition and two-stage fuzzy C clustering with a multi-task learning GRU neural network to perform state identification and active power decomposition of low-frequency industrial loads.

Benefits of technology

It improves the accuracy of low-frequency sampling non-intrusive industrial load decomposition, enabling more accurate identification of equipment status and decomposition of active power, providing reliable data support for energy-saving solutions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116522071B_ABST
    Figure CN116522071B_ABST
Patent Text Reader

Abstract

The application provides a low-frequency sampling non-intrusive industrial load decomposition method based on improved C-MTL-GRU, which first obtains short-term active power sample data sequences of each production device of a single bus of a factory, carries out difference on the active power sequences of each device after data cleaning to obtain difference sequences; the obtained active power difference sequences are decomposed by fast Fourier to obtain amplitude sequences, then, two-stage fuzzy C clustering is carried out on the obtained multi-section amplitude sequences of each device to identify the load state categories and corresponding power owned by each device, and a device state feature library is constructed; finally, the overall active power of the factory is input into a multi-task learning GRU neural network, the network bottom layer shares parameters, each load state category and corresponding power is taken as a model training supervision quantity, and device state monitoring and power decomposition are carried out, the application improves the precision of low-frequency sampling non-intrusive industrial load decomposition, thereby providing support for efficient power consumption equipment and energy-saving scheme making.
Need to check novelty before this filing date? Find Prior Art