Broadband oscillation detection and tracing method, storage medium and processor

Abstract

The invention discloses a broadband oscillation detection and tracing method, a storage medium and a processor, relates to the technical field of power systems, and solves the technical problem of low accuracy of detection and tracing of broadband oscillation signals in the prior art. The method comprises the following steps: constructing a mathematical model of a broadband oscillation signal; converting the broadband oscillation signal to a frequency domain based on discrete Fourier transform to obtain a plurality of frequency decomposition values; detecting the broadband oscillation signal based on Hilbert transform; respectively carrying out image matrix transformation on the plurality of coded data to obtain an oscillation characteristic matrix; and training the deep learning model based on the oscillation characteristic matrix and the corresponding oscillation position label to obtain an oscillation positioning model. According to the invention, through the strong sequence feature capturing capability of the convolutional neural network, each node feature vector and the system adjacency matrix are input to the oscillation positioning model, high-precision fitting of the broadband oscillation signal is realized, and the accuracy of detection and traceability of the broadband oscillation signal is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power systems, and specifically relates to a broadband oscillation detection and source tracing method, a storage medium and a processor. Background Art

[0002] As the degree of power electronics in the power system's "source-grid-load" increases year by year, the power system is gradually developing towards a high proportion of new energy generation and a high proportion of power electronic equipment ("double high"), and its dynamic characteristics have also undergone profound changes, resulting in new stability problems, such as broadband oscillation. Due to the influence of the topological structure and parameters of the power electronic device, the broadband oscillation mode will be unstable and the oscillation frequency will drift, making the broadband oscillation show significant time-varying characteristics. The frequency of broadband oscillation ranges from a few hertz to several thousand hertz, involving multiple regions and multiple electrical equipment. Multiple modes coexist in the oscillation, further enhancing its nonlinear characteristics.

[0003] In the context of increasingly complex operation and monitoring requirements of modern distribution networks, the data obtained by traditional data acquisition and monitoring control systems (SCADA) can no longer meet these high standards. In particular, traditional methods have difficulty overcoming the problem of low sampling rate of the phasor measurement unit (PMU) device itself, which limits its ability to cope with broadband monitoring requirements. To solve the above problems, the new method introduces a phasor measurement unit (PMU) with high precision and strong real-time performance, which can not only accurately measure the voltage phasor and branch current vector, but also monitor the bus voltage and generator power angle in real time, thereby significantly improving the accuracy of power system state estimation. In addition, combined with the new generation of artificial intelligence technology, this method makes full use of its powerful nonlinear fitting ability and feature expression ability to extract information that helps accurate classification from massive and complex data, and achieve faster and more accurate detection. By automatically learning input and output samples, artificial intelligence technology can fit complex nonlinear functions with arbitrary precision, and provide functions such as classification and dimensionality reduction, effectively solving the problem of insufficient fitting accuracy in traditional signal processing methods.

[0004] The present invention provides a broadband oscillation detection and source tracing method, a storage medium and a processor to solve the above technical problems. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a wide-band oscillation detection and tracing method, which is used to solve the technical problem that when detecting and tracing the wide-band oscillation signal in the prior art, the fitting accuracy of the signal processing method is not high, resulting in low accuracy in detecting and tracing the wide-band oscillation signal.

[0006] To achieve the above object, a first aspect of the present invention provides a broadband oscillation detection and source tracing method, comprising:

[0007] S1: Construct a mathematical model of broadband oscillation signal;

[0008] S2: Perform state estimation on the measurement data of the PMU device to obtain a state estimation result;

[0009] S3: Convert the broadband oscillation signal into the frequency domain based on discrete Fourier transform to obtain several frequency decomposition values;

[0010] S4: Detection of broadband oscillation signal based on Hilport transform;

[0011] S5: Perform self-encoding training on a number of frequency decomposition values ​​based on an intelligent encoding algorithm to obtain a number of encoding data;

[0012] S6: performing image matrix transformation on a plurality of coded data respectively to obtain an oscillation characteristic matrix;

[0013] S7: Train the deep learning model based on the oscillation feature matrix and the corresponding oscillation position labels to obtain an oscillation positioning model.

[0014] Preferably, the mathematical model of the broadband oscillation signal is:

[0015]

[0016] Where t is time, f is oscillation frequency, A i (t) is the function of the signal amplitude of the i-th sampling point changing with time, θ i (t) is a function of the phase angle of the i-th sampling point changing with time; wherein, i=1, 2, 3, ..., n, and n is the total number of sampling points.

