WO2020181804A1 - Method and apparatus for recognizing large power grid critical transient stability boundary state, and electronic device and storage medium - Google Patents

Abstract

Disclosed are a method and apparatus for recognizing a large power grid critical transient stability state, and an electronic device and a storage medium. The method comprises: constituting a power generator feature quantity set and a network feature quantity set; constituting a source matrix and a network matrix; constructing differences of Gaussian scale-space of the source matrix and the network matrix; performing extremum detection on the differences of Gaussian scale-space to acquire a local extremum point corresponding to the source matrix and a local extremum point corresponding to the network matrix; removing a local extremum point with a low contrast ratio and a low principal curvature, and taking a reserved local extremum point as a key point; allocating a direction value to the key point, and generating, according to the direction value, a feature vector corresponding to the key point; determining, according to the feature vector corresponding to the key point, feature vectors of the source matrix and the network matrix; matching the feature vectors of the source matrix and the network matrix, and determining a feature quantity matching degree index; and recognizing a large power grid critical transient state, wherein when the feature quantity matching degree index approaches 0, a large power grid approaches a critical transient stability state.

Description

Method, device, electronic equipment and storage medium for identifying critical transient stable boundary state of large power grid

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office with an application number of 201910185402.6 on March 12, 2019. The entire content of the application is incorporated into this application by reference.

Technical field

The present disclosure relates to the field of large power grid security applications, for example, to a method, device, electronic equipment, and storage medium for identifying critical transient stable states of large power grids.

Background technique

The development trend of the energy Internet with the power grid as the core and the highly developed information technology have put forward higher requirements for the safe and stable operation and intelligent prevention and control of large power grids. It is necessary to develop and establish a more refined grid evaluation system that is compatible with it. At the same time, the wide area measurement system/synchronous phasor measurement unit (Wide Area Measurement System/Phasor Measurement Unit, WAMS/PMU) measured information and fault set simulation results constitute the power grid spatio-temporal big data, how to use data mining technology to quickly and efficiently It is one of the core goals of smart grids to tap the ground and achieve refined evaluation of large power grids.

The transient problem of the power grid is one of the key problems that affect the stable operation of the power system, and it has been paid attention and attention by researchers at home and abroad. In the context of the Energy Internet, pattern recognition methods based on data mining have gradually provided new ideas for solving some traditional problems in the power system. The existing main methods of using data mining technology to solve the transient stability assessment (Transient Stability Assessment, TSA) of the power system include artificial neural network, principal component analysis, support vector machine, etc. Although the existing TSA method has many advantages, it only pays attention to the transient stability or transient instability that occurs in the normal state.

Summary of the invention

The present disclosure proposes a method for identifying critical transient stable states of large power grids, including:

Based on the actual measurement data on the power supply side of the large power grid, select the occurrence of the fault, when the fault is removed, and after the fault is removed, a total of 3 groups of 12 characteristic variables constitute the generator feature set. Based on the actual measurement data on the network side of the large power grid, select the fault occurrence When the fault is removed, a total of 3 groups of 12 feature quantities during and after the fault removal constitute the network feature quantity set;

Normalizing the generator feature set and the network feature set respectively, the processed data in the generator feature set constitutes a source matrix, and the processed network feature set data constitutes a network matrix;

Respectively constructing the Gaussian difference scale space of the source matrix and the net matrix;

In the Gaussian difference scale space of the source matrix, extremum detection is performed on the source matrix to obtain the local extremum points corresponding to the source matrix. In the Gaussian difference scale space of the net matrix, the Performing extreme value detection on the net matrix to obtain local extreme points corresponding to the net matrix;

Remove the local extreme points with low contrast and low principal curvature, and use the retained local extreme points as key points;

Assign a direction value to the key point, and generate a feature vector corresponding to the key point according to the direction value;

Determine the eigenvectors of the source matrix and the net matrix according to the eigenvectors corresponding to the key points;

The feature vectors of the source matrix and the net matrix are matched, and a feature quantity matching degree index is determined. The formula of the feature quantity matching degree index is as follows:

In the formula, Z is the feature quantity matching index, A is the smaller value of the number of key points of the source matrix and the number of key points of the net matrix, and B is the logarithm of the key points corresponding to the eigenvectors of the source matrix and the net matrix matching successfully;

The critical transient state of the large power grid is identified, and when the characteristic quantity matching index tends to 0, the large power grid tends to a critical transient stable state.

The present disclosure also proposes a critical transient stable state identification device for large power grids, including:

The feature input module is set to be based on the measured data on the power supply side of the large power grid. When a fault occurs, when the fault is removed, and after the fault is removed, a total of 3 groups of 12 feature variables constitute the generator feature set, based on the network side of the large power grid. Actually measured data, select the network characteristic quantity set of 3 groups of 12 characteristic quantities when the fault occurs, when the fault is removed, and after the fault is removed;

The matrix forming module is configured to perform normalization processing on the generator feature set and the network feature set respectively, the processed data in the generator feature set constitutes a source matrix, and the processed network feature set The data constitutes a network matrix;

A scale space construction module, configured to construct Gaussian difference scale spaces of the source matrix and the net matrix respectively;

The detection module is configured to perform extreme value detection on the source matrix in the Gaussian differential scale space of the source matrix to obtain the local extreme points corresponding to the source matrix, in the Gaussian differential scale space of the net matrix , Performing extreme value detection on the net matrix to obtain local extreme points corresponding to the net matrix;

The removal module is set to remove the local extreme points of low contrast and low main curve, and use the retained local extreme points as key points;

An allocation module, configured to allocate a direction value to the key point, and generate a feature vector corresponding to the key point according to the direction value;

A matching module, configured to determine the eigenvectors of the source matrix and the net matrix according to the eigenvectors corresponding to the key points, and match the eigenvectors of the source matrix and the net matrix;

The feature acquisition module is configured to determine the feature quantity matching degree index, and the formula of the feature quantity matching degree index is as follows:

In the formula, Z is the feature quantity matching index, A is the smaller value of the number of key points of the source matrix and the number of key points of the net matrix, and B is the logarithm of the key points corresponding to the eigenvectors of the source matrix and the net matrix matching successfully;

The identification module is configured to identify the critical transient state of the large power grid, and when the feature quantity matching index tends to 0, the large power grid tends to a critical transient stable state.

