Power distribution network fault mode identification method and device, terminal and medium

Abstract

The invention discloses a power distribution network fault mode identification method and device, a terminal and a medium. The power distribution network fault mode identification method comprises the following steps: preprocessing a power distribution network signal; mapping the preprocessed power distribution network signal to a Grassmann manifold space; constructing a recognition model, wherein the recognition model comprises a feature extraction layer, a first network, a second network, a feature fusion layer and an output layer; inputting the power distribution network signal and the preprocessed power distribution network signal into a trained identification model, and identifying a fault mode; according to the method, by constructing the regularized Grubrum matrix and mapping the regularized Grubrum matrix to the Grassmann manifold space, the inherent geometric structure and low-dimensional manifold features of the data can be better captured, so that the feature representation has higher discrimination capability and stability; moreover, the recognition model combines the characteristics of Grassmann manifold, introduces a Stiefel convolutional layer and a feature fusion strategy, and further enhances the extraction and recognition capabilities of fault mode features.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault processing, and in particular to a distribution network fault pattern identification method, device, terminal and medium. Background Art

[0002] As the intelligentization of power systems accelerates, data acquisition and sensor technology play an increasingly critical role in substation distribution networks. They monitor the grid's operating status in real time and collect multi-dimensional data, including current, voltage, and temperature. Furthermore, the in-depth application of big data and cloud computing technologies enables the efficient storage, processing, and analysis of massive amounts of data, providing solid data support and powerful computing power for fault analysis.

[0003] However, current distribution network fault detection and diagnosis technologies primarily focus on offline fault detection and diagnosis, which presents numerous limitations: First, their functionality is relatively limited, making it difficult to comprehensively and real-timely reflect the complex operational state of the power grid. Second, if detection equipment coverage is insufficient, fault information may not be captured promptly and accurately, delaying fault resolution and expanding the scope of the fault. Furthermore, existing traditional machine learning methods, such as support vector machines and random forests, rely on manual feature extraction, making them difficult to adapt to the complex and changing fault characteristics of distribution networks and exhibiting limited generalization capabilities. Summary of the Invention

[0004] In order to overcome the defects in the above-mentioned prior art, the present invention provides a distribution network fault pattern recognition method, including: a distribution network fault pattern recognition method, characterized in that it includes: preprocessing the distribution network signal; mapping the preprocessed distribution network signal to the Grassmann manifold space; constructing a recognition model, the recognition model including a feature extraction layer, a first network, a second network, a feature fusion layer and an output layer; inputting the distribution network signal and the preprocessed distribution network signal into the trained recognition model to identify the fault pattern.

[0005] As a preferred solution of the distribution network fault pattern identification method described in the present invention, the preprocessing of the distribution network signal includes: the distribution network signal is three-phase voltage, three-phase current, zero-sequence current, and zero-sequence voltage, and the sampling rate is set to 10kHz; a four-term third-order Blackman-Harris window function is used to perform windowing processing on the signal, wherein the window length is 128 points and the overlap rate is 50%; and the windowed signal is converted to generate a 30×65 complex matrix.

[0006] As a preferred solution of the distribution network fault pattern recognition method described in the present invention, it includes: mapping the preprocessed distribution network signal to the Grassmann manifold space includes: constructing a Gram matrix G according to the complex time-frequency matrix, adding a regularization term G1 to the Gram matrix G, and obtaining the regularized Gram matrix G reg , for the regularized Gram matrix G reg Perform Householder transformation to obtain matrix G trans ; According to the matrix G trans Gram-Schmidt orthogonalization is performed iteratively with the number of iterations set to 3. A dynamic weight adjustment mechanism is introduced in each iteration. The dynamic weight adjustment mechanism includes setting a dynamic weight coefficient according to the change in the module length of the vector before and after orthogonalization to optimize the orthogonalization result. The matrix after Gram-Schmidt orthogonalization is subjected to eigenvalue decomposition to obtain the eigenvalues ​​of the matrix and their corresponding eigenvectors. The eigenvalues ​​are sorted in descending order, and the corresponding eigenvectors are arranged accordingly. The eigenvectors corresponding to the first six largest singular values ​​are taken to form a six-dimensional Grassmann manifold point U as the input of the recognition model.

