# US 11,906,957 B1 · Method for assessing short-term voltage stability of power grid based on synchronous spatio-temporal information learning

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 11,906,957 B1
- Document ID: US11906957B1
- Status: Granted
- Kind: B1
- Application number: 18/223,559
- Filed: 2023-07-19
- Granted: 2024-02-20
- Earliest claimed: 2023-03-09
- Assignees: Hunan University
- Applicants: Hunan University
- Inventors: Lipeng Zhu; Zhikang Shuai; Jiayong Li; Bin Zhou; Cong Zhang; Yuhan Hu; Weizhi Huang; Limengqian Zheng
- Web page: https://patentgrep.com/patent/US11906957B1

## CPC

- G05B 23/0221: {Preprocessing measurements, e.g. data collection rate adjustment; Standardization of measurements; Time series or signal analysis, e.g. frequency analysis or wavelets; Trustworthiness of measurements; Indexes therefor; Measurements using easily measured parameters to estimate parameters difficult to measure; Virtual sensor creation; De-noising; Sensor fusion; Unconventional preprocessing inherently present in specific fault detection methods like PCA-based methods}
- H02J 3/001: Arrangements for handling faults or abnormalities, e.g. emergencies or contingencies
- H02J 3/0012: characterised by the contingency detection means in AC networks, e.g. using phasor measurement units [PMU], synchrophasors or contingency analysis
- G05B 2223/06: Remote monitoring
- H02J 2103/30: Simulating, planning, modelling, reliability check or computer assisted design [CAD] of electric power networks
- Y04S 10/50: Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
- G05B: CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS (systems for controlling or regulating non-electric variables G05D; systems for regulating electric or magnetic variables G05F; control devices or systems insofar as characterised by mechanical features only G05G)
- H02J: ELECTRIC POWER NETWORKS; CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- Y04S: SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS

## Abstract

A method for assessing short-term voltage stability of a power grid based on synchronous spatio-temporal information learning includes: performing a time-domain simulation for each transient fault of a power grid under each operating mode, and extracting temporal response trajectories of each monitoring bus and stability status information of the power grid to obtain a transient sample set; constructing a spatial adjacency matrix and a temporal adjacency matrix separately, and integrating the spatial adjacency matrix and the temporal adjacency matrix into a spatio-temporal adjacency matrix; performing synchronous spatio-temporal information learning by using a graph convolutional neural network algorithm, and performing training to obtain a short-term voltage stability assessment model driven by spatio-temporal information synchronization; and inputting a transient temporal responsive trajectory obtained into the short-term voltage stability assessment model, to obtain an assessment result of short-term voltage stability of the power grid.

## Claims

### Claim 1

A method for assessing a short-term voltage stability of a power grid based on a synchronous spatio-temporal information learning, comprising:
step S1: performing a time-domain simulation for each transient fault of the power grid under each operating mode, and extracting a temporal responsive trajectory of each monitoring bus and status information regarding the short-term voltage stability of the power grid from a time-domain simulation process to obtain a transient sample set; wherein an operating mode set and a transient fault set of the power grid are obtained from a dispatching platform of the power grid; N time-domain simulations are performed on each transient fault of the power grid under each operating mode by using an electromechanical transient time-domain simulation method, wherein N=a*b, a represents the number of operating modes, and b represents the number of transient faults; temporal response trajectories of a voltage V, a current I, an active power P, and a reactive power Q within L=50 time instants are recorded for each monitoring bus after the monitoring bus suffers from the transient fault, and a power grid status Y in each time-domain simulation process is recorded, wherein Y=1 represents a transient voltage stability, and Y=−1 represents a transient voltage instability; all temporal response trajectories and power grid statuses recorded in one time-domain simulation process form a transient sample; and all transient samples obtained from the N time-domain simulations are collected to form the transient sample set containing N samples;
step S2: constructing a spatial adjacency matrix and a temporal adjacency matrix separately by using the network impedance matrix of the power grid and correlations between power grid statuses at different time instants, and integrating the spatial adjacency matrix and the temporal adjacency matrix into a spatio-temporal adjacency matrix to comprehensively characterize a spatio-temporal correlation between monitoring buses within different time instants in a transient process of the power grid; wherein step S2 comprises:
step S21: constructing the spatial adjacency matrix AS for measuring an adjacency degree between the monitoring buses of the power grid, as shown in following formulae:
D
ij
=
|
(
Z
ii
-
Z
ij
)
-
(
Z
ji
-
Z
j
⁢
j
)
|
2
(
1
)
D
=
[
D
ij
]
n
*
⁢
n
(
2
)
D
ij
norm
=
D
ij
/
(
max
1
≤
u
,
v
≤
n
D
u
⁢
v
)
(
3
)
A
ij
S
=
1
1
+
D
ij
norm
/
(
max
1
≤
u
,
v
≤
n
1
1
+
D
uv
norm
)
(
4
)
A
S
=
[
A
ij
S
]
n
*
n
(
5
)
wherein Z represents an impedance value between the monitoring buses; i and j represent different monitoring buses, wherein 1≤i, and j≤n; n represents the total number of monitoring buses in the power grid; Dij represents an electrical distance between the monitoring bus i and the monitoring bus j; D represents an electrical distance matrix; max1≤u,v≤n Duv represents a maximum value of each element in the electrical distance matrix D, wherein u and v represent different monitoring buses, 1≤u, and v≤n; Dijnorm represents a normalized electrical distance between the monitoring bus i and the monitoring bus j; Duvnorm represents a normalized electrical distance between the monitoring bus u and the monitoring bus v; and AijS represents a metric of a spatial adjacency degree between the monitoring bus i and the monitoring bus j;
step S22: constructing the temporal adjacency matrix AT representing a temporal correlation between the monitoring buses within two consecutive time instants after the power grid suffers from the transient fault, as shown in following formulae:
A
ij
T
=
⁢
{
1
,
i
=
j
0
,
i
≠
j
(
6
)
A
T
=
[
A
ij
T
]
n
*
n
(
7
)
wherein AijT represents a metric of a temporal adjacency degree between the monitoring bus i and the monitoring bus j; and
step S23: constructing the spatio-temporal adjacency matrix AST representing a spatio-temporal correlation between the monitoring buses within three consecutive time instants after the power grid suffers from the transient fault, as shown in a following formula:
A
S
⁢
T
=
[
A
S
A
T
0
A
T
A
S
A
T
0
A
T
A
S
]
;
(
8
)
step S3: performing the synchronous spatio-temporal information learning on the transient sample set by using a graph convolutional neural network algorithm, and performing a training to obtain a short-term voltage stability assessment model driven by a spatio-temporal information synchronization; and
step S4: when the power grid suffers from the transient fault during an online operation, obtaining a transient temporal responsive trajectory of each monitoring bus in real time, inputting the transient temporal responsive trajectory into the short-term voltage stability assessment model, and outputting, by the short-term voltage stability assessment model, an online assessment result of the short-term voltage stability of the power grid in real time; and
taking an emergency control measure to restore the stable transient voltage for the power grid.

