Cable insulation layer defect type detection method

Abstract

The invention relates to a cable insulation layer defect type detection method. The cable insulation layer defect type detection method comprises the following steps: obtaining harmonic characteristic data of a to-be-detected cable; and inputting the harmonic characteristic data into a self-organizing mapping model constructed based on a self-organizing mapping grid to carry out cable defect identification processing to obtain a target defect type of the to-be-detected cable. The structural characteristics of the self-organizing mapping grid provided by the invention enable the self-organizing mapping grid to have good adaptability to defect identification of complex types of cables. When a self-organizing mapping model is constructed, a self-organizing mapping grid can be flexibly designed according to actual needs. For example, for situations having a variety of possible defect types (e.g., partial discharge, insulation damage, joint looseness, etc.) and each defect type having complex performance on harmonic features, the model may accommodate this complexity by adjusting the size of the mesh.

Description

Technical Field

[0001] The present application relates to the technical field of insulation detection, and in particular to a method for detecting defect types of cable insulation layers. Background Art

[0002] Cables are the core part of urban power transmission and distribution systems, and the health of cable insulation directly affects the safe operation of the power grid. However, as the cable ages, problems such as insulation aging and moisture gradually emerge, which can cause power system failures, resulting in large-scale power outages or fires. Therefore, accurate detection and classification of cable insulation defects are of great practical significance.

[0003] In the existing cable detection method, Fourier transform is usually used to decompose the current signal generated during the operation of the cable into multiple harmonic components, and these components are classified through clustering algorithms to identify the type of cable defects. Commonly used clustering algorithms need to model the data after dimensionality reduction to identify the aging defects and moisture defects of the cable.

[0004] However, it is difficult for the reduced-dimensional data to fully reflect the complex structure of the original data, which is not conducive to identifying complex defect types. Summary of the invention

[0005] Based on this, it is necessary to provide a cable insulation layer defect type detection method that can identify complex defect types in response to the above technical problems.

[0006] In a first aspect, the present application provides a method for detecting the type of defects in a cable insulation layer. The method comprises:

[0007] Obtain harmonic characteristic data of the cable to be tested;

[0008] The harmonic feature data is input into a pre-built self-organizing map model for cable defect recognition processing to obtain the target defect type of the cable to be tested; the self-organizing map model is built based on a self-organizing map grid, and the self-organizing map grid corresponds to the data category.

[0009] In one embodiment, the self-organizing map grid includes a plurality of target neurons, and the harmonic feature data is input into a pre-built self-organizing map model for cable defect recognition processing to obtain the target defect type of the cable to be tested, including:

[0010] Based on the self-organizing map model, the harmonic feature data and each target neuron are mapped to obtain the mapping result;

[0011] Determine the target defect type of the cable to be tested based on the mapping results.

[0012] In one embodiment, the mapping process of the harmonic feature data and each target neuron to obtain a mapping result includes:

[0013] Calculate the distance between the harmonic feature data and each target neuron, and determine the target neuron with the smallest distance as the winner neuron;

[0014] The harmonic feature data is mapped to the position corresponding to the winner neuron to obtain a mapping result; the dimension of the weight vector of the winner neuron is smaller than the dimension of the harmonic feature data.

[0015] In one embodiment, the process of constructing the self-organizing map grid includes:

[0016] Determine a basic structure of an initial mapping grid; the basic structure includes one of a two-dimensional rectangular grid and a hexagonal grid;

[0017] Initializing the weight vector of the initial mapping grid according to the sample feature data acquired in advance to obtain an intermediate mapping network;

[0018] The intermediate mapping network is trained according to the sample feature data to obtain a self-organizing mapping grid.

[0019] In one embodiment, the sample feature data includes a plurality of sample vectors, the intermediate mapping network includes a plurality of initial neurons, and the intermediate mapping network is trained according to the sample feature data to obtain a self-organizing mapping grid, including:

[0020] In each iteration, for each sample vector, the distance between the sample vector and each initial neuron is calculated, and the initial neuron with the smallest distance is determined as the intermediate neuron;

[0021] Obtaining domain neurons within a preset neighborhood range of the interneuron;

[0022] The first weight of the intermediate neuron and the second weight of the domain neuron are updated respectively to obtain an updated intermediate mapping network;

[0023] When the number of iterations meets a preset condition, the updated intermediate mapping network is determined as a self-organizing mapping grid.

[0024] In one embodiment, the updating process of the first weight of the intermediate neuron and the second weight of the domain neuron to obtain an updated intermediate mapping network includes:

[0025] Inputting the first weight and the second weight into the update model to perform weight calculation processing to obtain an update vector result;

[0026] The updated intermediate mapping network is determined according to the update vector result.

[0027] In one embodiment, the update vector result includes a first weight update result and a second weight update result; the first weight and the second weight are input into the update model for weight calculation processing to obtain the update vector result, including:

[0028] Get random sample vectors, neighborhood function values, and target learning rates;

[0029] Determine a first weight update result according to the first weight, the random sample vector, the neighborhood function value, and the target learning rate;

[0030] A second weight update result is determined according to the second weight, the random sample vector, the neighborhood function value, and the target learning rate.

[0031] In one embodiment, the process of determining the neighborhood function value includes:

[0032] Obtaining the neighborhood width, the first grid position of the interneuron, and the second grid position of the field neuron;

[0033] The neighborhood width, the first grid position, and the second grid position are input into a Gaussian function for calculation to obtain a neighborhood function value.

[0034] In one embodiment, the process of determining the target learning rate includes:

[0035] Get the initial learning rate, current number of iterations, and maximum number of iterations;

[0036] Calculate the quotient of the current number of iterations and the maximum number of iterations;

[0037] According to the quotient value and the initial learning rate, the target learning rate is determined.

[0038] In one embodiment, the process of determining the harmonic characteristic data includes:

[0039] Obtain the current signal of the cable to be tested;

[0040] Performing harmonic component extraction processing on the current signal to obtain harmonic component data;

[0041] The harmonic component data is preprocessed to obtain harmonic feature data; wherein the data preprocessing includes at least one of normalization processing, noise reduction processing and feature extraction processing.

[0042] In a second aspect, the present application also provides a device for detecting the type of defects in the cable insulation layer. The device comprises:

[0043] A data acquisition module, used to acquire harmonic characteristic data of the cable to be tested;

[0044] The type determination module is used to input the harmonic feature data into a pre-built self-organizing map model for cable defect identification processing to obtain the target defect type of the cable to be tested; the self-organizing map model is built based on the self-organizing map grid, and the self-organizing map grid corresponds to the data category.

[0045] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0046] Obtain harmonic characteristic data of the cable to be tested;

[0047] The harmonic feature data is input into a pre-built self-organizing map model for cable defect recognition processing to obtain the target defect type of the cable to be tested; the self-organizing map model is built based on a self-organizing map grid, and the self-organizing map grid corresponds to the data category.

[0048] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0049] Obtain harmonic characteristic data of the cable to be tested;

[0050] The harmonic feature data is input into a pre-built self-organizing map model for cable defect recognition processing to obtain the target defect type of the cable to be tested; the self-organizing map model is built based on a self-organizing map grid, and the self-organizing map grid corresponds to the data category.

[0051] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0052] Obtain harmonic characteristic data of the cable to be tested;

[0053] The harmonic feature data is input into a pre-built self-organizing map model for cable defect recognition processing to obtain the target defect type of the cable to be tested; the self-organizing map model is built based on a self-organizing map grid, and the self-organizing map grid corresponds to the data category.