[0007] As a preferred solution of the distribution network fault pattern recognition method described in the present invention, it includes: the first network includes a backbone network, a neck network and a detection head, wherein the backbone network includes a Stiefel convolution layer and several C3 modules, the neck network includes a bidirectional feature pyramid layer, and the detection head has dual output channels, wherein the first channel is used for bounding box prediction and the second channel is used for fault subspace feature prediction.

[0008] As a preferred solution of the distribution network fault pattern identification method described in the present invention, it includes: the second network includes a primary capsule layer, an intermediate capsule layer and an advanced capsule layer; the number of capsules in the primary capsule layer is 32, the subspace dimension is 4×6, and the number of routing iterations is 1; the number of capsules in the intermediate capsule layer is 16, the subspace dimension is 6×6, and the number of routing iterations is 2; the number of capsules in the advanced capsule layer is 12, the subspace dimension is 6×6, and the number of routing iterations is 3.

[0009] As a preferred solution of the distribution network fault pattern recognition method described in the present invention, the feature fusion layer includes: spatially aligning the output results of the first network and the output results of the second network through interpolation; calculating spatial attention on the aligned features and performing element-by-element fusion.

[0010] As a preferred solution of the distribution network fault pattern recognition method of the present invention, the method comprises: the C3 module includes a first convolutional layer, a harmonic attention mechanism layer, a second convolutional layer and a third convolutional layer;

[0011] Among them, the first convolutional layer is used to perform convolution operation on the input feature map to extract basic features;

[0012] The harmonic attention mechanism layer is used to fuse the basic features with the harmonic features, and the fused features are further convolved through the second convolutional layer. The third convolutional layer is used to connect the output of the second convolutional layer with the output of the first convolutional layer for feature fusion.

[0013] The present invention also provides a distribution network fault pattern recognition device for implementing the distribution network fault pattern recognition method as described in any of the above items, and the distribution network fault pattern recognition device includes: a signal processing module, configured to perform preprocessing of the distribution network signal; a mapping module, configured to perform mapping of the preprocessed distribution network signal to the Grassmann manifold space; a construction module, configured to perform construction of a recognition model, the recognition model including a feature extraction layer, a first network, a second network, a feature fusion layer and an output layer; and an identification module, configured to input the distribution network signal and the preprocessed distribution network signal into a trained recognition model to identify the fault pattern.

[0014] The present invention also provides a terminal device, comprising:

[0015] one or more processors;

[0016] a memory, coupled to the processor, for storing one or more programs;

[0017] When the one or more programs are executed by the one or more processors, the one or more processors implement the distribution network fault mode identification method as described in any one of the above items.

[0018] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the distribution network fault mode identification method as described in any one of the above items.

[0019] The beneficial effects of the present invention are as follows: By constructing a regularized Gram matrix and mapping it to the Grassmann manifold space, the present invention can better capture the intrinsic geometric structure and low-dimensional manifold features of the data, making the feature representation more discriminative and stable; and compared with other deep learning methods, the recognition model of the present invention combines the characteristics of the Grassmann manifold, introduces the Stiefel convolution layer and feature fusion strategy, and further enhances the extraction and recognition capabilities of fault mode features. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be derived from these drawings without inventive effort. Among them:

[0021] Figure 1 This is a flow chart of the method for identifying distribution network fault patterns according to the first embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of the process of mapping the preprocessed distribution network signal to the Grassmann manifold space according to the first embodiment of the present invention. DETAILED DESCRIPTION

[0023] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0024] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0026] The present invention is described in detail with reference to schematic diagrams. For ease of illustration, cross-sectional views of device structures may be partially enlarged and not to scale when describing embodiments of the present invention. Furthermore, the schematic diagrams are merely illustrative and should not limit the scope of the present invention. Furthermore, in actual production, the three-dimensional dimensions of length, width, and depth should be included.