[0017] Preferably, the state estimation of the measurement data of the PMU device includes:

[0018] S21: Extract the measurement data of the PMU device and use formula (1) to establish the objective function of distribution network state estimation:

[0019] minJ(x)=[zh(x)] T R -1 [zh(x)]; (2)

[0020] Wherein, z is the measurement vector of the power grid, x is the state vector of the power grid, h(x) is the nonlinear mapping from the state vector x to the measurement vector z, R is the covariance matrix of the measurement error, and is a diagonal matrix, the i-th diagonal element in R corresponds to the error variance of the i-th measurement value in z, and T represents the transpose;

[0021] S22: Solve the objective function minJ(x) to obtain the state estimation result.

[0022] Preferably, the performing precision analysis on the measurement results to obtain accurate status data includes:

[0023] S31: Calculate the average voltage amplitude estimation error e of the sampling point by formula mag :

[0024]

[0025] Among them, x i,true represents the true value of the voltage amplitude at the i-th sampling point, x i,est represents the estimated value of the voltage amplitude at the i-th sampling point;

[0026] S32: Calculate the average voltage phase angle estimation error e at the sampling point by using the formula ang :

[0027]

[0028] in, represents the true value of the voltage phase angle at the i-th sampling point, represents the estimated value of the voltage phase angle at the i-th sampling point;

[0029] S33: Select the minimum value of the average voltage amplitude estimation error and the average voltage phase angle estimation error of the sampling points and mark it as state data.

[0030] Preferably, the converting of the broadband oscillation signal into the frequency domain based on discrete Fourier transform comprises:

[0031] The broadband oscillation signal is extracted and marked as x(m). The broadband oscillation signal x(m) is decomposed by discrete Fourier transform. The formula is as follows:

[0032]

[0033] Wherein, m is the window length, the value range of m is 0≤m≤M-1, M is the total number of window lengths, and X(m) is the frequency domain decomposition value of the broadband oscillation signal x(m).

[0034] Preferably, the detecting of the broadband oscillation signal based on Hilport transform comprises:

[0035] The pure noise intrinsic mode function IMF is denoised by the autocorrelation coefficient algorithm to obtain the noisy IMF, and the noisy IMF is subjected to threshold denoising to obtain the effective IMF. The Hilbert transform is used to detect the oscillation of the frequency and amplitude of the effective IMF.

[0036] The present invention removes pure noise IMF by calculating the autocorrelation coefficient, and further performs threshold noise reduction processing on the remaining noisy IMF to obtain effective IMF; the frequency and amplitude of each effective IMF are measured by Hilbert transform, and the broadband oscillation signal is detected, which is beneficial to improve the accuracy of detecting the broadband oscillation signal.

[0037] Preferably, performing image matrix transformation on a plurality of coded data respectively comprises:

[0038] The electrical quantities of several sampling points are extracted and expressed in the form of an oscillation characteristic matrix Y. The oscillation characteristic matrix Y is shown as follows:

[0039]

[0040] Among them, y vr is the rth electrical quantity at the vth sampling point, and 1≦v≦q, 1≦r≦p; q is the encoding dimension of each electrical quantity after compression, and p is the total number of electrical quantities collected in the entire network.

[0041] It should be noted that y 11 It is the first electrical quantity collected at the first sampling point, and the rest are similar.

[0042] Preferably, the training of the deep learning model based on the oscillation feature matrix and the corresponding oscillation position labels includes:

[0043] The mapping relationship between the oscillation feature matrix and the corresponding oscillation source position label is obtained and integrated into a standard data set, and the standard data set is divided into several groups of training data, verification data and test data; the training data is used to train the deep learning model; the verification data is used to verify the trained deep learning model, and the test data is used to evaluate the model performance of the deep learning model; finally, an oscillation positioning model is obtained with the oscillation feature matrix as input and the oscillation source position as output; wherein, the deep learning model includes a graph neural network model and a convolutional neural network model.

[0044] The present invention obtains the mapping relationship between the oscillation feature matrix and the corresponding oscillation source position label, trains the deep learning model, and marks the trained deep learning model as an oscillation positioning model, so that the fitting accuracy of the broadband oscillation signal is higher, which is beneficial to improving the accuracy of detection and tracing of the broadband oscillation signal.

