A fault identification method for a high-voltage generator and related components

Abstract

This application discloses a high-voltage generator fault identification method and related components. When the zero-sequence voltage at the total output of M high-voltage generators operating in parallel exceeds a preset voltage, a fault feature vector is extracted from the zero-sequence voltage and current signals of the mth high-voltage generator. The vector with the greatest global correlation is determined as the data of the geometric center grid of the two-dimensional grid data. The remaining fault feature vectors with the highest scores are determined as the data of any blank grid closest to the geometric center grid, and this process continues until all fault feature vectors have been determined. The two-dimensional grid data is then input into a convolutional neural network model to determine whether the mth high-voltage generator has failed. The fault feature vector is one-dimensional data, and its correlation and spatial adjacent correlation between grids are mined to construct two-dimensional grid data. The one-dimensional data remains essentially unchanged, avoiding information redundancy and error accumulation caused by decomposing and transforming the one-dimensional data into multidimensional data and then fusing them, thereby improving judgment accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of high-voltage generator fault identification, and in particular to a high-voltage generator fault identification method and related components. Background Art

[0002] Traditional generator terminal voltages in power systems are below 20kV. To reduce losses during long-distance power transmission, a step-up transformer is required to boost the voltage. A high-voltage generator (Powerformer) is a new type of generator whose stator windings are wound with cross-linked polyethylene (XLPE) cable. The circular winding cross-section avoids the electric field inhomogeneity caused by the rectangular cross-section of traditional generator windings, allowing the output voltage of the high-voltage generator to jump from below 20kV to a theoretical 400kV, allowing direct connection to the high-voltage transmission network without the need for a step-up transformer. Single-phase stator grounding faults are the most potentially harmful and most frequent of all fault types in high-voltage generators. They can easily develop into more serious inter-turn and inter-phase faults, potentially damaging the iron core and burning out the stator windings, seriously impacting the safe and stable operation of power equipment and the power grid.

[0003] When a single-phase stator ground fault occurs in one of multiple high-voltage generators operating in parallel, the prior art extracts multiple features from the voltage at the total output of the multiple high-voltage generators and the transient current signals at the output of each high-voltage generator. These features are then decomposed and transformed to obtain multi-dimensional components, from which multi-dimensional information is extracted and fused. Finally, the fused information is compared with a set threshold to determine whether a single-phase stator ground fault has occurred in each high-voltage generator. However, when extracting and fusing multi-dimensional information, highly repetitive information may be extracted, resulting in information redundancy. Furthermore, errors are inevitably present in each fusion step, leading to the accumulation of errors, ultimately reducing the accuracy of determining a single-phase stator ground fault. Summary of the Invention

[0004] The purpose of this application is to provide a fault identification method for a high-voltage generator and related components. In this scheme, the fault feature vector is one-dimensional data. By mining its correlation and utilizing the spatial adjacent correlation relationship between grids to construct two-dimensional grid data, the essence of the one-dimensional data remains unchanged, thus avoiding the information redundancy and error accumulation caused by decomposing and transforming the one-dimensional data into multi-dimensional data and then extracting the multi-dimensional information for fusion, and can more accurately judge the fault.

[0005] To solve the above technical problems, the present application provides a method for identifying faults of a high-voltage generator, comprising:

[0006] When the zero-sequence voltage at the total output end of M high-voltage generators running in parallel is greater than a preset voltage, obtaining a zero-sequence voltage signal and a zero-sequence current signal of the m-th high-voltage generator within a preset time period, where the current time is within the preset time period, and 1≤m≤M;

[0007] Extracting a preset odd number of fault feature vectors from the zero-sequence voltage signal and the zero-sequence current signal, wherein the fault feature vector is a vector composed of fault features extracted from the zero-sequence voltage signal and the zero-sequence current signal at the same time;

[0008] Obtaining global correlations of the fault feature vectors and determining the fault feature vector with the largest global correlation as data of a geometric center grid of the two-dimensional grid data, wherein the global correlation is the sum of correlation coefficients between each fault feature vector and the remaining fault feature vectors;

[0009] Except for the fault feature vector that has been determined as the two-dimensional grid data, obtaining the fault feature vector with the highest score among the remaining fault feature vectors and determining it as the data of any blank grid in the two-dimensional grid data that has the smallest distance from the geometric center grid, and repeating this step until all the fault feature vectors are determined, wherein an average value of the relative distance-weighted correlation coefficients between the fault feature vector and all the determined fault feature vectors is positively correlated with the score;

[0010] The two-dimensional grid data is input into a pre-trained convolutional neural network model to determine whether a stator single-phase grounding fault occurs in the m-th high-voltage generator.

[0011] Preferably, the two-dimensional grid data is n*n two-dimensional grid data, where n is a positive odd number.

[0012] Preferably, after inputting the two-dimensional grid data into a pre-trained convolutional neural network model and determining whether a stator single-phase grounding fault occurs in the m-th high-voltage generator, the method further includes:

[0013] When it is determined that a stator single-phase grounding fault occurs in the m-th high-voltage generator, the alarm module is controlled to sound an alarm.

[0014] Preferably, extracting a preset odd number of fault feature vectors from the zero-sequence voltage signal and the zero-sequence current signal includes:

[0015] A preset odd number of fault feature vectors are extracted from the zero-sequence voltage signal and the zero-sequence current signal based on S-transformation.

