A cable early defect identification method, device, equipment and storage medium

Abstract

This invention discloses a method, apparatus, device, and storage medium for early cable defect identification. The method includes: acquiring a fault-state scattering matrix of a distribution network cable line under fault conditions and a reference scattering matrix under fault-free conditions; constructing a spectral response feature dataset based on the fault-state scattering matrix and the reference scattering matrix; training a pre-constructed Fourier step network model based on the spectral response feature dataset to obtain a trained Fourier step network model; and inputting the cable data to be tested into the trained Fourier step network model to obtain the cable defect identification result. This method, by training a Fourier step network model using the constructed spectral response feature dataset, can accurately identify early defects in cable lines.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of power distribution network operation and maintenance technology, and in particular to a method, device, equipment and storage medium for early defect identification of cables. Background Technology

[0002] As a critical energy transmission carrier connecting substations and users, the health of distribution network cables directly affects the safe and stable operation of the distribution network. However, cables are susceptible to latent defects caused by factors such as insulation aging and mechanical damage during long-term operation. Statistics show that cable faults caused by this factor account for more than 60% of distribution network faults. These defects are often accompanied by weak partial discharges or impedance changes, exhibiting characteristics such as high concealment and complex evolution processes. Once latent defects develop into permanent faults, they will lead to prolonged power outages and significant economic losses. Therefore, achieving early and accurate identification of latent cable defects is crucial.

[0003] Feature extraction is a core step in cable defect diagnosis and analysis. Time-domain feature extraction is the most direct and simple, but it requires the selection of appropriate feature quantities based on prior knowledge. Time-frequency domain feature extraction extracts signal features through various time-frequency analysis methods (such as Fourier transform and wavelet transform), and has good robustness when processing non-stationary signals. However, this method often relies on cumbersome time-frequency conversion, which can easily lead to information loss or computational redundancy.

[0004] In recent years, the development of deep learning technology has provided a new data-driven approach for fault diagnosis. However, when facing the specific problem of early-stage defects in distribution network cables, conventional deep learning methods still have the following limitations: on the one hand, conventional methods cannot directly utilize the global integral characteristics of the frequency domain for operator learning; on the other hand, in actual operation and maintenance, fault samples are scarce while normal monitoring data is massive, and existing supervised learning frameworks cannot effectively utilize the massive amount of unlabeled data. Therefore, in the diagnosis of early-stage defects in distribution network cables, how to achieve efficient feature learning in the frequency domain to accurately identify early-stage latent defects in cables under the condition of scarce actual operation and maintenance fault samples is an urgent problem to be solved. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and storage medium for identifying early-stage defects in cables, in order to solve the problem that existing technologies cannot accurately identify early-stage latent defects in cables.

[0006] According to one aspect of the present invention, a method for early defect identification in cables is provided, the method comprising: Obtain the fault-state scattering matrix of the distribution network cable line under fault conditions, and the reference scattering matrix under fault-free conditions; Based on the fault-state scattering matrix and the reference scattering matrix, a spectral response feature dataset is constructed; The pre-constructed Fourier ladder network model is trained based on the spectral response feature dataset to obtain the trained Fourier ladder network model. The cable data to be tested is input into the trained Fourier ladder network model to obtain the cable defect identification results.

[0007] According to another aspect of the present invention, a cable early defect identification device is provided, the device comprising: The acquisition module is used to acquire the fault-state scattering matrix of the distribution network cable line under fault conditions, and the reference scattering matrix under fault-free conditions. The construction module is used to construct a spectral response feature dataset based on the fault-state scattering matrix and the reference scattering matrix; The training module is used to train a pre-constructed Fourier ladder network model based on the spectral response feature dataset to obtain the trained Fourier ladder network model. The identification module is used to input the data of the cable under test into the trained Fourier ladder network model to obtain the cable defect identification result.

[0008] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: at least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the cable early defect identification method according to any embodiment of the present invention.

[0009] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the cable early defect identification method according to any embodiment of the present invention.

[0010] This invention discloses a method, apparatus, device, and storage medium for early cable defect identification. The method includes: acquiring a fault-state scattering matrix of a distribution network cable line under fault conditions and a reference scattering matrix under fault-free conditions; constructing a spectral response feature dataset based on the fault-state scattering matrix and the reference scattering matrix; training a pre-constructed Fourier step network model based on the spectral response feature dataset to obtain a trained Fourier step network model; and inputting the cable data to be tested into the trained Fourier step network model to obtain the cable defect identification result. This method, by training a Fourier step network model using the constructed spectral response feature dataset, can accurately identify early defects in cable lines, solving the problem in existing technologies that cannot accurately identify early latent defects in cables.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating an early defect identification method for cables provided in Embodiment 1 of the present invention. Figure 2 This is a schematic diagram of the structure of a Fourier ladder network model provided in an embodiment of the present invention; Figure 3 A flowchart illustrating an early defect identification method for cables provided in an embodiment of the present invention; Figure 4 A schematic diagram of a three-port distribution network of equal length provided in an embodiment of the present invention; Figure 5 A spatial spectrum of an AB circuit is provided as an embodiment of the present invention; Figure 6 A comparison chart of the prediction accuracy of a Fourier ladder network model and a regular convolutional neural network provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of an early cable defect identification device provided in Embodiment 2 of the present invention; Figure 8 This is a schematic diagram of the electronic device used in the cable early defect identification method according to an embodiment of the present invention. Detailed Implementation