### Claim 2 (depends on 1)

The method for assessing the short-term voltage stability of the power grid based on the synchronous spatio-temporal information learning according to claim 1, wherein step S3 comprises:
step S31: performing, by using the graph convolutional neural network algorithm, the synchronous spatio-temporal information learning on the transient sample set obtained in step S1, to match with the spatio-temporal adjacency matrix AST obtained in step S23, dividing the temporal response trajectories of the voltage V, the current I, the active power P, and the reactive power Q within L time instants contained in each sample in the transient sample set obtained in step S1 into L−2 trajectory slices consisting of three consecutive time instants, concisely denoting the temporal response trajectories of the voltage V, the current I, the active power P, and the reactive power Q within the three consecutive time instants in a trajectory slice as x, mapping the x to a hidden layer state h0 using a fully connected layer, and performing a spatio-temporal graph convolution operation on the hidden layer state h0 by using a graph convolution learning module to obtain h1:
h1=(ASTh0W1+b1)eσ(ASTh0W2+b2) (9)
wherein h1 represents a hidden layer state obtained by the spatio-temporal graph convolution operation, W1, W2, b1, and b2 represent to-be-learned parameters during the spatio-temporal graph convolution operation, σ(*) represents a sigmoid activation function, and e represents a matrix dot product; and
performing the spatio-temporal graph convolution operation for L−2 times during a graph convolution learning, learning L−2 trajectory slices separately, associating the hidden layer state h1 with the fully connected layer and a flattening layer sequentially after completing the graph convolution learning, denoting an output of the flattening layer as f, and associating the output f of the flattening layer with a discretized output of the power grid's predicted status Y′ by using the sigmoid activation function, to construct the short-term voltage stability assessment model driven by the spatio-temporal information synchronization; and
step S32: implementing, by using the transient sample set obtained in step S1, synchronous spatio-temporal information learning and performance test on the short-term voltage stability assessment model driven by spatio-temporal information synchronization in step S31, until a stability assessment accuracy of the short-term voltage stability assessment model meets a requirement.

### Claim 3 (depends on 2)

The method for assessing the short-term voltage stability of the power grid based on the synchronous spatio-temporal information learning according to claim 2, wherein in step S32, the performance test is performed on the short-term voltage stability assessment model by using a cross validation method; let Nf denote a sum the number of samples whose actual power grid statuses Y are 1 but predicted statuses are mistakenly assessed to be unstable (i.e., Y′=−1) and the number of samples whose actual power grid statuses Y are −1 but predicted statuses are mistakenly assessed to be stable (i.e., Y′=−1) in the transient sample set; the stability assessment accuracy of the short-term voltage stability assessment model is calculated according to a formula Acc=Nf/N×100%; and if Acc≥97%, stability assessment performance of the short-term voltage stability assessment model meets the requirement, and the synchronous spatio-temporal information learning training of the short-term voltage stability assessment model is completed, and then step S4 is performed; if Acc<97%, the synchronous spatio-temporal information learning is performed again by using the transient sample set obtained in step S1 until Acc≥97%, and then step S4 is performed.

### Claim 4 (depends on 3)

The method for assessing the short-term voltage stability of the power grid based on the synchronous spatio-temporal information learning according to claim 3, wherein in step S4, when the power grid suffers from the transient fault during the online operation, synchronous phasor measurement units deployed at individual monitoring buses in the power grid acquire the temporal data of responsive trajectories of the voltage V, the current I, the active power P, and the reactive power Q within the L time instants in real time after the power grid suffers from the transient fault, the temporal data of responsive trajectories of the voltage V, the current I, the active power P, and the reactive power Q are input into the short-term voltage stability assessment model that passes the performance test in step S32, and the short-term voltage stability assessment model outputs the online assessment result indicating whether the power grid is in a stable status or an unstable status.