[0054] The above-mentioned cable insulation layer defect type detection method, the cable insulation layer defect type detection method includes obtaining harmonic characteristic data of the cable to be tested; inputting the harmonic characteristic data into a self-organizing map model constructed based on the self-organizing map grid to perform cable defect identification processing, and obtaining the target defect type of the cable to be tested. The structural characteristics of the self-organizing map grid of the present application make it have good adaptability to complex types of cable defect identification. When constructing a self-organizing map model, the self-organizing map grid can be flexibly designed according to actual needs. For example, for a situation where there are multiple possible defect types (such as partial discharge, insulation damage, loose joints, etc.) and each defect type has a complex manifestation in the harmonic characteristics, the model can adapt to this complexity by adjusting the size of the grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A diagram showing an application environment of a method for detecting cable insulation layer defect types in an embodiment;

[0056] Figure 2 A schematic diagram of a flow chart of a method for detecting defect types of cable insulation layers in one embodiment;

[0057] Figure 3 A schematic diagram of a process for obtaining a target defect type of a cable to be tested in one embodiment;

[0058] Figure 4 A schematic diagram of a process for determining a mapping result in one embodiment;

[0059] Figure 5 A schematic diagram of a process of constructing a self-organizing map grid in one embodiment;

[0060] Figure 6 A schematic diagram of a process of a self-organizing map grid in one embodiment;

[0061] Figure 7 is a schematic diagram of a process of obtaining an updated intermediate mapping network in one embodiment;

[0062] Figure 8 A schematic diagram of a process for obtaining an update vector result in one embodiment;

[0063] Fig. 9 1 is a flow chart of a process for determining a neighborhood function value in one embodiment;

[0064] Fig.10 is a flow chart of a process for determining a target learning rate in one embodiment;

[0065] Fig.11 is a flow chart of a process for determining harmonic characteristic data in one embodiment;

[0066] Fig.12a This is one of the experimental verification results of a method for detecting cable insulation layer defect types in one embodiment;

[0067] Figure 12b This is the second experimental verification result of a method for detecting cable insulation layer defect types in an embodiment;

[0068] Fig.12c This is the third experimental verification result of a method for detecting cable insulation layer defect types in an embodiment;

[0069] Fig.12d This is the fourth experimental verification result of a method for detecting cable insulation layer defect types in an embodiment;

[0070] Fig.13 is a structural block diagram of a device for detecting defect types of cable insulation layers in one embodiment;

[0071] Fig.14 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0073] First of all, before specifically introducing the technical solution of the embodiments of the present application, the technical background based on the embodiments of the present application is introduced.

[0074] Cables are the core part of urban power transmission and distribution systems, and the health of cable insulation directly affects the safe operation of the power grid. However, as the cable ages, problems such as insulation aging and moisture gradually emerge, which can cause power system failures, resulting in large-scale power outages or fires. Therefore, accurate detection and classification of cable insulation defects are of great practical significance.

[0075] Among the existing cable detection methods, harmonic current analysis has been widely used due to its high sensitivity and non-destructive testing advantages. Traditional methods usually use Fourier transform to decompose the current signal generated during cable operation into multiple harmonic components, and classify these components through clustering algorithms to identify the type of cable defects. Commonly used clustering algorithms include principal component analysis combined with Gaussian mixture models, which can identify cable aging defects and moisture defects by modeling the reduced-dimensional data.

[0076] However, with the increasing complexity of the cable application environment and the increase in the number of years of operation, the single PCA-GMM model has exposed certain limitations when processing high-dimensional and complex defect data. First, PCA may lose some valuable information during the dimensionality reduction process, especially the topological structure information of the data, which is crucial for accurately identifying different types of defects. Second, the Gaussian mixture model has limited classification capabilities for complex data, especially in the presence of a large amount of noise or uneven data distribution. Therefore, existing methods are often not accurate enough when dealing with the identification of cable defect types in complex environments.

[0077] The closest existing technology is a cable insulation defect detection method that combines principal component analysis and Gaussian mixture model. The harmonic current data of the cable is obtained through Fourier transform, and principal component analysis is performed on these data to extract the main components, thereby reducing the data dimension. After that, the Gaussian mixture model is used to perform cluster analysis on the reduced-dimensional data to determine the type of cable defects.

[0078] The advantage of this method is that it can effectively reduce the data dimension, so that the computational efficiency of the model can be improved. However, since the principal component analysis only retains the main components in the data and ignores the minor components, this may lead to the loss of some key information, especially when the minor components contain information that helps to identify the defect type. In addition, the Gaussian mixture model assumes that the data conforms to the normal distribution. For actual complex cable defect data, this assumption is not always valid, resulting in reduced classification accuracy.

[0079] The existing cable defect detection methods based on PCA and GMM mainly have the following technical problems:

[0080] (1) Data topology loss: Although PCA retains the main information of the data during the dimensionality reduction process, it ignores the minor components, which may contain topological structure information that is valuable for defect identification. Therefore, some key information may be lost during the dimensionality reduction process, affecting the accuracy of defect identification.

[0081] (2) Insufficient high-dimensional data processing capabilities: When faced with high-dimensional data, traditional methods often simplify the data processing process by reducing the dimensionality. However, the reduced-dimensional data cannot fully reflect the complex structure of the original data, which is not conducive to identifying complex defect types, especially when the types of cable insulation defects are diverse.

[0082] (3) Insufficient classification accuracy: Since the Gaussian mixture model assumes that the data is normally distributed, and the distribution of cable harmonic data in practice is often more complex, the deviation from this assumption may lead to a decrease in classification accuracy. In particular, when there is a lot of noise or atypical data, the classification performance of the GMM model decreases significantly.

[0083] Based on this, the present application provides a method and device for detecting the defect type of a cable insulation layer, aiming to solve the above-mentioned technical problems.

[0084] The cable insulation layer defect type detection method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The server obtains the harmonic characteristic data of the cable to be tested; the harmonic characteristic data is input into the self-organizing map model constructed based on the self-organizing map grid for cable defect identification processing to obtain the target defect type of the cable to be tested. Among them, the terminal 102 can be but not limited to a smart meter, a power sensor, a data acquisition device, etc., and the server 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0085] In an exemplary embodiment, Figure 2 As shown, the present application provides a method for detecting the defect type of cable insulation layer, and the method is applied to Figure 1 The server in the example is used to illustrate the following steps:

[0086] S201, obtaining harmonic characteristic data of the cable to be tested.

[0087] Among them, harmonic characteristic data: During the operation of the cable, the current or voltage signal contains multiple frequency components. Harmonics refer to the components that are greater than integer multiples of the fundamental frequency obtained by Fourier series decomposition of periodic non-sinusoidal alternating current. The harmonic characteristic data of the cable are the data related to these different harmonic components, such as the amplitude and phase of each harmonic. These data can reflect the electrical characteristics of the cable, because different defect states of the cable insulation layer will affect the harmonic characteristics of the current or voltage. For example, the aging of the insulation layer may change the capacitance and inductance characteristics of the cable, thereby affecting the amplitude and phase of the harmonics.

[0088] In the embodiment of the present application, the server obtains the harmonic characteristic data of the cable to be tested by communicating with a special measuring device. These measuring devices can be high-precision power sensors, such as sensors that can measure the harmonic components in the cable current or voltage signal. After the sensor collects the electrical signal of the cable, it is filtered, amplified, analog-to-digital converted, and other operations are performed by the signal processing circuit, and the processed harmonic characteristic data is transmitted to the server.

[0089] S202, inputting the harmonic feature data into a pre-built self-organizing map model to perform cable defect recognition processing to obtain a target defect type of the cable to be tested.

[0090] Among them, SOM (Self-Organizing Map) is built based on the self-organizing map grid, and the self-organizing map grid corresponds to the data category.

[0091] The self-organizing map grid is the core structural part of the self-organizing map model. It is a grid structure composed of neurons, which can be a one-dimensional, two-dimensional or higher-dimensional grid. In this embodiment, the self-organizing map grid corresponds to the data category, which means that different grid areas can correspond to different cable insulation layer defect types or normal state categories. For example, a certain area in the grid may represent the moisture defect category of the cable insulation layer, and another area may represent the insulation layer damage defect category, etc. The distribution of neurons in the grid has a certain topological structure relationship, which helps to classify and map the input data.

[0092] In the embodiment of the present application, the pre-built self-organizing map model is obtained by training based on a large amount of harmonic feature data of known cable insulation layer states (including different defect types and normal states). During the training process, the harmonic feature data under different states are input into the self-organizing map model, and the neurons in the model gradually adjust their weights through competition and cooperation mechanisms, so that the model can classify the input data.

[0093] When the harmonic feature data of the cable to be tested is input into the trained self-organizing map model, the model will determine the corresponding category, that is, the target defect type of the cable to be tested, based on the mapping relationship of the input data in the self-organizing map grid. For example, if the input harmonic feature data is mapped to the grid area representing insulation damage in the model, then the target defect type of the cable to be tested is determined to be insulation damage.

[0094] The above-mentioned cable insulation layer defect type detection method and device, the cable insulation layer defect type detection method includes obtaining harmonic characteristic data of the cable to be tested; inputting the harmonic characteristic data into a self-organizing map model constructed based on the self-organizing map grid to perform cable defect identification processing, and obtaining the target defect type of the cable to be tested. The structural characteristics of the self-organizing map grid of the present application make it have good adaptability to complex types of cable defect identification. When constructing a self-organizing map model, the self-organizing map grid can be flexibly designed according to actual needs. For example, for a situation where there are multiple possible defect types (such as partial discharge, insulation damage, loose joints, etc.) and each defect type has a complex manifestation in the harmonic characteristics, the model can adapt to this complexity by adjusting the size of the grid.

