CN117110787B - Subsynchronous oscillation source positioning method of quaternary feature set convolutional neural network - Google Patents
Subsynchronous oscillation source positioning method of quaternary feature set convolutional neural network Download PDFInfo
- Publication number
- CN117110787B CN117110787B CN202311096456.8A CN202311096456A CN117110787B CN 117110787 B CN117110787 B CN 117110787B CN 202311096456 A CN202311096456 A CN 202311096456A CN 117110787 B CN117110787 B CN 117110787B
- Authority
- CN
- China
- Prior art keywords
- energy flow
- sso
- feature
- qfs
- power
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000010355 oscillation Effects 0.000 title claims abstract description 108
- 238000000034 method Methods 0.000 title claims abstract description 78
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 55
- 230000001052 transient effect Effects 0.000 claims abstract description 101
- 239000011159 matrix material Substances 0.000 claims abstract description 59
- 238000013136 deep learning model Methods 0.000 claims abstract description 12
- 238000000605 extraction Methods 0.000 claims abstract description 9
- 238000013507 mapping Methods 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 44
- 239000013598 vector Substances 0.000 claims description 29
- 230000002123 temporal effect Effects 0.000 claims description 11
- 238000013016 damping Methods 0.000 claims description 9
- 238000012806 monitoring device Methods 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 3
- 238000005265 energy consumption Methods 0.000 claims description 3
- 230000001360 synchronised effect Effects 0.000 claims description 3
- 230000036039 immunity Effects 0.000 abstract description 4
- 238000012360 testing method Methods 0.000 description 8
- 238000013135 deep learning Methods 0.000 description 6
- 238000009434 installation Methods 0.000 description 6
- 238000005259 measurement Methods 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000010248 power generation Methods 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000013526 transfer learning Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000005684 electric field Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000006698 induction Effects 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000001310 location test Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/08—Locating faults in cables, transmission lines, or networks
- G01R31/088—Aspects of digital computing
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/08—Locating faults in cables, transmission lines, or networks
- G01R31/081—Locating faults in cables, transmission lines, or networks according to type of conductors
- G01R31/086—Locating faults in cables, transmission lines, or networks according to type of conductors in power transmission or distribution networks, i.e. with interconnected conductors
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a subsynchronous oscillation source positioning method of a quaternary characteristic set convolutional neural network. The method comprises the following steps: acquiring voltage and current data of a port of a full-network observable element in a subsynchronous oscillation data sample from a power grid; according to the voltage and current data, calculating transient energy flow and energy flow power; according to the transient energy flow and the energy flow power, a two-dimensional SSO space-time feature extraction method based on the transient energy flow and the energy flow power is utilized to obtain a space-time feature matrix of an SSO sample; converting the feature matrix into a feature image by using color gamut mapping according to the space-time feature matrix of the SSO sample; and constructing a quaternary feature set convolutional neural network deep learning model based on the QFS data enhancement method by utilizing the quaternary feature set data enhancement method according to the time space feature image, and positioning the SSO oscillation source. The invention can rapidly position the SSO oscillation source, and has higher positioning precision and stronger noise immunity.
Description
Technical Field
The invention relates to the technical field of positioning of a subsynchronous oscillation source of an electric power system, in particular to a subsynchronous oscillation source positioning method of a quaternary characteristic set convolutional neural network.
Background
With the large-scale grid connection of renewable energy sources, a large number of power electronic devices are installed in the power system, and the transmission capacity of the power grid is improved. At the same time, however, the stability problems of the power system are also becoming increasingly pronounced, and subsynchronous oscillations (Sub-synchronous Oscillation, SSO) are becoming increasingly problematic. SSO may cause harmonic pollution, damage renewable energy equipment, and pose a threat to safe and stable operation of the power grid. The source of the SSO is rapidly and accurately positioned, and relevant parameters are timely adjusted, so that the method is important for eliminating the oscillation phenomenon.
The SSO can be locally distributed in a single wind power plant or can be spread to the whole power grid, and the analysis method of the numerical algorithm based on the traditional physical mechanism is applied to the novel SSO problem and has a certain limitation, however, the rapid development of artificial intelligence provides a new solution to the SSO problem of a complex power system.
Wind power system subsynchronous oscillation source positioning [ J ]. Electrician technical report containing flexible high-voltage direct current transmission based on countermeasure type transfer learning, 2021,36 (22): 4703-4715, establishes a linearization model of wind power plant grid connection with a power system through VSC-HVDC, and provides an oscillation source positioning method based on countermeasure type transfer learning. The method initially verifies the application feasibility of the artificial intelligence technology in system oscillation source positioning, but the method is shown to be further researched in a more complex large-scale wind power plant.
Y.Meng, Z.Yu, N.Lu, et al time Series Classification for Locating Forced Oscillation Sources [ J ]. IEEE Transactions on Smart Grid,2021,12 (2): 1712-1721, which proposes a machine learning based time series classification method to locate the source of forced oscillation. The method has the advantages of short calculation time, high precision and robustness. However, the proposed machine learning-based method cannot guarantee the accuracy of the result, especially when the forced oscillation source is driven by the load side and the forced oscillation source exhibits a dynamic response closely related to the load.
The positioning method of the subsynchronous oscillation source proposed by the two papers is a positioning method based on an artificial intelligence algorithm. A particular deep-learning network model needs to be trained relying on a large amount of sample data, however, the actual power system cannot collect oscillation data of the entire power grid, and there may not be enough marker data to support training of the deep-learning network model. Moreover, the monitored oscillation signal is usually unmarked and cannot be directly used for training of a deep learning network model. In addition, the method for positioning the oscillation source relies on the strong nonlinear fitting capability of deep learning, and ignores the physical mechanism of oscillation generation and propagation in the power system. Therefore, compared with the new searching method, the method lacks a certain interpretability, limits the positioning precision of the oscillation source and cannot adapt to more conditions.