[0045] Preferably, the deep learning model extracts features from the broadband oscillation signal, including:

[0046] The feature vectors of each sampling point are obtained and integrated into the information matrix X. The graph neural network model is used as the deep learning model, and the adjacency matrix B of the graph is expressed as follows:

[0047]

[0048] Among them, v1 is the number of rows of the adjacency matrix B, v2 is the number of columns of the adjacency matrix B, and E is the set of connection relationships between vertices in the graph;

[0049] The Graph Conv layer based on the graph convolutional neural network processes and outputs the adjacency matrix B. The processing formula is as follows:

[0050] H (x+c) =f(H (x) ,B); (8)

[0051] Among them, f() is a nonlinear function, c is the number of hidden layers;

[0052] When c is equal to 0, the matrix H is equal to the information matrix X; the values ​​of the adjacency matrix B at the diagonal position are all 0. When the adjacency matrix B and the matrix H are multiplied, only the weighted sum of the features of all adjacency matrices of the sampling point is calculated, and the features of the sampling point itself are ignored. The specific feature extraction rules are as follows:

[0053] H (x+1) =f(H (x) ,B)=σ(B (X) H (X) W (x) ); (9)

[0054] Among them, W is a trainable parameter matrix, σ() is a nonlinear function, and after the softmax() function operation, the label representing the location of the oscillation source is obtained.

[0055] A second aspect of the present invention provides a storage medium, which includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the above-mentioned wide-band oscillation detection and tracing method.

[0056] A third aspect of the present invention provides a processor, which is used to run a program, wherein the program executes the above-mentioned wide-band oscillation detection and tracing method when running.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] 1. The present invention accurately converts complex broadband oscillation signals from the time domain to the frequency domain through the discrete Fourier transform method, and further significantly improves the accuracy and comparability of the data by normalizing the original measurement data; on this basis, the decomposition values ​​of each frequency of the broadband signal are successfully decomposed, which is conducive to improving the accuracy of detecting the broadband oscillation signal.

[0059] 2. The broadband oscillation problem, due to its complex spatiotemporal characteristics and system topological structure, leads to the mutual coupling of electrical quantities at each sampling point, forming a difficult problem to solve in non-Euclidean space. The present invention introduces a convolutional neural network and utilizes its powerful sequence feature capture capability to input the feature vector of each node and the system adjacency matrix into the oscillation positioning model, thereby achieving high-precision fitting of broadband oscillation signals. This innovative method not only improves the fitting accuracy, but also significantly enhances the accuracy of signal tracing, which is conducive to improving the accuracy of tracing broadband oscillation signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0061] Figure 1 It is an overall flow chart of the broadband oscillation detection and tracing method based on deep learning of the present invention;

[0062] Figure 2 Schematic diagram of the principle of the graph convolutional neural network in the present invention. DETAILED DESCRIPTION

[0063] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] See also Figure 1-Figure 2 The first aspect of the present invention provides a broadband oscillation detection and source tracing method, comprising:

[0065] S1: Construct a mathematical model of broadband oscillation signal;

[0066] S2: Perform state estimation on the measurement data of the PMU device to obtain a state estimation result;

[0067] S3: Convert the broadband oscillation signal into the frequency domain based on discrete Fourier transform to obtain several frequency decomposition values;

[0068] S4: Detection of broadband oscillation signal based on Hilport transform;

[0069] S5: Perform self-encoding training on a number of frequency decomposition values ​​based on an intelligent encoding algorithm to obtain a number of encoding data;

[0070] S6: performing image matrix transformation on a plurality of coded data respectively to obtain an oscillation characteristic matrix;

[0071] S7: Train the deep learning model based on the oscillation feature matrix and the corresponding oscillation position labels to obtain an oscillation positioning model.

[0072] In this embodiment, a mathematical model of a broadband oscillation signal is constructed, including:

[0073] Firstly, according to the on-site signal analysis of broadband oscillation, the broadband oscillation signal has the processes of modulation, frequency ramp, linear amplitude change, etc., and the amplitude and phase angle have changed in time in these processes;

[0074] The mathematical model of the broadband oscillation signal is constructed by the following formula:

[0075]

[0076] Where t is time, f is oscillation frequency, A i (t) is the function of the signal amplitude of the i-th sampling point changing with time, θ i (t) is a function of the phase angle of the i-th sampling point changing with time; wherein, i=1, 2, 3, ..., n, and n is the total number of sampling points.