[0027] In the description of the present invention, it should be noted that the terms "upper, lower, inner, and outer" and other references to orientations or positional relationships are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first, second, or third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0028] In this disclosure, unless otherwise specified or limited, the terms "mounted," "connected," and "connected" should be interpreted broadly. For example, they may refer to fixed, removable, or integral connections. They may also refer to mechanical, electrical, or direct connections, indirect connections through an intermediary, or internal communication between two components. Those skilled in the art will understand the specific meanings of these terms in this disclosure.

[0029] Example 1

[0030] Reference Figures 1 and 2 , which is the first embodiment of the present invention, provides a method for identifying a distribution network fault pattern, comprising:

[0031] S1: Preprocess the distribution network signal.

[0032] The distribution network signals are three-phase voltage, three-phase current, zero-sequence current, and zero-sequence voltage, and the sampling rate is set to 10kHz;

[0033] Four third-order Blackman-Harris window functions were used for windowing, where the coefficients of the Blackman-Harris window were [0.35875, 0.48829, 0.14128, 0.01168], the window length was 128 points, and the overlap rate was 50%;

[0034] The windowed signal is transformed by short-time Fourier transform or wavelet transform, and a 30×65 (30 time frames × 65 frequency points) complex matrix is ​​generated for each signal to fully capture the time-frequency characteristics of the signal.

[0035] S2: Map the preprocessed distribution network signal to the Grassmann manifold space.

[0036] Reference Figure 2 , construct the Gram matrix G = C·C based on the complex time-frequency matrix H , where C H is the conjugate transposed matrix of the complex time-frequency matrix C;

[0037] In order to prevent the Gram matrix from having singularities and affecting the stability and accuracy of subsequent calculations, a regularization term G1 is added to the Gram matrix G. The regularization term G1 is: G1 = 0.05 〃 tr (G) 〃 I (to prevent singular matrices), where tr (G) represents the trace of the Gram matrix G, that is, the sum of its diagonal elements; I is a unit matrix of the same dimension as the Gram matrix G. Preferably, the regularization term can effectively increase the stability of the matrix by adjusting the diagonal elements of the matrix, while not causing excessive interference to the overall structure of the matrix.

[0038] Get the regularized Gram matrix G reg =G+0.05"tr(G)"I;

[0039] Furthermore, the regularized Gram matrix G reg Perform Householder transformation to obtain matrix G trans =H〃G reg , where H is the Householder reflection matrix;

[0040]

[0041] Where v is represented by the Gram matrix G reg The column vector of v T is the transpose of vector v, is the projection matrix.

[0042] It should be noted that the Householder transform is an orthogonal transform that can convert a matrix into an upper triangular matrix or a lower triangular matrix, thereby simplifying the subsequent eigenvalue decomposition and orthogonalization process.

[0043] According to the matrix G trans Iterate the Gram-Schmidt orthogonalization. The number of iterations is set to 3. In each iteration, a dynamic weight adjustment mechanism is introduced. The dynamic weight adjustment mechanism includes setting dynamic weight coefficients according to the change in the vector modulus before and after orthogonalization to optimize the orthogonalization result. The specific steps are as follows:

[0044] (1) Initialization: Let the original vector set be: {v1,v2,…,v n}, these vectors are matrices G trans Corresponding column space; initialize the orthogonal vector u1=v1, where the vector v n is the matrix G trans The nth column of , and then the nth orthogonal vector u in the orthogonalization process is calculated step by step through the orthogonalization steps n .

[0045] (2) First iteration (k=1), calculate the projection component of v2 on u1

[0046]

[0047] Initial orthogonalization yields

[0048] Calculate weight coefficient We get u2 = α2u′2;

[0049] Where k is the number of iterations, is the projection component of v2 on u1, which represents the component of v2 in the direction of u1; u′2 is the intermediate orthogonal vector after vector v2 removes the component in the direction of u1, which represents the vector after preliminary orthogonalization; α2 is the weight coefficient, which is used to measure the relative importance of u′2 in v2; u2 is the second orthogonal vector after weight adjustment.

[0050] (3) Second iteration (k=2), calculate the projection components of v3 on u1 and u2

[0051]

[0052] Initial orthogonalization yields

[0053] Calculate weight coefficient We get u3 = α3u′3;

[0054] Where, is the projection component of v3 on u1, is the projection component of v3 on u2; u′3 is the intermediate orthogonal vector after removing the directional components of u1 and u2 from vector v3; α3 is the weight coefficient used to measure the relative importance of u′3 in v3; u3 is the third orthogonal vector after weight adjustment.

[0055] (3) The third iteration (k=3) calculates the projection components of v4 on u1, u2 and u3

[0056]

[0057] Initial orthogonalization yields

[0058] Calculate weight coefficient We get u3 = α3u′3;

[0059] Where, is the projection component of v4 on u1, is the projection component of v4 on u2, is the projection component of v4 on u3; u′4 is the intermediate orthogonal vector after removing the directional components of u1, u2, and u3 from vector v4; α4 is the weight coefficient used to measure the relative importance of u′4 in v4; u4 is the fourth orthogonal vector after weight adjustment.

[0060] (5) Normalization: For the orthogonal vector set {u1,u2,…,u n} to obtain the normalized orthogonal vector set {e1,e2,…,e n},in e i is the normalized i-th orthogonal vector.

[0061] Preferably, this can more effectively eliminate the linear correlation between vectors, improve the orthogonalization effect, and enhance the discrimination and representativeness of the eigenvectors. In each iteration, the improved orthogonalization formula is used to gradually orthogonalize the vectors in the matrix to obtain a more accurate set of orthogonal vectors.

[0062] Furthermore, the matrix after Gram-Schmidt orthogonalization is subjected to eigenvalue decomposition to obtain the eigenvalues ​​λ1,λ2,…,λ n and its corresponding eigenvectors e1,e2,…,e n , sort the eigenvalues ​​in descending order, and arrange the corresponding eigenvectors accordingly;

[0063] Take the eigenvectors e1, e2,…, e6 corresponding to the first 6 largest singular values ​​to form a 6-dimensional Grassmann manifold point U as the input of the recognition model, where U represents the position of the data in the Grassmann manifold space.

[0064] Preferably, by selecting the eigenvectors corresponding to the first six largest singular values, the most important feature information in the data is retained, and at the same time, the high-dimensional data is mapped to a six-dimensional Grassmann manifold space, achieving effective dimensionality reduction and feature representation. This not only reduces the complexity of the data, but also highlights the features closely related to the fault mode, providing high-quality feature input for subsequent distribution network fault mode identification.

[0065] S3: Build a recognition model, which includes a feature extraction layer, a first network, a second network, a feature fusion layer, and an output layer.

[0066] The feature extraction layer includes three Stiefel convolution layers with a convolution kernel size of 3×3. The convolution kernel of the Stiefel convolution layer is constrained by the Stiefel manifold, and its parameters are located on the Stiefel manifold. Each time the parameters are updated, the updated convolution kernel is projected back onto the Stiefel manifold to maintain its orthogonality.

[0067] The first network includes a backbone network, a neck network and a detection head, wherein the backbone network includes a Stiefel convolution layer and several C3 modules. Specifically, the output of the Stiefel convolution layer is forced to be orthogonalized through QR decomposition to maintain the feature subspace structure, and the convolution kernel is updated through the Riemann gradient descent method, so that the output feature map automatically meets the Grassmann manifold requirements; the C3 module includes a first convolution layer, a harmonic attention mechanism layer, a second convolution layer and a third convolution layer; wherein, the first convolution layer is used to perform convolution operations on the input feature map to extract basic features; the harmonic attention mechanism layer is used to fuse the basic features with the harmonic features, and the harmonic features are extracted through a learnable frequency domain filter group, and the fused features are further convolved through the second convolution layer. The third convolution layer is used to connect the output of the second convolution layer with the output of the first convolution layer through parallel transmission on the manifold to perform feature fusion. The neck network consists of a bidirectional feature pyramid layer, with manifold alignment layers and lateral connections inserted in the top-down and bottom-up paths; the detection head has dual output channels, where the first channel is used for bounding box prediction and the second channel is used for fault subspace feature prediction.

[0068] The second network includes a primary capsule layer, an intermediate capsule layer, and an advanced capsule layer; the primary capsule layer has 32 capsules, a subspace dimension of 4×6, and 1 routing iteration; the intermediate capsule layer has 16 capsules, a subspace dimension of 6×6, and 2 routing iterations; the advanced capsule layer has 12 capsules, a subspace dimension of 6×6, and 3 routing iterations.

[0069] Optionally, an attention mechanism is introduced in the dynamic routing process to enable the capsule network to pay more attention to the feature channels related to the current fault mode, thereby improving the efficiency and accuracy of feature extraction;

[0070] In addition, the second network is pruned and quantized to remove redundant capsule units and connections, optimize the network structure, and improve the computational efficiency and real-time performance of the model.

[0071] The feature fusion layer includes: spatially aligning the output results of the first network with the output results of the second network through interpolation; calculating spatial attention on the aligned features and performing element-by-element fusion.

[0072] Preferably, the first network of this embodiment utilizes its multi-scale feature fusion capability to fuse the high-level semantic features extracted by the second network with the features extracted from the Grassmann manifold features by the feature extraction layer. This enables the recognition model to have the ability to express multi-scale features, thereby more accurately locating and identifying fault modes in the distribution network.

[0073] S4: Input the distribution network signal and the pre-processed distribution network signal into the trained recognition model to identify the fault mode.

[0074] The distribution network signal and the pre-processed distribution network signal are fed into a trained recognition model to identify the fault mode. The recognition model calculates and judges based on the characteristics of the input data and outputs the recognition result. For example, the collected signal shows: Phase A voltage drops to 5.8kV (nominal 10kV) and zero-sequence current is 3.2A. This is fed into the recognition model and the decision outputs are: Classification result: AG type fault (confidence level 92%); Location result: Branch line 3.2km from the substation; Response time: 9.6ms.

[0075] Example 2

[0076] The present invention provides a distribution network fault pattern recognition device, which is used to implement the steps of the distribution network fault pattern recognition method as described in any of the above embodiments. The distribution network fault pattern recognition device includes:

[0077] A signal processing module is configured to perform pre-processing on the power distribution network signal;

[0078] A mapping module configured to perform mapping of the preprocessed power distribution network signal to a Grassmann manifold space;

[0079] A construction module is configured to execute construction of a recognition model, where the recognition model includes a feature extraction layer, a first network, a second network, a feature fusion layer, and an output layer;

[0080] The identification module is configured to input the distribution network signal and the pre-processed distribution network signal into the trained identification model to identify the fault mode.

[0081] Example 3

[0082] This embodiment provides a terminal device, including:

[0083] one or more processors;

[0084] a memory, coupled to the processor, for storing one or more programs;

[0085] When the one or more programs are executed by the one or more processors, the one or more processors implement the distribution network fault pattern identification method as described above.

[0086] The processor is used to control the overall operation of the terminal device to complete all or part of the steps of the above-mentioned distribution network fault pattern identification method. The memory is used to store various types of data to support the operation of the terminal device. For example, these data may include instructions for any application or method used to operate on the terminal device, as well as application-related data. The memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0087] The terminal device can be implemented by one or more application-specific integrated circuits (AS1Cs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to execute the distribution network fault mode identification method described in any of the above embodiments and achieve the same technical effect as the above method.

[0088] Example 4

[0089] This embodiment provides a computer-readable storage medium, wherein the program instructions, when executed by a processor, implement the steps of the distribution network fault pattern identification method described in any of the above embodiments. For example, the computer-readable storage medium may be the aforementioned memory containing the program instructions, and the program instructions may be executed by a processor of a terminal device to perform the distribution network fault pattern identification method described in any of the above embodiments, thereby achieving the same technical effects as the above methods.

[0090] It should be appreciated that embodiments of the present invention can be implemented or practiced by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable memory. The methods can be implemented in a computer program using standard programming techniques, including a non-transitory computer-readable storage medium configured with a computer program, wherein the storage medium so configured causes the computer to operate in a specific and predefined manner, according to the methods and figures described in the specific embodiments. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if desired, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. In addition, the program can be run on a programmed application-specific integrated circuit for this purpose.

[0091] Furthermore, the operations of the processes described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by the context. The processes described herein (or variations and / or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) that is executed collectively on one or more processors, by hardware, or a combination thereof. The computer program includes a plurality of instructions that can be executed by one or more processors.

[0092] Furthermore, the methods can be implemented in any type of computing platform operably connected to a suitable computer, including but not limited to a personal computer, minicomputer, mainframe, workstation, network or distributed computing environment, standalone or integrated computer platform, or in communication with a charged particle tool or other imaging device, etc. Aspects of the present invention can be implemented as machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard drive, optical read and / or write storage medium, RAM, ROM, etc., such that it can be read by a programmable computer, and when the storage medium or device is read by the computer, can be used to configure and operate the computer to perform the processes described herein. Furthermore, the machine-readable code, or portions thereof, can be transmitted over wired or wireless networks. When such media includes instructions or programs for implementing the steps described above in conjunction with a microprocessor or other data processor, the invention described herein includes these and other different types of non-transitory computer-readable storage media. The invention also includes the computer itself, when programmed according to the methods and techniques described herein. The computer program can be applied to input data to perform the functions described herein, thereby converting the input data to generate output data that is stored in non-volatile memory. The output information can also be applied to one or more output devices, such as a display. In a preferred embodiment of the present invention, the converted data represents a physical and tangible object, including a specific visual depiction of the physical and tangible object produced on a display.

[0093] As used in this application, the terms "component", "module", "system" and the like are intended to refer to a computer-related entity, which can be hardware, firmware, a combination of hardware and software, software, or software in operation. For example, a component can be, but is not limited to: a process running on a processor, a processor, an object, an executable file, a thread in execution, a program and / or a computer. As an example, both an application running on a computing device and the computing device can be a component. One or more components can exist in an executing process and / or thread, and a component can be located in a computer and / or distributed between two or more computers. In addition, these components can be executed from various computer-readable media having various data structures thereon. These components can communicate in the form of local and / or remote processes, such as based on signals having one or more data packets (e.g., data from a component that interacts with another component in a local system, a distributed system, and / or interacts with other systems in the form of signals over a network such as the Internet).

[0094] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. 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 solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for identifying distribution network fault patterns, characterized in that: include: Preprocessing of distribution network signals; Mapping the preprocessed distribution network signal to the Grassmann manifold space; Constructing a recognition model, wherein the recognition model includes a feature extraction layer, a first network, a second network, a feature fusion layer, and an output layer; The distribution network signal and the preprocessed distribution network signal are input into the trained recognition model to identify the fault mode.

2. The method for identifying distribution network fault patterns according to claim 1, wherein: The preprocessing of the distribution network signal comprises: The distribution network signals are three-phase voltage, three-phase current, zero-sequence current, and zero-sequence voltage, and the sampling rate is set to 10kHz; Four third-order Blackman-Harris window functions were used to perform windowing processing, with a window length of 128 points and an overlap rate of 50%. The windowed signal is transformed to generate a 30×65 complex matrix.

3. The method for identifying distribution network fault patterns according to claim 2, wherein: include: Mapping the preprocessed distribution network signal to the Grassmann manifold space includes: The Gram matrix G is constructed according to the complex time-frequency matrix, and the regularization term G1 is added to the Gram matrix G to obtain the regularized Gram matrix G reg , For the regularized Gram matrix G reg Perform Householder transformation to obtain matrix G trans ; According to the matrix G trans Iteratively perform Gram-Schmidt orthogonalization with a number of iterations set to 3, and introduce a dynamic weight adjustment mechanism in each iteration. The dynamic weight adjustment mechanism includes setting dynamic weight coefficients based on the change in the modulus length of the vector before and after orthogonalization to optimize the orthogonalization result; Perform eigenvalue decomposition on the matrix after Gram-Schmidt orthogonalization to obtain the eigenvalues ​​of the matrix and their corresponding eigenvectors, sort the eigenvalues ​​in descending order, and arrange the corresponding eigenvectors accordingly; The eigenvectors corresponding to the first six largest singular values ​​are taken to form a 6-dimensional Grassmann manifold point U as the input of the recognition model.

4. The method for identifying distribution network fault patterns according to any one of claims 1 to 3, wherein: include: The first network includes a backbone network, a neck network and a detection head, wherein the backbone network includes a Stiefel convolution layer and several C3 modules, the neck network includes a bidirectional feature pyramid layer, and the detection head has dual output channels, wherein the first channel is used for bounding box prediction and the second channel is used for fault subspace feature prediction.

5. The method for identifying distribution network fault patterns according to claim 4, wherein: include: The second network includes a primary capsule layer, an intermediate capsule layer, and an advanced capsule layer; wherein the number of capsules in the primary capsule layer is 32, the subspace dimension is 4×6, and the number of routing iterations is 1; the number of capsules in the intermediate capsule layer is 16, the subspace dimension is 6×6, and the number of routing iterations is 2; the number of capsules in the advanced capsule layer is 12, the subspace dimension is 6×6, and the number of routing iterations is 3.

6. The method for identifying distribution network fault patterns according to claim 5, wherein: The feature fusion layer includes: spatially aligning the output result of the first network with the output result of the second network through interpolation; calculating spatial attention on the aligned features, and performing element-by-element fusion.

7. The method for identifying distribution network fault patterns according to claim 4, wherein: include: The C3 module includes the first convolutional layer, the harmonic attention mechanism layer, the second convolutional layer, and the third convolutional layer; Among them, the first convolutional layer is used to perform convolution operation on the input feature map to extract basic features; The harmonic attention mechanism layer is used to fuse the basic features with the harmonic features, and the fused features are further convolved through the second convolutional layer. The third convolutional layer is used to connect the output of the second convolutional layer with the output of the first convolutional layer for feature fusion.

8. A distribution network fault pattern recognition device, configured to implement the distribution network fault pattern recognition method according to any one of claims 1 to 7, the distribution network fault pattern recognition device comprising: A signal processing module is configured to perform pre-processing on the power distribution network signal; A mapping module configured to perform mapping of the preprocessed power distribution network signal to a Grassmann manifold space; A construction module is configured to execute construction of a recognition model, wherein the recognition model includes a feature extraction layer, a first network, a second network, a feature fusion layer, and an output layer; The identification module is configured to input the distribution network signal and the pre-processed distribution network signal into the trained identification model to identify the fault mode.

9. A terminal device, characterized in that: include: one or more processors; a memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the distribution network fault mode identification method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The computer program is executed by a processor to implement the distribution network fault pattern identification method according to any one of claims 1 to 7.

Images