[0095] In an exemplary embodiment, based on the above embodiment, see Figure 3The self-organizing map grid of the embodiment of the present application includes a plurality of target neurons, and the harmonic feature data is input into a pre-built self-organizing map model for cable defect recognition processing to obtain the target defect type of the cable to be tested, including the following steps:

[0096] S301, based on the self-organizing mapping model, mapping processing is performed on the harmonic feature data and each target neuron to obtain a mapping result.

[0097] Among them, the target neuron refers to the basic unit in the self-organizing map grid. Each target neuron has a certain weight vector, which is constantly adjusted during the training process of the model. When receiving the input harmonic feature data, the target neuron determines the degree of matching with the input data by comparing it with its own weight vector (such as calculating the distance). For example, when processing the harmonic feature data related to the moisture defect of the cable insulation layer, the target neuron that matches the moisture defect pattern better will show a stronger response in the mapping process.

[0098] After receiving the harmonic characteristic data of the cable to be tested, the self-organizing map model begins to map it with each target neuron in the self-organizing map grid. Specifically, for each target neuron, the distance metric between its weight vector and the input harmonic characteristic data is calculated. This distance metric can be a common distance calculation method such as Euclidean distance. For example, assuming that the weight vector of the target neuron is and the input harmonic characteristic data is, then the Euclidean distance. In this way, the distance between the harmonic characteristic data and each target neuron is calculated, and the target neuron with the smallest distance or the combination of several target neurons with the smallest distance within a certain range constitutes the mapping result.

[0099] In the embodiment of the present application, after receiving the harmonic characteristic data of the cable to be tested, the self-organizing map model begins to map it with each target neuron in the self-organizing map grid. Specifically, for each target neuron, the distance metric between its weight vector and the input harmonic characteristic data is calculated. This distance metric can be a common distance calculation method such as Euclidean distance.

[0100] S302: Determine the target defect type of the cable to be tested according to the mapping result.

[0101] Among them, the target defect types may include insulation layer damage, insulation layer aging, air gaps or voids inside the cable, etc.

[0102] In the embodiment of the present application, since the target neurons in the self-organizing map grid correspond to different cable insulation layer defect types or normal states, the target defect type of the cable to be tested can be determined according to the mapping result obtained in step S301. For example, if the mapping result shows that the distance to the target neuron representing the insulation layer damage defect is the closest, then the target defect type of the cable to be tested is determined to be insulation layer damage. This correspondence is established during the training process of the self-organizing map model. The model is trained through a large amount of harmonic feature data of cables with known defect types, so that the target neurons in the model can be accurately associated with different defect types.

[0103] The embodiment of the present application can more accurately detect the defect type of the cable insulation layer using the self-organizing map grid, thereby improving the safety and reliability of cable operation.

[0104] In an exemplary embodiment, based on the above embodiment, see Figure 4 The embodiment of the present application involves mapping harmonic feature data and target neurons to obtain a mapping result, including the following steps:

[0105] S401, calculating the distance between the harmonic feature data and each target neuron, and determining the target neuron with the smallest distance as the winner neuron.

[0106] Among them, Winner Neuron: In the self-organizing mapping model, when calculating the distance between the harmonic feature data and each target neuron, the target neuron with the smallest distance is determined as the winner neuron. This neuron plays a key role in the mapping process. It represents a state that best matches (closest distance) the input harmonic feature data. In the context of cable insulation defect type detection, the position corresponding to the winner neuron may be associated with a certain cable insulation defect type or normal state. For example, if the winner neuron is determined in a model that has been trained with a large amount of cable sample data containing insulation damage, then this winner neuron may correspond to the insulation damage defect type.

[0107] Weight Vector: An important attribute of the target neuron. Each target neuron has a weight vector, which contains a series of weight values. In the self-organizing map model, the weight vector is used to compare with the input harmonic feature data to calculate the distance. The dimension of the weight vector represents the number of elements in the vector. For example, in a simple two-dimensional space, the weight vector may be (w1, w2), where the dimension is 2. It is a feature of an embodiment of the present application that the dimension of the weight vector is smaller than the dimension of the harmonic feature data, which helps to simplify the calculation and extract key information from the harmonic feature data. For example, the harmonic feature data may contain information of multiple dimensions such as the amplitude and phase of multiple frequency components, and the weight vector may be a low-dimensional vector associated with this information after being refined and simplified.

[0108] Calculate the winner neuron: Calculate the input sample And the weight vector of each neuron in the SOM network The Euclidean distance between them is used to find the neuron with the smallest distance (the winner neuron) and record it as c, as shown in formula (1):

[0109] (1)

[0110] Among them, the Euclidean distance is shown in formula (2):

[0111] (2)

[0112] By calculating the distance between the harmonic feature data and each target neuron, the target neuron with the smallest distance is found, and this target neuron is the winner neuron. This calculation method can measure the similarity between the harmonic feature data and each target neuron, and the smallest distance means the highest similarity.

[0113] S402, mapping the harmonic feature data to the position corresponding to the winner neuron to obtain a mapping result.

[0114] Among them, the dimension of the weight vector of the winner neuron is smaller than the dimension of the harmonic feature data.

[0115] In the embodiment of the present application, once the winner neuron is determined, the harmonic feature data is mapped to the position corresponding to the winner neuron. This corresponding position has a specific meaning in the self-organizing map grid, and it is associated with a certain state of the cable insulation layer (such as a certain defect type or normal state). For example, in a two-dimensional self-organizing map grid, the winner neuron is located at a certain coordinate position. After mapping the harmonic feature data to this position, the mapping result can be determined based on the corresponding relationship between this position and the state of the cable insulation layer established in the previous training process. This mapping result will serve as an important basis for the subsequent determination of the target defect type of the cable to be tested.

[0116] The embodiments of the present application can more accurately and efficiently map the harmonic feature data and the target neurons, thereby improving the overall performance of the cable insulation layer defect type detection.

[0117] In an exemplary embodiment, based on the above embodiment, see Figure 5 The construction process of the self-organizing map grid in the embodiment of the present application includes the following steps:

[0118] S501, determining the basic structure of the initial mapping grid.

[0119] The initial mapping grid is the starting structure for building the self-organizing mapping grid. It is a grid with a basic structure, which includes one of a two-dimensional rectangular grid and a hexagonal grid. This initial grid structure is the basic framework for the subsequent construction process.

[0120] When constructing a self-organizing mapping grid, the embodiment of the present application first needs to determine the basic structure of the initial mapping grid. The selection of a two-dimensional rectangular or hexagonal grid as the basic structure is based on their respective geometric characteristics and potential advantages in data mapping. If a two-dimensional rectangular grid is selected, its grid points are arranged in a rectangular shape and have a regular row and column structure. This structure may be more suitable when processing data with obvious linear relationships or rectangular distribution characteristics. For example, in the detection of cable insulation defect types, if the harmonic feature data presents a rectangular distribution-like regularity in some dimensions, the two-dimensional rectangular grid may be able to better capture this regularity. The grid points of the hexagonal grid are arranged in a hexagonal shape, which has higher space filling efficiency and better symmetry. When the sample feature data has certain symmetry or continuity characteristics, the hexagonal grid may be more conducive to mapping and classifying the data. For example, when processing cable insulation layer moisture defect data, if there is a certain symmetry or continuous change relationship between the degree of moisture and certain harmonic feature data, the hexagonal grid may be able to more accurately reflect this relationship.

[0121] S502, performing weight vector initialization processing on the initial mapping grid according to the sample feature data obtained in advance to obtain an intermediate mapping network.

[0122] Among them, the sample feature data is used to construct and train the data samples of the self-organizing map grid. In the context of cable insulation defect type detection, the sample feature data is the relevant data obtained from the cable with known cable insulation state (including various defect types and normal state). For example, these data may contain the harmonic feature data of the cable, such as the amplitude and phase of different harmonic frequency components, and may also contain other feature information related to the cable insulation state. These sample feature data are the basis for constructing an accurate self-organizing map grid.

[0123] The intermediate mapping network is an intermediate mapping structure after the weight vector is initialized during the construction of the self-organizing mapping grid. It is a network structure based on the initial mapping grid and after the initial weight vector is assigned according to the sample feature data.

[0124] Once the basic structure of the initial mapping grid is determined in the embodiment of the present application, it is necessary to initialize the weight vector thereof according to the pre-acquired sample feature data. Specifically, for each element in the initial mapping grid (such as each neuron or grid point), an initial weight vector is determined according to the characteristics of the sample feature data. This process can be carried out in a variety of ways, such as random initialization combined with range adjustment of the sample feature data. Assuming that the value range of a certain feature dimension of the sample feature data is [a, b], when a corresponding weight value of the weight vector is randomly initialized, it can be made to be within this value range or to take values ​​near this value range according to a certain ratio. The intermediate mapping network thus obtained has the initial ability to process the sample feature data, and the weight vector of each element can be used as the initial state for subsequent training processing.