[0077] In this embodiment, the state estimation is performed on the measurement data of the PMU device, including:

[0078] S21: Extract the measurement data of the PMU device and use formula (1) to establish the objective function of distribution network state estimation:

[0079] minJ(x)=[zh(x)] T R -1 [zh(x)]; (2)

[0080] Wherein, z is the measurement vector of the power grid, x is the state vector of the power grid, h(x) is the nonlinear mapping from the state vector x to the measurement vector z, R is the covariance matrix of the measurement error, and is a diagonal matrix, the i-th diagonal element in R corresponds to the error variance of the i-th measurement value in z, and T represents the transpose;

[0081] S22: Solving the objective function minJ(x), including:

[0082] To obtain the optimal value, formula (2) must satisfy the following first-order optimal conditions:

[0083]

[0084] in, is the Jacobian matrix; the first-order optimal condition is an n-dimensional nonlinear algebraic equation system, which is generally solved by Newton's method; g(x) is calculated at x (k) Performing Taylor series expansion at , we can get:

[0085] g(x)=g(x (k) )+G(x (k) )(xx (k) )+L=0;

[0086] in, Ignoring the higher-order term L, the following linear form of the canonical equation can be obtained:

[0087] G(x (k) )Δx (k) =H T (x (k) )R -1 [zh(x (k) )];

[0088] Where k is the iteration counter, Δx (k) =x (k+1) -x (k) , G=H T R -1 H and G are symmetric positive definite matrices. When the system satisfies observability, G is decomposed by Cholesky, and the equation is solved by forward-backward substitution method to obtain the state estimation result.

[0089] In this embodiment, the measurement results are analyzed for accuracy to obtain accurate status data, including:

[0090] S31: Calculate the average voltage amplitude estimation error e of the sampling point by formula mag :

[0091]

[0092] Among them, x i,true represents the true value of the voltage amplitude at the i-th sampling point, x i,est represents the estimated value of the voltage amplitude at the i-th sampling point;

[0093] S32: Calculate the average voltage phase angle estimation error e at the sampling point by using the formula ang :

[0094]

[0095] in, represents the true value of the voltage phase angle at the i-th sampling point, represents the estimated value of the voltage phase angle at the i-th sampling point;

[0096] S33: Select the minimum value of the average voltage amplitude estimation error and the average voltage phase angle estimation error of the sampling points and mark it as state data.

[0097] In this embodiment, it is assumed that the data sampling rate of the PMU unit is f s , the data upload frequency is F s , under the existing communication bandwidth, if the autoencoder is used to encode and compress the PMU unit data u times, then f s With F s The mapping relationship between them is f s =F s ×u.

[0098] In this embodiment, the formula Normalize the broadband oscillation signal, where x max is the maximum value of the broadband oscillation signal, x min is the minimum value of the broadband oscillation signal.

[0099] In this embodiment, the broadband oscillation signal is converted into the frequency domain based on discrete Fourier transform, including:

[0100] The broadband oscillation signal is extracted and marked as x(m). The broadband oscillation signal x(m) is decomposed by discrete Fourier transform. The formula is as follows:

[0101]

[0102] Wherein, m is the window length, the value range of m is 0≤m≤M-1, M is the total number of window lengths, and X(m) is the frequency domain decomposition value of the broadband oscillation signal x(m).

[0103] In this embodiment, the broadband oscillation signal is detected based on the Hilport transform, including:

[0104] The pure noise intrinsic mode function IMF is denoised by the autocorrelation coefficient algorithm to obtain the noisy IMF, and the noisy IMF is subjected to threshold denoising to obtain the effective IMF. The Hilbert transform (HT transform) is used to detect the oscillation of the frequency and amplitude of the effective IMF.

[0105] The present invention eliminates pure noise IMFs by calculating the autocorrelation coefficient, and implements fine threshold noise reduction on the remaining noisy IMFs to obtain effective IMFs. The frequency and amplitude of each effective IMF are accurately measured by Hilbert transform, which significantly improves the detection accuracy of broadband oscillation signals, thereby enhancing the accuracy of tracing the source and providing a solid guarantee for the stable operation of the power system.

[0106] In this embodiment, image matrix transformation is performed on a plurality of coded data respectively, including:

[0107] The electrical quantities of several sampling points are extracted and expressed in the form of an oscillation characteristic matrix Y. The oscillation characteristic matrix Y is shown as follows:

[0108]

[0109] Among them, y vr is the rth electrical quantity at the vth sampling point, and 1≦v≦q, 1≦r≦p; q is the encoding dimension of each electrical quantity after compression, and p is the total number of electrical quantities collected in the entire network.

[0110] It should be noted that y 11 It is the first electrical quantity collected at the first sampling point, and the rest are similar.