[0014] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention. It should be understood that the various steps described in the method embodiments of the present invention can be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0015] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0016] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, any variations of the terms "comprising" and "having," etc., are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0017] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0018] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0019] Example 1 Figure 1This is a flowchart illustrating an early cable defect identification method provided in Embodiment 1 of the present invention. This method is applicable to identifying early defects in power distribution network cable lines. The method can be executed by an early cable defect identification device, which can be implemented by software and / or hardware and is generally integrated into an electronic device. In this embodiment, the electronic device includes, but is not limited to, devices such as computers.

[0020] like Figure 1 As shown, the cable early defect identification method provided in Embodiment 1 of the present invention includes the following steps: S110. Obtain the fault-state scattering matrix of the distribution network cable line under fault conditions, and the reference scattering matrix under fault-free conditions.

[0021] The fault-state scattering matrix can be a frequency domain scattering parameter matrix constructed based on data measured at observation ports of the distribution network cable line under fault conditions. The reference scattering matrix can be a frequency domain scattering parameter matrix constructed based on data measured at each observation port of the distribution network cable line under fault-free and normal operating conditions.

[0022] In this embodiment, the fault-state scattering matrix of the distribution network cable line under fault conditions and the reference scattering matrix under fault-free conditions can be obtained respectively.

[0023] In one embodiment, obtaining the fault-state scattering matrix of the distribution network cable line under fault conditions and the reference scattering matrix under fault-free conditions includes: for a distribution network cable network with multiple observation ports, when the cable network is in a fault state, acquiring test data through multi-port impulse response testing and reconstructing the fault-state scattering matrix from the test data; simulating the fault-free state of the cable network based on a fault-free cable line simulation model, and simulating the impulse injection and response process under the condition that each observation port of the fault-free cable line simulation model is connected to a matching impedance to obtain the reference scattering matrix.

[0024] In this context, the observation port can be an observation point on the cable line of the distribution network. Multi-port impulse response testing refers to the process of sequentially injecting pulse signals into multiple observation ports of the cable and simultaneously acquiring the time-domain response waveforms at all observation ports. Test data can refer to time-domain response waveform data. Reconstruction refers to the process of organizing the test data to obtain the scattering matrix. The fault-free cable line simulation model can be a simulation calculation model based on real cable parameters, scaled proportionally to the real cable line, and with consistent electrical characteristics. Matching impedance can be a terminating impedance equal to the characteristic impedance of the cable, used to eliminate port reflections. The pulse injection and response process can refer to the process of sequentially injecting narrow time-domain pulse signals into each observation port and simultaneously acquiring the time-domain response waveforms of all ports.

[0025] In this embodiment, for a distribution network cable network with multiple observation ports, test data can be collected through multi-port impulse response testing when the cable network is in a fault state, and the fault-state scattering matrix can be reconstructed from the test data. Based on a fault-free cable line simulation model, the fault-free state of the cable network is simulated. Under the condition that each observation port of the fault-free cable line simulation model is connected to a matching impedance, the pulse injection and response process is simulated to obtain a reference scattering matrix. It is understood that the pulse injected during the simulated pulse injection and response process can be the same as the pulse injected during the multi-port impulse response test.

[0026] For example, the fault-state scattering matrix of a cable line under fault conditions. The reference scattering matrix under fault-free conditions The construction process includes: targeting those with N The distribution cable network with each observation port is reconstructed through a multi-port impulse response test under fault operation conditions. N × N × T n Fault-state scattering matrix of dimension Simultaneously, based on a fault-free cable line simulation model, the pulse injection and response process is simulated under the condition that each port is connected with matched impedance, and the reference scattering matrix of the same dimension is calculated. .

[0027] S120. Based on the fault-state scattering matrix and the reference scattering matrix, construct a spectral response feature dataset.

[0028] The spectrum response feature dataset can be a collection of samples containing frequency domain features related to early cable defects. The spectrum response feature dataset can include multiple samples, which can be labeled or unlabeled.

[0029] In this embodiment, the fault-state scattering matrix and the reference scattering matrix can be processed to construct a spectral response feature dataset.

[0030] In one embodiment, constructing a spectral response feature dataset based on the fault-state scattering matrix and the reference scattering matrix includes: processing the fault-state scattering matrix and the reference scattering matrix using matrix difference operations to obtain an incremental scattering matrix; performing conjugate transpose processing on the incremental scattering matrix to obtain a differential time inversion operator matrix; performing eigenvalue decomposition on the differential time inversion operator matrix, and extracting the eigenvector corresponding to the largest eigenvalue from the decomposed eigenvalues ​​as a target eigenvector; reconstructing the frequency domain inversion signal of each observation port based on the target eigenvector and the initial pulse waveform spectrum, and injecting the frequency domain inversion signal into the fault-free cable line simulation model; collecting the frequency domain response signal of each observation port of the fault-free cable line simulation model after the frequency domain inversion signal is injected, and constructing a spectral response feature dataset based on each of the frequency domain response signals.