The present disclosure also provides an electronic device, including:

At least one processor;

Memory, set to store at least one computer program,

When the at least one computer program is executed by the at least one processor, the at least one processor implements the aforementioned method for identifying the critical transient stable state of the large power grid.

The present disclosure also provides a computer-readable storage medium that stores computer-executable instructions, and the computer-executable instructions are used to execute the aforementioned method for identifying a critical transient stable state of a large power grid.

Description of the drawings

FIG. 1 is a flowchart of a method for identifying critical transient and stable states of a large power grid according to an embodiment;

2 is a schematic diagram of a 10-machine 39-node system of a method for identifying critical transient and stable states of a large power grid according to an embodiment;

Fig. 3 is an embodiment of a method for identifying critical transient and stable states of a large power grid in a 10-machine 39-bus system with different fault removal times on the same line in the power angle time domain;

4 is an embodiment of a method for identifying critical transient stable states of large power grids, a 10-machine 39-node system with different fault removal time sources and changes in network matching degree;

FIG. 5 is a comparison diagram of transient stability matching results of multiple lines in a 10-machine 39-node system of a method for identifying critical transient and stable states of a large power grid according to an embodiment;

Fig. 6 is a structural diagram of a device for identifying critical transient and stable states of a large power grid according to an embodiment;

Fig. 7 is a schematic structural diagram of an electronic device according to an embodiment.

detailed description

Now, exemplary embodiments of the present disclosure will be described with reference to the accompanying drawings. The terms in the exemplary embodiments shown in the drawings do not limit the present disclosure. In the drawings, the same units/elements use the same reference signs.

Unless otherwise specified, the terms (including scientific and technological terms) used herein have the usual meanings to those skilled in the art. In addition, it is understandable that the terms defined in commonly used dictionaries should be understood as having consistent meanings in the context of their related fields, and should not be understood as idealized or overly formal meanings.

The existing methods of using data mining technology to solve the transient stability assessment (Transient Stability Assessment, TSA) of the power system often only pay attention to the transient stability or transient instability in the normal state, and ignore the critical stable state of the grid , Or can not accurately distinguish the critical stable state, which leads to fuzzy areas in the power grid safety assessment. How to find the critical area or appropriate stability parameters to evaluate the critical state becomes the key to using data mining technology to solve the TSA problem.

From the perspective of macro energy conversion, the essence of power system transient stability is the energy conversion and conservation between the injected mechanical kinetic energy and the electromagnetic energy absorbed by the network. At the same time, the power grid has a clear topological structure and inherent network attributes. As a strong nonlinear energy transmission system in which various components interact, there must be more or less interaction between them. Standing at a fully measurable perspective of the operating status of all components in the power grid, the topological relationship and interaction between the components must be contained in the wide-area space-time measurement information.

The power grid is a large-scale network energy transmission system based on the "source-network-load" model. The energy transmission system based on the "source-network-load" model is established to measure the space-time measurement of conventional transient stability or critical transient stability. Information, in-depth mining and analysis of the relationship between "source-network" is more in line with the overall stable behavior of the power grid as a nonlinear power system, and it is also increasingly used in actual power grid operation control.

The Scale-Invariant Feature Transform (SIFT) algorithm is derived from the field of computer image recognition, by obtaining the feature points (interest points, or corner points) in two pictures and the description of their positions and scale directions Obtain the feature vector and perform image feature point matching. According to the number of matching points, the purpose of image recognition is achieved. The method proposed in the present disclosure compares the “source” data and the “net” data obtained during the transient state of the power grid to two “pictures”, and achieves the purpose of identifying the state of the power grid through feature matching of the two “pictures”.

The embodiment of the present invention constructs a "source-network" matching index under the transient stability of the power grid, and uses the wide-area measurement information of the power grid to perform calculation and identification of the critical transient stable state of the power grid, and supports the refined evaluation of the transient stability of the power grid.

The embodiment of the present invention provides a method for identifying a critical transient stable state of a large power grid, as shown in FIG. 1, including:

Based on the actual measurement data on the power supply side of the large power grid, select the occurrence of the fault, when the fault is removed, and after the fault is removed, a total of 3 groups of 12 characteristic variables constitute the generator feature set. Based on the actual measurement data on the network side of the large power grid, select the fault occurrence When the fault is removed, a total of 3 groups of 12 feature quantities during and after the fault removal constitute the network feature quantity set.

The aforementioned input characteristics are selected from the measurement data of the synchronous phasor measurement unit (PMU) of the power grid wide area measurement system (WAMS). PMU can be simply regarded as a sensor, installed on the key bus/plant/transformer of the power grid and other equipment. Its function is to measure and collect power grid data, which is often said important information data such as voltage/current/power, and then transmit it to the grid for unified management In the system database, in order to distinguish the simulation data, the data obtained from the PMU is also called the measured data. In power system analysis, the actual measurement data comes from this.

Based on this, select the input characteristics of the large power grid source and network, and based on the PMU measurement data, based on the measured data on the power side of the large power grid provided by the PMU measurement data, select the total of 3 when the fault occurs, when the fault is removed, and after the fault is removed. Groups of 12 characteristic quantities constitute the generator characteristic quantity set. Based on the actual measurement data on the network side of the large power grid provided by the PMU measurement data, 3 groups of 12 characteristic quantities constitute the network when the fault occurs, when the fault is removed, and after the fault is removed. Feature set.

The generator feature quantity set and the network feature quantity set are respectively normalized, and the processed data in the generator feature quantity set constitutes a source matrix, and the processed data in the network feature quantity set constitutes a network matrix. Among them, the generator characteristic quantity set includes generator power angle, generator excitation voltage, generator electromagnetic power and generator reactive power. The network characteristic quantity set includes bus voltage amplitude, bus voltage phase angle, bus inflow active power, and Reactive power flows into the bus.

Constructing the difference of Gaussian scale-space (DOG scale-space) of the source matrix and the net matrix respectively, that is, using Gaussian difference kernels of different scales and matrix convolution to generate the Gaussian difference scale space. The step of separately constructing the Gaussian difference scale space of the source matrix and the net matrix includes:

Generate the Gaussian scale space of the source matrix, calculate the first difference of Gaussian scale space functions corresponding to two adjacent layers of the same order in the Gaussian scale space of the source matrix, and use the first difference as the source A layer of the Gaussian difference scale space of the matrix;

Generate the Gaussian scale space of the net matrix, calculate the second difference of the Gaussian scale space functions corresponding to two adjacent layers of the same order in the Gaussian scale space of the net matrix, and use the second difference as the net A layer of the Gaussian difference scale space of the matrix.