[0111] In this embodiment, the deep learning model is trained based on the oscillation feature matrix and the corresponding oscillation position labels, including:

[0112] The mapping relationship between the oscillation feature matrix and the corresponding oscillation source position label is obtained and integrated into a standard data set, and 60% of the standard data set is divided into several groups of training data, 20% is divided into verification data, and 20% is divided into test data; the training data is used to train the deep learning model; the verification data is used to verify the trained deep learning model, and the test data is used to evaluate the model performance of the deep learning model; finally, an oscillation positioning model with the oscillation feature matrix as input and the oscillation source position as output is obtained; wherein, the deep learning model includes a graph neural network model and a convolutional neural network model.

[0113] The present invention maps the relationship between the oscillation feature matrix and the oscillation source position label and uses it to train the deep learning model. After the training is completed, a high-precision oscillation positioning model is obtained; the oscillation positioning model can significantly improve the fitting accuracy of the broadband oscillation signal, thereby facilitating improving the accuracy of detecting and tracing the broadband oscillation signal.

[0114] In this embodiment, the deep learning model extracts features from the broadband oscillation signal, including:

[0115] The feature vectors of each sampling point are obtained and integrated into the information matrix X. The graph neural network model is used as the deep learning model, and the adjacency matrix B of the graph is expressed as follows:

[0116]

[0117] Among them, v1 is the number of rows of the adjacency matrix B, v2 is the number of columns of the adjacency matrix B, and E is the set of connection relationships between vertices in the graph;

[0118] The Graph Conv layer based on the graph convolutional neural network processes and outputs the adjacency matrix B. The processing formula is as follows:

[0119] H (x+c) =f(H (x) ,B); (8)

[0120] Among them, f() is a nonlinear function, c is the number of hidden layers;

[0121] When c is equal to 0, the matrix H is equal to the information matrix X; the values ​​of the adjacency matrix B at the diagonal position are all 0. When the adjacency matrix B and the matrix H are multiplied, only the weighted sum of the features of all adjacency matrices of the sampling point is calculated, and the features of the sampling point itself are ignored. The specific feature extraction rules are as follows:

[0122] H (x+1) =f(H (x) ,B)=σ(B (X) H (X) W (x) ); (9)

[0123] Among them, W is a trainable parameter matrix, σ() is a nonlinear function, and after the softmax() function operation, the label representing the location of the oscillation source is obtained.

[0124] It should be noted that the structure and parameters of the graph convolutional neural network generally require multiple training verifications to set. A simple graph convolutional neural network positioning model contains two graph convolutional layers and one fully connected layer. A Flatten layer is added between the graph convolutional layer and the fully connected layer. The purpose is to output the graph convolutional layer as a one-dimensional sequence, which can be used as the input of the fully connected layer. Finally, a 1×5 vector is output and processed by the softmax() function to obtain a label representing the location of the oscillation source.

[0125] It is worth noting that the broadband oscillation problem, due to its significant spatiotemporal characteristics and the complexity of the system topology, causes the electrical quantities of each sampling point to couple with each other, forming a difficult problem to solve in non-Euclidean space. The present invention cleverly utilizes the powerful sequence feature capture capability of the convolutional neural network, combines the feature vectors of each node and the system adjacency matrix, constructs an oscillation positioning model, and realizes the accurate fitting of broadband oscillation signals, thereby significantly improving the accuracy of oscillation source positioning.

[0126] In this embodiment, a compressed sensing method is used to achieve high-rate compression of the oscillation signal, and the compressed signal is transmitted to the master station under limited bandwidth. An ISTA signal recovery network is built at the master station to obtain an accurate broadband oscillation signal, thereby improving the accuracy of the broadband oscillation disturbance source positioning task.

[0127] A second aspect of the present invention provides a storage medium, which includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the above-mentioned wide-band oscillation detection and tracing method.

[0128] A third aspect of the present invention provides a processor, which is used to run a program, wherein the program executes the above-mentioned wide-band oscillation detection and tracing method when running.

[0129] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A broadband oscillation detection and source tracing method, characterized in that: S1: Construct a mathematical model of broadband oscillation signal; S2: Perform state estimation on the measurement data of the PMU device to obtain a state estimation result; S3: Convert the broadband oscillation signal into the frequency domain based on discrete Fourier transform to obtain several frequency decomposition values; S4: Detection of broadband oscillation signal based on Hilport transform; S5: Perform self-encoding training on a number of frequency decomposition values ​​based on an intelligent encoding algorithm to obtain a number of encoding data; S6: performing image matrix transformation on a plurality of coded data respectively to obtain an oscillation characteristic matrix; S7: Train the deep learning model based on the oscillation feature matrix and the corresponding oscillation position labels to obtain an oscillation positioning model.