Therefore, aiming at the SSO problem in the high-proportion renewable energy and power electronic equipment power system, the space-time feature map and QFS-CNN deep learning algorithm are utilized to rapidly and accurately position the SSO and timely cut off the SSO, and the method has important theoretical and practical significance.
Disclosure of Invention
The invention aims to provide a subsynchronous oscillation source positioning method of a quaternary characteristic set convolutional neural network, which aims to solve the problem that the traditional SSO positioning method is not suitable for the prior SSO positioning of high-proportion renewable energy sources and power electronic equipment power systems.
In order to solve the technical problems, the invention adopts the following technical scheme: a subsynchronous oscillation source positioning method of a quaternary feature set convolutional neural network comprises the following steps:
acquiring voltage and current data of a port of a full-network observable element in an SSO data sample from a power grid;
calculating transient energy flow and energy flow power according to the voltage and current data;
according to the transient energy flow and the energy flow power, a two-dimensional SSO space-time feature extraction method based on the transient energy flow and the energy flow power is utilized to obtain a space-time feature matrix of an SSO sample;
converting the feature matrix into a time space feature image by using color gamut mapping according to the space-time feature matrix of the SSO sample;
and constructing a QFS-CNN deep learning model based on the QFS data enhancement method by utilizing the QFS data enhancement method according to the time space feature image, and positioning an SSO oscillation source.
The technical scheme of the invention is further improved as follows: the calculating transient energy flow and energy flow power according to the voltage and current data comprises the following steps:
defining transient energy flow from a line start bus i to a line end bus j according to the voltage and current data transient values;
calculating transient energy flow power by utilizing derivative and combining three-phase voltage and current according to the transient energy flow;
and determining the energy flow power according to the transient energy flow power.
The technical scheme of the invention is further improved as follows: the transient energy flow is
Wherein W is ij Is a transient energy flow; the voltage and current in xy coordinate system can be converted from abc coordinate, and the coordinate conversion method is [ u ] x u y ] T ＝H[u a u b u c ] T ，[i x i y ] T ＝H[i a i b i c ] T Wherein the matrix is transformedWherein omega 0 Is the synchronous angular frequency; />The electrical angles of the x-axis lead the a-axis, the b-axis and the c-axis respectively, wherein the x-axis lags the y-axis by 90 degrees; i.e ij For line L ij Current on, represents conjugate, i ij,x And i ij,y Respectively is a line L ij Current in the x-axis and y-axis; u (u) i For the voltage of bus i, u i,x And u i,y The voltages of bus i on the x-axis and y-axis, u a ，u b And u c Instantaneous voltages on the a, b, c axes, i a ，i b And i c The instantaneous currents on the a, b, c axes, respectively.
The technical scheme of the invention is further improved as follows: the transient energy flow power is
p EF (t)＝-U ss I 0 (ω 0 -ω ss )cos((ω 0 -ω ss )t-α ss +β 0 )-U ss I ss (ω 0 -ω ss )cos(α ss -β ss )
Wherein p is EF (t) is the transient energy flow power; u (U) ss I 0 (ω 0 -ω ss )cos((ω 0 -ω ss )t-α ss +β 0 ) For oscillating terms, the term being related to the change in the transient energy of the element, U ss I ss (ω 0 -ω ss )cos(α ss -β ss ) The non-oscillation term corresponds to the consumption or generation of energy of the element and corresponds to the damping characteristic of the element; the non-oscillation term of the transient energy flow power is equivalent to the slope a of a transient energy streamline fitting curve W (t) =at+b, namely the energy flow power; wherein U is ss And alpha ss The amplitude and the phase of the subsynchronous frequency voltage are respectively; i 0 And beta 0 The amplitude and the phase of the fundamental frequency current are respectively; i ss And beta ss The amplitude and the phase of the sub-synchronous frequency current are respectively; omega 0 And omega ss The fundamental frequency power supply angular frequency and the subsynchronous frequency power supply angular frequency are respectively.
The technical scheme of the invention is further improved as follows: the method for extracting the space-time characteristic of the SSO sample by using the two-dimensional SSO space-time characteristic extraction method based on the transient energy flow and the energy flow power according to the transient energy flow and the energy flow power comprises the following steps:
constructing a characteristic vector of an observable bus of the whole power grid according to the transient energy flow and the energy flow power;
and constructing a space-time feature matrix of the SSO sample according to the feature vector of the observable bus of the whole power grid.
The technical scheme of the invention is further improved as follows: the feature vector of the whole power grid observable bus comprises: a transient energy flow vector and an energy flow power vector;
the transient energy flow vector is
W＝[w 1 w 2 … w j … w N ]
Wherein W represents t 0 From time to t N A temporal energy flow vector at a moment; w (w) j At t 0 From time to t j Transient energy at time; n represents the number of samples, j is 1,2, …, N;
the energy flow power vector is
P T ＝[a 1 a 2 … a i … a M ]
Wherein P is T Is a power vector of energy flow; a, a i Energy flow power on the ith bus; m is the number of buses for installing the power quality monitoring device.