[0125] S503, training the intermediate mapping network according to the sample feature data to obtain a self-organizing mapping grid.

[0126] In the embodiment of the present application, after the intermediate mapping network is obtained, it is necessary to further train it according to the sample feature data. This training process is an iterative process. In each iteration, a sample in the sample feature data is input into the intermediate mapping network. First, the distance (e.g., Euclidean distance) between this sample and each element in the intermediate mapping network (through its weight vector) is calculated, and the element with the smallest distance (called the winning element) is found.

[0127] Then, according to certain learning rules, adjust the weight vectors of the winning element and its neighboring elements. For example, a common learning rule is to move the weight vectors of the winning element and its neighboring elements in a small step toward the direction of the input sample. This step size can be determined according to the learning rate. As the number of iterations increases, the weight vectors of the elements in the intermediate mapping network are constantly adjusted, gradually forming the ability to classify different sample feature data, and finally obtaining a self-organizing mapping grid. This self-organizing mapping grid can accurately classify the different states of the cable insulation layer (including various defect types and normal states). When the harmonic feature data of the cable to be tested is input, the defect type of the cable can be effectively identified.

[0128] The self-organizing mapping grid constructed by the embodiment of the present application can provide a more accurate and effective mapping structure for the detection of cable insulation layer defect types, thereby improving the overall performance of cable insulation layer defect detection.

[0129] In an exemplary embodiment, based on the above embodiment, see Figure 6 The sample feature data of the embodiment of the present application includes a plurality of sample vectors, the intermediate mapping network includes a plurality of initial neurons, and the intermediate mapping network is trained according to the sample feature data to obtain a self-organizing mapping grid, including the following steps:

[0130] S601, in each iteration process, for each sample vector, the distance between the sample vector and each initial neuron is calculated, and the initial neuron with the smallest distance is determined as the intermediate neuron.

[0131] Sample vector: Sample feature data is composed of multiple sample vectors. In the context of cable insulation defect type detection, a sample vector is a set of feature values ​​extracted from a cable with known cable insulation state (including various defect types and normal state). For example, a sample vector may contain harmonic feature data of a cable sample, such as harmonic amplitude and phase at a specific frequency, and may also contain other feature values ​​related to the insulation state. These feature values ​​together constitute a vector, which is used to represent the state information of the cable sample during the training process.

[0132] In the embodiment of the present application, in the training process of the self-organizing mapping grid, iteration is an operation that is repeated multiple times. For each iteration, for each sample vector in the sample feature data, it is necessary to calculate its distance from each initial neuron in the intermediate mapping network. This distance can be calculated using a variety of methods, such as the Euclidean distance, as shown in the above formula (2). This method can determine which initial neuron best matches the sample vector in the current iteration.

[0133] S602, obtaining the domain neurons within the preset neighborhood range of the intermediate neuron.

[0134] Among them, domain neurons: after determining the intermediate neurons, the neurons within the preset neighborhood range of the intermediate neurons are obtained as domain neurons. The concept of domain neurons is based on the topological structure relationship between neurons in the self-organizing map model. In this model, neurons are not isolated, but interconnected, and the neurons around the intermediate neurons are considered to have a certain correlation with the intermediate neurons. This correlation is reflected in the training process, and their weights will be affected by the update of the intermediate neuron weights. For example, in a two-dimensional intermediate mapping network, if the preset neighborhood range of the intermediate neuron is the neurons in the nine-square grid area centered on it, then these neurons are the domain neurons of the intermediate neuron.

[0135] In an embodiment of the present application, after determining the intermediate neuron, it is necessary to obtain the domain neurons within its preset neighborhood range. The preset neighborhood range is a pre-set area, and the definition of this area can be determined according to the structure of the intermediate mapping network and actual needs. For example, in a two-dimensional intermediate mapping network, if the coordinates of the intermediate neuron are (i, j), the preset neighborhood range can be a circular area with a radius r centered on (i, j) or a square area with a side length of neurons. The purpose of obtaining domain neurons is because in the self-organizing mapping model, the weights of the intermediate neuron and its neighborhood neurons will be adjusted at the same time to reflect the correlation between neurons and the ability to coordinate and adapt to sample vectors.

[0136] S603, respectively updating the first weight of the intermediate neuron and the second weight of the domain neuron to obtain an updated intermediate mapping network.

[0137] First weight and second weight: Each neuron (including intermediate neurons and domain neurons) has its weight. The first weight refers to the weight of the intermediate neuron, and the second weight refers to the weight of the domain neuron. These weights are constantly updated during the training process so that the neurons can better classify the sample feature data. The weight is a vector containing multiple numerical values ​​(related to the dimension of the sample vector), which determines the degree of response of the neuron to the input sample vector. For example, the weight may be a randomly initialized numerical vector at the beginning. As the training process progresses, the weight will gradually adjust so that the neuron responds more accurately to different types of sample vectors (representing different cable insulation states).

[0138] In an embodiment of the present application, after determining the intermediate neuron and its domain neuron, their weights are updated. For the first weight of the intermediate neuron and the second weight of the domain neuron, the update is based on a certain learning rule. Through such an update process, the intermediate neuron and the domain neuron can better adapt to the state characteristics of the cable insulation layer represented by the sample vector, thereby obtaining an updated intermediate mapping network, which is structurally the same as before, but the weights of the neurons have changed, which is closer to the accurate classification of the sample feature data.

[0139] S604: When the number of iterations meets a preset condition, the updated intermediate mapping network is determined as a self-organizing mapping grid.

[0140] In the training process of the embodiment of the present application, a condition of the number of iterations is pre-set, for example, the maximum number of iterations is set to . As the iterations proceed, the intermediate mapping network is continuously updated. When the number of iterations reaches this preset condition (such as reaching the maximum number of iterations N), it is considered that the network has converged to a relatively stable state, and the updated intermediate mapping network is determined as a self-organizing mapping grid. This self-organizing mapping grid can be used to detect the defect type of the cable insulation layer, because it has been trained with a large amount of sample feature data and has the ability to classify different cable insulation layer states.

[0141] The embodiment of the present application constructs an accurate and effective self-organizing map grid to improve the accuracy and reliability of cable insulation layer defect detection.

[0142] In an exemplary embodiment, based on the above embodiment, see Figure 7 The embodiment of the present application is to update the first weight of the intermediate neuron and the second weight of the domain neuron respectively to obtain an updated intermediate mapping network, including the following steps:

[0143] S701, input the first weight and the second weight into the update model to perform weight calculation processing to obtain an update vector result.

[0144] Among them, the update model: This is a model used to calculate the first weight of the intermediate neuron and the second weight of the field neuron. It contains specific algorithms and rules for calculating according to the input weights (first weight and second weight) and other related information (such as sample vectors, etc.) to obtain the updated weight information (i.e., the update vector result).

[0145] Update vector result: This is the result obtained after the first weight and the second weight are input into the update model for weight calculation. The update vector result is a vector (or a set of vectors) containing the updated intermediate neuron weights and domain neuron weights. This result reflects how the neuron weights should be adjusted after considering the sample vector and the current weight state to better adapt to the cable insulation state characteristics represented by the sample feature data.

[0146] Update weight vector: Update the weight vector of the winner neuron and its neighborhood to make it closer to the current input sample The updated model is shown in formula (3):

[0147] (3)

[0148] in, is the learning rate, which decreases with the number of iterations t, It is a neighborhood function, which indicates the degree of neighborhood influence of neuron j on the winner neuron c.

[0149] S702, determining an updated intermediate mapping network according to the update vector result.

[0150] Once the update vector result is obtained in the embodiment of the present application, the updated intermediate mapping network can be determined based on this result. Specifically, the weights of the intermediate neurons and the domain neurons are updated according to the update vector result, thereby changing the properties of these neurons in the intermediate mapping network. Since the intermediate mapping network is composed of multiple neurons (including intermediate neurons and domain neurons), when the weights of these neurons are updated, the structure of the entire intermediate mapping network remains unchanged, but the responsiveness to the input sample feature data has changed. For example, if the update vector result increases the weight of certain neurons for specific sample features, then in subsequent processing, when sample vectors related to these features are input, these neurons will play a greater role, so that the intermediate mapping network can more accurately classify the types of cable insulation defects.

[0151] The embodiment of the present application can effectively update the weights of intermediate neurons and domain neurons to obtain an updated intermediate mapping network with better performance, laying the foundation for building an accurate self-organizing mapping grid, thereby improving the accuracy and reliability of cable insulation layer defect type detection.

[0152] In an exemplary embodiment, based on the above embodiment, see Figure 8 The update vector result of the embodiment of the present application includes a first weight update result and a second weight update result; the first weight and the second weight are input into the update model for weight calculation processing to obtain the update vector result, including the following steps:

[0153] S801, obtain a random sample vector, a neighborhood function value, and a target learning rate.