The Gaussian scale space function is expressed by the following formula:

L(x,y,σ)=G(x,y,σ)*I(x,y) (1)

Among them, L(x,y,σ) represents the Gaussian scale space, G(x,y,σ) represents the two-dimensional space Gaussian function, I(x,y) represents the source matrix or network matrix, and (x,y) is the matrix For points on I, σ is the scale space factor, and * means convolution calculation.

The function of Gaussian difference scale space is expressed by the following formula:

D(x,y,σ)=(G(x,y,kσ)-G(x,y,σ))*I(x,y)

=L(x,y,kσ)-L(x,y,σ) (3)

Among them, D (x, y, σ) represents the Gaussian difference scale space, and k is the multiple of the Gaussian scale space of two adjacent layers.

In the Gaussian difference scale space of the source matrix, extremum detection is performed on the source matrix to obtain the local extremum points corresponding to the source matrix. In the Gaussian difference scale space of the net matrix, the The net matrix performs extreme value detection to obtain the local extreme points corresponding to the net matrix. In order to detect the local extreme points in the Gaussian difference scale space, each sampling point must be compared with all its neighboring points. The sampling point of the intermediate detection needs to be compared with the 8 adjacent points in the same layer, and 9 points in the adjacent scales of the upper and lower layers to ensure that both the Gaussian difference scale space and the two-dimensional matrix where the sampling points are located are detected Extreme point. If the detected sampling point is the maximum or minimum in the 26 neighborhoods of this layer and the upper and lower layers of the Gaussian difference scale space, then the sampling point is a local extreme point of the matrix at this scale, such a local The extreme points can be identified as candidate key points.

Remove the local extreme points with low contrast and low principal curvature, and use the retained local extreme points as key points. Through extreme value detection, multiple local extreme points can be obtained, but not every local extreme point meets the requirements of subsequent processing. Therefore, two types of local extreme points need to be removed, one is a low-contrast local The extreme point, the other is the local extreme point with low principal curvature to enhance the matching stability and improve the anti-noise ability.

The step of removing the local extremum points of low contrast and low principal curvature includes: for each local extremum point, the function of the local extremum point corresponding to the Gaussian difference scale space is fitted with a three-dimensional quadratic function to determine the The location and scale of the local extreme point, according to the scale of the local extreme point, determine whether the local extreme point is a low-contrast point, remove all the local extreme points determined as low-contrast points, and remove the remaining The local extreme points of low principal curvature in the local extreme points.

Wherein, the step of performing three-dimensional quadratic function fitting on the function of the local extremum point corresponding to the Gaussian difference scale space to determine the position and scale of the local extremum point includes:

According to the Taylor expansion of the function D(x,y,σ) of the Gaussian difference scale space, the local extreme points are fitted with a three-dimensional quadratic function, and the Taylor expansion is expressed by the following formula:

Let the partial derivative of D(x,y,σ) with respect to x be equal to 0, and obtain the position of the local extreme point

Substitute formula (5) into formula (4) to obtain the scale of the local extreme point

among them,

Represents the first-order partial derivative of D ^{T with} respect to x,

Indicates the second-order partial derivative of D to x, the superscript T means transpose, and the superscript -1 means matrix inversion.

According to the scale of the local extreme point, the step of determining whether the local extreme point is a low-contrast point includes:

In the case of determining that the local extreme point is not a low-contrast point;

In the case of, it is determined that the local extreme point is a low-contrast point.

After removing all low-contrast local extreme points, among the remaining local extreme points, there may be local extreme points with low principal curvature. The extreme value of a poorly defined Gaussian difference operator is at the edge across the edge. Where there is a larger principal curvature, and a smaller principal curvature in the direction of the vertical edge, it will produce unstable edge response. Therefore, it is necessary to remove the local extreme points with low principal curvature among the remaining local extreme points. The curvature is obtained from the Hessian matrix.

For each remaining local extreme point, the principal curvature of the local extreme point is obtained according to the Hessian matrix, and the formula of the Hessian matrix is as follows:

Among them, H represents the Hessian matrix, D _xx , D _xy , D _yx and D _yy are elements of the Hessian matrix H with 2×2 dimensions, and the eigenvalues α and β of H represent the gradients in the x direction and the y direction.

The principal curvature of D is proportional to the eigenvalue of H. Let α be the maximum eigenvalue and β be the minimum eigenvalue, then:

T _r (H)=D _xx +D _yy ＝α+β (8)

D _et (H)=D _xx D _yy -(Dx _y ) ²

=αβ (9)

Among them, T _r (H) represents the sum of the diagonal elements of the matrix H, and D _et (H) represents the value of the determinant of the matrix H.

Let α=γβ, then the principal curvature

Determined by the following formula:

In the case that the main curvature is not less than (r+1) ² /r, determine that the local extreme point is a point with a low main curvature, and remove the local extreme point that is determined to be a point with a low main curvature;

In the case where the principal curvature is less than (r+1) ² /r, the local extreme point is retained as a key point.

The above-mentioned local extreme points with low principal curvature are unstable edge response points, which can be referred to as edge points for short. After removing both the low-contrast local extreme points and the local extreme points with low principal curvature, the remaining The local extreme point is the key point. In addition, key points can also be understood as feature points, that is, the information of key points can reflect the source matrix of the key point, such as the characteristics of the source matrix or the net matrix.

A direction value is assigned to the key point, and a feature vector corresponding to the key point is generated according to the direction value. The step of assigning direction values to the key points includes:

Calculate the direction value of the key point with the following formula:

Among them, m(x,y) represents the modulus of the gradient at (x,y), θ(x,y) represents the direction of the gradient at (x,y), L is the function corresponding to the Gaussian scale space where the key point is located, tan Said to calculate the tangent value. The gradient direction histogram is used to count the gradient direction of the neighborhood. The horizontal axis of the gradient direction histogram represents the magnitude of the gradient direction of the neighborhood key points, and the vertical axis represents the magnitude of the gradient magnitude of the neighborhood key points.