The technical scheme of the invention is further improved as follows: the space-time feature matrix of the SSO sample comprises: a temporal feature matrix and a spatial feature matrix;
the time characteristic matrix is
Wherein W is T Is a transient energy matrix; w (W) i For the transient energy flow vector on the ith bus, M is the number of buses for installing the power quality monitoring device, and w i,j For the ith bus t 0 From time to t j Transient energy at time; the obtained two-dimensional time characteristic matrix characterizes the time lag characteristic of transient energy, and when the SSO oscillation source occurs on an unobservable bus, the SSO oscillation source can be positioned according to the transient energy consumption changes of other observable buses; w (W) T A temporal feature representing transient energy flowing through the positive damping characteristic element port; n represents the number of samples, j is 1,2, …, N;
the space characteristic matrix is
Δa i,j ＝a i -a j ,a i ,a j ∈P T
Wherein P is D Is a power characteristic matrix of energy flow; Δa i,j The difference value of the energy flow power of the ith bus and the jth bus is obtained; matrix P D The elements in (a) represent the difference in bus energy flow power at two different spatial locations.
The technical scheme of the invention is further improved as follows: according to the time space feature image, a QFS-CNN deep learning model based on the QFS data enhancement method is constructed by using the QFS data enhancement method, and an SSO oscillation source is positioned, comprising the following steps:
generating a large number of QFS feature sets through a QFS data enhancement method according to the time space feature image;
constructing a QFS training sample set according to the QFS feature set as a training sample;
according to the QFS training sample set, combining super parameters trained by the QFS-CNN deep learning model, learning by using the QFS-CNN deep learning model, and training an SSO oscillation source positioning model;
and positioning the SSO oscillation source according to the SSO oscillation source positioning model.
The technical scheme of the invention is further improved as follows: the QFS feature set is
Wherein Q is i Representing a corresponding QFS feature set when the SSO oscillation source is located on busbar i;and->Two time feature images which are randomly and repeatedly selected in the time feature image set i; />And->Two spatial feature images which are randomly and repeatedly selected in the spatial feature image set i; thus, the QFS data enhancement method can generate n using SSO samples with n oscillation source positions as bus i 4 The QFS feature set is used as a training sample of QFS-CNN, so that the diversity of the training sample is effectively improved.
By adopting the technical scheme, the invention has the following technical progress:
the embodiment of the invention provides a subsynchronous oscillation source positioning method of a quaternary feature set convolutional neural network, which comprises the steps of obtaining voltage and current instantaneous value signals of a port of a full-network observable element in an SSO data sample from a power grid, calculating transient energy flow and energy flow power, extracting time features and space features of the SSO sample by using a two-dimensional SSO space-time feature extraction method based on the transient energy and the energy flow power, obtaining a space-time feature matrix of the SSO sample, representing SSO feature information of the whole power system, generating a feature image by using the feature matrix by a color gamut mapping method, constructing a QFS-CNN deep learning model based on the QFS data enhancement method by using the QFS data enhancement method, and accordingly positioning the SSO oscillation source.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments or the description of the prior art will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method for positioning a subsynchronous oscillation source of a quaternary feature set convolutional neural network provided by an embodiment of the invention;
FIG. 2 is a schematic diagram of generating a temporal feature image and a spatial feature image according to an embodiment of the present invention;
FIG. 3 is a flow chart of the calculated transient energy flow and energy flow power provided by an embodiment of the present invention;
FIG. 4 is a flow chart of a determining time-space feature matrix provided by an embodiment of the present invention;
FIG. 5 is a modified IEEE-39 node wind power generation system provided by an embodiment of the invention;
fig. 6 is a network structure and a specific hyper-parameter diagram of the QFS-CNN according to an embodiment of the present invention;
FIG. 7 is a graph of loss function of QFS-CNN at different SNRs provided by an embodiment of the present invention;
fig. 8 is a graph showing training accuracy of QFS-CNN at different SNRs provided by an embodiment of the present invention.
Detailed Description
The invention is further illustrated by the following examples:
in the following description, for purposes of explanation and not limitation, specific details are set forth such as the particular system architecture, techniques, etc., in order to provide a thorough understanding of the embodiments of the present invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the following description will be made by way of specific embodiments with reference to the accompanying drawings.
Fig. 1 is a flowchart of an implementation of a method for positioning a subsynchronous oscillation source of a quaternary feature set convolutional neural network according to an embodiment of the present invention, which is described in detail below:
step 101, acquiring voltage and current data of a port of a full-network observable element in an SSO data sample from a power grid.
Optionally, the voltage and current data in the acquired SSO data samples are instantaneous value signals.
Step 102, calculating transient energy flow and energy flow power according to the voltage and current data.
Optionally, in this step, a transient energy flow is defined from the line start bus i to the end bus j, based on the voltage and current data transients. And deriving time of transient energy flow, performing xy rotation coordinate system conversion on three-phase voltage and current in an abc coordinate system, substituting the converted voltage and current into the derived transient energy flow, calculating transient energy flow power, and further determining energy flow power.
And step 103, according to the transient energy flow and the energy flow power, acquiring a space-time feature matrix of the SSO sample by using a two-dimensional SSO space-time feature extraction method based on the transient energy flow and the energy flow power.
Optionally, in this step, a feature vector of the observable busbar of the whole power grid is constructed according to the transient energy flow and the energy flow power, and a space-time feature matrix of the SSO sample is constructed.
Step 104, converting the feature matrix into a feature image by using color gamut mapping according to the space-time feature matrix of the SSO sample.