[0016] Preferably, determining whether a stator single-phase grounding fault occurs in the m-th high-voltage generator includes:

[0017] When the result label output by the pre-trained convolutional neural network model is 0, determining that a stator single-phase grounding fault occurs in the m-th high-voltage generator;

[0018] When the result label output by the pre-trained convolutional neural network model is 1, it is determined that no stator single-phase grounding fault occurs in the m-th high-voltage generator.

[0019] Preferably, obtaining the zero-sequence voltage signal and the zero-sequence current signal of the m-th high-voltage generator within a preset time period includes:

[0020] A zero-sequence current signal within a preset time period is obtained by a current transformer provided at the output end of the mth high-voltage generator, and a zero-sequence voltage signal within a preset time period is obtained by a voltage transformer provided at the total output end of the M high-voltage generators running in parallel.

[0021] Preferably, obtaining the fault feature vector with the highest score among the remaining fault feature vectors and determining it as data of any blank grid in the two-dimensional grid data that has the smallest distance from the geometric center grid includes:

[0022] According to the scoring formula Score the remaining fault feature vectors, where T is any of the remaining fault feature vectors, and T e is the determined e-th fault feature vector, A is any blank grid in the two-dimensional grid data that has the smallest distance from the geometric center grid, and A e For the T e grid, n is the number of all the fault feature vectors that have been determined, and Corr(T, T e ) is the T and the T e The correlation coefficient, the Dist(A, A e ) is the A and the A e The length of the line connecting the geometric centers of

[0023] The fault feature vector with the highest score is obtained and determined as the data of A.

[0024] To solve the above technical problems, the present application also provides a high-voltage generator fault identification system, comprising:

[0025] a zero-sequence voltage signal and a zero-sequence current signal acquiring unit, configured to acquire the zero-sequence voltage signal and the zero-sequence current signal of the mth high-voltage generator within a preset time period when the zero-sequence voltage at the total output terminal of the M high-voltage generators operating in parallel is greater than a preset voltage, where the current time is within the preset time period and 1≤m≤M;

[0026] a fault feature vector acquisition unit, configured to extract a preset odd number of fault feature vectors from the zero-sequence voltage signal and the zero-sequence current signal, wherein the fault feature vector is a vector composed of fault features extracted from the zero-sequence voltage signal and the zero-sequence current signal at the same time;

[0027] a geometric center grid data determination unit, configured to obtain global correlations of the fault feature vectors and determine the fault feature vector with the largest global correlation as the data of the geometric center grid of the two-dimensional grid data, wherein the global correlation is the sum of correlation coefficients between each fault feature vector and the remaining fault feature vectors;

[0028] a data determination unit for the remaining grids, configured to obtain the fault feature vector with the highest score among the remaining fault feature vectors, except for the fault feature vector that has been determined as the two-dimensional grid data, and determine it as the data of any blank grid in the two-dimensional grid data that has the smallest distance from the geometric center grid, and repeat this step until all the fault feature vectors are determined, wherein an average value of the correlation coefficient weighted by the relative distance between the fault feature vector and all the determined fault feature vectors is positively correlated with the score;

[0029] A fault judgment unit is used to input the two-dimensional grid data into a pre-trained convolutional neural network model to determine whether a stator single-phase grounding fault occurs in the m-th high-voltage generator.

[0030] To solve the above technical problems, the present application also provides a high-voltage generator fault identification device, comprising:

[0031] memory for storing computer programs;

[0032] A processor is used to execute the computer program to implement the steps of the high-voltage generator fault identification method.

[0033] In order to solve the above technical problems, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the high-voltage generator fault identification method are implemented.

[0034] This application provides a high-voltage generator fault identification method and related components. In this solution, when the zero-sequence voltage at the total output of M high-voltage generators operating in parallel exceeds a preset voltage, a fault feature vector is extracted from the zero-sequence voltage and zero-sequence current signals of the m-th high-voltage generator. The fault feature vector with the highest global correlation is determined as the data of the geometric center grid of the two-dimensional grid data. The highest-scoring fault feature vector among the remaining fault feature vectors is determined as the data of any blank grid closest to the geometric center grid in the two-dimensional grid data, and this process continues until all fault feature vectors have been determined. The two-dimensional grid data is then input into a convolutional neural network model to determine whether the m-th high-voltage generator has a fault. The fault feature vector is one-dimensional data. By mining its correlation and utilizing the spatial adjacent correlation relationship between grids, the two-dimensional grid data is constructed. The essence of the one-dimensional data remains unchanged, thus avoiding the information redundancy and error accumulation caused by decomposing and transforming the one-dimensional data into multidimensional data and then extracting and fusing the multidimensional information, thereby enabling more accurate fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the prior art and the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0036] Figure 1 A flowchart of a high-voltage generator fault identification method provided in this application;

[0037] Figure 2 A schematic structural diagram of a high-voltage generator provided in this application;

[0038] Figure 3 This is an iterative diagram of a convolutional neural network model provided in this application;

[0039] Figure 4 An overall flow chart of fault identification provided by this application;

[0040] Figure 5 A schematic diagram of the structure of a high-voltage generator fault identification system provided in this application;

[0041] Figure 6 A schematic structural diagram of a high-voltage generator fault identification device provided in this application. DETAILED DESCRIPTION

[0042] The core of this application is to provide a fault identification method and related components for a high-voltage generator. In this solution, the fault feature vector is one-dimensional data. By mining its correlation and utilizing the spatial adjacent correlation relationship between grids to construct two-dimensional grid data, the essence of the one-dimensional data remains unchanged, thus avoiding the information redundancy and error accumulation caused by decomposing and transforming the one-dimensional data into multi-dimensional data and then extracting and fusing the multi-dimensional information, and enabling more accurate fault judgment.