When generating the feature vector corresponding to the key point, in order to further describe the information of the key point, it is important to determine the size of the neighborhood range of the key point. In order to enhance the anti-noise ability and the robustness of matching, the value range of the neighborhood near the key point is usually set to 16×16, divided into 4×4 sub-regions, and each sub-region is used as a seed point, then 4 ×4 seed points, each seed point has 8 directions, so the information of each key point is contained in a 4×4×8=128-dimensional vector. This 128-dimensional vector is the feature vector of the key point .

Determine the eigenvectors of the source matrix and the net matrix according to the eigenvectors corresponding to the key points. For example, the source matrix corresponds to k1 key points, and the net matrix corresponds to k2 key points. These k1 eigenvectors are the eigenvectors of the source matrix, and these k2 eigenvectors are the eigenvectors of the net matrix.

Matching the eigenvectors of the source matrix and the net matrix includes:

According to the Euclidean distance formula in the n-dimensional space, the eigenvectors of the source matrix and the net matrix are matched, and the Euclidean distance formula is as follows:

Among them, ρ represents Euclidean distance, a[i] represents the i-th element of the eigenvector of the source matrix, b[i] represents the i-th element of the eigenvector of the net matrix, i=1, 2,...,n .

After the above-mentioned feature matching "source-net" matrix, finally the key point logarithm corresponding to the successfully matched feature vector will be given according to the proportional threshold, that is, the number of matching point pairs. The more matching point pairs indicate "source-net" The more similarities; the fewer the number of matching points, the fewer the similarities, that is, the number of matching points is a quantitative measure of the "source-net" similarity.

The feature quantity matching degree index is determined, and the formula of the feature quantity matching degree index is as follows:

In the formula, Z is the feature quantity matching index, which can be understood as the matching index of the source and net identification, A is the smaller value of the key points of the source matrix and the key points of the net matrix, and B is the source matrix and net matrix matching successfully The feature vector of corresponds to the key point logarithm.

The critical transient state of the large power grid is identified. When the characteristic quantity matching index tends to 0, the large power grid tends to a critical transient stable state.

The embodiment of the present invention also provides a device 200 for identifying a critical transient and stable state of a large power grid. The device can be installed and used in computing equipment such as a personal computer.

As shown in FIG. 6, the device 200 includes:

The feature input module 201 is set to be based on the measured data on the power supply side of the large power grid. When a fault occurs, when the fault is removed, and after the fault is removed, a total of 3 groups of 12 feature quantities constitute a generator feature set, based on the network side of the large power grid According to the actual measurement data of the network, a total of 3 groups of 12 feature quantities are selected when the fault occurs, when the fault is removed, and after the fault is removed to form a network feature set.

The matrix forming module 202 is configured to perform normalization processing on the generator feature quantity set and the network feature quantity set respectively, and the processed data in the generator feature quantity set constitutes the source matrix and the processed network feature quantity The concentrated data constitutes a network matrix.

Among them, the generator characteristic quantity set includes generator power angle, generator excitation voltage, generator electromagnetic power and generator reactive power, and the network characteristic quantity set includes bus voltage amplitude, bus voltage phase angle, bus inflow active power and bus Reactive power flows in.

The scale space construction module 203 is configured to construct the Gaussian difference scale space of the source matrix and the net matrix respectively.

The scale space construction module 203 is configured to generate the Gaussian scale space of the source matrix, calculate the first difference of the Gaussian scale space functions corresponding to two adjacent layers of the same order in the Gaussian scale space of the source matrix, and convert the The first difference is used as a layer of the Gaussian difference scale space of the source matrix; the Gaussian scale space of the net matrix is generated, and the Gaussian scale space corresponding to two adjacent layers of the same order in the Gaussian scale space of the net matrix is calculated The second difference of the function takes the second difference as a layer of the Gaussian difference scale space of the net matrix. In other words, the first difference is the difference between the Gaussian scale space functions corresponding to two adjacent layers of the same order in the Gaussian scale space of the source matrix, and the second difference is the difference between two adjacent layers of the same order in the Gaussian scale space of the net matrix. The difference of the Gaussian scale space function.

The Gaussian scale space function is represented by the following formula:

L(x,y,σ)=G(x,y,σ)*I(x,y)

Among them, L(x,y,σ) represents the Gaussian scale space, G(x,y,σ) represents the two-dimensional space Gaussian function, I(x,y) represents the source matrix or network matrix, and (x,y) is the matrix For points on I, σ is the scale space factor, and * means convolution calculation.

The function of Gaussian difference scale space is expressed by the following formula:

D(x,y,σ)=(G(x,y,kσ)-G(x,y,σ))*I(x,y)

=L(x,y,kσ)-L(x,y,σ)

Among them, D (x, y, σ) represents the Gaussian difference scale space, and k is the multiple of the Gaussian scale space of two adjacent layers.

The detection module 204 is configured to perform extremum detection on the source matrix in the Gaussian difference scale space of the source matrix to obtain local extremum points corresponding to the source matrix. In the space, extremum detection is performed on the net matrix to obtain the local extremum points corresponding to the net matrix.

The removing module 205 is configured to remove the local extreme points of low contrast and low main curve, and use the retained local extreme points as key points.

The removal module 205 is configured to perform a three-dimensional quadratic function fitting for each local extreme point corresponding to the function of the Gaussian difference scale space to determine the location and scale of the local extreme point; The scale of the local extreme point determines whether the local extreme point is a low-contrast point; removes all the local extreme points determined as low-contrast points, and removes the remaining local extreme points with low principal curvature Local extreme points.

The removal module 205 is set to:

According to the Taylor expansion of the function D(x,y,σ) of the Gaussian difference scale space, the local extreme points are fitted with a three-dimensional quadratic function, and the Taylor expansion is expressed by the following formula:

Let the partial derivative of D(x,y,σ) with respect to x be equal to 0, and obtain the position of the local extreme point

Put

Substitute into the Taylor expansion of D(x,y,σ) to obtain the scale of the local extreme point

among them,

Represents the first-order partial derivative of DT to x,

Indicates the second-order partial derivative of D to x, the superscript T means transpose, and the superscript -1 means matrix inversion.

The removal module 205 is set to

In the case of determining that the local extreme point is not a low-contrast point,

In the case of, it is determined that the local extreme point is a low-contrast point.