Optionally, to more intuitively represent the spatiotemporal features of SSO, and facilitate learning of deep learning algorithms. Time feature matrix W T And a spatial feature matrix P D Can be converted into corresponding temporal and spatial feature images by the method shown in fig. 2. First, find the time feature matrix W T And a spatial feature matrix P D To specify a color range. Then W is T And P D The elements with different values in the color map are respectively corresponding to different colors by a linear mapping method. And finally, filling the positions corresponding to the time feature matrix and the space feature matrix with rectangular color blocks respectively to form a time feature image and a space feature image, wherein the feature image is more visual than the feature matrix, and can better represent main feature information.
And 105, constructing a QFS-CNN deep learning model based on the QFS data enhancement method by utilizing the QFS data enhancement method according to the time space feature image, and positioning the SSO oscillation source.
Alternatively, a large number of QFS feature sets are generated from the time-space feature images by a QFS data enhancing method using limited feature images, and then a QFS training sample set is constructed using the QFS feature sets as training samples. Setting hyper-parameters of QFS-CNN model training, learning a constructed QFS training sample set by using the QFS-CNN model, training an SSO oscillation source positioning model, judging the positioning accuracy of the SSO oscillation source positioning model of offline training, and if the positioning accuracy of the SSO oscillation source positioning model of offline training cannot reach the expected value, adjusting the sample number of the QFS training sample set, and re-training the positioning model until the positioning accuracy of the model reaches the expected value. And finally, positioning the SSO oscillation source through a trained SSO oscillation source positioning model.
Optionally, wherein the QFS feature set is
Wherein Q is i Representing a corresponding QFS feature set when the SSO oscillation source is located on busbar i;and->Two time feature images which are randomly and repeatedly selected in the time feature image set i; similarly, let go of>And->Two spatial feature images which are randomly and repeatedly selected in the spatial feature image set i; thus, the QFS data enhancement method can generate n using SSO samples with n oscillation source positions as bus i 4 The QFS feature set is used as a training sample of QFS-CNN, so that the diversity of the training sample is effectively improved.
In one embodiment, as shown in fig. 3, in this step, calculating the transient energy flow and the energy flow power according to the voltage and current data may include:
step 301, defining a transient energy flow from a line start bus i to an end bus j according to the voltage and current data transient values.
Alternatively, the transient energy flow is
Wherein W is ij Is a transient energy flow; the voltage and current in xy coordinate system can be converted from abc coordinate, and the coordinate conversion method is [ u ] x u y ] T ＝H[u a u b u c ] T ，[i x i y ] T ＝H[i a i b i c ] T Wherein the matrix is transformedWherein omega 0 Is the synchronous angular frequency; />The electrical angles of the x-axis lead the a-axis, the b-axis and the c-axis respectively, wherein the x-axis lags the y-axis by 90 degrees; i.e ij For line L ij Current on, represents conjugate, i ij,x And i ij,y Respectively is a line L ij Current in the x-axis and y-axis; u (u) i For the voltage of bus i, u i,x And u i,y The voltages of bus i on the x-axis and y-axis, u a ，u b And u c Instantaneous voltages on the a, b, c axes, i a ，i b And i c The instantaneous currents on the a, b, c axes, respectively.
Step 302, calculating the transient energy flow power by utilizing derivative and combining the three-phase voltage and the current according to the transient energy flow.
Alternatively, the transient energy flow power is
p EF (t)＝-U ss I 0 (ω 0 -ω ss )cos((ω 0 -ω ss )t-α ss +β 0 )-U ss I ss (ω 0 -ω ss )cos(α ss -β ss )
Wherein p is EF (t) is the transient energy flow power; u (U) ss I 0 (ω 0 -ω ss )cos((ω 0 -ω ss )t-α ss +β 0 ) For oscillating terms, the term being related to the change in the transient energy of the element, U ss I ss (ω 0 -ω ss )cos(α ss -β ss ) The non-oscillation term corresponds to the consumption or generation of energy of the element and corresponds to the damping characteristic of the element; the non-oscillation term of the transient energy flow power is equivalent to the slope a of a transient energy streamline fitting curve W (t) =at+b, namely the energy flow power; wherein U is ss And alpha ss The amplitude and the phase of the subsynchronous frequency voltage are respectively; i 0 And beta 0 The amplitude and the phase of the fundamental frequency current are respectively; i ss And beta ss The amplitude and the phase of the sub-synchronous frequency current are respectively; omega 0 And omega ss The fundamental frequency power supply angular frequency and the subsynchronous frequency power supply angular frequency are respectively.
Step 303, determining the energy flow power according to the transient energy flow power.
Optionally, the non-oscillation term of the transient energy flow power is equivalent to the slope a of the transient energy streamline fitting curve W (t) =at+b, namely the energy flow power.
In one embodiment, as shown in fig. 4, in this step, the method for extracting the space-time feature matrix of the SSO sample by using the two-dimensional SSO space-time feature extraction method based on the transient energy and the power may include
And step 401, constructing a characteristic vector of an observable bus of the whole power grid according to the transient energy flow and the energy flow power.
Alternatively, the transient energy flow vector is
W＝[w 1 w 2 … w j … w N ]
Wherein W represents t 0 From time to t N A temporal energy flow vector at a moment; w (w) j At t 0 From time to t j Transient energy at time; n represents the number of samples, j is 1,2, …, N;
alternatively, the energy flow power vector is
P T ＝[a 1 a 2 … a i … a M ]
Wherein P is T Is a power vector of energy flow; a, a i For energy on the ith busFlow power; m is the number of bus bars on which PQM is mounted.
And step 402, constructing a space-time feature matrix of the SSO sample according to the feature vector of the observable bus of the whole power grid.