[0154] Among them, random sample vector: This is a sample vector randomly selected from the sample feature data. During the training process of the self-organizing map grid, the random sample vector is used as a reference for calculating the weight update of the intermediate neuron and the domain neuron. Since the sample feature data contains sample vectors of multiple known cable insulation states (including various defect types and normal states), randomly selecting a sample vector can make the weight update process more random and representative, avoiding over-reliance on specific sample vectors, so that the intermediate mapping network can better adapt to various possible cable insulation states. For example, a random sample vector may contain harmonic feature data of a cable sample, such as amplitude, phase and other information at different harmonic frequencies, as well as other characteristic values ​​related to the insulation state.

[0155] Neighborhood function value: Neighborhood function is an important concept in the self-organizing map model. The neighborhood function value indicates the degree of relationship between the intermediate neuron and its domain neuron. It is a value calculated based on the positional relationship between the intermediate neuron and the domain neuron in the intermediate mapping network. This value will affect the magnitude of the domain neuron weight update and reflects the role of the topological structure relationship between neurons in the weight update process. For example, in a two-dimensional intermediate mapping network, if the intermediate neuron is close to a domain neuron, the corresponding neighborhood function value may be large, which means that when the weight is updated, the weight of the domain neuron will be greatly affected by the intermediate neuron weight update.

[0156] Target learning rate: The learning rate is an important parameter that controls the step size of weight adjustment during the weight update process. The target learning rate is a pre-set appropriate learning rate value, which determines the speed at which intermediate neurons and domain neurons update weights according to random sample vectors. If the target learning rate is set to a larger value, the step size of weight update will be larger, which may lead to rapid approach to the target during training, but may also miss the optimal solution; if the target learning rate is set to a smaller value, the step size of weight update will be smaller, and the training process will be more stable, but it may take more iterations to converge. For example, the target learning rate can be determined based on experience or through some preliminary experiments to balance training speed and accuracy.

[0157] In an embodiment of the present application, at the beginning of the weight calculation process, a random sample vector, a neighborhood function value and a target learning rate must first be obtained. The random sample vector is a sample vector randomly extracted from existing sample feature data, and these sample feature data are extracted from cable samples of many known cable insulation layer states. The neighborhood function value is calculated based on the positional relationship between the intermediate neuron and the domain neuron in the intermediate mapping network. For example, it can be calculated based on the distance between them through a specific neighborhood function (such as a Gaussian function, etc.). The target learning rate is a pre-set parameter that is determined based on experience or preliminary experiments to ensure that the weights of the intermediate neurons and the domain neurons can be adjusted at an appropriate speed in subsequent weight update calculations.

[0158] S802, determining a first weight update result according to the first weight, the random sample vector, the neighborhood function value and the target learning rate.

[0159] In the embodiment of the present application, the first weight of the intermediate neuron is the current weight of the intermediate neuron in the intermediate mapping network. In this step, the first weight, the random sample vector, the neighborhood function value and the target learning rate are used to determine the first weight update result. The specific calculation method can be based on the weight update principle of the self-organizing map model.

[0160] S803, determining a second weight update result according to the second weight, the random sample vector, the neighborhood function value and the target learning rate.

[0161] In the embodiment of the present application, the second weight of the domain neuron is the current weight of the domain neuron in the intermediate mapping network. Similar to the above-mentioned S802, the second weight update result is determined according to the second weight, the random sample vector, the neighborhood function value and the target learning rate. The calculation method here reflects the similarity in principle between the weight update of the domain neuron and the weight update of the intermediate neuron, that is, the difference between the random sample vector and the current weight, the influence of the neighborhood function value and the control of the update step length by the target learning rate are all considered. Through such calculation, the second weight update result of the domain neuron is obtained, which shows how the weight of the domain neuron needs to be adjusted so that it can better adapt to the classification requirements of the state characteristics of the cable insulation layer in the intermediate mapping network.

[0162] The embodiment of the present application can accurately calculate the first weight update result and the second weight update result based on the random sample vector, the neighborhood function value and the target learning rate, thereby obtaining the update vector result, further optimizing the weight update process of the intermediate neurons and the domain neurons, and improving the accuracy and reliability of the cable insulation layer defect detection.

[0163] In an exemplary embodiment, based on the above embodiment, see Fig. 9 The process of determining the neighborhood function value in the embodiment of the present application includes the following steps:

[0164] S901, obtaining the neighborhood width, the first grid position of the intermediate neuron, and the second grid position of the domain neuron.

[0165] Among them, neighborhood width: This is a parameter used to control the size of the neighborhood range. In the self-organizing map model, the neighborhood width determines which domain neurons around the intermediate neuron will be most affected by it. The larger the value of the neighborhood width, the wider the range of domain neurons affected by the intermediate neuron; conversely, the smaller the value of the neighborhood width, the narrower the range of domain neurons affected.

[0166] First grid position: This is the position coordinate of the intermediate neuron in the intermediate mapping network. The intermediate mapping network can be regarded as a grid structure composed of multiple neurons. Each neuron has its specific position in this grid. This position is represented by coordinates as the first grid position.

[0167] Second grid position: This is the position coordinate of the domain neuron in the intermediate mapping network. Similar to the first grid position, the second grid position also uses coordinates to represent the position of the domain neuron in the intermediate mapping network.

[0168] Before calculating the neighborhood function value, the embodiment of the present application needs to obtain relevant parameters and position information. The neighborhood width is a pre-set parameter, which is determined according to the characteristics and requirements of the intermediate mapping network. The first grid position of the intermediate neuron and the second grid position of the domain neuron are determined by the structure of the intermediate mapping network. In actual operation, these position information can be obtained through an index or coordinate system. For example, when constructing an intermediate mapping network, a unique index or coordinate can be assigned to each neuron. When the neighborhood function value needs to be calculated, the index or coordinate information corresponding to the intermediate neuron and the domain neuron is directly obtained as the first grid position and the second grid position.

[0169] S902, inputting the neighborhood width, the first grid position and the second grid position into a Gaussian function for calculation to obtain a neighborhood function value.

[0170] Among them, the Gaussian function is a common mathematical function, which is used to calculate the neighborhood function value in the embodiment of the present application.

[0171] The embodiment of the present application is based on the above formula (3) is a neighborhood function, which indicates the neighborhood influence of neuron j on the winner neuron c. It is usually in the form of a Gaussian function, as shown in formula (4):

[0172] (4)

[0173] in and are the positions of neurons c and j in the grid, is the neighborhood width, which decreases over time.

[0174] The neighborhood function value obtained above reflects the degree of relationship between the domain neuron and the intermediate neuron. In subsequent weight update and other operations, the magnitude of the domain neuron weight update will be determined based on this neighborhood function value.

[0175] Repeat the above steps and iteratively update the weight vector until the preset convergence condition is reached (such as the weight change is less than a certain threshold or the maximum number of iterations is reached). After the model converges, each input sample corresponds to a neuron position in the self-organizing map grid. At this point, the self-organizing map network has formed a low-dimensional representation that reflects the topological structure of the input data.

[0176] In practical applications, the newly input harmonic data feature vector is input into the trained self-organizing map model, and the model automatically maps the data to a neuron position in the grid. Adjacent neurons represent similar defect types, so the defect type of the cable can be identified and classified according to the position of the neuron.

[0177] The embodiment of the present application can accurately calculate the neighborhood function value, thereby more reasonably determining the relationship between the intermediate neuron and the domain neuron, optimizing the weight updating process, and improving the accuracy and reliability of cable insulation layer defect detection.

[0178] In an exemplary embodiment, based on the above embodiment, see Fig.10 , the process of determining the target learning rate in the embodiment of the present application includes the following steps:

[0179] S1001, obtain the initial learning rate, the current number of iterations and the maximum number of iterations.

[0180] Initial Learning Rate: This is the learning rate set at the beginning of the training process for the Self-Organizing Map grid construction. The value of the initial learning rate determines the step size of the neuron weight update at the beginning of training.

[0181] Before determining the target learning rate, the embodiment of the present application needs to obtain the three key parameters of the initial learning rate, the current number of iterations and the maximum number of iterations. The initial learning rate is set before the start of training based on experience or preliminary analysis of the model and data. The current number of iterations is recorded in real time during the training process, and the current number of iterations is increased by 1 for each training iteration. The maximum number of iterations is a fixed value pre-set before the start of training based on factors such as sample feature data, model accuracy requirements and computing resources.

[0182] S1002, calculating the quotient of the current number of iterations and the maximum number of iterations.