The removal module 205 is set to:

For each remaining local extreme point, the principal curvature of the local extreme point is obtained according to the Hessian matrix, and the formula of the Hessian matrix is as follows:

Among them, H represents the Hessian matrix, D _xx , D _xy , D _yx and D _yy are the elements of the Hessian matrix H with 2×2 dimensions, and the eigenvalues α and β of H represent the gradients in the x direction and the y direction;

The principal curvature of D is proportional to the eigenvalue of H. Let α be the maximum eigenvalue and β be the minimum eigenvalue, then:

T _r (H)=D _xx +D _yy ＝α+β

D _et (H)=D _xx D _yy -(D _xy ) ²

=αβ

Among them, T _r (H) represents the sum of the diagonal elements of the matrix H, and D _et (H) represents the value of the determinant of the matrix H.

Let α=γβ, then the principal curvature

Determined by the following formula:

In the case that the main curvature is not less than (r+1) ² /r, determine that the local extreme point is a point with a low main curvature, and remove the local extreme point that is determined to be a point with a low main curvature;

In the case where the principal curvature is less than (r+1) ² /r, the local extreme point is retained as a key point.

The allocation module 206 is configured to allocate a direction value to the key point, and generate a feature vector corresponding to the key point according to the direction value. The allocation module 206 is set to calculate the direction value of the key point using the following formula:

Among them, m(x,y) represents the modulus of the gradient at (x,y), θ(x,y) represents the direction of the gradient at (x,y), L is the function corresponding to the Gaussian scale space where the key point is located, tan Said to calculate the tangent value. Use the gradient direction histogram to count the gradient direction of the neighborhood. The horizontal axis of the gradient direction histogram represents the size of the gradient direction of the neighborhood key point, and the vertical axis represents the magnitude of the gradient of the neighborhood key point;

The matching module 207 is configured to determine the eigenvectors of the source matrix and the net matrix according to the eigenvectors corresponding to the key points, and match the eigenvectors of the source matrix and the net matrix. The matching module 207 is configured to match the eigenvectors of the source matrix and the net matrix according to the Euclidean distance formula in the n-dimensional space, and the Euclidean distance formula is as follows:

Among them, ρ represents Euclidean distance, a[i] represents the i-th element of the eigenvector of the source matrix, b[i] represents the i-th element of the eigenvector of the net matrix, i=1, 2,...,n .

The feature acquisition module 208 is configured to determine a feature quantity matching degree index, and the formula of the feature quantity matching degree index is as follows:

In the formula, Z is the feature quantity matching index, A is the smaller value of the number of key points of the source matrix and the number of key points of the net matrix, and B is the number of key point pairs corresponding to the eigenvectors of the source matrix and net matrix matching successfully.

The identification module 209 is configured to identify the critical transient state of the large power grid, and when the characteristic quantity matching index tends to 0, the large power grid tends to a critical transient stable state.

FIG. 7 is a schematic diagram of the hardware structure of an electronic device provided by an embodiment. As shown in FIG. 7, the electronic device includes: one or more processors 310 and a memory 320. One processor 310 is taken as an example in FIG. 7.

The electronic device may further include: an input device 330 and an output device 340.

The processor 310, the memory 320, the input device 330, and the output device 340 in the electronic device may be connected through a bus or other methods. In FIG. 7, the connection through a bus is taken as an example.

The memory 320, as a computer-readable storage medium, can be configured to store software programs, computer-executable programs, and modules. The processor 310 executes a variety of functional applications and data processing by running software programs, instructions, and modules stored in the memory 320 to implement any one of the methods in the foregoing embodiments.

The memory 320 may include a program storage area and a data storage area. The program storage area may store an operating system and an application program required by at least one function; the data storage area may store data created according to the use of the electronic device, and the like. In addition, the memory may include volatile memory such as Random Access Memory (RAM), and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage devices.

The memory 320 may be a non-transitory computer storage medium or a transitory computer storage medium. The non-transitory computer storage medium, for example, at least one magnetic disk storage device, flash memory device, or other non-volatile solid state storage device. In some embodiments, the memory 320 may optionally include a memory remotely provided with respect to the processor 310, and these remote memories may be connected to the electronic device through a network. Examples of the above-mentioned network may include the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

The input device 330 may be configured to receive inputted numeric or character information, and generate key signal input related to user settings and function control of the electronic device. The output device 340 may include a display device such as a display screen.

This embodiment also provides a computer-readable storage medium that stores computer-executable instructions, and the computer-executable instructions are used to execute the foregoing method.

All or part of the processes in the methods of the above-mentioned embodiments may be implemented by a computer program that executes the relevant hardware. The program may be stored in a non-transitory computer-readable storage medium. When the program is executed, it may include the methods described above. In the process of the embodiment of, the non-transitory computer-readable storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a RAM.

This disclosure uses the IEEE-39 node system as shown in Fig. 2 to describe the technical solution, and performs N-1 transient stability verification on 46 tie lines in turn. The fault type is three-phase short-circuit fault, so that each line is faulty. The removal time gradually increases until the system loses its stability for the first time, and then takes the system instability moment and the last system stability moment, both of which adopt the two-end approximation method to gradually determine the critical transient stable fault removal time, and the accuracy of the fault removal time is reached The approach is stopped at 0.005s, and the fault removal time at this time is selected as the critical removal time, and the state at this time is a critical transient stable state under a fault.

In order to achieve the purpose of comparison, under the benchmark of the critical removal time, 4 conventional transient stable fault removal times are selected, so that each line obtains 5 sample data under a single fault, and the other lines are processed in the same process , A total of 230 sample data were obtained, of which 46 were critical transient stable samples, accounting for 20% of the total number of samples.

In order to more intuitively describe the critical transient stability state of the power grid, one of the lines is selected, and through the above method, multiple samples of the same line under different fault removal times are obtained. The power angle time domain curve is shown in Figure 3.

Select the "source-net" input characteristics of the system, as shown in Table 1a or Table 1b;

Table 1a

Table 1b

Randomly select a set of fault data samples of the line. The set of data samples includes 4 conventional transient stability data and 1 critical transient stability data. The present disclosure selects the sample data of faults in lines 5-8, and conducts these samples. The matching results of SIFT algorithm are shown in Table 2;

Table 2

In order to more intuitively see the matching result trend of the algorithm to the "source-net" fault sample data, the matching degree of the "source-net" under different fault removal times can be formed into a matching degree change graph, and the result is shown in Figure 4:

For 46 lines of the IEEE-39 node system, each line selects a transient stability sample under non-critical fault removal time and critical fault actual time, and uses these 72 fault sample sets to do the feature matching of the SIFT algorithm. The matching result is shown in Figure 5.

As shown in Figure 3 and Figure 4, with the gradual increase of the fault removal time, the power angle time domain curve of the faulted line gradually tends to be unstable, and its characteristic matching index also gradually decreases, and at the critical fault removal time Time tends to 0.

As shown in Figure 5, after SIFT feature matching is performed on the fault sets of multiple lines, the matching degree under non-critical transient stability is significantly greater than that under critical transient stability, and the matching degree of critical transient stability tends to 0, indicating that the boundary characteristics of critical transient stability can be reflected by the index constructed by this algorithm, that is, when the matching degree index is close to 0, it can be considered that the transient stability is in a critical stable state at this time.

The present disclosure also provides a method for extracting critical transient stability boundary features of a large power grid, the method including:

Select the input characteristics of the source and network of the large power grid, according to the PMU measurement data, select when a fault occurs, when the fault is removed, and after the fault is removed, 3 groups of 12 feature variables constitute the generator feature variable set, and select when the fault occurs and when the fault is removed A total of 3 groups of 12 feature quantities after fault removal constitute a network feature quantity set;

Normalizing the generator feature set and network feature set level, and forming the processed data into a source matrix and a network matrix;

Construct the scale space of source matrix and net matrix;

Extremum detection of source matrix and net matrix in DOG space;

Filter feature points and locate key points, cut out low-contrast points, perform three-dimensional quadratic function fitting on local extreme points, and determine the location and scale of feature points;

Remove edge points according to the main curvature of the position and scale of the feature points, and assign direction values to the key points;

Generating a feature vector descriptor according to the direction value, and matching the feature vector;

Determine the feature quantity matching degree index, and obtain the feature matching degree index. The feature quantity matching degree index formula is as follows:

In the formula, H is the matching index of the source and network identification, A is the characteristic point of the source and network matrix, and B is the matching point of the source and network matrix, to identify the critical transient state of the large power grid. When H tends to 0, Large power grids tend to be in a critical transient state.

The present disclosure also provides a system for extracting critical transient stability boundary features of a large power grid, the system including:

The feature input module selects the input features of the large power grid source and network, and selects when the fault occurs, when the fault is removed, and after the fault is removed, 3 groups of 12 feature variables form the generator feature set, and select the time when the fault occurs , A total of 3 groups of 12 feature quantities during and after fault removal constitute a network feature quantity set;

A matrix forming module, which normalizes the generator feature set and network feature set level, and forms the source matrix and the network matrix with the processed data;

Construct the scale space module to construct the scale space of the source matrix and the net matrix;

Detection module, which performs extreme value detection on source matrix and net matrix in DOG space;

Filter module, filter feature points and locate key points, cut out low-contrast points, fit a three-dimensional quadratic function to local extreme points, and determine the location and scale of feature points;

An allocation module, removing edge points according to the main curvature of the position and scale of the feature points, and assigning direction values to the key points;

The matching module generates a feature vector descriptor according to the direction value, and matches the feature vector;

The feature acquisition module determines the feature matching degree index, and obtains the feature matching degree index. The feature matching degree index formula is as follows:

In the formula, H is the matching index of the source and network identification, A is the characteristic point of the source and network matrix, and B is the matching point of the source and network matrix, to identify the critical transient state of the large power grid. When H tends to 0, Large power grids tend to be in a critical transient state.

The present disclosure uses grid response information to construct indicators, and has less dependence on grid structure parameter information, and directly uses measurable grid response information, which makes the method strong in practicability and wide in range of use, and can be applied to multiple grid structure parameters Application scenarios.

The “source-net” input feature selected by the present disclosure is through repeated analysis of the correlation between the “source-net” state quantities, and finally the feature quantity with the best effect is selected, and the selected feature quantity is relatively more able to express the current The operating status of the power grid.

The present disclosure realizes the identification of critical states based on measurement data. Compared with traditional calculation based on modeling simulation, it only calculates power grid measurement information, avoids the complicated calculation process of modeling simulation method, and has faster recognition speed and timeliness. The performance is more in line with the requirements of modern power grids.

The present disclosure intuitively quantifies the boundary characteristics of critical transient stability through structural indicators of the boundary state of transient stability. The structured indicators can provide strong support for the realization of refined evaluation of transient stability, which has great academic research significance. And engineering use value.

Claims (26)

1. A method for identifying critical transient stable states of large power grids, the method comprising:

   Based on the actual measurement data on the power supply side of the large power grid, select the occurrence of the fault, when the fault is removed, and after the fault is removed, a total of 3 groups of 12 characteristic variables constitute the generator feature set. Based on the actual measurement data on the network side of the large power grid, select the fault occurrence When the fault is removed, a total of 3 groups of 12 feature quantities during and after the fault removal constitute the network feature quantity set;

   Normalizing the generator feature set and the network feature set respectively, the processed data in the generator feature set constitutes a source matrix, and the processed data in the network feature set constitutes a network matrix;

   Respectively constructing the Gaussian difference scale space of the source matrix and the net matrix;

   In the Gaussian difference scale space of the source matrix, extremum detection is performed on the source matrix to obtain the local extremum points corresponding to the source matrix. In the Gaussian difference scale space of the net matrix, the Performing extreme value detection on the net matrix to obtain local extreme points corresponding to the net matrix;

   Remove the local extreme points with low contrast and low principal curvature, and use the retained local extreme points as key points;

   Assign a direction value to the key point, and generate a feature vector corresponding to the key point according to the direction value;

   Determine the eigenvectors of the source matrix and the net matrix according to the eigenvectors corresponding to the key points;

   The feature vectors of the source matrix and the net matrix are matched, and a feature quantity matching degree index is determined. The formula of the feature quantity matching degree index is as follows:

   

   In the formula, Z is the feature quantity matching index, A is the smaller value of the number of key points of the source matrix and the number of key points of the net matrix, and B is the logarithm of the key points corresponding to the eigenvectors of the source matrix and the net matrix matching successfully;