Optionally, the time feature matrix is
Wherein W is T Is a transient energy matrix; w (W) i For the transient energy flow vector on the ith bus, M is the number of buses for installing PQM, w i,j For the ith bus t 0 From time to t j Transient energy at time; the obtained two-dimensional time characteristic matrix characterizes the time lag characteristic of transient energy, and when the SSO oscillation source occurs on an unobservable bus, the SSO oscillation source can be positioned according to the transient energy consumption changes of other observable buses; w (W) T A temporal feature representing transient energy flowing through the positive damping characteristic element port; n represents the number of samples, j is 1,2, …, N;
optionally, the spatial feature matrix is
Δa i,j ＝a i -a j ,a i ,a j ∈P T
Wherein P is D Is a power characteristic matrix of energy flow; Δa i,j The difference value of the energy flow power of the ith bus and the jth bus is obtained; matrix P D The elements in (a) represent the difference in bus energy flow power at two different spatial locations.
The embodiment of the invention provides a subsynchronous oscillation source positioning method of a quaternary characteristic set convolutional neural network, which is characterized in that voltage and current instantaneous value signals of a port of a full-network observable element in an SSO data sample are obtained from a power grid, transient energy flow is calculated, then the transient energy flow is derived over time and substituted into converted voltage and current, transient energy flow power is calculated, energy flow power is further determined, a two-dimensional SSO space-time characteristic extraction method based on the transient energy and the energy flow power is utilized to extract space-time characteristics of the SSO sample, characteristic vectors of observable buses of the whole power grid are constructed, a space-time characteristic matrix of the SSO sample is obtained, SSO characteristic information of the whole power system is characterized, a characteristic image is generated by using the characteristic matrix through a color gamut mapping method, a QFS data enhancement method is utilized to generate a large number of QFS characteristic sets, an S-CNN deep learning model based on the QFS data enhancement method is constructed, the SSO oscillation source is positioned accordingly, the SSO oscillation source can be positioned rapidly, and the SSO has high positioning accuracy and high noise immunity.
The implementation of the present scheme is described below in a specific example.
An improved IEEE-39 node wind power generation system as shown in FIG. 5. The wind turbine electric fields WF1-WF5 are respectively connected to the buses 4, 7, 16, 18 and 27. Each wind power plant comprises 66 doubly-fed wind turbines (Doubly Fed Induction Generator, DFIG), the rated power of each DFIG is 1.5MW, and the wind speed is set to be 15 meters/second. In order to extract the temporal and spatial signature of the SSO source by the transient energy absorbed by the positive damping characteristic element in the case that the SSO source is not observable. Accordingly, the PQM is mounted only on bus bars 30-39, which is the 10 positive damping characteristic element ports in the system. The number of each PQM is the same as the number of the generator connected to the bus bar. The system frequency was 50Hz and the sampling frequency of the PQM was set to 1000Hz.
(1) Training data generation
To generate different SSO spatiotemporal features, in [0.1,0.2]Randomly selected network side converter current regulator integral coefficient K for DFIG within range p And enabling the corresponding wind power plant to be an SSO oscillation source to trigger SSO of the power grid. The positioning result of the SSO oscillation source may be affected by power quality monitoring device (PQM) noise. In addition to measurement noise, the influence of the installation quantity of the PQMs in the power grid on the positioning result of the SSO oscillation source is considered. Two influencing factors are shown below:
case 1: measurement noise of PQM (gaussian white noise with signal-to-noise ratio of 10, 20, 30, 40, 50 dB) was added.
Case 2: the number of PQM installations and the locations of the installations are shown in table 1.
TABLE 1 installation of PQM when observability decreases
The present invention simulates 500 oscillation scenarios to represent different SSO events. Of these 500 scenarios, 300 scenarios will be used as training samples for the training of the QFS-CNN positioning model, and the remaining 200 scenarios not used for the QFS-CNN training will be used for the positioning test of the trained SSO oscillation source positioning model, as shown in table 2.
Table 2 number of SSO simulation scenarios
Wherein,indicating that the SSO oscillation sources are randomly present in n of the m wind farms. Simulation of oscillation scenario improved IEEE-39 node wind power generation system built in MATLAB/Simulink simulation software by varying net side converter current regulator integral coefficient K of DFIG in 5 wind farms (WF 1-WF 5) p The wind farm WF is made to be an SSO oscillation source, and SSO is generated. The oscillation scene comprises the conditions that a single SSO oscillation source appears, two SSO oscillation sources appear simultaneously and three SSO oscillation sources appear simultaneously, and the total number of the oscillation types is 25.
(2) Positioning results for different numbers of training samples
When the sampling time of the simulation system is 1s, the acquired data (voltage and current) occupy 728KB of storage space, the time feature matrix and the space feature matrix occupy 76KB of storage space, the pixel size of the extracted feature image is 224 multiplied by 224, and the time feature image and the space feature image occupy 16KB of storage space. Therefore, the time feature image and the space feature image occupy the minimum storage space.
Respectively using 50 scenes @Namely, the case of two single SSO oscillation sources, the case of two simultaneous occurrence of two SSO oscillation sources and the case of two simultaneous occurrence of three SSO oscillation sources, the following SSO scenes are the same), 100 scenes->200 scenes->And 300 scenes%I.e., training scenario in table 2), 400 QFS feature sets are generated as training samples for QFS-CNN by the QFS data enhancement method. The four trained SSO oscillation source positioning models use the positioning accuracy of the 200 test scene test models in table 2, and the positioning results are shown in table 3.