[0043] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0044] Figure 1 A flowchart of a high-voltage generator fault identification method provided in this application includes:

[0045] S11: When the zero-sequence voltage at the total output end of M high-voltage generators running in parallel is greater than a preset voltage, obtaining a zero-sequence voltage signal and a zero-sequence current signal of the m-th high-voltage generator within a preset time period, where the current time is within the preset time period, and 1≤m≤M;

[0046] S12: extracting a preset odd number of fault feature vectors from the zero-sequence voltage signal and the zero-sequence current signal, where the fault feature vector is a vector composed of fault features extracted from the zero-sequence voltage signal and the zero-sequence current signal at the same time;

[0047] S13: obtaining the global correlation of each fault feature vector and determining the fault feature vector with the largest global correlation as the data of the geometric center grid of the two-dimensional grid data, wherein the global correlation is the sum of the correlation coefficients between each fault feature vector and the remaining fault feature vectors;

[0048] S14: except for the fault feature vector that has been determined as the two-dimensional grid data, obtain the fault feature vector with the highest score among the remaining fault feature vectors and determine it as the data of any blank grid with the smallest distance from the geometric center grid in the two-dimensional grid data, and repeat this step until all fault feature vectors are determined, wherein the average value and score of the relative distance weighted correlation coefficient between the fault feature vector and all the determined fault feature vectors are positively correlated;

[0049] S15: Input the two-dimensional grid data into a pre-trained convolutional neural network model to determine whether a stator single-phase grounding fault occurs in the m-th high-voltage generator.

[0050] Traditional generator terminal voltages in power systems are below 20 kV. To reduce losses during long-distance power transmission, step-up transformers are required to boost the voltage. High-voltage generators, however, are a new type of generator. Their stator windings are wound with cross-linked polyethylene cable. The circular winding cross-section avoids the electric field inhomogeneity caused by the rectangular windings of traditional generators. This allows the output voltage of the high-voltage generator to jump from below 20 kV to a theoretical 400 kV, enabling direct connection to the high-voltage transmission network without the need for a step-up transformer. Single-phase stator grounding faults are the most potentially harmful and most frequent fault type in high-voltage generators. They can easily develop into more serious inter-turn and inter-phase faults, potentially damaging the iron core and burning the stator windings, seriously impacting the safe and stable operation of power equipment and the power grid. Accurately detecting and isolating single-phase stator grounding faults is a hot topic and a challenge in high-voltage generator protection.

[0051] Due to the complex operating conditions of the high-voltage generator grid-connected system and the weak fault signals generated by early stator single-phase grounding faults, the existing method for identifying such non-stationary transient weak fault signals when multiple high-voltage generators are running in parallel is to extract multiple features from the voltage at the total output end of the multiple high-voltage generators running in parallel and the transient current signal at the output end of each high-voltage generator, then decompose and transform the multiple features to obtain multi-dimensional components, and then extract multi-dimensional information from them for fusion. Finally, the fused information is compared with a set threshold to determine whether each high-voltage generator has a stator single-phase grounding fault. Among them, when extracting multi-dimensional information for fusion, highly repetitive information may be extracted, resulting in information redundancy; and errors are inevitably present in each step of the fusion, which leads to the accumulation of errors, ultimately reducing the accuracy of stator single-phase grounding fault determination.

[0052] To address the above technical issues, this application, in accordance with the principle of "the simpler the information, the more reliable the results," extracts multiple fault feature vectors from the zero-sequence voltage signal at the total output of multiple high-voltage generators running in parallel and the zero-sequence current signal at the output of each high-voltage generator. These fault feature vectors are one-dimensional data. Two-dimensional grid data is then constructed by combining the correlation between the multiple fault feature vectors and the spatial proximity correlation between grids in the two-dimensional blank grid data. Finally, a convolutional neural network model, which excels at processing two-dimensional grid data, is used to determine whether a stator single-phase grounding fault has occurred. This invention utilizes a data dimension conversion strategy to maximize the preservation of correlation information between one-dimensional fault feature vectors and maximizes the advantages of deep learning methods for processing weak electrical fault information, significantly improving the accuracy of discrimination.

[0053] Specifically, when a single-phase stator ground fault occurs among multiple high-voltage generators operating in parallel, a zero-sequence voltage is generated at their total output terminal. Therefore, the zero-sequence voltage at the total output terminal can be detected to determine whether a single-phase stator ground fault has occurred among the multiple high-voltage generators operating in parallel. When the detected zero-sequence voltage is greater than a preset voltage (set value), it is determined that a single-phase stator ground fault has occurred. At this point, subsequent steps are initiated to determine whether the mth high-voltage generator has a single-phase stator ground fault, where 1≤m≤M. This means that each of the M high-voltage generators must be determined to identify the generator with a single-phase stator ground fault. Otherwise, subsequent steps are not initiated. The preset voltage can be the sum of the zero-sequence fundamental voltage and the zero-sequence third harmonic voltage, i.e., the 15% phase voltage. This can be determined based on actual conditions and is not specifically limited here. Furthermore, when there is only one high-voltage generator, a single-phase stator ground fault can be determined simply by the zero-sequence voltage being greater than the preset voltage.