   Identify the critical transient state of the large power grid, and when the feature quantity matching index tends to 0, the large power grid tends to a critical transient stable state.
2. The method according to claim 1, wherein the generator characteristic quantity set includes: generator power angle, generator excitation voltage, generator electromagnetic power, and generator reactive power.
3. The method according to claim 1 or 2, wherein the set of network characteristics includes: bus voltage amplitude, bus voltage phase angle, bus inflow active power, and bus inflow reactive power.
4. The method according to claim 1, 2 or 3, wherein the step of separately constructing the Gaussian differential scale space of the source matrix and the net matrix comprises:

   Generate the Gaussian scale space of the source matrix, calculate the first difference of Gaussian scale space functions corresponding to two adjacent layers of the same order in the Gaussian scale space of the source matrix, and use the first difference as the source A layer of the Gaussian difference scale space of the matrix;

   Generate the Gaussian scale space of the net matrix, calculate the second difference of the Gaussian scale space functions corresponding to two adjacent layers of the same order in the Gaussian scale space of the net matrix, and use the second difference as the net A layer of the Gaussian difference scale space of the matrix.
5. The method according to claim 4, wherein the Gaussian scale space function is expressed by the following formula:

   L(x,y,σ)=G(x,y,σ)*I(x,y)

   

   Among them, L(x,y,σ) represents the Gaussian scale space, G(x,y,σ) represents the two-dimensional space Gaussian function, I(x,y) represents the source matrix or network matrix, and (x,y) is the matrix For points on I, σ is the scale space factor, and * means convolution calculation.
6. The method according to claim 5, wherein the function of the Gaussian difference scale space is expressed by the following formula:

   D(x,y,σ)=(G(x,y,kσ)-G(x,y,σ))*I(x,y)

   =L(x,y,kσ)-L(x,y,σ)

   Among them, D (x, y, σ) represents the Gaussian difference scale space, and k is the multiple of the Gaussian scale space of two adjacent layers.
7. The method according to claim 1 or 6, wherein the step of removing local extreme points with low contrast and low principal curvature comprises:

   For each local extreme point, perform a three-dimensional quadratic function fitting on the function of the local extreme point corresponding to the Gaussian difference scale space to determine the position and scale of the local extreme point;

   Determining whether the local extreme point is a low-contrast point according to the scale of the local extreme point;

   Remove all the local extreme points determined as low contrast points, and remove the local extreme points with low principal curvature among the remaining local extreme points.
8. 8. The method according to claim 7, wherein the step of performing three-dimensional quadratic function fitting of the function of the local extreme point corresponding to the Gaussian difference scale space to determine the location and scale of the local extreme point, include:

   According to the Taylor expansion of the function D(x,y,σ) of the Gaussian difference scale space, the local extreme points are fitted with a three-dimensional quadratic function, and the Taylor expansion is expressed by the following formula:

   

   Let the partial derivative of D(x,y,σ) with respect to x be equal to 0, and obtain the position of the local extreme point

   

   

   Put

   

   Substitute into the Taylor expansion of D(x,y,σ) to obtain the scale of the local extreme point

   

   

   among them,

   

   Represents the first-order partial derivative of D ^{T with} respect to x,

   

   Indicates the second-order partial derivative of D to x, the superscript T means transpose, and the superscript -1 means matrix inversion.
9. The method according to claim 8, wherein the step of determining whether the local extreme point is a low-contrast point according to the scale of the local extreme point comprises:

   in

   

   In the case of determining that the local extreme point is not a low-contrast point;

   in

   

   In the case of, it is determined that the local extreme point is a low-contrast point.
10. 8. The method according to claim 7, wherein the step of removing the local extreme points of low principal curvature in the remaining local extreme points comprises:

    For each remaining local extreme point, the principal curvature of the local extreme point is obtained according to the Hessian matrix, and the formula of the Hessian matrix is as follows:

    

    Among them, H represents the Hessian matrix, and D _xx , D _xy , D _yx and D _yy are the elements of the Hessian matrix H with 2×2 dimensions;

    Let α be the maximum eigenvalue and β be the minimum eigenvalue, then:

    T _r (H)=D _xx +D _yy ＝α+β

    D _et (H)=D _xx D _yy -(D _xy ) ²

    =αβ

    Let α=γβ, then the principal curvature

    

    Determined by the following formula:

    

    In the case that the main curvature is not less than (r+1) ² /r, determine that the local extreme point is a point with a low main curvature, and remove the local extreme point that is determined to be a point with a low main curvature;

    In the case where the principal curvature is less than (r+1) ² /r, the local extreme point is retained as a key point.
11. The method according to claim 1, 5 or 10, wherein the step of assigning a direction value to the key point comprises:

    Calculate the direction value of the key point with the following formula:

    

    

    Among them, m(x,y) represents the modulus of the gradient at (x,y), θ(x,y) represents the direction of the gradient at (x,y), L is the function corresponding to the Gaussian scale space where the key point is located, tan Said to calculate the tangent value.
12. The method according to claim 1, wherein the step of matching the eigenvectors of the source matrix and the net matrix comprises:

    According to the Euclidean distance formula in the n-dimensional space, the eigenvectors of the source matrix and the net matrix are matched, and the Euclidean distance formula is as follows:

    

    Among them, ρ represents Euclidean distance, a[i] represents the i-th element of the eigenvector of the source matrix, b[i] represents the i-th element of the eigenvector of the net matrix, i=1, 2,...,n .
13. A device for identifying critical transient and stable states of large power grids, the device comprising:

    The feature input module is set to be based on the measured data on the power supply side of the large power grid. When a fault occurs, when the fault is removed, and after the fault is removed, a total of 3 groups of 12 feature variables constitute the generator feature set, based on the network side of the large power grid. Actually measured data, select the network characteristic quantity set of 3 groups of 12 characteristic quantities when the fault occurs, when the fault is removed, and after the fault is removed;

    The matrix forming module is configured to perform normalization processing on the generator feature set and the network feature set respectively, the processed data in the generator feature set constitutes a source matrix, and the processed network feature set The data constitutes a network matrix;

    A scale space construction module, configured to construct Gaussian difference scale spaces of the source matrix and the net matrix respectively;