TABLE 3 positioning accuracy of test in different training scenario numbers (%)
The network structure and specific super parameters of QFS-CNN are shown in FIG. 6. Table 3 illustrates the important role played by the QFS data enhancement method in SSO oscillation source positioning. Under the condition of using the QFS data enhancement method, the positioning precision of the trained SSO oscillation source positioning model is higher than that of the SSO oscillation source positioning model without using the QFS data enhancement method. Even under the condition of fewer training scenes, the QFS data enhancement method can still enable the positioning model to keep higher positioning accuracy. Therefore, the QFS data enhancement method can effectively improve the positioning precision of the SSO oscillation source.
(3) Positioning results under different noise
In an actual power system, noise of the PQM measurement data may affect the positioning result of the SSO positioning model. The present invention thus analyzes the performance of the proposed SSO source localization method taking into account noise. Under different white gaussian noise level scenes (case 1), 500 SSO scenes are respectively set according to table 2, measurement data are collected, and a space-time characteristic image is generated as input sample data of QFS-CNN. The loss function versus training accuracy curves for QFS-CNN at different signal-to-noise ratio (Signal to Noise Ratio, SNR) values are shown in FIGS. 7 and 8.
The loss function decreases with increasing training iterations, indicating that the QFS-CNN network model has good convergence, as shown in fig. 7. As can be seen from fig. 8, in the training process, when the number of iterations is less than 100, the influence of noise on the training accuracy is significant. Under the condition of no measurement noise, the network model training is iterated for 50 times to achieve higher precision. When the training iteration number is more than 150, the training accuracy under different noise levels can reach more than 90%. After the training iteration number reaches 300, the QFS-CNN positioning model reaches higher training precision and tends to be stable. This shows that QFS-CNN has good robustness to different noise environments, and can be used for positioning SSO oscillation sources in the actual running state of the power system.
Table 4 summarizes the SSO source location test results of the QFS-CNN location model with different numbers of SSO sources. Although the measurement data of the PQM is severely contaminated by noise, the trained QFS-CNN positioning model performs well in positioning multiple SSO oscillation sources. The result shows that the noise has less influence on the SSO oscillation positioning accuracy of the method. On one hand, the spatial characteristic image has stronger noise immunity, and on the other hand, the structure of the CNN can also eliminate noise to a certain extent. Therefore, the SSO oscillation source positioning method provided by the invention can accurately position a plurality of SSO oscillation sources and has stronger noise immunity.
TABLE 4 test positioning accuracy (%)
(4) Positioning results under different observability
The SSO oscillation source positioning method provided by the invention is used for positioning the SSO oscillation source according to transient energy absorbed by the positive damping characteristic element in the power system. In the case of poor system observability, the space-time characteristic image can be extracted through the voltage and current data of the measurable bus, and then the SSO oscillation source is positioned.
Under the condition of different PQM installation quantity, SSO oscillation source positioning is carried out on 200 test scenes by using a trained QFS-CNN positioning model, and the test positioning accuracy is listed in Table 5.
TABLE 5 test positioning accuracy (%)
As shown in table 5, the QFS-CNN positioning model has good positioning accuracy when the number of PQMs installed in the system is 7 and 10. When the number of PQM installations is reduced to 4 and 2, SSO oscillation source positioning accuracy is lowered because SSO information observed in the system is reduced. However, even if only 2 PQMs are installed in the system, the positioning accuracy of the QFS-CNN positioning method provided by the present invention is 93.50%. The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention, and are intended to be included in the scope of the present invention.
Claims (9)
1. A subsynchronous oscillation source positioning method of a quaternary feature set convolutional neural network is characterized by comprising the following steps:
acquiring voltage and current data of a port of a full-network observable element in an SSO data sample from a power grid;
calculating transient energy flow and energy flow power according to the voltage and current data;
according to the transient energy flow and the energy flow power, a two-dimensional SSO space-time feature extraction method based on the transient energy flow and the energy flow power is utilized to obtain a space-time feature matrix of an SSO sample;
converting the feature matrix into a time space feature image by using color gamut mapping according to the space-time feature matrix of the SSO sample;
and constructing a QFS-CNN deep learning model based on the QFS data enhancement method by utilizing the QFS data enhancement method according to the time space feature image, and positioning an SSO oscillation source.
2. The method for positioning a subsynchronous oscillation source of a quaternary feature set convolutional neural network according to claim 1, wherein the calculating transient energy flow and energy flow power according to the voltage and current data comprises:
defining transient energy flow from a line start bus i to a line end bus j according to the voltage and current data transient values;
calculating transient energy flow power by utilizing derivative and combining three-phase voltage and current according to the transient energy flow;
and determining the energy flow power according to the transient energy flow power.
3. The method for positioning subsynchronous oscillation sources of a quaternary feature set convolutional neural network according to any one of claims 1-2, wherein the transient energy flow is
Wherein W is ij Is a transient energy flow; voltage and current in xy coordinate systemFrom abc coordinate, the coordinate conversion method is [ u ] x u y ] T ＝H[u a u b u c ] T ，[i x i y ] T ＝H[i a i b i c ] T Wherein the matrix is transformedWherein omega 0 Is the synchronous angular frequency; />The electrical angles of the x-axis lead the a-axis, the b-axis and the c-axis respectively, wherein the x-axis lags the y-axis by 90 degrees; i.e ij For line L ij Current on, represents conjugate, i ij,x And i ij,y Respectively is a line L ij Current in the x-axis and y-axis; u (u) i For the voltage of bus i, u i,x And u i,y The voltages of bus i on the x-axis and y-axis, u a ，u b And u c Instantaneous voltages on the a, b, c axes, i a ，i b And i c The instantaneous currents on the a, b, c axes, respectively.
