Fault recognition method, recognition device, electronic equipment and readable storage medium

Abstract

The invention discloses a fault recognition method and device, electronic equipment and a readable storage medium, and relates to the technical field of electronic power. The fault identification method comprises the following steps: acquiring a first parameter in the operation process of a distribution line; processing the first parameter to generate a first image according to Grubrum angle field transformation; the first image is input to a first neural network, fault features are generated, and an attention module is arranged in the first neural network; the fault features are input to a second neural network to determine probabilities of the plurality of fault types. According to the fault identification method, the accuracy and the response speed of fault detection can be remarkably improved. Meanwhile, in combination with an image processing technology, fault positions and types can be displayed more visually, and operation and maintenance personnel can take corresponding measures in time.

Description

Technical Field

[0001] The present invention relates to the field of electronic power technology, and in particular to a fault identification method, an identification device, an electronic device and a readable storage medium. Background Art

[0002] With the increase in the penetration rate of distributed source loads, mainly wind power, solar power generation, and new energy charging piles. The power system shows a typical trend of power electronics, and the fault characteristics of the distribution network have shown some new characteristics compared with the past. The typical characteristics mainly include two points: 1) The expansion of the scale of interference sources in the new power system; 2) The structure of the distribution network is complex, and the branches affect each other. When a distribution network fails, it further affects the power quality of the power grid, which will not only cause damage to sensitive equipment, data loss, and increased equipment energy consumption for end users, but may even lead to large-scale power outages.

[0003] In the related technologies, the fault diagnosis scheme for power distribution system line short circuit mainly includes fault feature analysis method, which is based on manual feature extraction, requires expert experience, is difficult to model, has high complexity of diagnosis scheme, and has low efficiency of fault diagnosis. Therefore, how to improve the fault diagnosis efficiency of power distribution lines has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The present invention aims to at least solve the technical problem of low efficiency in fault diagnosis of power distribution lines existing in the prior art or related art.

[0005] To this end, a first aspect of the present invention proposes a fault identification method.

[0006] A second aspect of the present invention provides a fault identification device.

[0007] A third aspect of the present invention provides an electronic device.

[0008] A fourth aspect of the present invention provides a readable storage medium.

[0009] In view of this, a first aspect of the present invention proposes a fault identification method for identifying the fault type of a distribution line, the fault identification method comprising: obtaining a first parameter during the operation of the distribution line; processing the first parameter according to the Gram angular field transform to generate a first image; inputting the first image into a first neural network to generate a fault feature, wherein an attention module is provided in the first neural network; inputting the fault feature into a second neural network to determine the probabilities of multiple fault types.

[0010] The fault identification method provided by the present invention can be used to identify the fault type existing in the distribution line, that is, to identify what kind of fault has occurred in the current distribution line. Specifically, the current in the distribution line is usually three-phase electricity, that is, phase A, phase B and phase C. Accordingly, the fault types of the distribution line may include: phase A grounding, phase B grounding, phase C grounding, AB short circuit, BC short circuit, AC short circuit, AB phase grounding, BC phase grounding, AC phase grounding, ABC short circuit, phase A intermittent grounding, phase B intermittent grounding, phase C intermittent grounding and lightning fault.

[0011] Furthermore, in the process of identifying the fault type of the distribution line, the first parameter in the operation process of the distribution line can be obtained first. Specifically, the first parameter of the distribution line can be obtained by using a data acquisition module. The first parameter can include three-phase current information and three-phase voltage information in the distribution line. The data acquisition module can include a monitoring device, which monitors the three-phase current information and three-phase voltage information in real time by installing the monitoring device in the distribution line. High-precision monitoring devices are reasonably deployed at the key nodes and branch points of each distribution line. Each monitoring device contains advanced sensors and data acquisition units, which can accurately capture the current and voltage changes in the line. The selection of the installation location is scientifically calculated and tested to ensure that all important detection points can be covered, providing basic data for comprehensive fault detection. After the monitoring device is installed, the data acquisition module will start to monitor the three-phase current and three-phase voltage information in real time. Highly sensitive sensors can capture tiny current and voltage changes and record them in real time through the data acquisition unit. The real-time monitoring function ensures the continuity and timeliness of the data, so that the system can obtain the latest electrical parameters in the first time.

[0012] Furthermore, after obtaining the first parameter during the operation of the distribution line, the first parameter can be processed to identify the fault type of the distribution line according to the first parameter. Specifically, the first parameter can be processed according to the Gram angular field transformation, so as to transform the first parameter into a corresponding first image. That is, digital information is transformed into image information by means of the Gram angular field transformation.

[0013] Furthermore, after the first image corresponding to the first parameter is generated, the first image can be input into the first neural network to perform feature extraction on the first image. Specifically, the first neural network can be a convolutional neural network, and the unique feature extraction capability of the convolutional neural network gives it a significant advantage in processing complex image data. It can not only identify significant features in the image, but also learn implicit patterns and laws from the data, with high generalization and robustness. With its excellent feature extraction capability, the convolutional neural network can better learn and capture deep information in the image. Through a combination of a series of convolutional layers, pooling layers, and nonlinear activation functions, the convolutional neural network can extract multi-scale information of the image layer by layer.

[0014] Specifically, convolutional neural networks can effectively process high-dimensional image data through local connections and shared weights. Its core operations include convolution, pooling, and activation: Convolution layer: Use the convolution kernel to slide on the image and extract local features through the local receptive field. The convolution operation can capture basic features such as edges and textures in the image. Pooling layer: Downsample the features extracted by the convolution layer to reduce the data dimension and retain important information. Common pooling operations include maximum pooling and average pooling. Activation function: Through nonlinear activation functions, nonlinearity is introduced, allowing the network to learn more complex feature representations. These operations are superimposed on multiple layers, allowing convolutional neural networks to extract and integrate multi-scale information in the image layer by layer, thereby performing well in feature learning.