    The detection module is configured to perform extreme value detection on the source matrix in the Gaussian differential scale space of the source matrix to obtain the local extreme points corresponding to the source matrix, in the Gaussian differential scale space of the net matrix , Performing extreme value detection on the net matrix to obtain local extreme points corresponding to the net matrix;

    The removal module is set to remove the local extreme points of low contrast and low main curve, and use the retained local extreme points as key points;

    An allocation module, configured to allocate a direction value to the key point, and generate a feature vector corresponding to the key point according to the direction value;

    A matching module, configured to determine the eigenvectors of the source matrix and the net matrix according to the eigenvectors corresponding to the key points, and match the eigenvectors of the source matrix and the net matrix;

    The feature acquisition module is configured to determine the feature quantity matching degree index, and the formula of the feature quantity matching degree index is as follows:

    

    In the formula, Z is the feature quantity matching index, A is the smaller value of the number of key points of the source matrix and the number of key points of the net matrix, and B is the logarithm of the key points corresponding to the eigenvectors of the source matrix and the net matrix matching successfully;

    The identification module is configured to identify the critical transient state of the large power grid, and when the feature quantity matching index tends to 0, the large power grid tends to a critical transient stable state.
14. The device according to claim 13, wherein the generator characteristic quantity set includes: generator power angle, generator excitation voltage, generator electromagnetic power and generator reactive power.
15. The device according to claim 13 or 14, wherein the set of network characteristic quantities comprises: bus voltage amplitude, bus voltage phase angle, bus inflow active power, and bus inflow reactive power.
16. The device according to claim 13, 14 or 15, wherein the scale space construction module is set to:

    Generate the Gaussian scale space of the source matrix, calculate the first difference of Gaussian scale space functions corresponding to two adjacent layers of the same order in the Gaussian scale space of the source matrix, and use the first difference as the source A layer of the Gaussian difference scale space of the matrix;

    Generate the Gaussian scale space of the net matrix, calculate the second difference of the Gaussian scale space functions corresponding to two adjacent layers of the same order in the Gaussian scale space of the net matrix, and use the second difference as the net A layer of the Gaussian difference scale space of the matrix.
17. The device according to claim 16, wherein the Gaussian scale space function is expressed by the following formula:

    L(x,y,σ)=G(x,y,σ)*I(x,y)

    

    Among them, L(x,y,σ) represents the Gaussian scale space, G(x,y,σ) represents the two-dimensional space Gaussian function, I(x,y) represents the source matrix or network matrix, and (x,y) is the matrix For points on I, σ is the scale space factor, and * means convolution calculation.
18. The apparatus according to claim 17, wherein the function of the Gaussian difference scale space is expressed by the following formula:

    D(x,y,σ)=(G(x,y,kσ)-G(x,y,σ))*I(x,y)

    =L(x,y,kσ)-L(x,y,σ)

    Among them, D (x, y, σ) represents the Gaussian difference scale space, and k is the multiple of the Gaussian scale space of two adjacent layers.
19. The device according to claim 13 or 18, wherein the removal module is configured to:

    For each local extreme point, perform a three-dimensional quadratic function fitting on the function of the local extreme point corresponding to the Gaussian difference scale space to determine the position and scale of the local extreme point;

    Determining whether the local extreme point is a low-contrast point according to the scale of the local extreme point;

    Remove all the local extreme points determined as low contrast points, and remove the local extreme points with low principal curvature among the remaining local extreme points.
20. The device according to claim 19, wherein the removal module is configured to:

    According to the Taylor expansion of the function D(x,y,σ) of the Gaussian difference scale space, the local extreme points are fitted with a three-dimensional quadratic function, and the Taylor expansion is expressed by the following formula:

    

    Let the partial derivative of D(x,y,σ) with respect to x be equal to 0, and obtain the position of the local extreme point

    

    

    Put

    

    Substitute into the Taylor expansion of D(x,y,σ) to obtain the scale of the local extreme point

    

    

    among them,

    

    Represents the first-order partial derivative of D ^{T with} respect to x,

    

    Indicates the second-order partial derivative of D to x, the superscript T means transpose, and the superscript -1 means matrix inversion.
21. The device according to claim 20, wherein the removal module is configured to:

    in

    

    In the case of determining that the local extreme point is not a low-contrast point;

    in

    

    In the case of, it is determined that the local extreme point is a low-contrast point.
22. The device according to claim 19, wherein the removal module is configured to:

    For each remaining local extreme point, the principal curvature of the local extreme point is obtained according to the Hessian matrix, and the formula of the Hessian matrix is as follows:

    

    Among them, H represents the Hessian matrix, and D _xx , D _xy , D _yx and D _yy are the elements of the Hessian matrix H with 2×2 dimensions;

    Let α be the maximum eigenvalue and β be the minimum eigenvalue, then:

    T _r (H)=D _xx +D _yy ＝α+β

    D _et (H)=D _xx D _yy -(D _xy ) ²

    =αβ

    Let α=γβ, then the principal curvature

    

    Determined by the following formula:

    

    In the case that the main curvature is not less than (r+1) ² /r, determine that the local extreme point is a point with a low main curvature, and remove the local extreme point that is determined to be a point with a low main curvature;

    In the case where the principal curvature is less than (r+1) ² /r, the local extreme point is retained as a key point.
23. The device according to claim 13, 17 or 22, wherein the distribution module is configured to:

    Calculate the direction value of the key point with the following formula:

    

    

    Among them, m(x,y) represents the modulus of the gradient at (x,y), θ(x,y) represents the direction of the gradient at (x,y), L is the function corresponding to the Gaussian scale space where the key point is located, tan Said to calculate the tangent value.
24. The device according to claim 13, wherein the matching module is configured to:

    According to the Euclidean distance formula in the n-dimensional space, the eigenvectors of the source matrix and the net matrix are matched, and the Euclidean distance formula is as follows:

    

    Among them, ρ represents Euclidean distance, a[i] represents the i-th element of the eigenvector of the source matrix, b[i] represents the i-th element of the eigenvector of the net matrix, i=1, 2,...,n .
25. An electronic device including:

    At least one processor;

    Memory, set to store at least one computer program,

    When the at least one computer program is executed by the at least one processor, the at least one processor realizes the method for identifying a critical transient stable state of a large power grid according to any one of claims 1-12.
26. A computer-readable storage medium storing computer-executable instructions for executing the method for identifying critical transient stable states of large power grids according to any one of claims 1-12.

Images