2. A broadband oscillation detection and source tracing method according to claim 1, characterized in that: The mathematical model of the broadband oscillation signal is: Where t is time, f is oscillation frequency, A i (t) is the function of the signal amplitude of the i-th sampling point changing with time, θ i (t) is a function of the phase angle of the i-th sampling point changing with time; wherein, i=1, 2, 3, ..., n, and n is the total number of sampling points.

3. A broadband oscillation detection and source tracing method according to claim 2, characterized in that: The state estimation of the measured data of the PMU device includes: S21: Extract the measurement data of the PMU device and use formula (1) to establish the objective function of distribution network state estimation: minJ(x)=[zh(x)] T R -1 [zh(x)]; (2) Wherein, z is the measurement vector of the power grid, x is the state vector of the power grid, h(x) is the nonlinear mapping from the state vector x to the measurement vector z, R is the covariance matrix of the measurement error, and is a diagonal matrix, the i-th diagonal element in R corresponds to the error variance of the i-th measurement value in z, and T represents the transpose; S22: Solve the objective function minJ(x) to obtain the state estimation result.

4. A broadband oscillation detection and source tracing method according to claim 1, characterized in that: The precision analysis of the measurement results to obtain accurate status data includes: S31: Calculate the average voltage amplitude estimation error e of the sampling point by formula mag : Among them, x i,true represents the true value of the voltage amplitude at the i-th sampling point, x i,est represents the estimated value of the voltage amplitude at the i-th sampling point; S32: Calculate the average voltage phase angle estimation error e at the sampling point by formula ang : in, represents the true value of the voltage phase angle at the i-th sampling point, represents the estimated value of the voltage phase angle at the i-th sampling point; S33: Select the minimum value of the average voltage amplitude estimation error and the average voltage phase angle estimation error of the sampling points and mark it as state data.

5. A broadband oscillation detection and source tracing method according to claim 1, characterized in that: The method of converting the broadband oscillation signal into the frequency domain based on discrete Fourier transform includes: The broadband oscillation signal is extracted and marked as x(m). The broadband oscillation signal x(m) is decomposed by discrete Fourier transform. The formula is as follows: Wherein, m is the window length, the value range of m is 0≤m≤M-1, M is the total number of window lengths, and X(m) is the frequency domain decomposition value of the broadband oscillation signal x(m).

6. A broadband oscillation detection and source tracing method according to claim 1, characterized in that: The detecting of the broadband oscillation signal based on Hilport transform comprises: The pure noise intrinsic mode function IMF is denoised by the autocorrelation coefficient algorithm to obtain the noisy IMF, and the noisy IMF is subjected to threshold denoising to obtain the effective IMF. The Hilbert transform is used to detect the oscillation of the frequency and amplitude of the effective IMF.

7. A broadband oscillation detection and source tracing method according to claim 2, characterized in that: The performing image matrix transformation on the plurality of coded data respectively comprises: The electrical quantities of several sampling points are extracted and expressed in the form of an oscillation characteristic matrix Y. The oscillation characteristic matrix Y is shown as follows: Among them, y vr is the rth electrical quantity at the vth sampling point, and 1≦v≦q, 1≦r≦p; q is the encoding dimension of each electrical quantity after compression, and p is the total number of electrical quantities collected in the entire network.

8. A broadband oscillation detection and source tracing method according to claim 1, characterized in that: The deep learning model is trained based on the oscillation feature matrix and the corresponding oscillation position label, including: The mapping relationship between the oscillation feature matrix and the corresponding oscillation source position label is obtained and integrated into a standard data set, and the standard data set is divided into several groups of training data, verification data and test data; the training data is used to train the deep learning model; the verification data is used to verify the trained deep learning model, and the test data is used to evaluate the model performance of the deep learning model; finally, an oscillation positioning model is obtained with the oscillation feature matrix as input and the oscillation source position as output; wherein, the deep learning model includes a graph neural network model and a convolutional neural network model.

9. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute a broadband oscillation detection and tracing method as described in any one of claims 1 to 8.

10. A processor, characterized in that: The processor is used to run a program, wherein the program, when running, executes a wide-band oscillation detection and source tracing method as described in any one of claims 1 to 8.

Images