[0054] When the zero-sequence voltage is greater than a preset voltage, subsequent steps are initiated to determine whether a stator single-phase grounding fault has occurred in the mth high-voltage generator. The subsequent steps are: obtaining the zero-sequence voltage signal and zero-sequence current signal within a preset time period including the current time. The zero-sequence voltage signal and zero-sequence current signal can also be stored while being obtained, so that the zero-sequence voltage signal and zero-sequence current signal before the current time can be obtained from the pre-stored information, and the zero-sequence voltage signal and zero-sequence current signal after the current time can continue to be obtained; then, sampling is performed from the zero-sequence voltage signal and zero-sequence current signal obtained within the preset time period to extract a preset odd number of fault feature vectors, each of which is composed of fault features extracted from the zero-sequence voltage signal and zero-sequence current signal at the same time. When a stator single-phase grounding fault occurs in a high-voltage generator and when a stator single-phase grounding fault does not occur, the zero-sequence voltage signal and zero-sequence current signal have different fault features, and the fault is determined based on this.

[0055] Afterwards, the extracted preset odd number of fault feature vectors are used to construct two-dimensional grid data, and the two-dimensional grid data is constructed based on the correlation between the fault feature vectors and the spatial adjacent correlation between the grids. First, the correlation coefficient (which can be the Pearson product-moment correlation coefficient) can be used to calculate the correlation between each two fault feature vectors, and the correlation coefficient matrix of the one-dimensional fault feature vector can be constructed based on this, and then the global correlation of each fault feature vector can be obtained. The fault feature vector with the largest global correlation is compared and selected and determined as the data of the geometric center grid of the two-dimensional grid data. The two-dimensional grid is composed of an odd number of grid units, and the grid units correspond one to one to the fault feature vectors. Among them, the Pearson product-moment relationship between the fault feature vectors is used to quantify their correlation, and the calculation formula is: Cov(Ti , T j ) is the i-th fault feature vector T i and the jth fault feature vector T j The covariance of i and σ j They are the fault feature vectors T i and fault feature vector T j The standard deviation of .

[0056] After determining the data of the geometric center grid of the two-dimensional grid data, the fault feature vector with the highest score among the remaining fault feature vectors is selected, and a radial filling strategy is used to fill the fault feature vector into the two-dimensional grid (that is, the data of the blank grids of the two-dimensional grid data are determined in order from small to large according to the distance from the geometric center grid. When multiple blank grids are at the same distance from the geometric center grid, any blank grid can be selected to fill the data) until the remaining fault feature vectors are filled into the two-dimensional grid, and finally the two-dimensional grid data is formed, thereby avoiding information redundancy caused by multi-dimensional information with a high degree of repetition when the one-dimensional data is decomposed and transformed into multi-dimensional data and then the multi-dimensional information is extracted.

[0057] This score is positively correlated with the average of the relative distance-weighted correlation coefficients of all identified fault feature vectors. This means that the correlation between fault feature vectors and the spatial proximity between grids reflect the implicit correlation between fault feature vectors, converting one-dimensional fault feature vectors into two-dimensional grid data. This two-dimensional grid data has more stable and specific features, and the two-dimensional grid data is still composed of one-dimensional fault feature vectors, thus avoiding the error accumulation caused by the fusion of multidimensional data into one-dimensional data. Furthermore, a pre-trained convolutional neural network model can be used to determine the attributes of the two-dimensional grid data (fault or non-fault), thereby tracing whether the mth high-voltage generator to which the two-dimensional grid data belongs has a stator single-phase grounding fault.

[0058] In summary, the present application provides a method for identifying faults in high-voltage generators. When the zero-sequence voltage at the total output of M high-voltage generators operating in parallel is greater than a preset voltage, a fault feature vector is extracted from the zero-sequence voltage and zero-sequence current signals of the m-th high-voltage generator. The fault feature vector with the greatest global correlation is determined as the data of the geometric center grid of the two-dimensional grid data. The highest-scoring of the remaining fault feature vectors is determined as the data of any blank grid closest to the geometric center grid in the two-dimensional grid data, until all fault feature vectors are determined. The two-dimensional grid data is input into a convolutional neural network model to determine whether the m-th high-voltage generator has a fault. The fault feature vector is one-dimensional data. By mining its correlation and utilizing the spatial adjacent correlation relationship between grids to construct two-dimensional grid data, the essence of the one-dimensional data remains unchanged. This avoids information redundancy and error accumulation caused by decomposing and transforming the one-dimensional data into multidimensional data and then extracting and fusing the multidimensional information. This method enables more accurate fault diagnosis, improves the reliability of single-phase ground fault protection for the stator winding of the high-voltage generator, and ensures the safe and stable operation of the power supply system of the high-voltage generator. The method has far-reaching significance and broad application prospects.

