Transformer failure identification and location diagnosis method based on multi-stage transfer learning

Patent metadata

Field	Value
Publication number	US-11619682-B2
Application number	US-202017105616-A
Country	US
Kind code	B2
Filing date	Nov 26, 2020
Priority date	Dec 10, 2019
Publication date	Apr 4, 2023
Grant date	Apr 4, 2023

Abstract

A transformer failure identification and location diagnosis method based on a multi-stage transfer learning theory is provided. Simulation is set up first, a winding parameter of a transformer to be tested is calculated, and a winding equivalent circuit is accordingly built. Different failures are configured for the equivalent circuit, and simulation is performed to obtain a large number of sample data sets. A sweep frequency response test is performed on the transformer to be tested, and detection data sets are obtained. Initial network training is performed on simulation data sets by using the transfer learning method, and the detection data sets are further trained accordingly. A failure support matrix obtained through diagnosis is finally fused. The multi-stage transfer learning theory is provided by the disclosure.

First claim

What is claimed is:

1. A transformer failure identification and location diagnosis method based on multi-stage transfer learning, comprising:
1) Establishing a finite element model according to a structure and a material property of a transformer to be tested and simulating and calculating a winding parameter of transformer;
2) Performing a sweep frequency response test on the transformer to be tested and simulating different failure situations of the transformer; wherein a plurality of taps are selected to act as detection points of the sweep frequency response test if the taps are present and obtaining detection data sets comprising information of the detection points, and history detection data of the transformer to be tested is added to the detection data sets if the history detection data is provided;
3) Building a winding equivalent circuit of the transformer, inputting the winding parameter obtained through calculating into the equivalent circuit, and performing programming to accomplish a sweep frequency response analysis of the equivalent circuit;
4) Setting up a loop in a program and simulating and obtaining a large number of simulation data sets for the detection points and failure situations in 2);
5) Constructing a convolutional neural network of multi-stage transfer learning, dividing each of the simulation data sets and the detection data sets into a training set and a validation set, and performing data enhancement on all data;
6) Performing initial network training on the simulation data sets by using a transfer learning method;
7) Keeping a one-stage trained network and accordingly performing multi-stage training on the detection data sets; and
8) Diagnosing the transformer to be tested by using the trained network and finally fusing the detection data sets comprising the information of the detection points with a failure support matrix obtained through network diagnosis,
wherein the step of simulating different failure situations in step 2) further comprises: adding a pad between different windings or between a winding and an iron core and simulating a winding pitch failure and a winding ground failure; connecting the taps to a resistor, a capacitor, or an inductor in parallel to simulate a failure if the taps are present; and selecting the taps to act as the detection points of the sweep frequency response test when the taps are present and obtaining the detection data sets comprising the information of the detection points.

2. The transformer failure identification and location diagnosis method based on multi-stage transfer learning according to claim 1 , wherein the structure of the transformer provided in step 1) comprises an axial height, a winding thickness, a winding radial width, an iron core thickness, an iron core outer diameter, a winding inner diameter, an insulation paper outer diameter, an end ring thickness, a pad thickness, and a stay thickness of the transformer to be tested, the material property of the transformer comprises relative dielectric constants of insulation paper, a pad, a stay, an end ring, and a phenolic paper tube, simulating and obtaining the parameter of the transformer are performed for a normal state, resistance and self-inductance are calculated by using 1 disk in a 3D model, mutual inductance and capacitance are calculated by using 2 disks, earth capacity Cg is calculated by using an iron core and one disk of winding, 2nd-order mutual inductance is calculated by using 3 disks of winding, and winding parameters of orders in an approximate equivalent circuit are identical in the equivalent circuit.

3. The transformer failure identification and location diagnosis method based on multi-stage transfer learning according to claim 1 , wherein the winding equivalent circuit of the transformer in step 3) is an N-order lumped parameter equivalent circuit, circuit parameters: ground resistance Cg, inter-winding capacitance Cs, self-inductance Ls, mutual inductance Mi(i+1), and resistance R calculated and obtained in step 1) are substituted into the equivalent circuit, and the sweep frequency response analysis is accomplished through programming.

4. The transformer failure identification and location diagnosis method based on multi-stage transfer learning according to claim 1 , wherein the step of obtaining the simulation data sets in step 4) further comprises: setting up a loop based on a normal state value and selecting parameters of ground resistance Cg, inter-winding capacitance Cs, self-inductance Ls, mutual inductance Mi(i+1), and resistance R of the equivalent circuit provided to be 1-2 times the normal state value according to detection content, wherein selection of a loop step length and an abnormal parameter is determined according to a required number of the data sets and a failure type needed to be detected.

5. The transformer failure identification and location diagnosis method based on multi-stage transfer learning according to claim 1 , wherein the step of constructing the convolutional neural network based on multi-stage transfer learning for performing two-stage transfer learning in step 5) specifically comprises:
1) Marking public data sets as D 1 , the simulation data sets as D 2 , and detection data as D 3 , such that first-stage transfer learning comprises:

Source
:

{

D
s

=

D
1

=

{

χ
1

,

P
1

(

X
1

)

}

;

T
s

=

T
1

=

Assignees

• Univ Wuhan

Inventors

• He Yigang
• Duan Jiajun
• Wu Xiaoxin
• He Liulu
• Zhang Hui

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/096

  Transfer learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/08

  Learning methods · CPC title