Traveling wave head fault positioning method and system based on modal decomposition and neural network

Abstract

The invention discloses a traveling wave head fault positioning method and system based on modal decomposition and a neural network, and the method comprises the steps: obtaining a first traveling wave signal on a power transmission line, and extracting a waveform feature corresponding to the first traveling wave signal; inputting the waveform characteristics into a preset parameter optimization model, determining a corresponding optimal decomposition parameter, performing multi-scale variational mode decomposition on the first traveling wave signal by using the optimal decomposition parameter to obtain a plurality of initial modes, and determining a corresponding key mode based on each initial mode; inputting the key mode into a preset distortion elimination model, determining a distortion result of each sampling point in the key mode, and filtering a distortion section of the first traveling wave signal based on the distortion result and the key mode to obtain a second traveling wave signal; and determining a traveling wave head abrupt change point based on the second traveling wave signal, and determining a corresponding fault position based on the traveling wave head abrupt change point. According to the invention, the accuracy of traveling wave head fault positioning can be improved.

Description

Technical Field

[0001] This application relates to the field of fault location, and particularly to a traveling wave head fault location method and system based on modal decomposition and neural network. Background Art

[0002] During the operation of the power grid, external factors such as lightning strikes, wildfires, icing, and tree obstacles often trigger transmission and distribution line fault tripping accidents. These accidents not only occur frequently but also pose a great threat to the operation safety and power supply reliability of the power grid. Therefore, how to accurately locate the fault location has become the key to ensuring the stable operation of the power grid.

[0003] Currently, in the industry, mainly methods such as wavelet transform and wavelet packet transform are used to achieve accurate positioning of traveling wave heads. However, these traditional methods gradually expose some inevitable defects in practical applications. First, before a transmission line fault tripping accident occurs, the arc discharge near the fault point will release a large amount of high-frequency energy in a short time. This process will cause the traveling wave signal to have a distorted signal with a small amplitude but a high frequency before the wave head mutation point. Since the distorted signal often overlaps with the wave head signal, it brings great difficulties to accurately distinguish the true wave head signal and the arc distortion signal, thereby affecting the positioning accuracy of the fault point. Second, the wavelet transform and its derivative methods are prone to modal aliasing problems during the decomposition process. Specifically, the energies of different frequency components cannot be completely separated. This problem will cause the mutation characteristics of the traveling wave head to be partially masked or dispersed into multiple frequency bands during the decomposition process, making the characteristics of the mutation point unable to be concentrated in one frequency band, thereby affecting the accuracy of wave head extraction. Finally, the performance of the wavelet transform and its derivative algorithms highly depends on parameters such as the selection of the mother wavelet type, the number of decompositions, and the definition of the threshold. However, these parameters are usually determined through experiments or artificial experience and cannot be dynamically adjusted under different line conditions. This limitation often leads to an increase in positioning errors in practical applications, thereby affecting the reliability of fault location.

[0004] Application content

[0005] This application provides a traveling wave head fault location method and system based on modal decomposition and neural network, which solves the problems of positioning being interfered by distorted waveforms, modal aliasing, and inability to achieve parameter adaptive adjustment in the prior art through modal decomposition and neural network, and improves the accuracy of traveling wave head fault location.

[0006] In a first aspect, this application provides a traveling wave head fault location method based on modal decomposition and neural network, including:

[0007] Obtain a first traveling wave signal on a transmission line and extract waveform features corresponding to the first traveling wave signal;

[0008] Input the waveform features into a preset parameter optimization model to determine the corresponding optimal decomposition parameters, and use the optimal decomposition parameters to perform multi-scale variational mode decomposition on the first traveling wave signal to obtain a number of initial modes, and determine the corresponding key modes based on each of the initial modes. Among them, the parameter optimization model is trained using optimal waveform parameters, and the optimal waveform parameters are determined by performing parameter optimization on a traveling wave waveform dataset;

[0009] Input the key modes into a preset distortion rejection model to determine the distortion results of each sampling point in the key modes, and filter out the distorted sections of the first traveling wave signal based on the distortion results and the key modes to obtain a second traveling wave signal;

[0010] Determine the traveling wave head mutation point based on the second traveling wave signal, and determine the corresponding fault location based on the traveling wave head mutation point.

[0011] In the embodiment of the present application, by extracting the waveform features corresponding to the first traveling wave signal, it is possible to provide comprehensive signal information for the subsequent parameter optimization model, which helps to accurately determine the optimal decomposition parameters; through the parameter optimization model, it is possible to automatically and quickly determine the optimal decomposition parameters according to the input waveform features, solving the problem that it is difficult to adaptively adjust parameters in the prior art; using the optimal decomposition parameters to perform multi-scale variational mode decomposition on the traveling wave signal can effectively separate different frequency components in the signal, avoiding the phenomenon of mode mixing, facilitating the subsequent rapid and accurate identification of the mutation point of the traveling wave head, and improving the accuracy of wave head positioning; determining the corresponding key modes based on each of the initial modes can focus more on the key modes containing the information of the wave head mutation point, providing a more accurate signal basis for subsequent distortion rejection and fault location; through the distortion rejection model, it is possible to accurately identify the distortion conditions of each sampling point in the key modes and filter out the distorted sections in the first traveling wave signal accordingly, effectively solving the problem of inaccurate wave head positioning caused by distorted waveform interference in the prior art; by processing the second traveling wave signal, it is possible to accurately determine the mutation point of the traveling wave head, and thus accurately calculate the fault location. Compared with the prior art, the present application solves the problems existing in the prior art such as positioning being interfered by distorted waveforms, mode mixing, and inability to achieve parameter adaptive adjustment through mode decomposition and neural networks, improving the accuracy of traveling wave head fault location.