[0059] Based on the above embodiment:

[0060] As a preferred embodiment, the two-dimensional grid data is n*n two-dimensional grid data, where n is a positive odd number.

[0061] In this embodiment, when constructing two-dimensional grid data, the square with the strongest spatial adjacent correlation between the geometric center grid and the surrounding grids is selected, that is, the size of the two-dimensional grid data is n*n, where n is a positive odd number. Accordingly, a preset odd number of fault feature vectors are extracted from the zero-sequence voltage signal and the zero-sequence current signal. The preset odd number corresponds to the size of the two-dimensional grid data, which facilitates the convolutional neural network to determine the fault based on the constructed two-dimensional grid data.

[0062] As a preferred embodiment, after inputting the two-dimensional grid data into a pre-trained convolutional neural network model and determining whether a stator single-phase grounding fault occurs in the m-th high-voltage generator, the method further includes:

[0063] When it is determined that a stator single-phase grounding fault occurs in the mth high-voltage generator, the alarm module is controlled to sound an alarm.

[0064] In this embodiment, when it is determined that a single-phase grounding fault has occurred in the stator of the mth high-voltage generator, an alarm can be issued by controlling the alarm module. For example, a warning signal is issued to indicate that the mth high-voltage generator has a single-phase grounding fault in the stator. Other forms of alarm prompts can also be used to enable operators to discover and handle the fault in a timely manner to avoid causing greater losses.

[0065] As a preferred embodiment, extracting a preset odd number of fault feature vectors from the zero-sequence voltage signal and the zero-sequence current signal includes:

[0066] A preset odd number of fault feature vectors are extracted from the zero-sequence voltage signal and the zero-sequence current signal based on S-transform.

[0067] In this embodiment, S transform is used to extract a preset odd number of fault feature vectors, wherein the method used and the specific fault feature vectors extracted are not limited, as long as they can characterize the fault characteristics of the zero-sequence voltage signal and the zero-sequence current signal.

[0068] The specific extraction method and content can be:

[0069] First, the zero-sequence current signal at the output of the high-voltage generator is decomposed by S transform to obtain the complex amplitude-time-frequency matrix S. ij It is the element in the i-th row and j-th column of the matrix S, whose real part is X and imaginary part is Y, corresponding to the j-th time period of the i-th frequency, and its amplitude Defined as:

[0070]

[0071] Where Λ is the phase polarity under time-frequency; i = 1, 2, 3, ..., N / 2 + 1; j = 1, 2, 3, ..., N; According to the above formula, all amplitude-time-frequency matrix elements A are obtained ij That is to say Then we get the amplitude-time-frequency matrix A:

[0072]

[0073] In order to better reflect the differences between the amplitude-time-frequency matrices A of each high-voltage generator, a correlation coefficient formula is defined as follows:

[0074]

[0075] Among them, β ab A represents the correlation coefficient of the amplitude-time-frequency matrix of the a-th and b-th high-voltage generators; a and A b are the amplitude-time-frequency matrices of the a-th and b-th high-voltage generators, respectively. Taking the a-th high-voltage generator as an example, a comprehensive correlation coefficient ρ is defined a as follows:

[0076]

[0077] Where k is the total number of high-voltage generators.

[0078] 2. Let the transient energy of the a-th high-voltage generator in the characteristic frequency band be E a :

[0079]

[0080] In order to eliminate the dimension, normalization can be performed, then the transient energy characteristic e a It can be defined as:

[0081]

[0082] 3. The average active power component is the average value of the product of the zero-sequence current signal and the zero-sequence voltage signal within one power frequency cycle T. The average active power component can be obtained by averaging the zero-sequence active component within unit time. Its expression is as follows:

[0083]

[0084] Where Δp a is the average active power component at the output of the a-th high-voltage generator; Δu0(t) is the zero-sequence voltage signal component at the output of the high-voltage generator; Δi 0a (t) is the zero-sequence current signal component at the output of the a-th high-voltage generator. Define a zero-sequence active power characteristic λ a :

[0085]

[0086] In summary, the comprehensive correlation coefficient ρ of the a-th high-voltage generator is finally defined a , transient energy characteristics e a and zero-sequence active power characteristic λ a， This constitutes the fault characteristic vector (ρ a , e a ,λ a ).

[0087] As a preferred embodiment, determining whether a stator single-phase grounding fault occurs in the m-th high-voltage generator includes:

[0088] When the result label output by the pre-trained convolutional neural network model is 0, it is determined that the stator single-phase grounding fault occurs in the m-th high-voltage generator;

[0089] When the result label output by the pre-trained convolutional neural network model is 1, it is determined that the m-th high-voltage generator does not have a stator single-phase grounding fault.

[0090] In this embodiment, a pre-trained convolutional neural network model is used to determine whether a stator single-phase grounding fault occurs in the mth high-voltage generator. The model is not easily affected by transition resistance, fault location, and neutral point grounding method, and has strong robustness and generalization capabilities.

[0091] Specifically, the training process of the convolutional neural network model can be: S transform is used to extract the fault characteristics: the comprehensive correlation coefficient ρ of the a-th high-voltage generator a , transient energy characteristics e a and zero-sequence active power characteristic λ a , constitutes the fault characteristic vector of the a-th high-voltage generator (ρ a , e a ,λ a ), and then obtain the fault feature vectors simulated in different scenarios. After converting them into two-dimensional grid data using the data dimension conversion strategy, 80% of the two-dimensional grid data can be used as a training set to train the convolutional neural network model, and the remaining 20% of the two-dimensional grid data can be used as a test set to test the feasibility of this application.

[0092] Take three high-voltage generators running in parallel as an example. To identify the high-voltage generator with a stator single-phase grounding fault among the three generators, please refer to Figure 2 , three high-voltage generators are operated in parallel. Each high-voltage generator includes a three-phase stator winding (phase a, phase b, phase c). High-voltage generators 1-3 can be set to a rated voltage of 150kV and a rated frequency of 50Hz. The capacitance to ground of the three-phase stator winding is 0.555μF, 0.5774μF, and 0.555μF respectively. The capacitance to ground of the external system is 0.15μF. The fault closing angle is 0°, the inductance Ln is 1838mH, and the damping resistance R n is 57Ω, the neutral point grounding resistance R l The stator single-phase grounding fault of the high-voltage generator under different fault conditions is simulated, and the specific setting of the transition resistance R g The values of the stator single-phase grounding fault position α are 5Ω, 100Ω, 500Ω, 750Ω, 1000Ω, ... 7500Ω, 7750Ω, and 8000Ω respectively; the position of the stator single-phase grounding fault is set to 0%, 25%, 50%, 75%, and 100%; the neutral point grounding method is set to high-resistance grounding or arc suppression coil grounding; the partial fault characteristic vectors obtained are shown in Table 1. Among them, the fault position can be F1. When a stator single-phase grounding fault occurs here, the connection relationship after F1 can be replaced by the transition resistor R g .

[0093] Table 1

[0094]

[0095] Taking a simulation experiment as an example, a preset odd number (e.g., 9) of fault feature vectors can be extracted from the zero-sequence voltage and current signals of the mth high-voltage generator and grouped together. This simulation experiment yielded a total of 432 fault feature vectors as raw data, of which 297 were obtained when the high-voltage generator was in a faulty operating state and 135 were obtained when it was in a normal operating state. The Pearson product-moment correlation coefficients between each pair of fault feature vectors in each group were calculated, resulting in 48 correlation matrices for the sample set. The correlation matrix for one group of fault feature vectors is shown in Table 2, where G1, G2, G3, G4, G5, G6, G7, G8, and G9 represent nine fault feature vectors in each group.

[0096] Table 2

[0097]

[0098] Then, the data dimension conversion strategy of this application can be used to convert the historical one-dimensional fault feature vector into two-dimensional grid data, and obtain 48 3*3 two-dimensional grid data. 80% of the two-dimensional grid data is used as the training set, and the remaining 20% is used as the test set. The training set is input into the CNN (Convolutional Neural Networks) model to train the optimization parameters. The iteration of the model is shown as follows: Figure 3 As shown in Figure 3, the accuracy of the model gradually increases with the increase in the number of iterations. The test set is input into the trained and optimized CNN model, and the discrimination results are shown in Table 3, where the result label 0 represents a faulty high-voltage generator and the result label 1 represents a non-faulty high-voltage generator.

[0099] Table 3

[0100]

[0101] The simulation results show that the fault identification scheme of the high-voltage generator based on the data dimension conversion strategy in this application can accurately identify the faulty high-voltage generator, thereby improving the accuracy and reliability of fault identification.

[0102] In addition, referring to all the above embodiments, it is possible to adopt Figure 4 The overall flow chart shown is used to determine the stator unidirectional grounding fault of the high voltage generator. N is the phase voltage, and the preset voltage is 15% of the phase voltage; the comprehensive correlation coefficient ρ, the transient energy characteristic e and the zero-sequence active power characteristic λ constitute the fault characteristic vector of a certain high-voltage generator.

[0103] As a preferred embodiment, obtaining the zero-sequence voltage signal and the zero-sequence current signal of the m-th high-voltage generator within a preset time period includes:

[0104] A zero-sequence current signal within a preset time period is obtained by a current transformer arranged at the output end of the mth high-voltage generator, and a zero-sequence voltage signal within a preset time period is obtained by a voltage transformer arranged at the total output end of the M high-voltage generators running in parallel.

[0105] This embodiment can refer to Figure 2 , Figure 2 CT is a current transformer that can sample the zero-sequence current signal i at the output of the high-voltage generator 1. 01 , the zero sequence current signal i at the output end of the high voltage generator 2 02 , the zero sequence current signal i at the output end of the high voltage generator 3 03 ; PT is a voltage transformer, which can sample the zero-sequence voltage signal u0 of the bus Bus at the total output end of the three high-voltage generators; it is easy to implement and has a simple circuit structure.

[0106] As a preferred embodiment, obtaining the fault feature vector with the highest score among the remaining fault feature vectors and determining it as the data of any blank grid with the smallest distance from the geometric center grid in the two-dimensional grid data includes:

[0107] According to the scoring formula Score the remaining fault feature vectors, where T is any remaining fault feature vector, T e is the determined e-th fault feature vector, A is any blank grid with the smallest distance from the geometric center grid in the two-dimensional grid data, A e T e grid, n is the number of all fault feature vectors that have been determined, Corr(T, T e ) for T and T e The correlation coefficient of Dist(A, A e ) for A and A e The length of the line connecting the geometric centers of

[0108] Obtain the fault feature vector with the highest score and determine it as the data of A.

[0109] In this embodiment, the fault feature vector with the highest score among the remaining fault feature vectors is determined as the data of any blank grid in the two-dimensional grid data that has the smallest distance from the geometric center grid, and then all the fault feature vectors are determined in sequence, that is, the fault feature vector with the highest score among the remaining fault feature vectors is filled in order of distance from the center of the geometric grid from near to far, and the remaining fault feature vectors are radially filled into the remaining grids in the two-dimensional grid data one by one in order.

[0110] The average value of the relative distance weighted Pearson product-moment correlation coefficient can be selected as the scoring standard for the fault feature vector selection evaluation mechanism. The specific scoring formula can be Scoring and evaluation are performed using the spatial adjacent correlation between the grids of the determined fault feature vectors and the blank grids and the correlation between the determined fault feature vectors and the remaining fault feature vectors; the set of determined fault feature vectors can be Q, and the set of remaining fault feature vectors can be P, T belongs to P, T e Belong to Q; Dist(A, A e ) for A and A e The length of the line connecting the geometric centers of , that is, the relative straight-line distance between them.

[0111] In summary, the two-dimensional grid data established using the scoring formula retains the correlation information between the one-dimensional fault feature vectors to the greatest extent, and then the advantages of the deep learning method based on the convolutional neural network model in processing weak electrical fault information can be maximized to improve the discrimination accuracy.

[0112] Please refer to Figure 5 , Figure 5 A schematic structural diagram of a high-voltage generator fault identification system provided in this application includes:

[0113] a zero-sequence voltage signal and zero-sequence current signal acquisition unit 21 for acquiring the zero-sequence voltage signal and zero-sequence current signal of the m-th high-voltage generator within a preset time period when the zero-sequence voltage at the total output terminal of the M high-voltage generators operating in parallel is greater than a preset voltage, where the current time is within the preset time period and 1≤m≤M;

[0114] A fault feature vector acquisition unit 22 is configured to extract a preset odd number of fault feature vectors from the zero-sequence voltage signal and the zero-sequence current signal. The fault feature vector is a vector composed of fault features extracted from the zero-sequence voltage signal and the zero-sequence current signal at the same time.

[0115] The geometric center grid data determination unit 23 is used to obtain the global correlation of each fault feature vector and determine the fault feature vector with the largest global correlation as the data of the geometric center grid of the two-dimensional grid data. The global correlation is the sum of the correlation coefficients between each fault feature vector and the remaining fault feature vectors.

[0116] The data determination unit 24 for the remaining grids is configured to obtain the fault feature vector with the highest score from the remaining fault feature vectors, except for the fault feature vector that has been determined as the two-dimensional grid data, and determine it as the data of any blank grid with the smallest distance from the geometric center grid in the two-dimensional grid data, and repeat this step until all fault feature vectors are determined, wherein the average value of the relative distance-weighted correlation coefficient between the fault feature vector and all the determined fault feature vectors is positively correlated with the score;

[0117] The fault judgment unit 25 is used to input the two-dimensional grid data into a pre-trained convolutional neural network model to determine whether a stator single-phase grounding fault occurs in the m-th high-voltage generator.

[0118] For an introduction to a high-voltage generator fault identification system provided in this application, please refer to the above-mentioned embodiment, and this application will not go into details here.

[0119] Please refer to Figure 6 , Figure 6 A schematic diagram of the structure of a high-voltage generator fault identification device provided in this application includes:

[0120] Memory 31, for storing computer programs;

[0121] The processor 32 is configured to execute a computer program to implement the steps of the high-voltage generator fault identification method.

[0122] For an introduction to a fault identification device for a high-voltage generator provided in this application, please refer to the above embodiment, and this application will not go into details here.

[0123] The present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the high-voltage generator fault identification method are implemented.

[0124] For an introduction to a computer-readable storage medium provided in this application, please refer to the above embodiments, and this application will not go into details here.

[0125] It should be noted that, in this specification, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0126] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying a fault of a high-voltage generator, characterized in that: include: When the zero-sequence voltage at the total output end of M high-voltage generators running in parallel is greater than a preset voltage, obtaining a zero-sequence voltage signal and a zero-sequence current signal of the m-th high-voltage generator within a preset time period, where the current time is within the preset time period, and 1≤m≤M; Extracting a preset odd number of fault feature vectors from the zero-sequence voltage signal and the zero-sequence current signal, wherein the fault feature vector is a vector composed of fault features extracted from the zero-sequence voltage signal and the zero-sequence current signal at the same time; Obtaining global correlations of the fault feature vectors and determining the fault feature vector with the largest global correlation as data of a geometric center grid of the two-dimensional grid data, wherein the global correlation is the sum of correlation coefficients between each fault feature vector and the remaining fault feature vectors; Except for the fault feature vector that has been determined as the two-dimensional grid data, obtaining the fault feature vector with the highest score among the remaining fault feature vectors and determining it as the data of any blank grid in the two-dimensional grid data that has the smallest distance from the geometric center grid, and repeating this step until all the fault feature vectors are determined, wherein an average value of the relative distance-weighted correlation coefficients between the fault feature vector and all the determined fault feature vectors is positively correlated with the score; Inputting the two-dimensional grid data into a pre-trained convolutional neural network model to determine whether a stator single-phase grounding fault occurs in the m-th high-voltage generator; According to the scoring formula Score the remaining fault feature vectors, where T is any of the remaining fault feature vectors, and T e is the determined e-th fault feature vector, A is any blank grid in the two-dimensional grid data that has the smallest distance from the geometric center grid, and A e For the T e grid, n is the number of all the fault feature vectors that have been determined, and Corr(T, T e ) is the T and the T e The correlation coefficient, the Dist(A, A e ) is the A and the A e The length of the line connecting the geometric centers.

2. The fault identification method of a high-voltage generator according to claim 1, characterized in that: The two-dimensional grid data is n*n two-dimensional grid data, where n is a positive odd number.

3. The fault identification method of a high-voltage generator according to claim 1, characterized in that: After inputting the two-dimensional grid data into a pre-trained convolutional neural network model to determine whether a stator single-phase grounding fault occurs in the m-th high-voltage generator, the method further includes: When it is determined that a stator single-phase grounding fault occurs in the m-th high-voltage generator, the alarm module is controlled to sound an alarm.

4. The method for identifying a fault of a high-voltage generator according to claim 1, wherein: Extracting a preset odd number of fault feature vectors from the zero-sequence voltage signal and the zero-sequence current signal includes: A preset odd number of fault feature vectors are extracted from the zero-sequence voltage signal and the zero-sequence current signal based on S-transformation.

5. The fault identification method of a high-voltage generator according to claim 1, characterized in that: Determining whether a stator single-phase grounding fault occurs in the m-th high-voltage generator includes: When the result label output by the pre-trained convolutional neural network model is 0, determining that a stator single-phase grounding fault occurs in the m-th high-voltage generator; When the result label output by the pre-trained convolutional neural network model is 1, it is determined that no stator single-phase grounding fault occurs in the m-th high-voltage generator.

6. The high-voltage generator fault identification method according to claim 1, characterized in that: Obtaining the zero-sequence voltage signal and the zero-sequence current signal of the m-th high-voltage generator within a preset time period, including: A zero-sequence current signal within a preset time period is obtained by a current transformer provided at the output end of the mth high-voltage generator, and a zero-sequence voltage signal within a preset time period is obtained by a voltage transformer provided at the total output end of the M high-voltage generators running in parallel.

7. The high-voltage generator fault identification method according to any one of claims 1 to 6, characterized in that: Obtaining the fault feature vector with the highest score among the remaining fault feature vectors and determining it as data of any blank grid in the two-dimensional grid data that has the smallest distance from the geometric center grid, including: Scoring the remaining fault feature vectors according to a scoring formula; The fault feature vector with the highest score is obtained and determined as the data of A.

8. A fault identification system for a high-voltage generator, characterized in that: include: a zero-sequence voltage signal and a zero-sequence current signal acquiring unit, configured to acquire the zero-sequence voltage signal and the zero-sequence current signal of the mth high-voltage generator within a preset time period when the zero-sequence voltage at the total output terminal of the M high-voltage generators operating in parallel is greater than a preset voltage, where the current time is within the preset time period and 1≤m≤M; Fault a feature vector acquisition unit, configured to extract a preset odd number of fault feature vectors from the zero-sequence voltage signal and the zero-sequence current signal, wherein the fault feature vector is a vector composed of fault features extracted from the zero-sequence voltage signal and the zero-sequence current signal at the same time; a geometric center grid data determination unit, configured to obtain global correlations of the fault feature vectors and determine the fault feature vector with the largest global correlation as the data of the geometric center grid of the two-dimensional grid data, wherein the global correlation is the sum of correlation coefficients between each fault feature vector and the remaining fault feature vectors; The data determination unit for the remaining grids is used to obtain the fault feature vector with the highest score among the remaining fault feature vectors, except for the fault feature vector that has been determined as the two-dimensional grid data, and determine it as the data of any blank grid with the smallest distance from the geometric center grid in the two-dimensional grid data, and repeat this step until all the fault feature vectors are determined, wherein the average value of the correlation coefficient weighted by the relative distance between the fault feature vector and all the determined fault feature vectors is positively correlated with the score; wherein, according to the scoring formula Score the remaining fault feature vectors, where T is any of the remaining fault feature vectors, and T e is the determined e-th fault feature vector, A is any blank grid in the two-dimensional grid data that has the smallest distance from the geometric center grid, and A e For the T e grid, n is the number of all the fault feature vectors that have been determined, and Corr(T, T e ) is the T and the T e The correlation coefficient, the Dist(A, A e ) is the A and the A e The length of the line connecting the geometric centers of A fault judgment unit is used to input the two-dimensional grid data into a pre-trained convolutional neural network model to determine whether a stator single-phase grounding fault occurs in the m-th high-voltage generator.

9. A fault identification device for a high-voltage generator, characterized in that: include: Memory for storing computer programs; A processor, configured to execute the computer program to implement the steps of the high-voltage generator fault identification method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the high-voltage generator fault identification method according to any one of claims 1 to 7 are implemented.

Images