[0183] The embodiment of the present application calculates the quotient of the current number of iterations and the maximum number of iterations, that is, the current number of iterations / the maximum number of iterations. This quotient reflects the progress ratio of the training process. For example, if the current number of iterations is 50 and the maximum number of iterations is 100, then the quotient is 0.5, indicating that the training has been carried out to half of the process.

[0184] S1003, determining a target learning rate according to the quotient value and the initial learning rate.

[0185] Target learning rate: This is a learning rate that is dynamically determined based on the relationship between the current number of iterations and the maximum number of iterations combined with the initial learning rate. The target learning rate has different values ​​at different stages of the training process, and it is dynamically adjusted as the training progresses to meet the training needs at different stages.

[0186] The learning rate is usually gradually reduced during the training process to ensure that the model gradually converges to a stable state. The target learning rate is shown in formula (5):

[0187] (5)

[0188] in is the initial learning rate, t is the current number of iterations, and T is the maximum number of iterations.

[0189] The embodiment of the present application can dynamically determine the target learning rate according to the training process, improve the efficiency and accuracy of model training, and thus improve the accuracy and reliability of cable insulation layer defect detection.

[0190] In an exemplary embodiment, based on the above embodiment, see Fig.11 , the process of determining the harmonic characteristic data in the embodiment of the present application includes:

[0191] During operation, the cable is affected by defects such as aging and moisture, which will lead to the enhancement of the nonlinear characteristics of the current signal, which is manifested as changes in high-order harmonic components. Therefore, harmonic data is an important information carrier for identifying defects in the cable insulation layer. The process of obtaining harmonic data includes the following steps:

[0192] S1101, obtaining a current signal of the cable to be tested.

[0193] In the embodiment of the present application, the current signal of the cable to be tested is obtained by a suitable current sensor. The current sensor should have sufficient accuracy and bandwidth to accurately measure the current in the cable. For example, a Hall effect current sensor can be used, which can convert the current signal into a measurable voltage signal based on the Hall effect principle, and then convert it into a current signal in the form of a digital signal suitable for subsequent processing through circuits such as amplification and filtering.

[0194] S1102, performing harmonic component extraction processing on the current signal to obtain harmonic component data.

[0195] Use appropriate harmonic component extraction methods to process the acquired current signal. For example, use discrete Fourier transform (DFT) or its fast algorithm fast Fourier transform (FFT) to convert the current signal from the time domain to the frequency domain, so as to obtain information such as the component amplitude and phase at each harmonic frequency to form harmonic component data. In this process, it is necessary to determine the appropriate sampling frequency and number of sampling points to ensure that the required harmonic components can be accurately extracted. For example, according to the Nyquist sampling theorem, the sampling frequency should be at least twice the highest frequency of the signal to avoid aliasing. Typically, harmonic analysis involves 1st (fundamental), 3rd, 5th, 7th, and other harmonic components.

[0196] S1103, preprocessing the harmonic component data to obtain harmonic characteristic data.

[0197] The data preprocessing includes at least one of normalization processing, noise reduction processing and feature extraction processing.

[0198] Specifically, the collected harmonic data is converted into a format suitable for algorithm processing, usually in the form of a multidimensional vector, represented as , where each component Represents a specific harmonic amplitude or phase.

[0199] Data labeling: Prepare labeled data for the subsequent training process, that is, label the harmonic data of known defect types. These data will be input into the SOM model as training data sets.

[0200] Data preprocessing: It is crucial to preprocess the harmonic data before formally inputting it into the SOM model. The preprocessing steps include:

[0201] Normalization: Normalize all harmonic components and standardize the data so that they fall within a uniform range (such as [0, 1] or [-1, 1]). The normalization formula is formula (6):

[0202] (6)

[0203] in, is the original data, is the normalized data, and are the minimum and maximum values ​​of the dataset in the jth dimension, respectively.

[0204] Noise reduction: Use methods such as wavelet transform or fast Fourier transform (FFT) to denoise the harmonic data, reduce the impact of external interference signals, and retain the harmonic features that are critical to defect type identification.

[0205] Feature extraction: Extract effective features from harmonic data, such as the amplitude, phase, harmonic distortion (THD) of each harmonic, and form a high-dimensional feature vector. For example, for a data set containing n harmonic components, the feature vector can be expressed as formula (7):

[0206] (7)

[0207] Among them, m is the number of samples and n is the feature dimension of each sample.

[0208] Wavelet transform processing: According to the characteristics of the harmonic component data and the type of noise to be removed, select the appropriate wavelet basis function. Use the selected wavelet basis function to perform wavelet decomposition and reconstruction on the harmonic component data, remove noise or extract local features, and obtain the processed harmonic component data.

[0209] Fast Fourier transform processing (optional): If you need to further optimize the frequency domain representation, you can perform fast Fourier transform processing again. According to the fast Fourier transform algorithm flow, the previously processed harmonic component data is processed to improve the efficiency and accuracy of data processing.

[0210] The embodiments of the present application can accurately determine harmonic characteristic data, provide a reliable data basis for cable insulation layer defect detection, and thus improve the accuracy and reliability of cable insulation layer defect detection.

[0211] Furthermore, based on the target defect identification results of the SOM model, a detailed cable defect detection report can be generated. The report contains the following contents:

[0212] (1) Defect type identification: Clearly identify the specific defect type of the cable to be tested and quantify the probability of each defect.

[0213] (2) Visual output: Using the visualization capability of the SOM model, the detection results are presented in a graphical manner to help users intuitively understand the defect distribution of the cable.

[0214] (3) Maintenance recommendations: Based on the test results, corresponding maintenance recommendations are provided, such as replacing cables and regular monitoring, to ensure the safe operation of the power system.

[0215] During the experimental verification process, based on the actual operating power system environment, the harmonic current components of damp cables and normal cables when experimental currents of different amplitudes were applied were collected, and verified based on the clustering method proposed in this application. The experimental objects include two types of cable samples: one is normal cables that are not damp or damaged, and the other is damp cables whose insulation is damaged due to excessive environmental humidity or water immersion. In order to ensure the accuracy of the experiment, we used a harmonic current acquisition device to collect multiple times the harmonic responses of the cable after applying currents of different amplitudes, covering various working conditions that the cable may encounter. The collected data was input into the clustering analysis platform described in this patent, and the cable harmonic data was classified and identified through the self-organizing mapping model. The experimental results obtained are as follows. Fig.12a , 12b , 12c, and 12d.

[0216] In an exemplary embodiment, based on the above embodiment, the method of the embodiment of the present application further includes the following steps:

[0217] Step 1: Determine the basic structure of the initial mapping grid; perform weight vector initialization processing on the initial mapping grid according to the pre-acquired sample feature data to obtain an intermediate mapping network;

[0218] Step 2: In each iteration process, for each sample vector in the sample feature data, the distance between the sample vector and each initial neuron in the initial mapping grid is calculated, and the initial neuron with the smallest distance is determined as the intermediate neuron;

[0219] Step 3: Obtaining the domain neurons within the preset neighborhood of the intermediate neuron;

[0220] Step 4: Obtain the neighborhood width, the first grid position of the intermediate neuron, and the second grid position of the domain neuron; input the neighborhood width, the first grid position, and the second grid position into the Gaussian function for calculation to obtain the neighborhood function value;

[0221] Step 5: Obtain the initial learning rate, the current number of iterations, and the maximum number of iterations; calculate the quotient of the current number of iterations and the maximum number of iterations; determine the target learning rate based on the quotient and the initial learning rate;

[0222] Step 6: Get the random sample vector, neighborhood function value and target learning rate;

[0223] Step 7: Determine the first weight update result according to the first weight, the random sample vector, the neighborhood function value and the target learning rate; determine the second weight update result according to the second weight, the random sample vector, the neighborhood function value and the target learning rate;

[0224] Step 8: determining an updated intermediate mapping network according to the update vector result; and determining the updated intermediate mapping network as a self-organizing mapping grid when the number of iterations meets a preset condition;

[0225] Step 9: Acquire the current signal of the cable to be tested; perform harmonic component extraction processing on the current signal to obtain harmonic component data; perform data preprocessing on the harmonic component data to obtain harmonic characteristic data;

[0226] Step 10: Calculate the distance between the harmonic characteristic data and each target neuron, and determine the target neuron with the smallest distance as the winner neuron; map the harmonic characteristic data to the position corresponding to the winner neuron to obtain the mapping result; determine the target defect type of the cable to be tested according to the mapping result.

[0227] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0228] Based on the same inventive concept, the embodiment of the present application also provides a cable insulation layer defect type detection device for implementing the cable insulation layer defect type detection method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more cable insulation layer defect type detection device embodiments provided below can refer to the limitations of the cable insulation layer defect type detection method above, and will not be repeated here.