[0012] Further, the optimal decomposition parameters include the number of modes, the bandwidth penalty parameter, the modal center frequency, and the noise tolerance. The use of the optimal decomposition parameters to perform multi-scale variational mode decomposition on the first traveling wave signal to obtain a number of initial modes is specifically:

[0013] Perform signal preprocessing on the first traveling wave signal to obtain a third traveling wave signal;

[0014] In each iteration process, the third traveling wave signal is iteratively and optimally decomposed based on the number of modes, the bandwidth penalty parameter, and the modal center frequency, and the noise suppression ability of the decomposition process is adjusted according to the noise tolerance to obtain a first decomposition result. When the update amount of the iterative and optimal decomposition is less than a preset threshold, the corresponding initial mode is determined based on the obtained first decomposition result.

[0015] In this way, by using the optimal decomposition parameters to perform multi-scale variational mode decomposition on the traveling wave signal, different frequency components in the signal can be effectively separated, the mode mixing phenomenon is avoided, the mutation point of the traveling wave head can be quickly and accurately identified subsequently, and the accuracy of wave head positioning is improved.

[0016] Furthermore, the optimal waveform parameters are determined by parameter optimization on the traveling wave waveform data set, specifically:

[0017] Perform multi-scale variational mode decomposition on the traveling wave waveform data set to obtain a second decomposition result, and determine the corresponding number of modes based on the traveling wave waveform data set and the second decomposition result;

[0018] Based on the number of modes and the second decomposition result, iterative optimization is performed with the goal of minimizing the bandwidth objective function to determine the corresponding bandwidth penalty parameter;

[0019] Calculate the main frequency peak of the traveling wave waveform data set, use the main frequency peak as the initial center frequency, determine the spectral energy density of each decomposition based on the initial center frequency and the bandwidth penalty parameter, and perform iterative optimization based on the spectral energy density to determine the modal center frequency;

[0020] Determine the corresponding bandwidth based on the modal center frequency, determine the frequency band overlap degree of adjacent modes in the variational mode decomposition process based on the bandwidth, and perform iterative optimization based on the frequency band overlap degree to determine the noise tolerance.

[0021] In this way, by performing parameter optimization on the traveling wave waveform data set, a parameter optimization model can be conveniently trained quickly and accurately subsequently, and the accuracy of wave head positioning is further improved.

[0022] Furthermore, the calculation formula for performing iterative optimization based on the spectral energy density to determine the modal center frequency is specifically:

[0023]

[0024] In the formula, ω k is the modal center frequency, U k(ω) is the spectral energy density, F(ω) is the spectral function obtained by Fourier transform or wavelet transform of the traveling wave waveform dataset, Λ(ω) is the Lagrange multiplier term, α is the bandwidth penalty parameter, i is the iteration number, k is the k-th mode, and ω is the initial center frequency.

[0025] Further, the training process of the parameter optimization model is specifically as follows:

[0026] Input the traveling wave waveform dataset into a preset neural network model, and calculate the predicted values of the modal parameters through forward propagation;

[0027] Calculate the loss value between the predicted parameter values and the optimal waveform parameters, and perform backpropagation through the Adam algorithm to update the weights of the neural network model according to the loss value until the loss value meets the preset conditions, then determine the optimal weights, and determine the parameter optimization model based on the optimal weights.

[0028] In this way, the parameter optimization model obtained through training can automatically and quickly determine the optimal decomposition parameters according to the input waveform characteristics, solving the problem that the parameters are difficult to adaptively adjust in the prior art.

[0029] Further, determining the corresponding key mode based on each of the initial modes is specifically as follows:

[0030] Calculate the high-frequency energy ratio, time-domain variance, and sample entropy of each of the initial modes;

[0031] Based on the high-frequency energy ratio, the time-domain variance, and the sample entropy, score each of the initial modes, and use the initial mode with the highest score as the key mode.

[0032] In this way, determining the corresponding key mode based on each of the initial modes can focus more on the key mode containing the information of the wavefront mutation point, providing a more accurate signal basis for subsequent distortion elimination and fault location.

[0033] Further, the relevant formula for determining the corresponding key mode based on each of the initial modes is specifically as follows:

[0034] Score(IMF k ) = ω1·E k +ω2·V k -ω3·H k ;

[0035]

[0036] In the formula, is the score of the k-th initial mode, ω1, ω2, and ω3 are the weight coefficients of each characteristic, and Ek is the high-frequency energy proportion of the k-th initial mode, V k is the time-domain variance of the k-th initial mode, H k is the sample entropy of the k-th initial mode, ω max and ω min are the upper and lower limits of the frequency range respectively, ω high is the starting point of the high-frequency range, U k S(ω) is the spectral energy density, T is the duration of the initial mode, μ k is the average value of the initial mode, M is the total number of sampling points of the initial mode, P i is the probability distribution of the i-th sample.

[0037] Further, inputting the key mode into a preset distortion removal model to determine the distortion results of each sampling point in the key mode specifically includes:

[0038] Inputting the key mode into the first convolutional layer and the first pooling layer of the distortion removal model to obtain a first output feature;

[0039] Inputting the first output feature into the second convolutional layer and the second pooling layer to obtain a second output feature;

[0040] Inputting the second output feature into the third convolutional layer and the third pooling layer to obtain a third output feature, and inputting the third output feature into the global pooling layer to obtain a global feature vector;

[0041] Mapping the global feature vector to a high-dimensional feature space through a first fully connected layer and a second fully connected layer to obtain a discrimination boundary between the wavefront and the distortion, and outputting the distortion results of each sampling point in the key mode based on the output layer.