[0015] Furthermore, an attention module is also provided in the first neural network. Through the attention module, the weight of the features can be dynamically adjusted during the process of extracting the features of the first image, so that the first neural network can pay more attention to the key areas and important features in the first image, thereby further improving the feature extraction performance of the first neural network and further improving the accuracy of fault identification.

[0016] Furthermore, after the first neural network extracts features from the first image, the fault features can be input into the second neural network, so that the probabilities of all possible fault types of the distribution line can be determined based on the fault features, and then the fault of the distribution line can be determined. That is, the fault type with the highest probability can be determined as the fault of the current distribution line.

[0017] Specifically, the second neural network can be a multi-layer neural network, which can be used in conjunction with the Softmax function to classify fault features. Specifically, the Softmax function calculates the probability of each fault type and normalizes the result into a probability distribution. Finally, the second neural network determines the most likely fault type based on the calculated probability and outputs the classification result.

[0018] The fault identification method provided in the present application utilizes the Gram angular field transform to transform the first parameter of the distribution line obtained into a first image, that is, to transform the digital information into image information, and then extract features of the first image through a first neural network to generate fault features. At the same time, an attention module is provided in the first neural network to further improve the feature extraction effect of the first neural network. Finally, the fault features are processed through a second neural network to determine the type of fault existing in the distribution line. That is, by combining the Gram angular field transform with a neural network based on an attention mechanism, the fault type of the distribution line can be identified, which can significantly improve the accuracy and response speed of fault detection. At the same time, combined with image processing technology, the fault location and type can be displayed more intuitively, which is convenient for operation and maintenance personnel to take corresponding measures in a timely manner.

[0019] In addition, the identification method in the above technical solution provided by the present invention may also have the following additional technical features:

[0020] In some technical schemes, optionally, the first parameter is processed according to the Gram angular field transform to generate the first image, including: encoding the first parameter in polar coordinate space to generate coordinate information corresponding to the first parameter, wherein the coordinate information includes angle information and radius information corresponding to the first parameter; processing the coordinate information according to the Gram angular field transform to generate the Gram angular sum field, Gram angular difference field and radius matrix corresponding to the first parameter; determining the Gram angular sum field matrix corresponding to the first parameter according to the Gram angular sum field and radius matrix; determining the Gram angular difference field matrix corresponding to the first parameter according to the Gram angular difference field and radius matrix; and generating the first image according to the Gram angular sum field matrix and the Gram angular difference field matrix.

[0021] In some technical solutions, optionally, before encoding the first parameter in polar coordinate space and generating coordinate information corresponding to the first parameter, the fault identification method further includes: preprocessing the first parameter; and scaling the preprocessed first parameter so that the first parameter is within a preset range.

[0022] In some technical solutions, optionally, the attention module includes a spatial attention module and a channel attention module, and the first image is input into the first neural network to generate fault features, including: inputting the first image into the first neural network, and determining the spatial attention weight of the first image through the spatial attention module; determining the channel attention weight of the first image through the channel attention module; and performing feature extraction on the first image according to the spatial attention weight and the channel attention weight to generate fault features.

[0023] In some technical solutions, optionally, the first parameter includes three-phase voltages and three-phase currents of multiple detection points in the distribution line. A self-attention feature fusion module is also provided in the first neural network. Feature extraction of the first image is performed according to the spatial attention weight and the channel attention weight. Before generating the fault feature, the fault identification method also includes: fusing the voltage parameters and current parameters of each phase through the self-attention feature fusion module.

[0024] In some technical schemes, optionally, before obtaining the first parameter during the operation of the distribution line, the fault identification method also includes: obtaining multiple second parameters during the operation of the distribution line within a preset time period; dividing the multiple second parameters into multiple training data sets according to multiple fault types, wherein each training data set includes an operating parameter corresponding to at least one fault type; processing the multiple training data sets according to the Gram angular field transform to generate multiple second images; and training the first neural network according to the multiple second images.

[0025] In some technical solutions, optionally, the first neural network is trained according to the multiple second images, including: inputting the multiple second images into the first neural network, and adjusting the parameters of the first neural network based on an adaptive moment estimation optimizer.

[0026] According to a second aspect of the present invention, a fault identification device is proposed for identifying a fault type of a distribution line, the fault identification device comprising: an acquisition unit, for acquiring a first parameter during the operation of the distribution line; a processing unit, for processing the first parameter according to a Gram angular field transform to generate a first image; an input unit, for inputting the first image into a first neural network to generate a fault feature, wherein an attention module is provided in the first neural network; the input unit is also used to input the fault feature into a second neural network to determine the probabilities of multiple fault types.

[0027] The fault identification device provided by the present application utilizes the Gram angular field transform to transform the first parameter of the distribution line obtained into a first image, that is, to transform the digital information into image information, and then extracts features of the first image through the first neural network to generate fault features. At the same time, an attention module is provided in the first neural network to further improve the feature extraction effect of the first neural network. Finally, the fault features are processed through the second neural network to determine the type of fault existing in the distribution line. That is, by combining the Gram angular field transform with a neural network based on an attention mechanism, the fault type of the distribution line can be identified, which can significantly improve the accuracy and response speed of fault detection. At the same time, combined with image processing technology, the fault location and type can be displayed more intuitively, which is convenient for operation and maintenance personnel to take corresponding measures in a timely manner.

[0028] Furthermore, the processing unit is specifically used to: perform polar coordinate space encoding on the first parameter to generate coordinate information corresponding to the first parameter, wherein the coordinate information includes angle information and radius information corresponding to the first parameter; process the coordinate information according to the Gram angle field transform to generate a Gram angle sum field, a Gram angle difference field and a radius matrix corresponding to the first parameter; determine the Gram angle sum field matrix corresponding to the first parameter according to the Gram angle sum field and radius matrix; determine the Gram angle difference field matrix corresponding to the first parameter according to the Gram angle difference field and radius matrix; and generate a first image according to the Gram angle sum field matrix and the Gram angle difference field matrix.

[0029] Furthermore, the processing unit is specifically used to: preprocess the first parameter; and scale the preprocessed first parameter so that the first parameter is within a preset range.

[0030] Furthermore, the attention module includes a spatial attention module and a channel attention module, and the input unit is specifically used to: input the first image into the first neural network, and determine the spatial attention weight of the first image through the spatial attention module; determine the channel attention weight of the first image through the channel attention module; extract features of the first image according to the spatial attention weight and the channel attention weight to generate fault features.

[0031] Furthermore, the first parameter includes three-phase voltage and three-phase current of multiple detection points in the distribution line. A self-attention feature fusion module is also provided in the first neural network, and the processing unit is also used to fuse the voltage parameters and current parameters of each phase through the self-attention feature fusion module.

[0032] Furthermore, the acquisition unit is also used to acquire multiple second parameters during the operation of the distribution line within a preset time period; the processing unit is also used to divide the multiple second parameters into multiple training data sets according to multiple fault types, wherein each training data set includes an operating parameter corresponding to at least one fault type; according to the Gram angular field transform, the multiple training data sets are processed to generate multiple second images; and the first neural network is trained according to the multiple second images.

[0033] Furthermore, the processing unit is specifically used to: input the multiple second images into the first neural network, and adjust the parameters of the first neural network based on the adaptive moment estimation optimizer.

[0034] According to a third aspect of the present application, an electronic device is proposed, comprising a processor and a memory, wherein the memory stores programs or instructions that can be executed on the processor, and when the programs or instructions are executed by the processor, the steps of the fault identification method of the first aspect are implemented.

[0035] According to a fourth aspect of the present application, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the fault identification method of the first aspect are implemented.

[0036] Additional aspects and advantages of the present invention will become apparent from the following description or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0038] Figure 1 A schematic diagram showing a flow chart of a fault identification method according to an embodiment of the present invention;

[0039] Figure 2 A structural principle diagram of a fault identification method according to an embodiment of the present invention is shown;

[0040] Figure 3 A circuit diagram of a power distribution circuit according to an embodiment of the present invention is shown;

[0041] Figure 4 The operating principle diagram of the attention module in the fault identification method according to the embodiment of the present invention is shown;

[0042] Figure 5 A waveform diagram of a first parameter and a corresponding first image when a lightning fault occurs in a power distribution line according to an embodiment of the present invention are shown;

[0043] Figure 6 A waveform diagram of a first parameter and a corresponding first image when a phase A instantaneous grounding fault occurs in a power distribution line according to an embodiment of the present invention are shown;

[0044] Figure 7 A waveform diagram of a first parameter and a corresponding first image when a permanent ground fault of phase A occurs in a power distribution line according to an embodiment of the present invention are shown;

[0045] Figure 8 Shows Figure 2 A partial enlarged view of point D in the middle;

[0046] Fig. 9 Shows Figure 2 A partial enlarged view of point E in the middle;

[0047] Fig.10 shows the loss curve during the training of the first neural network in the embodiment of the present application;

[0048] Fig.11 The accuracy curve of the first neural network test process in the embodiment of the present application is shown;

[0049] Fig.12 The accuracy curve of the first neural network in the photovoltaic multi-node access situation in the embodiment of the present application is shown;

[0050] Fig.13 A structural block diagram of a fault identification device according to an embodiment of the present application is shown;

[0051] Fig.14 A structural block diagram of an electronic device according to an embodiment of the present application is shown.

[0052] Reference numerals:

[0053] 1300 fault identification device, 1302 acquisition unit, 1304 processing unit, 1306 input unit, 1400 electronic device, 1402 processor, 1404 memory. DETAILED DESCRIPTION

[0054] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0055] 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. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0056] Refer to the following Figures 1 to 14 A fault identification method, an identification device, an electronic device, and a readable storage medium according to some embodiments of the present invention are described.

[0057] like Figure 1 As shown, according to an embodiment of the present invention, a fault identification method is proposed for identifying the fault type of a distribution line. The fault identification method includes:

[0058] S102, obtaining a first parameter during the operation of the distribution line;

[0059] S104, processing the first parameter according to Gram angular field transform to generate a first image;

[0060] S106, inputting the first image into a first neural network to generate a fault feature, wherein the first neural network is provided with an attention module;

[0061] S108, inputting the fault characteristics into a second neural network to determine the probabilities of multiple fault types.

[0062] The fault identification method provided by the present invention can be used to identify the fault type existing in the distribution line, that is, to identify what kind of fault has occurred in the current distribution line. Specifically, the current in the distribution line is usually three-phase electricity, that is, phase A, phase B and phase C. Accordingly, the fault types of the distribution line may include: phase A grounding, phase B grounding, phase C grounding, AB short circuit, BC short circuit, AC short circuit, AB phase grounding, BC phase grounding, AC phase grounding, ABC short circuit, phase A intermittent grounding, phase B intermittent grounding, phase C intermittent grounding, and lightning fault.

[0063] Furthermore, in the process of identifying the fault type of the distribution line, the first parameter of the distribution line operation process can be obtained first. Specifically, the data acquisition module can be used to obtain the first parameter of the distribution line. The first parameter can include three-phase current information and three-phase voltage information in the distribution line. Figure 3 As shown, the data acquisition module may include a monitoring device ( Figure 3 By installing monitoring devices in the distribution line, different positions in the distribution line ( Figure 3 The three-phase current information and three-phase voltage information of the fault location 1 to the fault location 5 in the distribution line. Figure 3 Fault location 1 to fault location 5), reasonably deploy high-precision monitoring devices. Each monitoring device contains advanced sensors and data acquisition units, which can accurately capture the current and voltage changes in the line. The selection of the installation location has been scientifically calculated and tested to ensure that all important detection points can be covered to provide basic data for comprehensive fault detection. After the monitoring device is installed, the data acquisition module will start to monitor the three-phase current and three-phase voltage information in real time. Highly sensitive sensors can capture tiny current and voltage changes and record them in real time through the data acquisition unit. The real-time monitoring function ensures the continuity and timeliness of the data, so that the system can obtain the latest electrical parameters in the first time. In addition, it should be noted that the distribution line is connected to the substation through a transformer to receive electrical energy. In the process of obtaining the first parameter, the protection device simulation device (protection device simulation device 1 to protection device simulation device 3) is used for simulation operation.

[0064] Furthermore, after obtaining the first parameter during the operation of the distribution line, the first parameter can be processed to identify the fault type of the distribution line according to the first parameter. Specifically, the first parameter can be processed according to the Gram angular field transformation, so as to transform the first parameter into a corresponding first image. That is, digital information is transformed into image information by means of the Gram angular field transformation.

[0065] It is understandable that for different fault types, the specific values ​​of the first parameter are different, that is, under different fault conditions, the values ​​presented by the three-phase current and the three-phase voltage are different, and accordingly, the display content of the first image is also different, such as Figure 5 , Figure 6 and Figure 7 shown.

[0066] Furthermore, if Figure 2 As shown, after the first image corresponding to the first parameter is generated, the first image can be input into the first neural network to perform feature extraction on the first image. Specifically, the first neural network can be a convolutional neural network, and the unique feature extraction capability of the convolutional neural network gives it a significant advantage in processing complex image data. It can not only identify significant features in the image, but also learn implicit patterns and laws from the data, with high generalization and robustness. With its excellent feature extraction capability, the convolutional neural network can better learn and capture deep information in the image. Through a combination of a series of convolutional layers, pooling layers, and nonlinear activation functions, the convolutional neural network can extract multi-scale information of the image layer by layer.

[0067] Specifically, convolutional neural networks can effectively process high-dimensional image data through local connections and shared weights. Its core operations include convolution, pooling, and activation: Convolution layer: Use the convolution kernel to slide on the image and extract local features through the local receptive field. The convolution operation can capture basic features such as edges and textures in the image. Pooling layer: Downsample the features extracted by the convolution layer to reduce the data dimension and retain important information. Common pooling operations include maximum pooling and average pooling. Activation function: Through nonlinear activation functions, nonlinearity is introduced, allowing the network to learn more complex feature representations. These operations are superimposed on multiple layers, allowing convolutional neural networks to extract and integrate multi-scale information in the image layer by layer, thereby performing well in feature learning.

[0068] Furthermore, an attention module is also provided in the first neural network. Through the attention module, the weight of the features can be dynamically adjusted during the process of extracting the features of the first image, so that the first neural network can pay more attention to the key areas and important features in the first image, thereby further improving the feature extraction performance of the first neural network and further improving the accuracy of fault identification.

[0069] Furthermore, if Figure 2As shown, after the first neural network extracts features from the first image, the fault features can be input into the second neural network, so that the probabilities of all possible fault types of the distribution line can be determined according to the fault features, and then the fault of the distribution line can be determined. That is, the fault type with the highest probability can be determined as the fault of the current distribution line.

[0070] Specifically, the second neural network can be a multi-layer neural network, which can be used in conjunction with the Softmax function to classify fault features. Specifically, the Softmax function calculates the probability of each fault type and normalizes the result into a probability distribution. Finally, the second neural network determines the most likely fault type based on the calculated probability and outputs the classification result.

[0071] The fault identification method provided in the present application utilizes the Gram angular field transform to transform the first parameter of the distribution line obtained into a first image, that is, to transform the digital information into image information, and then extract features of the first image through a first neural network to generate fault features. At the same time, an attention module is provided in the first neural network to further improve the feature extraction effect of the first neural network. Finally, the fault features are processed through a second neural network to determine the type of fault existing in the distribution line. That is, by combining the Gram angular field transform with a neural network based on an attention mechanism, the fault type of the distribution line can be identified, which can significantly improve the accuracy and response speed of fault detection. At the same time, combined with image processing technology, the fault location and type can be displayed more intuitively, which is convenient for operation and maintenance personnel to take corresponding measures in a timely manner.

[0072] In some embodiments, optionally, the first parameter is processed according to the Gram angular field transform to generate a first image, including: encoding the first parameter in polar coordinate space to generate coordinate information corresponding to the first parameter, wherein the coordinate information includes angle information and radius information corresponding to the first parameter; processing the coordinate information according to the Gram angular field transform to generate a Gram angular sum field, a Gram angular difference field, and a radius matrix corresponding to the first parameter; determining the Gram angular sum field matrix corresponding to the first parameter according to the Gram angular sum field and radius matrix; determining the Gram angular difference field matrix corresponding to the first parameter according to the Gram angular difference field and radius matrix; and generating the first image according to the Gram angular sum field matrix and the Gram angular difference field matrix.

[0073] In this embodiment, in the process of processing the first parameter according to the Gram angular field transform and generating the first image, firstly, the first parameter may be transformed into polar coordinates to generate coordinate information of the first parameter, wherein the coordinate information includes angle information and radius information corresponding to the first parameter. Specifically, the preset formula may be used:

[0074]

[0075] Among them, X is the first parameter, is the angle information corresponding to the first parameter, r is the radius information corresponding to the first parameter, i represents the sequence number of the first parameter, that is, the first parameter can be multiple, i represents the sequence number of the first parameter, and T is the acquisition time of the first parameter. Through polar coordinate transformation, the first parameter can be converted into angle information and radius information under polar coordinates.

[0076] Further, the coordinate information is processed according to the Gram angle field transformation to generate a Gram angle sum field, a Gram angle difference field and a radius matrix corresponding to the first parameter;

[0077] Specifically, according to the formula:

[0078]

[0079] Determine the Gram angle and field corresponding to the first parameter.

[0080] According to the formula:

[0081]

[0082] Determine the Gram angle difference field corresponding to the first parameter.

[0083] According to the formula:

[0084]

[0085] Determine the radius matrix corresponding to the first parameter.

[0086] Wherein, n represents an integer from 1 to i, and the Gram angle sum field matrix, the Gram angle difference field matrix and the radius matrix are all i×i.

[0087] Furthermore, according to the formula:

[0088] GADF_R=GADF·R GAF

[0089] Determine the Gram angle and field matrix corresponding to the first parameter.

[0090] According to the formula:

[0091] GASF_R=GASF·R GAF

[0092] Determine the Gram angle difference field matrix corresponding to the first parameter.

[0093] Finally, the first image can be generated according to the Gram angle sum field matrix and the Gram angle difference field matrix. Specifically, the GASF_R and GADF_R matrices can be fused. The two matrices are spliced ​​in a certain dimension (for example, horizontally or vertically) to generate a matrix containing more features, that is, the first image.

[0094] In some embodiments, optionally, before encoding the first parameter in polar coordinate space and generating coordinate information corresponding to the first parameter, the fault identification method further includes: preprocessing the first parameter; and scaling the preprocessed first parameter so that the first parameter is within a preset range.

[0095] In this embodiment, before polar coordinate encoding is performed on the first parameter, the first parameter may be preprocessed. Specifically, the preprocessing of the first parameter may include steps such as denoising and normalization to ensure the quality and consistency of the data.

[0096] Furthermore, after preprocessing the first parameter, the first parameter may be scaled so that the value of the first parameter is within a preset range, thereby reducing the value of the first parameter and thereby reducing the amount of calculation in the subsequent processing of the first parameter. Specifically, according to the formula:

[0097]

[0098] The first parameter is scaled to between [-1, 1]. Where X is the first parameter, and i is the serial number of the first parameter. Here, the range of i can be selected according to the needs, that is, the number of the first parameter can be set. It can be understood that since the convolutional neural network is complex for large input images, in order to reduce the amount of calculation of the convolutional neural network, for example, the value range of i can be between 1 and 100. Accordingly, after the first image is generated, the size of the first image is 100×100.

[0099] Furthermore, after the first image is generated, since the size of the first image is 100×100, the first neural network can use the VGG16 (visual geometry group 16) deep convolutional neural network to extract features from the first image. In addition, the VGG16 network was originally designed to process input images of 224×224 scale. In order to adapt to the needs of the first image, the input parameters of the VGG16 network can be modified to meet the processing requirements of the first image size. Specifically, part of the architecture of the VGG16 network can be cropped to retain only its feature extraction part. That is, the proven convolutional layers and pooling layers in VGG16 are used intensively to give full play to its powerful ability in extracting image features.

[0100] In some embodiments, optionally, the attention module includes a spatial attention module and a channel attention module, and inputs the first image into the first neural network to generate fault features, including: inputting the first image into the first neural network, and determining the spatial attention weight of the first image through the spatial attention module; determining the channel attention weight of the first image through the channel attention module; and performing feature extraction on the first image according to the spatial attention weight and the channel attention weight to generate fault features.

[0101] In this embodiment, if Figure 4 As shown, the attention module may include a spatial attention module and a channel attention module, that is, in the first neural network, a hybrid attention mechanism of spatial attention and channel attention is added, so as to enhance the ability to extract important information and local features in the first image and improve the ability of the first neural network to capture details in the first image.

[0102] Specifically, the attention mechanism is a method that imitates the human visual and cognitive system, which allows the neural network to focus on relevant parts when processing input data. By introducing the attention mechanism, the neural network can automatically learn and selectively focus on important information in the input, improving the performance and generalization ability of the model. The application of the attention mechanism in the first neural network can be divided into the following steps: Weight calculation: Generate an attention weight matrix by calculating the relevance of the input features. This matrix reflects the importance of each feature in the current task. Weighted fusion: According to the attention weight, the features are weighted averaged or weighted summed. This step can highlight key features and suppress irrelevant information.

[0103] Among them, Figure 8 As shown, the spatial attention module emphasizes the contextual information of the image features by calculating the correlation between the pixels of the first image. A convolution operation is performed on the first image in the spatial dimension to obtain the spatial attention weight.

[0104] Specifically, the spatial attention weight Attention(X) = σ(W s ×X), where X is the first input image, W s is the weight of the spatial convolutional layer, and σ is the activation function.

[0105] The channel attention module performs a global average pooling (GAP) operation on the first input image to reduce the dimension. The features after pooling are fully connected (FC) to obtain the channel attention weight.

[0106] Specifically, the attention weight Attention(X) = σ(W α ·X). Where X is the first input image, W α is the weight of the fully connected layer, and σ is the activation function.

[0107] Furthermore, in the process of extracting features of the first image by the first neural network, features of the first image can be extracted according to the spatial attention weight and the channel attention weight to generate fault features. The first neural network can more intelligently allocate feature weights when processing the first image, significantly improving the overall performance of the first neural network. The attention mechanism not only enhances the feature extraction capability of the first neural network, but also improves its adaptability and robustness to complex image data.

[0108] That is, the first neural network can efficiently learn and extract multi-scale features in the first image through multi-layer convolution, pooling and activation operations. The introduction of the attention mechanism further optimizes the distribution of feature weights, allowing the first neural network to pay more attention to key areas in the first image, improving the effect of feature extraction and overall performance. This method that combines convolutional neural networks and attention mechanisms provides strong technical support for image processing and deep learning tasks.

[0109] In some embodiments, optionally, the first parameter includes three-phase voltages and three-phase currents of multiple detection points in the distribution line. A self-attention feature fusion module is also provided in the first neural network. Feature extraction of the first image is performed according to the spatial attention weight and the channel attention weight. Before generating the fault feature, the fault identification method also includes: fusing the voltage parameters and current parameters of each phase through the self-attention feature fusion module.

[0110] In this embodiment, the first parameter includes three-phase voltages and three-phase currents of multiple detection points in the distribution line. Through the self-attention feature fusion module, the voltage features and current features of each phase can also be fused. The six input channels are converted into three channels for subsequent processing. This is achieved through the convolution operation transformation to ensure that the converted feature representation still maintains the integrity of the original information.

[0111] Specifically, Fig. 9As shown in the figure, the feature fusion of the self-attention mechanism makes full use of the voltage feature image, and the self-attention mechanism weight image is obtained after the calculation of the self-attention mechanism. These input channels will be processed separately and embedded into a feature space, which is completed through a 1×1 convolution layer (Conv1×1). The self-attention mechanism generates a weight matrix by calculating the correlation of the input features, thereby focusing on the most important part of the data. For each input feature, the query, key, and value vectors are calculated. The attention weight matrix is ​​obtained by calculating the dot product of the query and the key. The value vector is weighted summed using the attention weight matrix to obtain the output feature. The features processed by the self-attention mechanism are fused. The fusion strategy can be weighted summation, splicing, or other methods to integrate the information of each channel to generate a more expressive feature representation. The generated attention weight matrix can enhance the local information in the image, allowing the model to capture important features more accurately. In current and voltage signals, this local feature enhancement helps to identify faults and abnormal patterns.

[0112] Specifically, the self-attention feature fusion module takes the feature image of the same-phase current and voltage information as the input of the module, such as Fig. 9 shown.

[0113] Q i =W q U i ;

[0114] K i =W K U i ;

[0115] V i =W V I i ;

[0116] The vectors of query, key and value are calculated as shown above, where Q i is the query vector, K i is the key vector, V i is a value vector, U is voltage, I is current, i is the serial number of each value, W q , W K , W V Represents a trainable weight matrix. Then the attention weight matrix is ​​obtained by taking the dot product of the query vector and the key vector.

[0117]

[0118] Among them, A ij represents the current attention score, d kThe dimension of the key vector is used to scale the result of the dot product. i and j represent the number of columns and rows of the attention weight matrix, respectively. j K i is the transpose of Q i Vector and each K i The similarity between vectors.

[0119]

[0120] These attention weights are weighted and summed with the corresponding Value vector Vj to obtain a weighted sum vector. j is an integer from 1 to n, and n is the number of the first parameter. This weighted sum vector will be used as a new representation of the current vector, containing richer context information. The output feature O is obtained i .

[0121] In some embodiments, optionally, before obtaining the first parameter during the operation of the distribution line, the fault identification method also includes: obtaining multiple second parameters during the operation of the distribution line within a preset time period; dividing the multiple second parameters into multiple training data sets according to multiple fault types, wherein each training data set includes operating parameters corresponding to at least one fault type; processing the multiple training data sets according to the Gram angular field transform to generate multiple second images; and training the first neural network according to the multiple second images.

[0122] In this embodiment, the first neural network can be trained by acquiring multiple second parameters during the operation of the distribution line within a preset time period to improve the feature extraction capability of the first neural network, that is, to improve the accuracy of the first neural network.

[0123] Specifically, firstly, a plurality of second parameters of the operation process of the distribution line in a preset time period are obtained. For example, the distribution line can be simulated to run for 1.6s at a sampling frequency of 20kHz, and then the second parameters are collected for each different fault type. The second parameters are the three-phase current and three-phase voltage at different positions in the distribution line. Specifically, a total of 16,000 data samples can be collected for each three-phase signal of the transmission fault type.

[0124] Furthermore, according to the multiple fault types, the multiple second parameters are divided into multiple training data sets, wherein each training data set includes an operating parameter corresponding to at least one fault type.

[0125] It is understandable that in order to improve the accuracy of sample information, the data samples can be divided into multiple different time periods when the fault occurs using a time window according to the fault type, that is, divided into multiple training data sets.

[0126] Furthermore, according to the Gram angular field transform, multiple training data sets are processed to generate multiple second images; and the first neural network is trained according to the multiple second images.

[0127] It should be noted that 80% of the multiple second images can be used to train the first neural network, and 20% of the multiple second images can be used to test the performance of the first neural network to ensure the feature extraction capability of the first neural network.

[0128] In some embodiments, optionally, training the first neural network according to the plurality of second images includes: inputting the plurality of second images into the first neural network, and adjusting parameters of the first neural network based on an adaptive moment estimation optimizer.

[0129] In this embodiment, during the training of the first neural network, a plurality of second images may be input into the first neural network, and the parameters of the first neural network may be adjusted based on the adaptive moment estimation optimizer.

[0130] It is understood that neural networks mimic the application of neurons present in the human brain. When neurons are connected, the strength of the electrical signal is known as weights. The weight of a particular neuron is multiplied by a particular value that is the input to the neuron. The sum of all weighted values ​​associated with the next layer is applied to the input as the activation function. In this type of network, when new data is inserted into the network, the interaction between neurons affects the output neuron. The new data needs to be optimized to conform to the previous input, and the back-propagation algorithm can be used to optimize the neuron weights in this neural network architecture. The adaptive moment estimation (Adam) optimizer is then used to achieve the learning of the target features by the network.

[0131] Specifically, at the beginning of the training process, the input data of the first neural network is pre-labeled as the correct output. The labeled data samples are passed through the first neural network. The actual output from the first neural network and the correct output can be different or the same. If there are differences between the outputs, the learning rate is multiplied by a weighting parameter that reflects these differences. When the new weight values ​​are updated, the resulting results are applied. Through multiple experiments, the optimal parameters of the Adam optimizer were determined, resulting in the lowest loss value and the highest model accuracy.

[0132] The model was trained for 10 cycles using the Adam optimization algorithm and the optimal result was obtained using the categorical cross entropy cost function. After the training phase, the first neural network had the same performance as the second image obtained by the Gram angular field transform. Fig.11 As shown, the training method achieves an accuracy of 99.98% and a loss of 0.0004. This loss value is the optimal loss value for the proposed model.

[0133] In some specific embodiments, the distribution line may include a photovoltaic single-node access form and a photovoltaic multi-node access form.

[0134] like Figure 3 As shown in the figure, the access points of photovoltaic are photovoltaic 1, photovoltaic 2 and photovoltaic 3, respectively, for the form of photovoltaic single node access, that is, when only one access point is considered each time. The obtained fault data set is input into the neural network after Gram angle field transformation. Under the condition of single node access of photovoltaic, the accuracy of fault recognition of distribution line can reach 99.97%. Therefore, neural network is feasible for distribution network fault identification and classification. Due to the obvious distinction of distribution line fault characteristics, the first neural network can be stabilized after 10 iterations, and the experimental accuracy is close to 100%.

[0135] Furthermore, the hybrid attention mechanism is placed in the last layer of the first neural network to improve the first neural network's ability to extract image detail features, and then the two-dimensional feature map is converted into a one-dimensional vector feature as the input of the second neural network. Fig.10 and Fig.11 As shown in the figure, after adding mixed attention, the first neural network can better learn new features in the data set at each iteration, avoiding falling into the local optimal solution. After adding mixed attention, the training process is shortened and the accuracy is higher.

[0136] When three photovoltaics are connected, that is, when a high proportion of distributed power sources are connected in the form of photovoltaic multi-node access, energy flows in both directions, and the distribution network fault characteristics are complex, which is quite different from the traditional distribution network. The training method of the first neural network provided by this application can ensure the image feature ability of the first neural network, ensure that the weight and bias of the first neural network remain unchanged, and shorten the time of secondary training. Fig.12 As shown in the figure, after the photovoltaic multi-nodes are connected to the distribution line, the first neural network can achieve a fault recognition accuracy of 98.48% after 10 cycles of data learning. That is, the first neural network can be applied to the distribution network with large-capacity photovoltaic access.

[0137] According to one embodiment of the present invention, Fig.13As shown, a fault identification device 1300 is proposed for identifying the fault type of a distribution line. The fault identification device 1300 includes: an acquisition unit 1302, for acquiring a first parameter during the operation of the distribution line; a processing unit 1304, for processing the first parameter according to the Gram angular field transform to generate a first image; an input unit 1306, for inputting the first image into a first neural network to generate a fault feature, wherein an attention module is provided in the first neural network; the input unit 1306 is also used to input the fault feature into a second neural network to determine the probabilities of multiple fault types.

[0138] The fault identification device 1300 provided in the present application uses the Gram angular field transform to transform the first parameter of the distribution line obtained into a first image, that is, to transform the digital information into image information, and then extract the features of the first image through the first neural network to generate fault features. At the same time, an attention module is provided in the first neural network to further improve the feature extraction effect of the first neural network. Finally, the fault features are processed by the second neural network to determine the type of fault existing in the distribution line. That is, by combining the Gram angular field transform with the neural network based on the attention mechanism, the fault type of the distribution line can be identified, which can significantly improve the accuracy and response speed of fault detection. At the same time, combined with image processing technology, the fault location and type can be displayed more intuitively, which is convenient for operation and maintenance personnel to take corresponding measures in time.

[0139] Further, the processing unit 1304 is specifically used to: perform polar coordinate space encoding on the first parameter to generate coordinate information corresponding to the first parameter, wherein the coordinate information includes angle information and radius information corresponding to the first parameter; process the coordinate information according to the Gram angle field transform to generate a Gram angle sum field, a Gram angle difference field and a radius matrix corresponding to the first parameter; determine the Gram angle sum field matrix corresponding to the first parameter according to the Gram angle sum field and radius matrix; determine the Gram angle difference field matrix corresponding to the first parameter according to the Gram angle difference field and radius matrix; and generate a first image according to the Gram angle sum field matrix and the Gram angle difference field matrix.

[0140] Furthermore, the processing unit 1304 is specifically configured to: preprocess the first parameter; and scale the preprocessed first parameter so that the first parameter is within a preset range.

[0141] Furthermore, the attention module includes a spatial attention module and a channel attention module, and the input unit 1306 is specifically used to: input the first image into the first neural network, and determine the spatial attention weight of the first image through the spatial attention module; determine the channel attention weight of the first image through the channel attention module; extract features of the first image according to the spatial attention weight and the channel attention weight to generate fault features.

[0142] Furthermore, the first parameter includes three-phase voltages and three-phase currents of multiple detection points in the distribution line. A self-attention feature fusion module is also provided in the first neural network, and the processing unit 1304 is also used to fuse the voltage parameters and current parameters of each phase through the self-attention feature fusion module.

[0143] Furthermore, the acquisition unit 1302 is also used to obtain multiple second parameters during the operation of the distribution line within a preset time period; the processing unit 1304 is also used to divide the multiple second parameters into multiple training data sets according to multiple fault types, wherein each training data set includes an operating parameter corresponding to at least one fault type; according to the Gram angular field transform, the multiple training data sets are processed to generate multiple second images; and the first neural network is trained according to the multiple second images.

[0144] Furthermore, the processing unit 1304 is specifically used to: input the multiple second images into the first neural network, and adjust the parameters of the first neural network based on the adaptive moment estimation optimizer.

[0145] According to one embodiment of the present invention, Fig.14 As shown, an electronic device 1400 is proposed, including a processor 1402 and a memory 1404, wherein the memory 1404 stores programs or instructions that can be executed on the processor 1402, and when the programs or instructions are executed by the processor 1402, the steps of the fault identification method as in any one of the above embodiments are implemented.

[0146] The electronic device 1400 provided by the present invention implements the steps of the fault identification method in any one of the above embodiments when the program or instruction is executed by the processor 1402. Therefore, the electronic device 1400 has all the beneficial effects of the steps of the fault identification method in any one of the above embodiments, which will not be repeated here.

[0147] According to one embodiment of the present invention, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the fault identification method as described in any one of the above embodiments are implemented.

[0148] The readable storage medium provided by the present invention implements the steps of the fault identification method in any one of the above-mentioned embodiments when the program or instruction is executed by the processor. Therefore, the readable storage medium has all the beneficial effects of the steps of the fault identification method in any one of the above-mentioned embodiments, which will not be repeated here.

[0149] In the present invention, the term "plurality" means at least two or more than two, unless otherwise clearly defined. The terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; "connected" can be a direct connection or an indirect connection through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0150] In the description of this specification, the description of the terms "one embodiment", "some embodiments", "specific embodiments", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0151] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may be modified and transformed in various ways. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A fault identification method for identifying the fault type of a power distribution line, characterized in that: The fault identification method comprises: Acquire a first parameter during the operation of the power distribution line; Processing the first parameter according to Gram angular field transform to generate a first image; Inputting the first image into a first neural network to generate a fault feature, wherein the first neural network is provided with an attention module; The fault signature is input into a second neural network to determine the probability of a plurality of the fault types.

2. The fault identification method according to claim 1, characterized in that: The step of processing the first parameter according to the Gram angular field transform to generate a first image includes: Perform polar coordinate space encoding on the first parameter to generate coordinate information corresponding to the first parameter, wherein the coordinate information includes angle information and radius information corresponding to the first parameter; Processing the coordinate information according to the Gram angle field transformation to generate a Gram angle sum field, a Gram angle difference field and a radius matrix corresponding to the first parameter; Determine a Gram angle and field matrix corresponding to the first parameter according to the Gram angle and field and the radius matrix; Determine a Gram angle difference field matrix corresponding to the first parameter according to the Gram angle difference field and the radius matrix; The first image is generated according to the Gram angle sum field matrix and the Gram angle difference field matrix.

3. The fault identification method according to claim 2, characterized in that: Before performing polar coordinate space encoding on the first parameter to generate coordinate information corresponding to the first parameter, the fault identification method further includes: Preprocessing the first parameter; The preprocessed first parameter is scaled so that the first parameter is within a preset range.

4. The fault identification method according to claim 1, characterized in that: The attention module includes a spatial attention module and a channel attention module, and the first image is input into a first neural network to generate a fault feature, including: Inputting the first image into a first neural network, and determining a spatial attention weight of the first image through the spatial attention module; Determining, by the channel attention module, a channel attention weight of the first image; Feature extraction is performed on the first image according to the spatial attention weight and the channel attention weight to generate the fault feature.

5. The fault identification method according to claim 4, characterized in that: The first parameter includes three-phase voltages and three-phase currents at multiple detection points in the distribution line. The first neural network is further provided with a self-attention feature fusion module. The feature extraction of the first image is performed according to the spatial attention weight and the channel attention weight. Before generating the fault feature, the fault identification method further includes: The voltage parameters and current parameters of each phase are fused through the self-attention feature fusion module.

6. The fault identification method according to any one of claims 1 to 5, characterized in that: Before obtaining the first parameter in the operation process of the distribution line, the fault identification method further includes: Acquiring a plurality of second parameters during operation of the power distribution line within a preset time period; According to multiple fault types, the multiple second parameters are divided into multiple training data sets, wherein each of the training data sets includes an operating parameter corresponding to at least one fault type; Processing the plurality of training data sets according to the Gram angular field transform to generate a plurality of second images; The first neural network is trained based on a plurality of second images.

7. The fault identification method according to claim 6, characterized in that: The training of the first neural network according to the plurality of second images comprises: A plurality of the second images are input into the first neural network, and parameters of the first neural network are adjusted based on an adaptive moment estimation optimizer.

8. A fault identification device for identifying the fault type of a power distribution line, characterized in that: The fault identification device comprises: An acquisition unit, used for acquiring a first parameter during the operation of the power distribution line; A processing unit, configured to process the first parameter according to Gram angular field transform to generate a first image; An input unit, used for inputting the first image into a first neural network to generate a fault feature, wherein the first neural network is provided with an attention module; The input unit is also used to input the fault characteristics into a second neural network to determine the probabilities of the multiple fault types.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a program or instruction that can be run on the processor, and when the program or instruction is executed by the processor, the steps of the fault identification method according to any one of claims 1 to 7 are implemented.

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

Images