[0229] In one embodiment, Fig.13 As shown, the present application also provides a cable insulation layer defect type detection device 1200, comprising:

[0230] The data acquisition module 1201 is used to acquire the harmonic characteristic data of the cable to be tested;

[0231] The type determination module 1202 is used to input the harmonic feature data into a pre-built self-organizing map model for cable defect identification processing to obtain the target defect type of the cable to be tested; the self-organizing map model is built based on the self-organizing map grid, and the self-organizing map grid corresponds to the data category.

[0232] In one embodiment, the self-organizing map grid includes a plurality of target neurons, and the type determination module includes:

[0233] A mapping unit, used for mapping the harmonic feature data and each target neuron based on a self-organizing mapping model to obtain a mapping result;

[0234] The type determination unit is used to determine the target defect type of the cable to be tested according to the mapping result.

[0235] In one embodiment, the mapping unit includes:

[0236] A neuron determination subunit is used to calculate the distance between the harmonic feature data and each target neuron, and determine the target neuron with the smallest distance as the winner neuron;

[0237] The mapping subunit is used to map the harmonic feature data to the position corresponding to the winner neuron to obtain a mapping result; the dimension of the weight vector of the winner neuron is smaller than the dimension of the harmonic feature data.

[0238] In one embodiment, the above device further comprises:

[0239] A structure determination module, used for determining a basic structure of an initial mapping grid; the basic structure includes one of a two-dimensional rectangular grid and a hexagonal grid;

[0240] An initialization module is used to perform weight vector initialization processing on the initial mapping grid according to the sample feature data obtained in advance to obtain an intermediate mapping network;

[0241] The training module is used to train the intermediate mapping network according to the sample feature data to obtain the self-organizing mapping grid.

[0242] In one embodiment, the sample feature data includes a plurality of sample vectors, the intermediate mapping network includes a plurality of initial neurons, and the training module includes:

[0243] The intermediate neuron determination unit is used to calculate the distance between the sample vector and each initial neuron for each sample vector in each iteration process, and determine the initial neuron with the smallest distance as the intermediate neuron;

[0244] A domain neuron determination unit, used for obtaining domain neurons within a preset neighborhood range of the intermediate neuron;

[0245] A network determination unit, used to update the first weight of the intermediate neuron and the second weight of the domain neuron respectively to obtain an updated intermediate mapping network;

[0246] The target network determination unit is used to determine the updated intermediate mapping network as a self-organizing mapping grid when the number of iterations meets a preset condition.

[0247] In one embodiment, the network determination unit includes:

[0248] A weight calculation subunit, used for inputting the first weight and the second weight into the update model to perform weight calculation processing to obtain an update vector result;

[0249] The network updating subunit is used to determine the updated intermediate mapping network according to the update vector result.

[0250] In one of the embodiments, the above-mentioned update vector result includes a first weight update result and a second weight update result; the weight calculation subunit is specifically used to: obtain a random sample vector, a neighborhood function value and a target learning rate; determine the first weight update result according to the first weight, the random sample vector, the neighborhood function value and the target learning rate; determine the second weight update result according to the second weight, the random sample vector, the neighborhood function value and the target learning rate.

[0251] In one embodiment, the above device further comprises:

[0252] a position acquisition unit, used to acquire the neighborhood width, the first grid position of the intermediate neuron and the second grid position of the domain neuron;

[0253] The function value determination unit is used to input the neighborhood width, the first grid position and the second grid position into a Gaussian function for calculation to obtain a neighborhood function value.

[0254] In one embodiment, the above device further comprises:

[0255] The acquisition unit is used to obtain the initial learning rate, the current number of iterations and the maximum number of iterations;

[0256] A quotient value determination unit, used for calculating a quotient value of a current number of iterations and a maximum number of iterations;

[0257] The learning rate determination unit is used to determine a target learning rate according to the quotient value and the initial learning rate.

[0258] In one embodiment, the above device further comprises:

[0259] A signal acquisition unit, used for acquiring a current signal of the cable to be tested;

[0260] A component extraction unit, used for performing harmonic component extraction processing on the current signal to obtain harmonic component data;

[0261] The data processing module is used to perform data preprocessing on the harmonic component data to obtain harmonic feature data; wherein the data preprocessing includes at least one of normalization processing, noise reduction processing and feature extraction processing.

[0262] Each module in the above-mentioned cable insulation layer defect type detection device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0263] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.14 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for detecting the type of defects in the cable insulation layer is implemented.

[0264] Those skilled in the art will understand that Fig.14 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0265] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0266] Obtain harmonic characteristic data of the cable to be tested;

[0267] The harmonic feature data is input into a pre-built self-organizing map model for cable defect recognition processing to obtain the target defect type of the cable to be tested; the self-organizing map model is built based on a self-organizing map grid, and the self-organizing map grid corresponds to the data category.

[0268] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0269] Based on the self-organizing map model, the harmonic feature data and each target neuron are mapped to obtain the mapping result;

[0270] Determine the target defect type of the cable to be tested based on the mapping results.

[0271] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0272] Calculate the distance between the harmonic feature data and each target neuron, and determine the target neuron with the smallest distance as the winner neuron;

[0273] The harmonic feature data is mapped to the position corresponding to the winner neuron to obtain a mapping result; the dimension of the weight vector of the winner neuron is smaller than the dimension of the harmonic feature data.

[0274] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0275] Determine a basic structure of an initial mapping grid; the basic structure includes one of a two-dimensional rectangular grid and a hexagonal grid;

[0276] Initializing the weight vector of the initial mapping grid according to the sample feature data acquired in advance to obtain an intermediate mapping network;

[0277] The intermediate mapping network is trained according to the sample feature data to obtain a self-organizing mapping grid.

[0278] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0279] In each iteration, for each sample vector, the distance between the sample vector and each initial neuron is calculated, and the initial neuron with the smallest distance is determined as the intermediate neuron;

[0280] Obtaining domain neurons within a preset neighborhood range of the interneuron;

[0281] The first weight of the intermediate neuron and the second weight of the domain neuron are updated respectively to obtain an updated intermediate mapping network;

[0282] When the number of iterations meets a preset condition, the updated intermediate mapping network is determined as a self-organizing mapping grid.

[0283] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0284] Inputting the first weight and the second weight into the update model to perform weight calculation processing to obtain an update vector result;

[0285] The updated intermediate mapping network is determined according to the update vector result.

[0286] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0287] Get random sample vectors, neighborhood function values, and target learning rates;

[0288] Determine a first weight update result according to the first weight, the random sample vector, the neighborhood function value, and the target learning rate;

[0289] A second weight update result is determined according to the second weight, the random sample vector, the neighborhood function value, and the target learning rate.

[0290] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0291] Obtaining the neighborhood width, the first grid position of the interneuron, and the second grid position of the field neuron;

[0292] The neighborhood width, the first grid position, and the second grid position are input into a Gaussian function for calculation to obtain a neighborhood function value.

[0293] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0294] Get the initial learning rate, current number of iterations, and maximum number of iterations;

[0295] Calculate the quotient of the current number of iterations and the maximum number of iterations;

[0296] According to the quotient value and the initial learning rate, the target learning rate is determined.

[0297] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0298] Obtain the current signal of the cable to be tested;

[0299] Performing harmonic component extraction processing on the current signal to obtain harmonic component data;

[0300] The harmonic component data is preprocessed to obtain harmonic feature data; wherein the data preprocessing includes at least one of normalization processing, noise reduction processing and feature extraction processing.

[0301] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0302] Obtain harmonic characteristic data of the cable to be tested;

[0303] The harmonic feature data is input into a pre-built self-organizing map model for cable defect recognition processing to obtain the target defect type of the cable to be tested; the self-organizing map model is built based on a self-organizing map grid, and the self-organizing map grid corresponds to the data category.

[0304] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0305] Based on the self-organizing map model, the harmonic feature data and each target neuron are mapped to obtain the mapping result;

[0306] Determine the target defect type of the cable to be tested based on the mapping results.

[0307] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0308] Calculate the distance between the harmonic feature data and each target neuron, and determine the target neuron with the smallest distance as the winner neuron;

[0309] The harmonic feature data is mapped to the position corresponding to the winner neuron to obtain a mapping result; the dimension of the weight vector of the winner neuron is smaller than the dimension of the harmonic feature data.

[0310] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0311] Determine a basic structure of an initial mapping grid; the basic structure includes one of a two-dimensional rectangular grid and a hexagonal grid;

[0312] Initializing the weight vector of the initial mapping grid according to the sample feature data acquired in advance to obtain an intermediate mapping network;

[0313] The intermediate mapping network is trained according to the sample feature data to obtain a self-organizing mapping grid.