[0042] In this way, through the distortion removal model, the distortion conditions of each sampling point in the key mode can be accurately identified, and the distorted sections in the first traveling wave signal can be filtered accordingly, effectively solving the problem of inaccurate wavefront positioning caused by distorted waveform interference in the prior art.

[0043] Further, determining the traveling wavefront mutation point based on the second traveling wave signal specifically includes:

[0044] Performing a first-order difference operation on the second traveling wave signal to obtain a first difference sequence, and performing a smoothing process on the first difference sequence to obtain a second difference sequence;

[0045] Screen the second difference sequence based on a dynamic threshold to obtain candidate wavefront mutation points, and further screen the candidate wavefront mutation points based on a local threshold to determine the traveling wavefront mutation points, where the dynamic threshold is determined according to the mean and standard deviation of the second difference sequence, and the local threshold is determined according to the maximum value in the neighborhood of the second difference sequence.

[0046] By processing the second traveling wave signal in this way, the mutation points of the traveling wavefront can be accurately determined, so as to accurately calculate the fault location.

[0047] In a second aspect, the present application provides a traveling wavefront fault location system based on modal decomposition and neural network, including: an acquisition module, a decomposition module, a rejection module, and a location module;

[0048] The acquisition module is configured to acquire a first traveling wave signal on a transmission line and extract waveform features corresponding to the first traveling wave signal;

[0049] The decomposition module is configured to input the waveform features into a preset parameter optimization model to determine corresponding optimal decomposition parameters, and perform multi-scale variational mode decomposition on the first traveling wave signal by using the optimal decomposition parameters to obtain a plurality of initial modes, and determine corresponding key modes based on each of the initial modes, where the parameter optimization model is trained by using optimal waveform parameters, and the optimal waveform parameters are determined after parameter optimization on a traveling wave waveform data set;

[0050] The rejection module is configured to input the key modes into a preset distortion rejection model to determine the distortion results of each sampling point in the key modes, and filter out the distorted sections of the first traveling wave signal based on the distortion results and the key modes to obtain a second traveling wave signal;

[0051] The location module is configured to determine the traveling wavefront mutation points based on the second traveling wave signal and determine the corresponding fault locations based on the traveling wavefront mutation points.

[0052] In the embodiment of the present application, by extracting the waveform features corresponding to the first traveling wave signal, comprehensive signal information can be provided for the subsequent parameter optimization model, which helps to accurately determine the optimal decomposition parameters; through the parameter optimization model, the optimal decomposition parameters can be automatically and quickly determined according to the input waveform features, solving the problem that parameters are difficult to adaptively adjust in the prior art; using the optimal decomposition parameters to perform multi-scale variational mode decomposition on the traveling wave signal can effectively separate different frequency components in the signal, avoiding the mode mixing phenomenon, facilitating the subsequent rapid and accurate identification of the mutation points of the traveling wave head, and improving the accuracy of wave head positioning; based on each of the initial modes to determine the corresponding key modes can focus more on the key modes containing the information of the wave head mutation points, providing a more accurate signal basis for subsequent distortion rejection and fault location; through the distortion rejection model, the distortion conditions of each sampling point in the key mode can be accurately identified, and accordingly the distorted sections in the first traveling wave signal can be filtered out, effectively solving the problem of inaccurate wave head positioning caused by distorted waveform interference in the prior art; by processing the second traveling wave signal, the mutation points of the traveling wave head can be accurately determined, thereby accurately calculating the fault location. Compared with the prior art, the present application solves the problems existing in the prior art, such as positioning being interfered by distorted waveforms, mode mixing, and inability to achieve parameter adaptive adjustment, through mode decomposition and neural networks, improving the accuracy of traveling wave head fault location. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 FIG. is a schematic flow chart of an embodiment of a traveling wave head fault location method based on mode decomposition and neural network provided by the present application;

[0054] Figure 2 FIG. is a schematic diagram of the model structure of the parameter optimization model provided by the present application;

[0055] Figure 3 FIG. is a schematic diagram of the model structure of the distortion rejection model provided by the present application;

[0056] Figure 4 FIG. is a schematic diagram of the structure of another embodiment of a traveling wave head fault location system based on mode decomposition and neural network provided by the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0058] It should be understood that the step numbers used in the text are only for convenience of description and do not limit the order of execution of the steps.

[0059] It should be understood that the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0060] The terms "comprising" and "including" indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their combinations.

[0061] The term "and / or" refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0062] In the operation of the power grid, external factors such as lightning strikes, wildfires, icing, and tree obstacles often cause faults and tripping of transmission and distribution lines, thus threatening the safety of power grid operation and power supply reliability. Therefore, accurately locating the fault location is crucial for ensuring the stable operation of the power grid. At present, the industry mainly uses methods such as wavelet transform and wavelet packet transform to achieve accurate positioning of the traveling wave head. However, these methods have defects: the high-frequency distortion signals generated by the arc discharge before the fault will overlap with the wave head signals, thus affecting the fault location accuracy; modal aliasing is prone to occur during the decomposition process of wavelet transform, thus masking the mutation characteristics of the wave head and leading to an increase in the positioning error.