[0031] Matrix difference operations can be subtraction operations on matrices. Conjugate transpose processing refers to taking the complex conjugate of each element in a complex matrix based on its transpose. Eigenvalue decomposition is the process of splitting a matrix into a set of eigenvalues ​​and their corresponding eigenvectors. The initial pulse waveform spectrum can be the waveform spectrum corresponding to the pulse injected to generate the fault-state scattering matrix and the reference scattering matrix. The frequency domain inversion signal can be the conversion of the cable's frequency domain response data into a time-domain or spatial-domain defect characteristic signal using an inverse problem solution method. The frequency domain response signal can be the response signal of the cable network to the injected pulse.

[0032] In this embodiment, the fault-state scattering matrix and the reference scattering matrix can be processed based on matrix difference operations to obtain the incremental scattering matrix. The incremental scattering matrix is ​​then subjected to conjugate transpose processing to obtain the differential time inversion operator matrix. The differential time inversion operator matrix is ​​then subjected to eigenvalue decomposition to obtain multiple eigenvalues ​​and eigenvectors corresponding to each eigenvalue. The eigenvector corresponding to the largest eigenvalue is extracted from the decomposed eigenvalues ​​as the target eigenvector. Based on the target eigenvector and the initial pulse waveform spectrum, the frequency domain inversion signal of each observation port is reconstructed. The frequency domain inversion signal is then injected into the fault-free cable line simulation model. The frequency domain response signal of each observation port of the fault-free cable line simulation model after the frequency domain inversion signal is injected is collected. A spectral response feature dataset is constructed based on each frequency domain response signal.

[0033] For example, after obtaining the fault-state scattering matrix and the reference scattering matrix, the incremental scattering matrix is ​​extracted using matrix difference operations. : ; Perform conjugate transpose on the incremental scattering matrix The differential-time inversion operator matrix is ​​obtained. and to Perform eigenvalue decomposition to extract the eigenvector corresponding to the largest eigenvalue. v ; Using feature vectors v The frequency domain inversion signal of each port is reconstructed by combining the spectrum of the initial pulse waveform, and the frequency domain inversion signal is injected into the simulation model of the faultless cable line; The frequency domain response signals of each detection port in the simulation model of a faultless cable line are collected, truncated, and sampled to obtain a length of... L (like L Spectral response sequence of (=1024) x The spectral response sequence x As a dataset of spectral response features, it serves as the input features for subsequent neural networks.

[0034] S130. Train the pre-constructed Fourier ladder network model based on the spectral response feature dataset to obtain the trained Fourier ladder network model.

[0035] Among them, the Fourier ladder network model can be a network model that integrates frequency domain feature transformation and deep feature learning.

[0036] In this embodiment, a pre-built Fourier ladder network model can be constructed, and the pre-built Fourier ladder network model can be trained based on the spectral response feature dataset to obtain a trained Fourier ladder network model.

[0037] In one embodiment, training a pre-constructed Fourier step network model based on the spectral response feature dataset to obtain a trained Fourier step network model includes: inputting the spectral response feature dataset into the pre-constructed Fourier step network model to obtain predicted labels; determining the current loss value of the Fourier step network model based on the predicted labels, the true labels, and the loss function; updating the neural network parameters of the Fourier step network model when the loss value is greater than a preset threshold, and continuing to train the updated Fourier step network model using the spectral response feature dataset; and using the current Fourier step network model as the trained Fourier step network model when the loss value is less than the preset threshold or the current iteration count reaches the maximum iteration count.

[0038] In this model, the predicted label can be the label of the sample predicted by the model, and the true label can be the actual label of the sample. The preset threshold and maximum number of iterations can be set according to actual conditions; this embodiment does not limit these. Neural network parameters can include the number of layers, the number of neurons, the learning rate, etc.

[0039] In this embodiment, the spectral response feature dataset can be input into a pre-constructed Fourier step network model to obtain the predicted labels of each sample. Based on the predicted labels, true labels, and predefined loss functions of each sample, the current loss value of the Fourier step network model is calculated. When the loss value is greater than a preset threshold, the neural network parameters of the Fourier step network model can be updated, and the updated Fourier step network model can be trained again using the spectral response feature dataset. When the loss value is less than the preset threshold or the current iteration count reaches the maximum iteration count, the current Fourier step network model can be used as the trained Fourier step network model.

[0040] S140. Input the data of the cable to be tested into the trained Fourier ladder network model to obtain the cable defect identification result.

[0041] The data of the cable under test can be data from a cable whose defects need to be identified. The cable defect identification results can include the defect details, such as whether a defect exists and the type of defect, and can also include the features extracted from the cable data under test.

[0042] In this embodiment, the data of the cable under test can be input into a trained Fourier ladder network model to obtain the cable defect identification result corresponding to the data of the cable under test.

[0043] This invention provides a method for identifying early defects in cables, comprising: acquiring a fault-state scattering matrix of a distribution network cable line under fault conditions and a reference scattering matrix under fault-free conditions; constructing a spectral response feature dataset based on the fault-state scattering matrix and the reference scattering matrix; training a pre-constructed Fourier step network model based on the spectral response feature dataset to obtain a trained Fourier step network model; and inputting the cable data to be tested into the trained Fourier step network model to obtain the cable defect identification result. This method, by training a Fourier step network model using the constructed spectral response feature dataset, can accurately identify early defects in cable lines, solving the problem in existing technologies that cannot accurately identify early latent defects in cables.

[0044] Based on the above embodiments, modified embodiments of the above embodiments are proposed. It should be noted that, in order to keep the description brief, only the differences from the above embodiments are described in the modified embodiments.

[0045] In one embodiment, the Fourier step network model includes a noisy encoder, a noiseless encoder, and a decoder. The predicted label is the label output by the noisy encoder. Accordingly, determining the current loss value of the Fourier step network model based on the predicted label, the true label, and the loss function includes: determining a supervised loss value based on the predicted label, the true label, and the supervised loss function; determining an unsupervised loss value based on the noiseless feature representation output by the noiseless encoder, the reconstructed features output by the decoder, and the unsupervised loss function; and determining the current loss value of the Fourier step network model based on the supervised loss value, the unsupervised loss value, and the hybrid loss function.

[0046] In this system, a noisy encoder can be defined as an encoder that maps input data with added artificial noise to a low-dimensional latent space and extracts robust feature representations. A noiseless encoder can be defined as an encoder that takes clean, un-noiseed original data as input. A decoder can be defined as a module that reconstructs the original input data space based on the latent features output by the encoder through inverse mapping, making the reconstructed result as close as possible to the clean original input. A supervised loss function can be a loss function used in supervised learning, an unsupervised loss function can be a loss function used in unsupervised learning, and a hybrid loss function can be a loss function used in semi-supervised learning. A noiseless feature representation can be the low-dimensional latent space features output by the noiseless encoder, and the reconstructed features can be the feature data output by the decoder.

[0047] In this embodiment, the Fourier step network model may include a noisy encoder, a noiseless encoder, and a decoder. The predicted label can be the label output by the noisy encoder. When calculating the loss value, the supervised loss value can be calculated based on the predicted label, the true label, and the supervised loss function. The unsupervised loss value can be calculated based on the noiseless feature representation output by the noiseless encoder, the reconstructed features output by the decoder, and the unsupervised loss function. Finally, the current loss value of the Fourier step network model is calculated based on the supervised loss value, the unsupervised loss value, and the mixed loss function.

[0048] For example, when training the model, the parameters of the Fourier staircase network model can be randomly initialized, the training data can be input into the model, the forward propagation results can be calculated, and the mixture loss function can be calculated. C The value of the loss function is used to update the network parameters through backpropagation using the gradient descent algorithm. This process of calculating the forward propagation result and updating the network parameters is repeated until the loss function value falls below a preset threshold or the maximum number of iterations is reached, at which point training stops. Finally, the cable data to be tested is input into the trained Fourier stepwise network model. The output of the Fourier stepwise network model is the cable defect diagnosis result, while the output of the highest-level noisy encoder is... The extracted feature vector.

[0049] In one embodiment, each layer of the noisy encoder includes frequency domain linear transformation, complex batch normalization, spectral noise injection, frequency domain affine transformation, and nonlinear activation; each layer of the noiseless encoder includes frequency domain linear transformation, complex batch normalization, frequency domain affine transformation, and nonlinear activation; and each layer of the decoder includes frequency domain linear transformation, normalization, and spectral noise reduction operations. The frequency domain linear transformation of each layer utilizes Fourier neural operators to perform feature transformation in the frequency domain.

[0050] Among these, frequency domain linear transformation refers to the operation of linearly mapping the output features of the previous layer to the current layer using a learnable frequency domain weight matrix (Fourier operator kernel); complex batch normalization refers to the operation of standardizing the statistical distribution of the real and imaginary parts of the features; spectral noise injection refers to the operation of injecting Gaussian noise into the standardized feature spectrum to generate noisy features; frequency domain affine transformation refers to the operation of performing affine transformation on noisy features to restore the expressive power of the features; and nonlinear activation refers to the operation of performing nonlinear transformation on features. Fourier neural operators can be deep neural network operators used to learn function space-to-function-space mappings.

[0051] In this embodiment, the noisy encoder, the noiseless encoder, and the decoder can all contain multi-layer structures. Each layer of the noisy encoder can include five operations: frequency domain linear transformation, complex batch normalization, spectral noise injection, frequency domain affine transformation, and nonlinear activation. The highest layer of the noisy encoder uses the Softmax activation function, while the remaining layers use the Rectified Linear Unit (ReLU) activation function. Finally, the output of the highest-layer noisy encoder is used to calculate supervised classification prediction. The noiseless encoder can share the frequency domain weight matrix of each layer with the noisy encoder, but does not perform noise injection. The output of the noiseless encoder can be used as the target benchmark for reconstruction. Each layer of the decoder can include frequency domain linear transformation, normalization, and spectral noise reduction operations. The decoder can receive feature information from the corresponding layer of the noisy encoder through lateral connections and combine it with the output of the upper-layer decoder to reconstruct layer by layer to approximate the output of the noiseless encoder.

[0052] The method of this invention effectively overcomes the effects of signal dispersion and attenuation, achieving high feature extraction efficiency. By utilizing the global convolution operation of Fourier neural operators in the frequency domain, it solves the problem that existing time-domain analysis methods or local convolutional networks struggle to extract minute defect features due to high-frequency component attenuation and dispersion effects when processing long-distance cable signals. This method eliminates the need for cumbersome time-frequency transformations, directly capturing global features reflecting impedance changes along the line from end-measured signals, significantly improving the sensitivity for early identification of minute defects. Compared to traditional fully supervised learning, which heavily relies on massive amounts of labeled samples, this invention employs a semi-supervised learning strategy, reducing dependence on large amounts of labeled data and resulting in strong model generalization ability. This solves the problems of poor generalization performance and the need for repeated training in existing deep learning models. It can learn the mapping laws of physical fields using limited samples, adapting to varying distribution network conditions and possessing stronger engineering practicality.

[0053] For example, Figure 2 This is a schematic diagram of the structure of a Fourier ladder network model provided in an embodiment of the present invention, as shown below. Figure 2 As shown, each layer of the constructed noisy encoder comprises five operations: frequency domain linear transformation, complex batch normalization, spectral noise injection, frequency domain affine transformation, and nonlinear activation. The frequency domain linear transformation operation linearly maps the output features of the previous layer to the current layer through a learnable frequency domain weight matrix (Fourier operator kernel). The frequency domain weight matrix can be user-defined. The mathematical mechanism of the Fourier integral layer (i.e., the frequency domain linear transformation layer utilizing Fourier neural operators) is detailed below: Global feature extraction is performed using the integral operator defined by the Green's function, for the feature function of the input spectral response feature dataset. The output of the (t+1)th layer Defined as: ; in, It is a non-linear activation function. It is a linear transformation matrix. It is a kernel integral operator; definition for: ; in, For the kernel function that the neural network needs to learn, ( x, y ) is the domain D points within; Using the convolution theorem, the above integral operation is transformed into a multiplication operation in the frequency domain, let... The above integral can then be simplified to a convolution form, which can be achieved through Fourier transform: ; in, Represents the Fast Fourier Transform. This represents the inverse fast Fourier transform. It is a learnable complex weight matrix in the frequency domain, i.e., the frequency domain weight matrix.

[0054] Complex batch normalization can standardize the statistical distribution of the real and imaginary parts of features; spectral noise injection can inject Gaussian noise into the standardized feature spectrum to generate noisy features; frequency domain affine transformation can perform affine transformation on noisy features to restore their expressive power; nonlinear activation can perform nonlinear transformation on features.

[0055] The constructed noiseless encoder shares the frequency domain weight matrix of each layer with the noisy encoder, but does not perform noise injection operations; its output serves as the target benchmark for reconstruction. The constructed decoder, with each layer containing frequency domain linear transformation, normalization, and spectral denoising operations, receives feature information from the corresponding layer's noisy encoder through lateral connections and combines it with the output of the upper-layer decoder to reconstruct layer by layer to approximate the output of the noiseless encoder.

[0056] In one embodiment, the supervised loss function for; ; in, M The number of samples in the spectral response feature dataset, including both unlabeled and labeled samples. The first in the spectral response feature dataset n One sample, For the first n The true label of each sample The predicted label is the output of the noisy encoder; Indicates sample Corresponding predicted labels Belongs to real labels The probability value; The unsupervised loss function for: ; in, m l For the first l The number of neurons in a layered neural network. For the first l The output of the first layer of the neural network n Noise-free feature representation of each sample For the first l The output of the first layer of the neural network n Reconstructed features of each sample; The hybrid loss function C for: ; in, l For the first l The weights of the layered neural network in the unsupervised loss. L This represents the total number of layers in the ladder network.

[0057] In this embodiment, a hybrid loss function can be constructed using supervised and unsupervised loss functions.

[0058] For example, to construct a 4-layer Fourier ladder network model containing a noisy encoder, a noiseless encoder, and a decoder, the process of constructing the hybrid loss function is as follows: Based on predicted labels and real labels t The cross-entropy between them is used to construct the loss function for the supervised learning process. : ; Based on the reconstructed features output by the decoder Noise-free feature representation of the output of the noise-free decoder The mean squared error between them is used to construct the loss function for the unsupervised learning process. : Indicates the input sample Under certain conditions, the probability that a neural network model predicts its class label, i.e., the probability that the model determines the sample belongs to the true label. The probability value; ; Based on the supervised and unsupervised loss functions mentioned above, a hybrid loss function is constructed: .

[0059] Next, by selecting appropriate neural network parameters, the parameters of a four-layer Fourier step network can be set. The first layer of the encoder uses a Fourier Neural Operator (FNO) that retains 32 low-frequency modes and a feature width of 8, resulting in an output size of 256×8 after 4x downsampling (Stride=4). The second layer has 16 modes and a width of 16, resulting in an output size of 128×16 after 2x downsampling. The third layer has 16 modes and a width of 32, resulting in an output size of 64×32 after 2x downsampling. The fourth layer has 8 modes and a width of 64, resulting in an output size of 16×64 after 4x downsampling. At the encoder end, the features are flattened, mapped to the four output nodes via a fully connected layer, and the cable defect diagnosis result is calculated using the Softmax function. The decoder recovers the signal layer by layer through upsampling and FNO operations, and fuses the noisy features of the corresponding coding layers using lateral connections, ultimately reconstructing the original spectrum. The Fourier ladder network model is optimized using the Adaptive Moment Estimation (Adam) algorithm. The initial learning rate can be set to 0.02, and a decay strategy is adopted, with the learning rate decaying starting from the 15th epoch. The total number of training iterations can be set to 1500. Set the encoder noise amplitude and the reconstruction loss weights for each decoder layer: After normalization of each encoder layer and before the activation function, inject the amplitude value... A noise =0.6 Gaussian noise; input layer reconstruction weights input Set to 10, first layer weight L1 Set to 1.0, and the remaining high-level weights ( L2 ~ L4 All values ​​are set to 0.1.

[0060] Based on the technical solutions of the above embodiments, this invention provides several specific implementation methods.

[0061] As one specific implementation method of this embodiment. Figure 3 This is a flowchart illustrating a method for identifying early defects in cables according to an embodiment of the present invention, as shown below. Figure 3As shown, the frequency domain scattering matrix (fault-state scattering matrix and reference scattering matrix) of distribution network cable lines under fault and fault-free states can be measured, and the frequency domain response feature dataset of the distribution network cable lines can be determined based on the frequency domain scattering matrix. The dataset contains labeled and unlabeled samples. Based on the frequency domain response feature dataset, a Fourier ladder network model is constructed. This model includes a noisy encoder, a noiseless encoder, and a decoder, where each layer of the network uses Fourier neural operators to perform feature transformation in the frequency domain. Based on the Fourier ladder network model, a hybrid loss function containing supervised and unsupervised loss terms is constructed, and appropriate neural network parameters are selected. Finally, the constructed Fourier ladder network model is subjected to semi-supervised iterative training until the loss function value decreases to a given threshold, thereby realizing the feature extraction of early cable defects. After inputting into the test task, the output of the neural network is obtained, which is the distribution network cable defect diagnosis result. The method of this embodiment of the invention can directly perform operator learning in the frequency domain through Fourier operators without the need for cumbersome time-frequency conversion; compared with traditional supervised and unsupervised learning, the semi-supervised learning method has good generalization ability.

[0062] Figure 4 A schematic diagram of a distribution network equal-length three-port line provided for an embodiment of the present invention, such as... Figure 4 As shown, this study focuses on a three-port distribution network of equal length, with the fault uniformly set at point AB, 800m from point B. For a distribution cable network with three observation ports, a 3×3× fault can be reconstructed through multi-port impulse response testing under three fault conditions: short circuit, floating potential, and open circuit. T n Fault-state scattering matrix of dimension Simultaneously, based on a proportional numerical simulation model, the pulse injection and response process is simulated under the condition that each port is connected with matched impedance, and the reference scattering matrix of the same dimension is calculated. .

[0063] After the dataset is constructed, it contains three types of fault signals: short circuit, floating potential, and open circuit. Each type contains 60 samples, for a total of 180 samples. Figure 5 A spatial spectrum of an AB circuit is provided as an embodiment of the present invention, such as Figure 5 As shown, the figure includes spatial spectra of short-circuit, floating potential, and open-circuit fault signals. 150 samples can be used for training, and 30 samples for testing. A total of 6 sets of experiments were conducted. In each set, a subset (30, 60, 90, 120, and 150 samples) of training samples were extracted and labeled for supervised learning. All labeled and unlabeled samples were used for unsupervised learning. Six rounds of experiments were conducted, employing a cross-validation approach, and the average accuracy across the six rounds was used as the final result.

[0064] Figure 6This is a comparison chart showing the prediction accuracy of a Fourier ladder network model and a regular convolutional neural network, provided in an embodiment of the present invention. Figure 6 As shown, although both Fourier step network and convolutional neural network can obtain relatively accurate simulation results, the Fourier step network has a significantly improved diagnostic accuracy for cable defects compared to the convolutional neural network because it performs operator learning directly in the frequency domain, eliminating the need for cumbersome time-frequency conversion, and it incorporates additional information from unlabeled samples. Secondly, the 150-label count demonstrates that even with sufficient labeled samples, introducing unsupervised learning can still improve the generalization performance of the neural network.

[0065] Example 2 Figure 7 This is a schematic diagram of the structure of an early cable defect identification device provided in Embodiment 2 of the present invention. The device is applicable to the identification of early defects in power distribution network cable lines. The device can be implemented by software and / or hardware and is generally integrated on electronic equipment.

[0066] like Figure 7 As shown, the device includes: The acquisition module 210 is used to acquire the fault-state scattering matrix of the distribution network cable line under fault conditions and the reference scattering matrix under fault-free conditions. Construction module 220 is used to construct a spectral response feature dataset based on the fault-state scattering matrix and the reference scattering matrix; Training module 230 is used to train a pre-constructed Fourier ladder network model based on the spectral response feature dataset to obtain a trained Fourier ladder network model. The identification module 240 is used to input the data of the cable to be tested into the trained Fourier ladder network model to obtain the cable defect identification result.

[0067] This embodiment provides a cable early defect identification device, comprising: an acquisition module for acquiring the fault-state scattering matrix of a distribution network cable line under fault conditions and the reference scattering matrix under fault-free conditions; a construction module for constructing a spectral response feature dataset based on the fault-state scattering matrix and the reference scattering matrix; a training module for training a pre-constructed Fourier step network model based on the spectral response feature dataset to obtain a trained Fourier step network model; and an identification module for inputting the cable data to be tested into the trained Fourier step network model to obtain the cable defect identification result. By training the Fourier step network model using the constructed spectral response feature dataset, early defects in cable lines can be accurately identified, solving the problem in existing technologies that cannot accurately identify early latent defects in cables.

[0068] Furthermore, the acquisition module 210 includes: For a distribution network cable network with multiple observation ports, when the cable network is in a fault state, test data is collected through multi-port impulse response testing, and the fault state scattering matrix is ​​reconstructed from the test data; Based on the simulation model of a faultless cable line, the faultless state of the cable network is simulated. Under the condition that each observation port of the faultless cable line simulation model is connected to a matching impedance, the pulse injection and response process is simulated to obtain the reference scattering matrix.

[0069] Furthermore, module 220 includes: The fault-state scattering matrix and the reference scattering matrix are processed by matrix difference operations to obtain the incremental scattering matrix; The incremental scattering matrix is ​​conjugate transposed to obtain the differential-time inversion operator matrix; The differential time inversion operator matrix is ​​decomposed into eigenvalues, and the eigenvector corresponding to the largest eigenvalue is extracted from the decomposed eigenvalues ​​as the target eigenvector. Based on the target feature vector and the initial pulse waveform spectrum, the frequency domain inversion signal of each observation port is reconstructed, and the frequency domain inversion signal is injected into the fault-free cable line simulation model. After the frequency domain inversion signal is injected into the simulation model of the faultless cable line, the frequency domain response signal of each observation port is collected, and a spectrum response feature dataset is constructed based on the frequency domain response signal.

[0070] Furthermore, training module 230 includes: The spectral response feature dataset is input into a pre-constructed Fourier ladder network model to obtain predicted labels; The current loss value of the Fourier ladder network model is determined based on the predicted label, the true label, and the loss function. When the loss value is greater than a preset threshold, the neural network parameters of the Fourier ladder network model are updated, and the updated Fourier ladder network model is trained again using the spectral response feature dataset. When the loss value is less than a preset threshold or the current iteration count reaches the maximum iteration count, the current Fourier step network model is used as the trained Fourier step network model.

[0071] Furthermore, the Fourier step network model includes a noisy encoder, a noiseless encoder, and a decoder, and the predicted label is the label output by the noisy encoder. Correspondingly, determining the current loss value of the Fourier step network model based on the predicted label, the true label, and the loss function includes: Based on the predicted labels and the true labels, as well as the supervised loss function, the supervised loss value is determined. Based on the noiseless feature representation output by the noiseless encoder, the reconstructed features output by the decoder, and the unsupervised loss function, the unsupervised loss value is determined. Based on the supervised loss value, the unsupervised loss value, and the hybrid loss function, the current loss value of the Fourier ladder network model is determined.

[0072] Furthermore, the supervised loss function for; ; in, M The number of all samples in the spectral response feature dataset. The first in the spectral response feature dataset n One sample, For the first n The true label of each sample The predicted label is the output of the noisy encoder; Indicates sample Corresponding predicted labels Belongs to real labels The probability value; The unsupervised loss function for: ; in, m l For the first l The number of neurons in a layered neural network. For the first l The output of the first layer of the neural network n Noise-free feature representation of each sample For the first l The output of the first layer of the neural network n Reconstructed features of each sample; The hybrid loss function C for: ; in, l For the first l The weights of the layered neural network in the unsupervised loss. L This represents the total number of layers in the ladder network.

[0073] Furthermore, each layer of the noisy encoder includes frequency domain linear transformation, complex batch normalization, spectral noise injection, frequency domain affine transformation, and nonlinear activation; each layer of the noiseless encoder includes frequency domain linear transformation, complex batch normalization, frequency domain affine transformation, and nonlinear activation; and each layer of the decoder includes frequency domain linear transformation, normalization, and spectral noise reduction operations. The frequency domain linear transformation of each layer utilizes Fourier neural operators to perform feature transformation in the frequency domain.

[0074] The aforementioned cable early defect identification device can execute the cable early defect identification method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method.

[0075] Example 3 Figure 8 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0076] like Figure 8 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0077] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0078] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as early defect identification methods for cables.

[0079] In some embodiments, the cable early defect identification method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or mounted on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the cable early defect identification method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the cable early defect identification method by any other suitable means (e.g., by means of firmware).

[0080] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0081] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0082] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0083] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0084] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0085] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0086] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0087] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for early defect identification in cables, characterized in that, The method includes: Obtain the fault-state scattering matrix of the distribution network cable line under fault conditions, and the reference scattering matrix under fault-free conditions; Based on the fault-state scattering matrix and the reference scattering matrix, a spectral response feature dataset is constructed; The pre-constructed Fourier ladder network model is trained based on the spectral response feature dataset to obtain the trained Fourier ladder network model. The cable data to be tested is input into the trained Fourier ladder network model to obtain the cable defect identification results.

2. The method according to claim 1, characterized in that, The acquisition of the fault-state scattering matrix of the distribution network cable line under fault conditions and the reference scattering matrix under fault-free conditions includes: For a distribution network cable network with multiple observation ports, when the cable network is in a fault state, test data is collected through multi-port impulse response testing, and the fault state scattering matrix is ​​reconstructed from the test data; The fault-free state of the cable network is simulated based on the fault-free cable line simulation model. Under the condition that each observation port of the fault-free cable line simulation model is connected to the matching impedance, the pulse injection and response process is simulated to obtain the reference scattering matrix.

3. The method according to claim 1, characterized in that, The construction of a spectral response feature dataset based on the fault-state scattering matrix and the reference scattering matrix includes: The fault-state scattering matrix and the reference scattering matrix are processed by matrix difference operations to obtain the incremental scattering matrix; The incremental scattering matrix is ​​subjected to conjugate transpose to obtain the differential time inversion operator matrix; The differential time inversion operator matrix is ​​decomposed into eigenvalues, and the eigenvector corresponding to the largest eigenvalue is extracted from the decomposed eigenvalues ​​as the target eigenvector. Based on the target feature vector and the initial pulse waveform spectrum, the frequency domain inversion signal of each observation port is reconstructed, and the frequency domain inversion signal is injected into the fault-free cable line simulation model. After the frequency domain inversion signal is injected into the simulation model of the faultless cable line, the frequency domain response signal of each observation port is collected, and a spectrum response feature dataset is constructed based on the frequency domain response signal.

4. The method according to claim 1, characterized in that, The process of training a pre-constructed Fourier ladder network model based on the spectral response feature dataset to obtain the trained Fourier ladder network model includes: The spectral response feature dataset is input into a pre-constructed Fourier ladder network model to obtain the predicted labels; The current loss value of the Fourier ladder network model is determined based on the predicted label, the true label, and the loss function; When the loss value is greater than a preset threshold, the neural network parameters of the Fourier ladder network model are updated, and the updated Fourier ladder network model is trained again using the spectral response feature dataset. When the loss value is less than a preset threshold or the current iteration count reaches the maximum iteration count, the current Fourier step network model is used as the trained Fourier step network model.

5. The method according to claim 4, characterized in that, The Fourier step network model includes a noisy encoder, a noiseless encoder, and a decoder. The predicted label is the label output by the noisy encoder. Correspondingly, determining the current loss value of the Fourier step network model based on the predicted label, the true label, and the loss function includes: Based on the predicted labels and the true labels, as well as the supervised loss function, the supervised loss value is determined. Based on the noiseless feature representation output by the noiseless encoder, the reconstructed features output by the decoder, and the unsupervised loss function, the unsupervised loss value is determined. Based on the supervised loss value, the unsupervised loss value, and the hybrid loss function, the current loss value of the Fourier ladder network model is determined.

6. The method according to claim 5, characterized in that, The supervised loss function for; ； in, M The number of all samples in the spectral response feature dataset. The first in the spectral response feature dataset n One sample, For the first n The true label of each sample The predicted label is the output of the noisy encoder; Indicates sample Corresponding predicted labels Belongs to real labels The probability value; The unsupervised loss function for: ； in, m l For the first l The number of neurons in a layered neural network. For the first l The output of the first layer of the neural network n Noise-free feature representation of each sample For the first l The output of the first layer of the neural network n Reconstructed features of each sample; The hybrid loss function C for: ； in, l For the first l The weights of the layered neural network in the unsupervised loss. L This represents the total number of layers in the ladder network.

7. The method according to claim 5, characterized in that, Each layer of the noisy encoder includes frequency domain linear transformation, complex batch normalization, spectral noise injection, frequency domain affine transformation, and nonlinear activation. Each layer of the noiseless encoder includes frequency domain linear transformation, complex batch normalization, frequency domain affine transformation, and nonlinear activation. Each layer of the decoder includes frequency domain linear transformation, normalization, and spectral noise reduction operations. The frequency domain linear transformation of each layer utilizes Fourier neural operators to perform feature transformation in the frequency domain.

8. A device for identifying early defects in cables, characterized in that, The device includes: The acquisition module is used to acquire the fault-state scattering matrix of the distribution network cable line under fault conditions, and the reference scattering matrix under fault-free conditions. The construction module is used to construct a spectral response feature dataset based on the fault-state scattering matrix and the reference scattering matrix; The training module is used to train a pre-constructed Fourier ladder network model based on the spectral response feature dataset to obtain the trained Fourier ladder network model. The identification module is used to input the data of the cable under test into the trained Fourier ladder network model to obtain the cable defect identification result.

9. An electronic device, characterized in that, The device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the cable early defect identification method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the cable early defect identification method according to any one of claims 1-7.

Images