Method and device for labeling transient voltage stability samples in power grid based on semi-supervised learning

Patent metadata

Field	Value
Publication number	US-12307378-B1
Application number	US-202418829358-A
Country	US
Kind code	B1
Filing date	Sep 10, 2024
Priority date	Apr 2, 2024
Publication date	May 20, 2025
Grant date	May 20, 2025

Abstract

A method and device for labeling transient voltage stability samples in a power grid based on semi-supervised learning are provided. The method includes: S 1 : obtaining a transient voltage time series trajectory V formed for each load bus in the power grid under N transient operating scenarios; S 2 : preliminarily labeling the stability status of each transient operating scenario with a voltage time series dataset V, and integrating the labeling result Y i into a class label dataset Y; S 3 : constructing a voltage stability sample set S={(V i , Y i )|1≤i≤N}, dividing S into sample subsets S u and S k ; S 4 : labeling samples in S u by using a semi-supervised clustering learning method and a semi-supervised classification learning method to obtain result datasets Y u1 and Y u2 respectively; S 5 : performing interactive verification on Y u1 and Y u2 , and updating S u and S k ; and S 6 : performing repeated iteration on the S 4 and the S 5.

First claim

The invention claimed is:

1. A method for labeling transient voltage stability samples in a power grid based on semi-supervised learning, comprising following steps:
S 1 : obtaining a transient voltage time series trajectory V i formed for each load bus in the power grid under N transient operating scenarios, wherein 1≤i≤N; and integrating all V i into a voltage time series dataset V;
S 2 : preliminarily labeling the stability status of each transient operating scenario based on the voltage time series dataset V, and integrating the labeling result Y i into a class label dataset Y;
S 3 : using V and Y as the input data and output data respectively to construct a voltage stability sample set S={(V i , Y i )|1≤i≤N}, integrating a sample whose class information is unknown in the S into a sample subset S u , with a sample quantity denoted as N u , and integrating remaining data in the S into a sample subset S k , with a sample quantity denoted as N k , wherein N k +N u =N;
S 4 : labeling stability statuses of the N u samples in S u by using a semi-supervised clustering learning method and a semi-supervised classification learning method, to obtain labeling results Y 1j and Y 2j respectively, and integrating the labeling results Y 1j and Y 2j into result datasets Y u1 and Y u2 respectively, wherein 1≤j≤N u ;
wherein the procedure of obtaining the result dataset Y u1 in the S 4 comprises following specific content:
performing unsupervised learning on input data of all samples in the S by using an autoencoder algorithm based on convolutional neurons, and extracting feature information of a hidden layer from an autoencoder based on convolutional neurons;
performing clustering-enabled labeling on all the samples in the S u by using a semi-supervised clustering algorithm with the extracted feature information of the hidden layer as an input and class information of all samples in the S k as supervised information:
if the clustering-enabled labeling result is stable, labeling the class of a current sample as Y 1j =1;
if the clustering-enabled labeling result is unstable, labeling the class of the current sample as Y 1j =−1; and
after completing the clustering-enabled labeling for the N u samples in the S u , integrating all labeling results Y 1j into the result dataset Y u1 ;

S 5 : performing interactive verification on the Y u1 and the Y u2 , moving all samples that pass the interactive verification from the S u to the S k , and updating the S u and the S k ; and
S 6 : performing repeated iteration on the S 4 and the S 5 until all samples in the S u pass the interactive verification or the repeated iteration is performed for M times, using Y u1 obtained through the last-round iteration as the final labeling result of the S u , merging finally updated sample subsets S u and S k , and exporting a complete sample set S′=S u ∪S k as the final sample set with definite transient voltage stability classes.

2. The method for labeling the transient voltage stability samples in the power grid based on semi-supervised learning according to claim 1 , wherein the S 1 comprises following specific content:
collecting the N transient operating scenarios of the power grid that possibly occur in future T hours, performing N time-domain simulations on the collected N transient operating scenarios by using an electromechanical transient time-domain simulation method, and collecting the transient voltage time series trajectory of each load bus in the power grid after each time-domain simulation ends.

3. The method for labeling the transient voltage stability samples in the power grid based on semi-supervised learning according to claim 1 , wherein the S 2 comprises following specific content:
S 21 : calculating the maximum Lyapunov exponent λ i for each transient operating scenario based on the V,
S 22 : preliminarily labeling the stability status of each transient operating scenario based on the λ i ; and
S 23 : integrating the labeling result Y i into the class label dataset Y.

4. The method for labeling the transient voltage stability samples in the power grid based on semi-supervised learning according to claim 3 , wherein the S 21 comprises following specific content:
within a monitoring time window ΔT, the maximum Lyapunov exponent of an i th transient operating scenario is:

λ
i

=

1

n
⁢
K
⁢
Δ
⁢
t

·

∑

m
=
1

n

ln
⁢



(

V

(

K
+
m

)

⁢
Δ
⁢
t

i

-

V

(

K
+
m
-
1

)

⁢
Δ
⁢
t

i

)

/

(

V

m
⁢
Δ
⁢
t

i

-

V

(

m
-
1

)

⁢
Δ
⁢
t

Assignees

• Univ Hunan

Inventors

• Zhu Lipeng
• Wen Weijia
• Huang Weizhi
• Shuai Zhikang
• Li Jiayong
• Zhang Cong
• Zhou Bin
• Hou Jie

Classifications

• H02J2103/30

  Simulating, planning, modelling, reliability check or computer assisted design [CAD] of electric power networks · CPC title

• G06N3/088

  Non-supervised learning, e.g. competitive learning · CPC title

• G06N3/08

  Learning methods · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/096

  Transfer learning · CPC title