[0063] Next, the nouns involved in this application are analyzed:

[0064] The multi-scale variational mode decomposition algorithm (Multi-scale Variational Modal Decomposition), abbreviated as MSVMD, is an advanced signal processing technology. It is improved and extended on the basis of the traditional variational mode decomposition (VMD) algorithm to meet the needs of more complex signal analysis. The core idea is to introduce multi-scale analysis, decompose and process the signal at different scales, and decompose the complex signal into multiple intrinsic mode functions. Each intrinsic mode function has narrowband frequency characteristics, and the frequency distributions between the modes do not overlap with each other, avoiding the problem of modal aliasing.

[0065] Based on this, the embodiments of this application provide a traveling wave head fault location method and system based on modal decomposition and neural network, which solve the problems existing in the prior art such as the positioning being interfered by distorted waveforms, modal aliasing, and the inability to achieve parameter adaptive adjustment through modal decomposition and neural network, and improve the accuracy of traveling wave head fault location.

[0066] The traveling wave head fault location method and system based on modal decomposition and neural network provided by the embodiments of the present application will be specifically described through the following embodiments. First, the traveling wave head fault location based on modal decomposition and neural network in the embodiments of the present application will be described.

[0067] The traveling wave head fault location based on modal decomposition and neural network provided by the embodiments of the present application relates to the field of fault location. The traveling wave head fault location based on modal decomposition and neural network provided by the embodiments of the present application can be applied to terminals, can also be applied to server sides, or can also be software running on terminals or server sides. In some embodiments, the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, etc.; the server side can be configured as an independent physical server, can also be configured as a server cluster or distributed system composed of multiple physical servers, or can also be configured as a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the traveling wave head fault location based on modal decomposition and neural network, etc., but is not limited to the above forms.

[0068] The present application can also be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, small computers, large computers, distributed computing environments including any of the above systems or devices, and so on. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in a distributed computing environment, where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0069] Embodiment 1

[0070] Please refer to Figure 1 , Figure 1 , which is a schematic flow chart of an embodiment of the traveling wave head fault location method based on modal decomposition and neural network provided by the present application, including steps S101 to S104;

[0071] Step S101, obtain the first traveling wave signal on the transmission line, and extract the waveform features corresponding to the first traveling wave signal;

[0072] In some embodiments, obtaining the first traveling wave signal on the transmission line specifically includes: installing traveling wave signal sensors such as high-precision current transformers (CTs) and potential transformers (PTs) on the transmission line to obtain the current traveling wave signal and voltage traveling wave signal on the transmission line in real time, that is, the first traveling wave signal. It should be noted that the first traveling wave signal is the transient change of electrical quantities caused by a fault point on the transmission line.

[0073] In some embodiments, extracting the waveform features corresponding to the first traveling wave signal specifically means performing feature extraction on the preprocessed first traveling wave signal to obtain waveform features, where the waveform features include time domain features, frequency domain features, time-frequency features, and entropy features.

[0074] It should be noted that the time-frequency features mainly focus on the statistical features of the signal on the time axis and are suitable for describing the overall characteristics of the signal, including but not limited to: signal mean, variance, maximum value, minimum value, peak position, etc.; the frequency domain features mainly analyze the frequency components and distribution of the signal by converting the signal from the time domain to the frequency domain, including but not limited to main frequency, bandwidth, high-frequency energy ratio, the number of main peaks in the spectrum, etc.; the time-frequency features are the time-frequency distribution energy extracted by short-time Fourier transform; the entropy features include sample entropy and energy entropy.

[0075] It should be noted that the specific method of feature extraction is not the focus of this application, so it will not be elaborated here.

[0076] Step S102: Input the waveform features into a preset parameter optimization model to determine the corresponding optimal decomposition parameters, and use the optimal decomposition parameters to perform multi-scale variational mode decomposition on the first traveling wave signal to obtain several initial modes, and determine the corresponding key modes based on each initial mode, where the parameter optimization model is trained using optimal waveform parameters, and the optimal waveform parameters are determined by performing parameter optimization on the traveling wave waveform dataset;

[0077] In some embodiments, before inputting the waveform features into a preset parameter optimization model, it is necessary to train the parameter optimization model, and before training, it is necessary to first perform parameter optimization on the traveling wave waveform dataset to obtain optimal waveform parameters, and then train the parameter optimization model based on the optimal waveform parameters.

[0078] It should be noted that the optimal waveform parameters include several key parameters such as the number of modes K, the bandwidth penalty parameter α, the modal center frequency ω k and the noise tolerance τ. The number of modes K is used to specify how many narrowband frequency components the signal is decomposed into; the bandwidth penalty parameter α is used to control the width of the modal bandwidth; the modal center frequency ωk is the initial center frequency of each of the modes; the noise tolerance τ is used to balance the noise suppression ability and the decomposition speed of the decomposition. Generally speaking, if the frequency components of the traveling wave waveform dataset x(t) are few, the value of the mode number K should be small; if the traveling wave waveform dataset x(t) contains complex frequency characteristics, the value of the mode number K should be large. If the value of the bandwidth penalty parameter α is larger, the bandwidth of each mode is narrower and the decomposition result is smoother; if the value of the bandwidth penalty parameter α is smaller, the bandwidth of the mode is wider and the decomposition result is rougher;

[0079] In some embodiments, parameter optimization is performed on the traveling wave waveform dataset to obtain optimal waveform parameters, including: performing multi-scale variational mode decomposition on the traveling wave waveform dataset to obtain a second decomposition result, and determining the corresponding mode number based on the traveling wave waveform dataset and the second decomposition result; based on the mode number and the second decomposition result, performing iterative optimization with the goal of minimizing the bandwidth objective function to determine the corresponding bandwidth penalty parameter; calculating the main frequency peak of the traveling wave waveform dataset, using the main frequency peak as the initial center frequency, and determining the spectral energy density of each decomposition based on the initial center frequency and the bandwidth penalty parameter, and performing iterative optimization based on the spectral energy density to determine the mode center frequency; determining the corresponding bandwidth based on the mode center frequency, determining the frequency band overlap degree of adjacent modes in the variational mode decomposition process based on the bandwidth, and performing iterative optimization based on the frequency band overlap degree to determine the noise tolerance.

[0080] In some embodiments, the process of determining the corresponding mode number K is specifically as follows: First, initialize the traversal range of the mode number K; second, for each mode number K, use the multi-scale variational mode decomposition algorithm to decompose the traveling wave waveform dataset x(t) to obtain K intrinsic mode functions, that is, the second decomposition result μ k (t). After that, based on the traveling wave waveform dataset x(t) and the second decomposition result μ k (t), calculate the signal reconstruction error E. The calculation formula of the signal reconstruction error E is: In the formula, x(t) is the traveling wave waveform dataset, μ k (t) is the second decomposition result, and K is the mode number; finally, traverse the value range of the mode number K to find the optimal mode number K corresponding to the minimum signal reconstruction error E.

[0081] In some embodiments, the process of determining the corresponding bandwidth penalty parameter α is specifically as follows: When the calculated mode number K and the second decomposition result μ kAfter (t), traverse the value range of the bandwidth penalty parameter α. For each bandwidth penalty parameter α, calculate the minimized bandwidth objective function J(α). The calculation formula for the minimized bandwidth objective function J(α) is as follows: In the formula, K is the number of the modes, μ k (t) is the second decomposition result, Overlap(ω k ) is the overlapping degree of frequencies between each mode, ω k is the center frequency of each mode, which is used to avoid mode aliasing. After that, find the optimal bandwidth penalty parameter α corresponding to the minimum of the minimized bandwidth objective function J(α) according to the above formula.

[0082] In some embodiments, the process of determining the center frequency ω k of the mode is specifically as follows: Obtain the main frequency peak of the traveling wave waveform dataset x(t) through fast Fourier transform or wavelet transform. After that, use the main frequency peak as the initial center frequency ω of the second decomposition result μ k (t). Determine the spectral energy density U k (ω) of each decomposition based on the initial center frequency ω and the bandwidth penalty parameter α. Finally, perform iterative optimization based on the spectral energy density U k (ω) to update the center frequency ω k of the mode until it no longer changes significantly, then determine the optimal center frequency ω k .

[0083] In some embodiments, the calculation formula for performing iterative optimization based on the spectral energy density U k (ω) to determine the center frequency ω k of the mode is specifically as follows:

[0084]

[0085] In the formula, ω k is the center frequency of the mode, U k (ω) is the spectral energy density, F(ω) is the spectral function obtained by performing Fourier transform or wavelet transform on the traveling wave waveform dataset, Λ(ω) is the Lagrange multiplier term, α is the bandwidth penalty parameter, i is the number of iterations, k is the kth mode, and ω is the initial center frequency.

[0086] It should be noted that when optimizing the bandwidth penalty parameter α, the center frequency ω k of the mode needs to be temporarily fixed, and when optimizing the center frequency ω k of the mode, the bandwidth penalty parameter α needs to be temporarily fixed. Iterate repeatedly in this way to obtain the optimal α and ω k .

[0087] In some embodiments, the process of determining the noise tolerance τ is specifically as follows: When determining the modal center frequency ω k after that, it is necessary to determine the corresponding bandwidth B based on the modal center frequency ω k Then, based on the bandwidth B k the frequency band overlap degree δ between adjacent modes in the variational mode decomposition process is determined k The calculation formula of the frequency band overlap degree δ is as follows: k where δ k In the formula, B and B k are the bandwidths of the k-th and (k + 1)-th modal center frequencies respectively, and f k+1 and f k are the center frequencies of the k-th and (k + 1)-th modes respectively; finally, based on the frequency band overlap degree δ k+1 iterative optimization is performed to determine the optimal noise tolerance τ k where the formula for iterative optimization is: τ new = τ·(1 + β·δ new ), in the formula, τ max is the optimal noise tolerance, β is a preset coefficient that can be custom-set, and δ new is the maximum frequency band overlap degree, and δ max = max(δ max ). k )

[0088] It should be noted that both f k and ω k are center frequencies, but the unit of f k is Hz, and the unit of ω k is rad / s.

[0089] In this way, by optimizing the parameters of the traveling wave waveform dataset, it is convenient to quickly and accurately train the parameter optimization model subsequently, thereby improving the accuracy of wavefront positioning.

[0090] In some embodiments, since the above optimization process is relatively slow and not suitable for the scenario of precise traveling wavefront positioning with high timeliness requirements, a parameter optimization model is trained using the optimal waveform parameters. Among them, the training process of the parameter optimization model is specifically as follows: The traveling wave waveform dataset is input into a preset neural network model, and the modal parameter prediction values are obtained through forward propagation calculation; the mean square error loss value between the parameter prediction values and the optimal waveform parameters is calculated, and backpropagation is performed through the Adam algorithm to update the weights of the neural network model according to the loss value until the loss value meets the preset conditions, then the optimal weights are determined, and the parameter optimization model is determined based on the optimal weights.