4. The method for positioning a subsynchronous oscillation source of a quaternary feature set convolutional neural network according to claim 2, wherein the transient energy flow power is
p EF (t)＝-U ss I 0 (ω 0 -ω ss )cos((ω 0 -ω ss )t-α ss +β 0 )-U ss I ss (ω 0 -ω ss )cos(α ss -β ss )
Wherein p is EF (t) is the transient energy flow power; u (U) ss I 0 (ω 0 -ω ss )cos((ω 0 -ω ss )t-α ss +β 0 ) For oscillating terms, the term being related to the change in the transient energy of the element, U ss I ss (ω 0 -ω ss )cos(α ss -β ss ) Is not aAn oscillation term, which corresponds to the consumption or generation of energy of the element, corresponding to the damping characteristic of the element; the non-oscillation term of the transient energy flow power is equivalent to the slope a of a transient energy streamline fitting curve W (t) =at+b, namely the energy flow power; wherein U is ss And alpha ss The amplitude and the phase of the subsynchronous frequency voltage are respectively; i 0 And beta 0 The amplitude and the phase of the fundamental frequency current are respectively; i ss And beta ss The amplitude and the phase of the sub-synchronous frequency current are respectively; omega 0 And omega ss The fundamental frequency power supply angular frequency and the subsynchronous frequency power supply angular frequency are respectively.
5. The method for positioning a subsynchronous oscillation source of a quaternary feature set convolutional neural network according to claim 1, wherein the obtaining a space-time feature matrix of an SSO sample by using a two-dimensional SSO space-time feature extraction method based on transient energy and energy flow power according to the transient energy flow and energy flow power comprises:
constructing a characteristic vector of an observable bus of the whole power grid according to the transient energy flow and the energy flow power;
and constructing a space-time feature matrix of the SSO sample according to the feature vector of the observable bus of the whole power grid.
6. The method for positioning a subsynchronous oscillation source of a quaternary feature set convolutional neural network according to claim 5, wherein the feature vector of the whole power grid observable bus comprises: a transient energy flow vector and an energy flow power vector;
the transient energy flow vector is
W＝[w 1 w 2 …w j …w N ]
Wherein W represents t 0 From time to t N A temporal energy flow vector at a moment; w (w) j At t 0 From time to t j Transient energy at time; n represents the number of samples, j is 1,2, …, N;
the energy flow power vector is
P T ＝[a 1 a 2 …a i …a M ]
Wherein P is T Is a power vector of energy flow; a, a i Energy flow power on the ith bus; m is the number of buses for installing the power quality monitoring device.
7. The method for positioning a subsynchronous oscillation source of a quaternary feature set convolutional neural network of claim 6, wherein the space-time feature matrix of SSO samples comprises: a temporal feature matrix and a spatial feature matrix;
the time characteristic matrix is
Wherein W is T Is a transient energy matrix; w (W) i For the transient energy flow vector on the ith bus, M is the number of buses for installing the power quality monitoring device, and w i,j For the ith bus t 0 From time to t j Transient energy at time; the obtained two-dimensional time characteristic matrix characterizes the time lag characteristic of transient energy, and when the SSO oscillation source occurs on an unobservable bus, the SSO oscillation source can be positioned according to the transient energy consumption changes of other observable buses; w (W) T A temporal feature representing transient energy flowing through the positive damping characteristic element port; n represents the number of samples, j is 1,2, …, N;
the space characteristic matrix is
Δa i,j ＝a i -a j ,a i ,a j ∈P T
Wherein P is D Is a power characteristic matrix of energy flow; Δa i,j The difference value of the energy flow power of the ith bus and the jth bus is obtained; matrix P D The elements in (a) represent bus energy flows at two different spatial positionsDifference in power.
8. The method for positioning a subsynchronous oscillation source of a quaternary feature set convolutional neural network according to claim 1, wherein the constructing a QFS-CNN deep learning model based on a QFS data enhancement method according to the time-space feature image, positioning the SSO oscillation source comprises:
generating a large number of QFS feature sets through a QFS data enhancement method according to the time space feature image;
constructing a QFS training sample set according to the QFS feature set as a training sample;
according to the QFS training sample set, combining super parameters trained by the QFS-CNN deep learning model, learning by using the QFS-CNN deep learning model, and training an SSO oscillation source positioning model;
and positioning the SSO oscillation source according to the SSO oscillation source positioning model.