[0314] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0315] In each iteration, for each sample vector, the distance between the sample vector and each initial neuron is calculated, and the initial neuron with the smallest distance is determined as the intermediate neuron;

[0316] Obtaining domain neurons within a preset neighborhood range of the interneuron;

[0317] The first weight of the intermediate neuron and the second weight of the domain neuron are updated respectively to obtain an updated intermediate mapping network;

[0318] When the number of iterations meets a preset condition, the updated intermediate mapping network is determined as a self-organizing mapping grid.

[0319] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0320] Inputting the first weight and the second weight into the update model to perform weight calculation processing to obtain an update vector result;

[0321] The updated intermediate mapping network is determined according to the update vector result.

[0322] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0323] Get random sample vectors, neighborhood function values, and target learning rates;

[0324] Determine a first weight update result according to the first weight, the random sample vector, the neighborhood function value, and the target learning rate;

[0325] A second weight update result is determined according to the second weight, the random sample vector, the neighborhood function value, and the target learning rate.

[0326] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0327] Obtaining the neighborhood width, the first grid position of the interneuron, and the second grid position of the field neuron;

[0328] The neighborhood width, the first grid position, and the second grid position are input into a Gaussian function for calculation to obtain a neighborhood function value.

[0329] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0330] Get the initial learning rate, current number of iterations, and maximum number of iterations;

[0331] Calculate the quotient of the current number of iterations and the maximum number of iterations;

[0332] According to the quotient value and the initial learning rate, the target learning rate is determined.

[0333] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0334] Obtain the current signal of the cable to be tested;

[0335] Performing harmonic component extraction processing on the current signal to obtain harmonic component data;

[0336] The harmonic component data is preprocessed to obtain harmonic feature data; wherein the data preprocessing includes at least one of normalization processing, noise reduction processing and feature extraction processing.

[0337] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0338] Obtain harmonic characteristic data of the cable to be tested;

[0339] The harmonic feature data is input into a pre-built self-organizing map model for cable defect recognition processing to obtain the target defect type of the cable to be tested; the self-organizing map model is built based on a self-organizing map grid, and the self-organizing map grid corresponds to the data category.

[0340] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0341] Based on the self-organizing map model, the harmonic feature data and each target neuron are mapped to obtain the mapping result;

[0342] Determine the target defect type of the cable to be tested based on the mapping results.

[0343] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0344] Calculate the distance between the harmonic feature data and each target neuron, and determine the target neuron with the smallest distance as the winner neuron;

[0345] The harmonic feature data is mapped to the position corresponding to the winner neuron to obtain a mapping result; the dimension of the weight vector of the winner neuron is smaller than the dimension of the harmonic feature data.

[0346] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0347] Determine a basic structure of an initial mapping grid; the basic structure includes one of a two-dimensional rectangular grid and a hexagonal grid;

[0348] Initializing the weight vector of the initial mapping grid according to the sample feature data acquired in advance to obtain an intermediate mapping network;

[0349] The intermediate mapping network is trained according to the sample feature data to obtain a self-organizing mapping grid.

[0350] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0351] In each iteration, for each sample vector, the distance between the sample vector and each initial neuron is calculated, and the initial neuron with the smallest distance is determined as the intermediate neuron;

[0352] Obtaining domain neurons within a preset neighborhood range of the interneuron;

[0353] The first weight of the intermediate neuron and the second weight of the domain neuron are updated respectively to obtain an updated intermediate mapping network;

[0354] When the number of iterations meets a preset condition, the updated intermediate mapping network is determined as a self-organizing mapping grid.

[0355] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0356] Inputting the first weight and the second weight into the update model to perform weight calculation processing to obtain an update vector result;

[0357] The updated intermediate mapping network is determined according to the update vector result.

[0358] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0359] Get random sample vectors, neighborhood function values, and target learning rates;

[0360] Determine a first weight update result according to the first weight, the random sample vector, the neighborhood function value, and the target learning rate;

[0361] A second weight update result is determined according to the second weight, the random sample vector, the neighborhood function value, and the target learning rate.

[0362] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0363] Obtaining the neighborhood width, the first grid position of the interneuron, and the second grid position of the field neuron;

[0364] The neighborhood width, the first grid position, and the second grid position are input into a Gaussian function for calculation to obtain a neighborhood function value.

[0365] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0366] Get the initial learning rate, current number of iterations, and maximum number of iterations;

[0367] Calculate the quotient of the current number of iterations and the maximum number of iterations;

[0368] According to the quotient value and the initial learning rate, the target learning rate is determined.

[0369] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0370] Obtain the current signal of the cable to be tested;

[0371] Performing harmonic component extraction processing on the current signal to obtain harmonic component data;

[0372] The harmonic component data is preprocessed to obtain harmonic feature data; wherein the data preprocessing includes at least one of normalization processing, noise reduction processing and feature extraction processing.

[0373] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0374] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0375] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0376] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for detecting cable insulation defect types, It is characterized in that The method comprises: Obtain harmonic characteristic data of the cable to be tested; The harmonic feature data is input into a pre-built self-organizing map model for cable defect recognition processing to obtain the target defect type of the cable to be tested; the self-organizing map model is built based on a self-organizing map grid, and the self-organizing map grid corresponds to the data category.

2. The method according to claim 1, It is characterized in that The self-organizing map grid includes a plurality of target neurons, and the harmonic feature data is input into a pre-built self-organizing map model for cable defect recognition processing to obtain the target defect type of the cable to be tested, including: Based on the self-organizing map model, mapping processing is performed on the harmonic feature data and each of the target neurons to obtain a mapping result; The target defect type of the cable to be tested is determined according to the mapping result.

3. The method according to claim 2, It is characterized in that The mapping process is performed on the harmonic feature data and each of the target neurons to obtain a mapping result, including: Calculating the distance between the harmonic feature data and each of the target neurons, and determining the target neuron with the smallest distance as the winner neuron; The harmonic feature data is mapped to the position corresponding to the winner neuron to obtain the mapping result; the dimension of the weight vector of the winner neuron is smaller than the dimension of the harmonic feature data.

4. The method according to claim 1, It is characterized in that The construction process of the self-organizing map grid includes: Determine a basic structure of an initial mapping grid; the basic structure comprises one of a two-dimensional rectangular grid and a hexagonal grid; Performing weight vector initialization processing on the initial mapping grid according to the sample feature data acquired in advance to obtain an intermediate mapping network; The intermediate mapping network is trained according to the sample feature data to obtain the self-organizing mapping grid.

5. The method according to claim 4, It is characterized in that The sample feature data includes a plurality of sample vectors, the intermediate mapping network includes a plurality of initial neurons, and the training process is performed on the intermediate mapping network according to the sample feature data to obtain the self-organizing mapping grid, including: In each iteration process, for each sample vector, the distance between the sample vector and each of the initial neurons is calculated, and the initial neuron with the smallest distance is determined as the intermediate neuron; Acquire domain neurons within a preset neighborhood range of the intermediate neuron; Respectively updating the first weight of the intermediate neuron and the second weight of the domain neuron to obtain an updated intermediate mapping network; When the number of iterations meets a preset condition, the updated intermediate mapping network is determined as the self-organizing mapping grid.

6. The method according to claim 5, It is characterized in that The updating process of the first weight of the intermediate neuron and the second weight of the domain neuron to obtain an updated intermediate mapping network comprises: Inputting the first weight and the second weight into an update model to perform weight calculation processing to obtain an update vector result; The updated intermediate mapping network is determined according to the update vector result.

7. The method according to claim 6, It is characterized in that The update vector result includes a first weight update result and a second weight update result; the first weight and the second weight are input into the update model for weight calculation processing to obtain the update vector result, including: Get random sample vectors, neighborhood function values, and target learning rates; Determine the first weight update result according to the first weight, the random sample vector, the neighborhood function value and the target learning rate; The second weight update result is determined according to the second weight, the random sample vector, the neighborhood function value and the target learning rate.

8. The method according to claim 7, It is characterized in that The process of determining the neighborhood function value includes: Obtaining a neighborhood width, a first grid position of the intermediate neuron, and a second grid position of the field neuron; The neighborhood width, the first grid position, and the second grid position are input into a Gaussian function for calculation to obtain the neighborhood function value.

9. The method according to claim 7, It is characterized in that The process of determining the target learning rate includes: Get the initial learning rate, current number of iterations, and maximum number of iterations; Calculating a quotient of the current number of iterations and the maximum number of iterations; The target learning rate is determined according to the quotient and the initial learning rate.

10. The method according to any one of claims 1 to 9, It is characterized in that The process of determining the harmonic characteristic data includes: Acquiring a current signal of the cable to be tested; Performing harmonic component extraction processing on the current signal to obtain harmonic component data; The harmonic component data is preprocessed to obtain the harmonic feature data; wherein the data preprocessing includes at least one of normalization processing, noise reduction processing and feature extraction processing.

Images