[0091] It should be noted that during the model training process, it is also necessary to evaluate the model prediction error of the parameter optimization model on the validation set. When the validation loss has not decreased for 10 consecutive epochs, the early stopping mechanism is triggered to save the best weights. Finally, when the accuracy of the number of modes K of the parameter optimization model on the test set is ≥ 95% and the mean square error of the penalty factor α is ≤ 0.1, the training is stopped to obtain the trained parameter optimization model. The schematic diagram of the model structure of the parameter optimization model is as shown in Figure 2 shown, which consists of an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer. Among them, the structures of the 3 hidden layers are the same, and the number of their neurons is 128. The ReLU activation function is used, and the Dropout parameter is set to 0.2. The dimension of the output layer is 4 (the number of modes K, the bandwidth penalty parameter α, the modal center frequency ωk, and the noise tolerance τ), and the linear activation function is used.

[0092] In this way, by training the parameter optimization model, it can automatically and quickly determine the optimal decomposition parameters according to the input waveform features, solving the problem that the parameters in the prior art are difficult to adaptively adjust.

[0093] In some embodiments, after training the parameter optimization model, it is necessary to input the waveform features into the preset parameter optimization model, and based on its model parameters, the optimal decomposition parameters of the multi-scale variational mode decomposition can be determined instantly.

[0094] In some embodiments, the multi-scale variational mode decomposition of the first traveling wave signal using the optimal decomposition parameters to obtain several initial modes includes: performing signal preprocessing on the first traveling wave signal to obtain a third traveling wave signal; in each iteration process, based on the number of modes, the bandwidth penalty parameter, and the modal center frequency, performing iterative optimization decomposition on the third traveling wave signal, and adjusting the noise suppression ability of the decomposition process according to the noise tolerance until the update amount of the iterative optimization decomposition is less than a preset threshold, then determining the corresponding initial modes based on the obtained first decomposition result. Specifically, first, after obtaining the first traveling wave signal, it is necessary to perform preprocessing operations such as filtering, normalization, and denoising on the first traveling wave signal to obtain a third traveling wave signal; secondly, use the number of modes K, the bandwidth penalty parameter α, and the modal center frequency ω in the optimal decomposition parameters k to perform multi-scale variational mode decomposition on the third traveling wave signal, and in each iteration, adjust the noise suppression ability of the decomposition process according to the noise tolerance τ. If the update amount of the iterative optimization decomposition is less than a preset threshold, it is considered that the decomposition converges. When the convergence condition is satisfied, the iteration is stopped to obtain the first decomposition result. Finally, several initial modes are extracted from the first decomposition result.

[0095] It should be noted that signal preprocessing can improve the signal quality and ensure the accuracy of subsequent analysis. However, signal preprocessing is not the focus of this application, so it will not be elaborated here.

[0096] By using the optimal decomposition parameters to perform multi-scale variational mode decomposition on the traveling wave signal, different frequency components in the signal can be effectively separated, avoiding the mode mixing phenomenon, facilitating the subsequent rapid and accurate identification of the mutation points of the traveling wave head, and improving the accuracy of wave head positioning.

[0097] In some embodiments, determining the corresponding key mode based on each of the initial modes specifically includes: calculating the high-frequency energy ratio E k of each of the initial modes, the time-domain variance V k and the sample entropy H k ; based on the high-frequency energy ratio E k , the time-domain variance V k and the sample entropy H k , scoring each of the initial modes, and taking the initial mode with the highest score as the key mode.

[0098] In some embodiments, the relevant formula for determining the corresponding key mode based on each of the initial modes specifically is:

[0099] Score(IMF k ) = ω1·E k + ω2·V k - ω3·H k ;

[0100]

[0101] In the formula, is the score of the kth initial mode, ω1, ω2, and ω3 are the weight coefficients of each characteristic respectively, E k is the high-frequency energy ratio of the kth initial mode, V k is the time-domain variance of the kth initial mode, H k is the sample entropy of the kth initial mode, ω max and ω min are the upper and lower limits of the frequency range respectively, ω high is the starting point of the high-frequency range, U k (ω) is the spectral energy density, T is the duration of the initial mode, μ k is the average value of the initial mode, M is the total number of sampling points of the initial mode, and P i is the probability distribution of the ith sample.

[0102] Determining the corresponding key modes based on each of the initial modes can focus more on the key modes containing the information of the wavefront mutation points, providing a more accurate signal basis for subsequent distortion rejection and fault location.

[0103] Step S103: Input the key mode into a preset distortion rejection model, determine the distortion results of each sampling point in the key mode, and filter out the distorted section of the first traveling wave signal based on the distortion results and the key mode to obtain a second traveling wave signal.

[0104] It should be noted that due to the existence of arc discharge before the fault trip, and there is partial overlap between the distorted waveform characteristics and the wavefront characteristics caused by arc discharge, so the distorted waveform will also be transmitted to the key mode. Therefore, it is necessary to input the key mode into the distortion rejection model to remove the distorted part, so as to improve the extraction accuracy of the wavefront mutation point, and further effectively improve the wavefront positioning accuracy.

[0105] In some embodiments, the specific operation of inputting the key mode into a preset distortion rejection model and determining the distortion results of each sampling point in the key mode is as follows: Input the key mode into the first convolutional layer and the first pooling layer of the distortion rejection model to obtain a first output feature; input the first output feature into the second convolutional layer and the second pooling layer to obtain a second output feature; input the second output feature into the third convolutional layer and the third pooling layer to obtain a third output feature, and input the third output feature into the global pooling layer to obtain a global feature vector; map the global feature vector to a high-dimensional feature space through the first fully connected layer and the second fully connected layer to obtain the discrimination boundary between the wavefront and the distortion, and output the distortion results of each sampling point in the key mode based on the output layer.