9. The method for positioning a subsynchronous oscillation source of a quaternary feature set convolutional neural network according to claim 8, wherein the QFS feature set is
Wherein Q is i Representing a corresponding QFS feature set when the SSO oscillation source is located on busbar i;and->Two time feature images which are randomly and repeatedly selected in the time feature image set i; />And->Two spatial feature images which are randomly and repeatedly selected in the spatial feature image set i; thus, the QFS data enhancement method can generate n using SSO samples with n oscillation source positions as bus i 4 The QFS feature set is used as a training sample of QFS-CNN, so that the diversity of the training sample is effectively improved.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311096456.8A CN117110787B (en) | 2023-08-29 | 2023-08-29 | Subsynchronous oscillation source positioning method of quaternary feature set convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311096456.8A CN117110787B (en) | 2023-08-29 | 2023-08-29 | Subsynchronous oscillation source positioning method of quaternary feature set convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117110787A CN117110787A (en) | 2023-11-24 |
| CN117110787B true CN117110787B (en) | 2024-02-09 |
Family
ID=88799694
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311096456.8A Active CN117110787B (en) | 2023-08-29 | 2023-08-29 | Subsynchronous oscillation source positioning method of quaternary feature set convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117110787B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113285471A (en) * | 2021-06-25 | 2021-08-20 | 广东电网有限责任公司 | Method, device and equipment for sensing and positioning sub-supersynchronous oscillation source of offshore wind power plant |
| CN114792984A (en) * | 2022-04-18 | 2022-07-26 | 国网山东省电力公司莱芜供电公司 | Quick positioning method for subsynchronous/supersynchronous oscillation source |
| CN115187154A (en) * | 2022-09-14 | 2022-10-14 | 华中科技大学 | Neural network-based regional power grid oscillation source risk prediction method and system |
| CN115238980A (en) * | 2022-07-13 | 2022-10-25 | 湖南大学 | Micro-grid transient power angle trajectory prediction method based on attention mechanism and space-time diagram convolutional neural network |
| CN115377999A (en) * | 2022-09-26 | 2022-11-22 | 华北电力大学 | Subsynchronous oscillation identification method based on broadband measurement data |
| WO2023207124A1 (en) * | 2022-04-25 | 2023-11-02 | 中国电力科学研究院有限公司 | Power grid oscillation source distributed positioning method and system, device, storage medium, computer program, and computer program product |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220115871A1 (en) * | 2020-10-08 | 2022-04-14 | Zhe Yu | Power System Low-Frequency Oscillation Mechanism Identification with CNN and Transfer Learning |
-
2023
- 2023-08-29 CN CN202311096456.8A patent/CN117110787B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113285471A (en) * | 2021-06-25 | 2021-08-20 | 广东电网有限责任公司 | Method, device and equipment for sensing and positioning sub-supersynchronous oscillation source of offshore wind power plant |
| CN114792984A (en) * | 2022-04-18 | 2022-07-26 | 国网山东省电力公司莱芜供电公司 | Quick positioning method for subsynchronous/supersynchronous oscillation source |
| WO2023207124A1 (en) * | 2022-04-25 | 2023-11-02 | 中国电力科学研究院有限公司 | Power grid oscillation source distributed positioning method and system, device, storage medium, computer program, and computer program product |
| CN115238980A (en) * | 2022-07-13 | 2022-10-25 | 湖南大学 | Micro-grid transient power angle trajectory prediction method based on attention mechanism and space-time diagram convolutional neural network |
| CN115187154A (en) * | 2022-09-14 | 2022-10-14 | 华中科技大学 | Neural network-based regional power grid oscillation source risk prediction method and system |
| CN115377999A (en) * | 2022-09-26 | 2022-11-22 | 华北电力大学 | Subsynchronous oscillation identification method based on broadband measurement data |
Non-Patent Citations (2)
| Title |
|---|
| 基于对抗式迁移学习的含柔性高压直流输电的风电系统次同步振荡源定位;陈剑 等;《电工技术学报》;第36卷(第22期);4703-4715 * |
| 风电系统次同步振荡检测及定位技术研究;吕万;《中国优秀硕士全文论文数据库 工程科技II辑》(第6期);第4.2节 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117110787A (en) | 2023-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Bie et al. | Reliability evaluation of active distribution systems including microgrids | |
| CN112636341B (en) | Power system inertia spatial distribution estimation method and device based on multiple innovation identification | |
| CN103441527B (en) | A kind of wind power integration system model based on measured data | |
| CN110854884B (en) | Wind power collection area subsynchronous oscillation risk online assessment and early warning method | |
| CN111103503B (en) | Micro-grid fault positioning method based on simulation | |
| CN114123344A (en) | Power system inertia evaluation method and device based on adaptive recursive least squares | |
| Wu et al. | Development of an equivalent wind farm model for frequency regulation | |
| CN112072692A (en) | Impedance equivalence method and device for new energy power generation station | |
| Rao et al. | Wideband impedance online identification of wind farms based on combined data-driven and knowledge-driven | |
| CN117110787B (en) | Subsynchronous oscillation source positioning method of quaternary feature set convolutional neural network | |
| Morren et al. | Grid interaction of offshore wind farms. Part 1. Models for dynamic simulation | |
| Rudion et al. | MaWind-tool for the aggregation of wind farm models | |
| Jadidi et al. | Fault detection and diagnosis in power electronic converters at microgrid level based on filter bank approach | |
| CN109657380A (en) | A kind of double-fed fan motor field Dynamic Equivalence based on Extended Kalman filter | |
| CN115377999A (en) | Subsynchronous oscillation identification method based on broadband measurement data | |
| CN108964044B (en) | Ocean micro-grid group equivalent modeling method interconnected through submarine cables | |
| Tian et al. | Simulation study on direct‐drive wind power system | |
| CN111817359A (en) | Micro-grid solvable boundary analysis method based on equivalent circuit | |
| Abdalla et al. | Aggregation of a Wind Farm Model for Grid Connection Planning Studies | |
| Ma et al. | Cyber-physical Modeling Technique based Dynamic Aggregation of Wind Farm Considering LVRT Characteristic | |
| Preda et al. | Dynamic equivalents of active distribution power systems for investigation of transient stability | |
| Wang et al. | Research on Wind Farm Clustering Modeling for Power Grid Dynamic Equivalence | |
| Li et al. | Online Frequency Response Analysis of Multi-energy System Based on CloudPSS Real-time Simulation | |
| Zhang et al. | Research on equivalent electromechanical transient modeling of PMSG-based wind farms | |
| Yin et al. | Analysis and research on short-circuit current characteristics and grid access faults of wind farms with multi-type fans |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |