Fault diagnosis method and device for distribution network based on spectral residual and random matrix theory

Abstract

The present application relates to a kind of distribution network fault diagnosis method and equipment based on spectral residual and random matrix theory, the method includes the following steps: collecting low-voltage distribution network voltage current recording wave data of different dimensions;Dimension is calculated to the low-voltage distribution network voltage current recording wave data collected by spectrum residual, and obtains significant atlas;According to distribution network topological information, space combination is carried out to significant atlas, and time-space data set matrix is constructed;Based on random matrix theory, using moving sliding window method, matrix is selected on the time-space data set matrix in turn, it is judged whether there is abnormal value, realizes distribution network fault diagnosis and positioning.Compared with prior art, the present application has the advantages of high accuracy, high efficiency etc..

Description

Technical Field

[0001] This invention relates to the field of distribution network fault diagnosis technology, and in particular to a distribution network fault diagnosis method and equipment based on spectral residual and random matrix theory. Background Technology

[0002] With the continuous improvement of power grid technology, the construction of various power grid infrastructures has brought data-driven situational awareness capabilities to the power grid. This provides conditions for assessing the operating status of the distribution network and using these statuses to conduct relevant scientific research. The diagnosis of distribution network faults is a crucial task in distribution network operation; however, the causes, phenomena, and processes of faults are complex and intertwined, posing significant challenges to fault diagnosis and analysis.

[0003] Traditional distribution network status analysis typically employs mechanistic model-based methods. However, each data analysis requires determining all parameter models, leading to significant challenges in information acquisition, computational complexity, and handling the impact of flawed data. Furthermore, the large-scale grid connection of distributed generation (DG) significantly increases the uncertainty of distribution network power flow and the difficulty in obtaining topology information; the large-scale charging and discharging of electric vehicles (EVs) further increases load-side randomness. These massive uncertainties and their coupling effects make distribution network behavior more complex, potentially rendering traditional research assumptions invalid. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a highly accurate and efficient method and device for diagnosing distribution network faults based on spectral residuals and random matrix theory.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A distribution network fault diagnosis method based on spectral residual and random matrix theory includes the following steps:

[0007] Collect voltage and current waveform data of low-voltage distribution networks from different dimensions;

[0008] The spectral residuals of the collected low-voltage distribution network voltage and current waveform data are calculated in different dimensions, and a significance spectrum is obtained.

[0009] Spatial combination of saliency maps based on distribution network topology information to construct a spatiotemporal dataset matrix;

[0010] Based on the theory of random matrices, the moving sliding window method is used to sequentially select matrices on the spatiotemporal dataset matrix to determine whether there are outliers, thereby realizing the diagnosis and location of distribution network faults.

[0011] Furthermore, the dimensions include three-phase voltage and current and neutral point current.

[0012] Furthermore, the process of obtaining the saliency map is as follows:

[0013] The spectral residuals of the acquired distribution network voltage and current waveform data are calculated in different dimensions. The normal information is filtered out by using two-dimensional discrete Fourier transform to obtain the saliency spectrum.

[0014] Furthermore, the construction of the spatiotemporal dataset matrix specifically involves:

[0015] Based on the saliency maps of different dimensions of each device, a multidimensional time dataset matrix for a single device is established;

[0016] Based on the distribution network topology information, the multidimensional time dataset matrices of multiple devices are stacked to form a spatiotemporal dataset matrix.

[0017] Furthermore, the determination of whether outliers exist specifically involves:

[0018] Set up a sliding window, and select the sliding window matrix in the spatiotemporal dataset matrix based on the sliding window;

[0019] The sliding window matrix is ​​standardized to generate a standardized matrix;

[0020] Find the singular value equivalent matrix of the normalized matrix and perform normalization to obtain the normalized matrix;

[0021] Empirical spectral distribution analysis of the normalized matrix is ​​performed using the circular law to determine whether outliers occur.

[0022] Furthermore, the standardized matrix is ​​a matrix with a mean of 0 and a variance of 1.

[0023] Furthermore, the method also includes:

[0024] Calculate the linear eigenvalue statistics of the matrix selected by the moving sliding window method to assess the fault status of the distribution network.

[0025] Furthermore, a linear characteristic value index system for the distribution network is constructed based on the calculated linear characteristic value statistics. This linear characteristic value index system is then compared with a preset index system to assess the fault status of the distribution network.

[0026] Furthermore, the linear eigenvalue statistics are obtained based on the average spectral radius.

[0027] The present invention also provides an electronic device, including one or more processors, a memory, and one or more programs stored in the memory, said one or more programs including instructions for executing the distribution network fault diagnosis method based on spectral residual and random matrix theory as described above.

[0028] Compared with existing technologies, this invention utilizes high-dimensional statistical characteristic analysis and evaluation techniques to perform feature monitoring, analysis, and fault diagnosis on spatiotemporal datasets. Through in-depth mining of distribution network waveform information, it can provide effective monitoring and evaluation methods for distribution network operation status, and has the following beneficial effects:

[0029] 1) Based on the effective coding idea, after calculating the spectral residual of the distribution network waveform data, the corresponding saliency spectrum in the time domain can be obtained based on the two-dimensional discrete Fourier inverse transform. It contains only fault features, while normal periodic waves are effectively filtered out, thereby achieving the purpose of fault segmentation and improving the sensitivity of subsequent fault judgment.

[0030] 2) By constructing a spatiotemporal dataset matrix from multiple device time datasets based on topology information, this method efficiently utilizes prior information about the distribution network topology. While improving fault diagnosis and location capabilities, this method reduces the impact of external environment and the condition of the sampling equipment itself on the diagnostic results, thus enhancing the robustness of the diagnostic method.

[0031] 3) Extract saliency map features, construct a spatiotemporal dataset, and use the circular law and linear feature statistics of random matrix theory to make full use of the data flow information of the distribution network operation, and realize efficient and accurate monitoring of the distribution network status based on the data-driven approach.

[0032] 4) This invention helps to improve the efficiency and accuracy of distribution network fault monitoring, and timely and accurately identify changes in the operating status of the distribution network, which is of great significance for carrying out low-voltage distribution network inspection and maintenance. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0034] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0035] This invention provides a distribution network fault diagnosis method based on spectral residual and random matrix theory. It uses high-dimensional statistical characteristic analysis and evaluation technology to carry out feature monitoring, analysis and fault diagnosis on spatiotemporal datasets. Through in-depth mining of distribution network waveform information, the fault diagnosis and location accuracy is high.

[0036] First, the theoretical basis, data model, and algorithm involved in this invention will be introduced.

[0037] 1. Spectral Residual Algorithm

[0038] The spectral residual model is an effective visual saliency detection model in frequency domain analysis. Based on the idea of ​​efficient coding, a fundamental principle of the visual system is to suppress recurring uniform texture features while preserving irregular feature information that deviates from the principle. This allows image information to be... Decomposed into two parts:

[0039]

[0040] In the formula, This indicates the unique information that needs to be retained. The background represents redundant information. Given an image I(x), we first calculate the two-dimensional discrete Fourier transform of the image, transforming it from the spatial domain to the frequency domain. Taking the logarithm of the amplitude yields the log spectrum L(f):

[0041]

[0042]

[0043]

[0044] Where F represents the two-dimensional discrete Fourier transform, and |·| represents its magnitude. Let P(f) represent the phase, and P(f) represent the phase spectrum of the image. Since the log curve satisfies the local linearity condition, it is smoothed using a local averaging filter hn(f) to obtain the log spectrum, which can be expressed as:

[0045]

[0046] Here, the symbol "*" represents convolution, and hn(f) is an n×n mean filter template. If the value of n is too small, the significance is not obvious, leading to an increased false negative rate for strip defects; if the value of n is too large, it will introduce too much noise, increasing the false positive rate. Through multiple experiments, it was found that the detection effect is better when n is 10. hn(f) is defined as:

[0047]

[0048] Therefore, the spectral residual visual attention model is expressed as:

[0049]

[0050] The above model is obtained by the difference between the log spectrum and the value after mean filtering.

[0051] To reconstruct the saliency map, a two-dimensional discrete inverse Fourier transform is performed on the spectral residual visual attention model R(f) and the phase spectrum P(f):

[0052]

[0053] In the formula, S(x) represents the saliency map obtained through the spectral residual visual attention model.

[0054] 2. Theoretical Model of Random Matrix

[0055] By using a random matrix model to analyze the inherent uncertainty and periodicity in the data, relevant indicators can be extracted for subsequent signal detection, correlation analysis, etc.

[0056] For engineering big data, its spatial dimension N and time dimension T These characteristics are often not identical. Research is conducted on Wishart matrices that conform to these properties, along with their linear eigenvalue statistics (LES) and their laws of large numbers and central limit theorem. Based on this research, the limit value of LES is obtained a priori, and the convergence rate of the empirical spectral density of the data matrix is ​​evaluated. These can serve as comparative references and analytical bases for experimental data, providing a starting point for achieving data-driven object cognition.

[0057] The collected spatiotemporal data are analyzed using the circular law to monitor the occurrence of abnormal states, and a linear eigenvalue evaluation system is introduced to evaluate the abnormal states.

[0058] Circular Law: Consider Singular value equivalent matrix cumulative product of independent random matrices ,in Furthermore, it is possible to... Normalization Then when and hour, The empirical spectral density function almost certainly converges to

[0059] .

[0060] Develop analytical algorithms for data models, extract statistics from the data models and calculate their theoretical limits, and then combine hypothesis testing and other methods to extract the statistical information contained in the data.

[0061] 3. Linear Eigenvalue Statistics (LES)

[0062] The definition of the linear eigenvalue statistic LES is given, and its statistical properties are studied—the law of large numbers and the central limit theorem of LES.

[0063] definition: ,

[0064] in, This is a continuous testing function. These are the eigenvalues ​​of the matrix.

[0065] 4. Test function Design

[0066] test function The design is central to optimizing LES performance. Sufficient continuity is sufficient. Common test functions are as follows:

[0067] Mean spectral radius (MSR) .in: (i = 1, 2, ..., N) is eigenvalues, for The radius of distribution on the complex plane.

[0068] Similar to filters, different Different perceptual effects can be obtained to adapt to specific applications, thereby constructing an effective criterion.

[0069] Based on the above theoretical foundation, the specific description of the distribution network fault diagnosis method based on spectral residual and random matrix theory provided by this invention is as follows. Figure 1 As shown, the method includes the following steps:

[0070] S1. Collect voltage and current waveform data of low-voltage distribution network in different dimensions.

[0071] Preferably, in order to improve fault diagnosis performance, the dimensions considered in the waveform data of this invention include three-phase voltage and current and neutral point current.

[0072] S2. Calculate the spectral residuals of the acquired distribution network voltage and current waveform data in different dimensions, and use two-dimensional discrete Fourier transform to filter out normal information and obtain a saliency spectrum.

[0073] S3. Spatially combine the saliency map data according to topological information to construct a spatiotemporal dataset matrix. Specifically: for a single device, construct a multidimensional time dataset matrix based on the waveform recording dimension; for time datasets of multiple devices, construct a spatiotemporal dataset matrix based on topological information.

[0074] S4. Based on random matrix theory, a moving sliding window method is used to sequentially select matrices on the spatiotemporal dataset matrix to determine if outliers occur and their locations, using the average spectral radius. A linear eigenvalue statistic is constructed, and the statistic of the matrix enclosed by the sliding window is calculated to assess the distribution network operation status. Specifically, the empirical spectral distribution of the spatiotemporal matrix of the saliency map is analyzed using a random matrix model, and the presence of outliers is determined using the circular law. A distribution network operation status indicator is selected from the pre-constructed linear eigenvalue indicator system, and it is determined whether the indicator exceeds a threshold when an abnormal state occurs.

[0075] The above method introduces tools such as spectral residual algorithm, random matrix model, and eigenvalue (spectral) analysis to deeply mine the statistical information of the spatiotemporal dataset, and then associates it with the fault status of the distribution network to design a corresponding indicator system to assist in the functional design.

[0076] If the above methods are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0077] Example 1

[0078] This embodiment provides a distribution network fault diagnosis method based on spectral residual and random matrix theory, including the following steps:

[0079] Step 1: Collect data from the distribution network The waveform recording data at each moment of operation, with the waveform recording dimensions including three-phase voltage and current and neutral point current;

[0080] Step 2: Calculate the saliency map for each recorded dimension based on the spectral residual algorithm, and construct a multi-dimensional time data matrix A for each device. n×T , where n=7.

[0081] Step 3: Stack the multidimensional time data matrix calculated in Step 2 according to the distribution network topology to form a spatiotemporal dataset matrix X. N×T Where N = n × number of devices;

[0082] Step 4: Use a width of The moving sliding window, in step 2, the spatiotemporal dataset matrix X N×T Select X from the top boxN×t And determine the movement of the window;

[0083] Step 5: Analyze the X obtained in Step 4. N×t Matrix standardization can be achieved using basic transformation formulas. ,in, , , Transform it into a standardized matrix with a mean of 0 and a variance of 1. ;

[0084] Step 6: Standardize the matrix According to the formula Find its singular value equivalent matrix, where, It is a unitary matrix;

[0085] Step 7: For singular value equivalent matrices Obtain the normalized matrix by performing normalization. The calculation formula is: ;

[0086] Step 8: Apply the circular law of the random matrix model to the normalized matrix. Perform empirical spectrum distribution analysis to detect any outliers.

[0087] Step 9: Calculate the matrix Linear eigenvalue index system The fault status of the distribution network is monitored and evaluated by combining the distribution network operation status threshold.

[0088] Example 2

[0089] This embodiment provides an electronic device, including one or more processors, a memory, and one or more programs stored in the memory, the one or more programs including instructions for executing the distribution network fault diagnosis method based on spectral residual and random matrix theory as described in Embodiment 1.

[0090] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A power distribution network fault diagnosis method based on spectral residual error and random matrix theory, characterized in that, Includes the following steps: Collect voltage and current waveform data of low-voltage distribution networks from different dimensions; The spectral residuals of the collected low-voltage distribution network voltage and current waveform data are calculated in different dimensions, and a significance spectrum is obtained. Spatial combination of saliency maps based on distribution network topology information to construct a spatiotemporal dataset matrix; Based on the theory of random matrices, the moving sliding window method is used to sequentially select matrices on the spatiotemporal dataset matrix to determine whether there are outliers, thereby realizing the diagnosis and location of distribution network faults.

2. The distribution network fault diagnosis method based on spectral residuals and random matrix theory according to claim 1, characterized in that, The dimensions include three-phase voltage and current and neutral point current.

3. The distribution network fault diagnosis method based on spectral residuals and random matrix theory according to claim 1, characterized in that, The process of obtaining the saliency map is as follows: The spectral residuals of the acquired distribution network voltage and current waveform data are calculated in different dimensions. The normal information is filtered out by using two-dimensional discrete Fourier transform to obtain the saliency spectrum.

4. The distribution network fault diagnosis method based on spectral residuals and random matrix theory according to claim 1, characterized in that, The specific steps for constructing the spatiotemporal dataset matrix are as follows: Based on the saliency maps of different dimensions of each device, a multidimensional time dataset matrix for a single device is established; Based on the distribution network topology information, the multidimensional time dataset matrices of multiple devices are stacked to form a spatiotemporal dataset matrix.

5. The distribution network fault diagnosis method based on spectral residuals and random matrix theory according to claim 1, characterized in that, The specific steps for determining whether outliers exist are as follows: Set up a sliding window, and select the sliding window matrix in the spatiotemporal dataset matrix based on the sliding window; The sliding window matrix is ​​standardized to generate a standardized matrix; Find the singular value equivalent matrix of the normalized matrix and perform normalization to obtain the normalized matrix; Empirical spectral distribution analysis of the normalized matrix is ​​performed using the circular law to determine whether outliers occur.

6. The distribution network fault diagnosis method based on spectral residuals and random matrix theory according to claim 5, characterized in that, The standardized matrix is ​​a matrix with a mean of 0 and a variance of 1.

7. The distribution network fault diagnosis method based on spectral residuals and random matrix theory according to claim 1, characterized in that, The method also includes: Calculate the linear eigenvalue statistics of the matrix selected by the moving sliding window method to assess the fault status of the distribution network.

8. The distribution network fault diagnosis method based on spectral residuals and random matrix theory according to claim 7, characterized in that, Based on the calculated linear characteristic value statistics, a linear characteristic value index system for the distribution network is constructed. This linear characteristic value index system is then compared with a preset index system to assess the fault status of the distribution network.

9. The distribution network fault diagnosis method based on spectral residuals and random matrix theory according to claim 7, characterized in that, The linear eigenvalue statistics are obtained based on the average spectral radius.

10. An electronic device, characterized in that, It includes one or more processors, a memory, and one or more programs stored in the memory, said one or more programs including instructions for executing the distribution network fault diagnosis method based on spectral residual and random matrix theory as described in any one of claims 1-9.

Images