[0106] It should be noted that the schematic diagram of the model structure of the distortion rejection model is as Figure 3 shown. Among them, the distortion rejection model consists of 1 input layer, 3 convolutional layers + max pooling layers (the first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer, the third convolutional layer and the third pooling layer), 1 global pooling layer (global pooling layer), 2 fully connected layers (the first fully connected layer and the second fully connected layer) and 1 output layer. The Adam optimizer is selected during the training process of the distortion rejection model, the initial learning rate is set to 1×10-3, and the binary cross-entropy loss is used as the loss function of the network to train and obtain.

[0107] It should be noted that the first convolutional layer is mainly used for local feature extraction, the second and third convolutional layers are mainly used for deep pattern mining, and the first and second fully connected layers are mainly used for high-dimensional mapping of global features. The convolutional kernel size of the first convolutional layer is 5, the stride is 1, and the number of output channels is 64. The ReLU activation function is used. Through 64 convolutional kernels with a width of 5, it slides along the time axis to capture the high-frequency oscillation features of the wavefront mutation points and the low-frequency disturbance patterns of the arc distortion, and outputs a feature map with a dimension of 996×64. The pooling size of the first pooling layer is 2, the stride is 2, and the max pooling method is used. By compressing the 996×64 feature map into a 498×64 feature map, it retains the local extrema and suppresses small-amplitude noise to obtain the first output feature. The convolutional kernel sizes of the second and third convolutional layers are 3, and the numbers of output channels are 128 and 256 respectively. The other parameters are the same as those of the first convolutional layer. The second convolutional layer further extracts multi-scale time-frequency features through 128 convolutional kernels with a width of 3, identifies the differences in the frequency band energy distribution between the wavefront and the distorted waveform, and outputs a feature map with a dimension of 498×128. The parameters of the second pooling layer are the same as those of the first pooling layer. By compressing the 498×128 feature map to 248×128, it enhances the feature translational invariance. The third convolutional layer fuses cross-layer features through 256 convolutional kernels with a width of 3, establishes the association pattern between the wavefront mutation points and the context waveforms, and outputs a feature map with a dimension of 246×256. The parameters of the third pooling layer are the same as those of the first pooling layer, and compress the 246×256 feature map into a 123×256 feature map. The global pooling layer takes the maximum value of each channel along the time axis to generate a 256-dimensional global feature vector, which characterizes the overall distortion degree of the signal. The first fully connected layer contains 256 neurons, uses the ReLU activation function, and the Dropout parameter is set to 0.5, mapping the 256-dimensional global features to a 256-node high-dimensional space. The second fully connected layer condenses the global features to 128-dimensional features, encoding the discrimination boundary between the wavefront and the distortion. The output layer restores the original signal length through time interpolation, outputs the distortion probability point by point, and performs binarization processing using a threshold of 0.5 to obtain the distortion discrimination results of each point of the key mode signal vector (0 indicates that the sampling point is in the distorted section, and 1 indicates that the sampling point is not in the distorted section). This distortion discrimination result can be transferred to the original traveling wave waveform.

[0108] In this way, through the distortion rejection model, the distortion conditions of each sampling point in the key mode can be accurately identified, and the distorted sections in the first traveling wave signal can be filtered accordingly, effectively solving the problem of inaccurate wavefront positioning caused by the interference of distorted waveforms in the prior art.

[0109] In some embodiments, based on the distortion result and the key mode, the distorted section of the first traveling wave signal is filtered to obtain a second traveling wave signal. Specifically, the key mode signal vectors corresponding to each waveform data and the distortion discrimination results of each sampling point are input into a convolutional neural network for training. The trained neural network model can be used to filter the distorted section to obtain the second traveling wave result. The specific training process is not the focus of this application, so it will not be elaborated here.

[0110] Step S104: Determine the traveling wave head mutation point based on the second traveling wave signal, and determine the corresponding fault location based on the traveling wave head mutation point.

[0111] In some embodiments, the determining the traveling wave head mutation point based on the second traveling wave signal is specifically as follows: perform a first-order difference operation on the second traveling wave signal to obtain a first difference sequence, and perform a smoothing process on the first difference sequence to obtain a second difference sequence; screen the second difference sequence based on a dynamic threshold to obtain candidate wave head mutation points, and further screen the candidate wave head mutation points based on a local threshold to determine the traveling wave head mutation point, where the dynamic threshold is determined according to the mean and standard deviation of the second difference sequence, and the local threshold is determined according to the maximum value in the neighborhood of the second difference sequence. Specifically, first, perform a first-order difference operation on the second traveling wave signal to highlight the mutation characteristics of the wave head mutation point. The formula is Δx(t)′ = |x(t + 1)′ - x(t)′| (t = 0, 1,..., N - 1), where Δx(t)′ is the first difference sequence, N is the total length of the area to be calculated of the second traveling wave signal, and x(t) is the second traveling wave signal; secondly, use a sliding mean filter to perform a smoothing process on the difference sequence to suppress the pseudo-gradient fluctuation caused by random noise to determine the second difference sequence; then, determine the dynamic threshold by calculating the mean and standard deviation of the second difference sequence. The calculation formula of the dynamic threshold is: T h = μ Δ + 3σ Δ , where T h is the dynamic threshold, μ Δ is the mean of the second difference sequence, σ Δ is the standard deviation of the second difference sequence, and extract all candidate point sets C = {t1, t2,..., tm} greater than the dynamic threshold T h to obtain a candidate pool that forms a preliminary wave head mutation point area, that is, candidate wave head mutation points. Finally, intercept the difference signal segment of the neighborhood [ti - 5, ti + 5] of each candidate point ti in the candidate wave head mutation points, and use the following expression to calculate the local threshold: T2 = max(Δx′(t i - 5: t i((T2 + 5)) × 0.8), where T2 is the local threshold and Δx′ is the second difference sequence; and candidate wavefront mutation points greater than the local threshold are screened, and the continuously occurring candidate wavefront mutation points are merged, and the candidate wavefront mutation point with the earliest time is retained, that is, the traveling wave front mutation point.

[0112] In this way, by processing the second traveling wave signal, the mutation point of the traveling wave front can be accurately determined, so as to accurately calculate the fault location.

[0113] In some embodiments, based on the traveling wave front mutation point, the corresponding fault location is determined. Specifically, when the traveling wave front mutation point is determined, by using the time information of the wave front mutation point and combining the traveling wave propagation speed and line parameters, the fault location can be calculated. Exemplarily, since the propagation speed V of the traveling wave on the transmission line is known, assuming the time of the wave front mutation point is tp, the fault location L can be calculated by the following formula: L = v × tp.

[0114] In the embodiment of the present application, by extracting the waveform features corresponding to the first traveling wave signal, comprehensive signal information can be provided for the subsequent parameter optimization model, which helps to accurately determine the optimal decomposition parameters; through the parameter optimization model, the optimal decomposition parameters can be automatically and quickly determined according to the input waveform features, solving the problem that the parameters are difficult to adaptively adjust in the prior art; using the optimal decomposition parameters to perform multi-scale variational mode decomposition on the traveling wave signal can effectively separate different frequency components in the signal, avoiding the mode mixing phenomenon, facilitating the subsequent rapid and accurate identification of the mutation points of the traveling wave front, and improving the accuracy of wave front positioning; based on each of the initial modes, the corresponding key modes are determined, which can focus more on the key modes containing the information of the wave front mutation points, providing a more accurate signal basis for subsequent distortion rejection and fault location; through the distortion rejection model, the distortion conditions of each sampling point in the key mode can be accurately identified, and accordingly, the distorted sections in the first traveling wave signal can be filtered out, effectively solving the problem of inaccurate wave front positioning caused by distorted waveform interference in the prior art; by processing the second traveling wave signal, the mutation point of the traveling wave front can be accurately determined, so as to accurately calculate the fault location. Compared with the prior art, the present application solves the problems existing in the prior art such as positioning being interfered by distorted waveforms, mode mixing, and inability to achieve parameter adaptive adjustment through mode decomposition and neural networks, and improves the accuracy of traveling wave front fault location.

[0115] Embodiment 2

[0116] Please refer to Figure 4 , Figure 4 which is a schematic structural diagram of another embodiment of the traveling wave front fault location system based on mode decomposition and neural network provided by the present application, including: an acquisition module 100, a decomposition module 200, a rejection module 300, and a location module 400;

[0117] The obtaining module 100 is configured to obtain a first traveling wave signal on a transmission line and extract waveform features corresponding to the first traveling wave signal;

[0118] The decomposition module 200 is configured to input the waveform features into a preset parameter optimization model to determine corresponding optimal decomposition parameters, and perform multi-scale variational mode decomposition on the first traveling wave signal by using the optimal decomposition parameters to obtain a plurality of initial modes, and determine corresponding key modes based on each of the initial modes, wherein the parameter optimization model is trained by using optimal waveform parameters, and the optimal waveform parameters are determined by performing parameter optimization on a traveling wave waveform data set;

[0119] The rejection module 300 is configured to input the key modes into a preset distortion rejection model to determine distortion results of each sampling point in the key modes, and filter out the distorted sections of the first traveling wave signal based on the distortion results and the key modes to obtain a second traveling wave signal;

[0120] The positioning module 400 is configured to determine a traveling wave head mutation point based on the second traveling wave signal and determine a corresponding fault location based on the traveling wave head mutation point.

[0121] For the information interaction, execution process, etc. between the modules in the above traveling wave head fault location system based on mode decomposition and neural network, since they are based on the same concept as the embodiments of the traveling wave head fault location based on mode decomposition and neural network in the first aspect of the present invention, the achieved technical effects are basically the same. For specific content, reference can be made to the description in Embodiment 1 of the method of the present invention, and details are not repeated here.

[0122] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separated, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the method of this embodiment.

[0123] Those of ordinary skill in the art can understand that all or part of the processes of implementing the above method embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above method embodiments. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0124] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present application. It should be understood that the above description is only for the specific embodiments of the present application and is not intended to limit the protection scope of the present application.

[0125] It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A traveling wave head fault location method based on modal decomposition and neural network, characterized in that Including: Obtain a first traveling wave signal on a transmission line and extract waveform features corresponding to the first traveling wave signal; Input the waveform features into a preset parameter optimization model to determine corresponding optimal decomposition parameters, and use the optimal decomposition parameters to perform multi-scale variational mode decomposition on the first traveling wave signal to obtain a number of initial modes, and determine corresponding key modes based on each of the initial modes. Among them, the parameter optimization model is trained using optimal waveform parameters, and the optimal waveform parameters are determined by performing parameter optimization on a traveling wave waveform dataset; Input the key modes into a preset distortion removal model to determine the distortion results of each sampling point in the key modes, and filter out the distorted sections of the first traveling wave signal based on the distortion results and the key modes to obtain a second traveling wave signal; Determine the traveling wave head mutation point based on the second traveling wave signal, and determine the corresponding fault location based on the traveling wave head mutation point.

2. The traveling wave head fault location method based on modal decomposition and neural network according to claim 1, characterized in that The optimal decomposition parameters include the number of modes, the bandwidth penalty parameter, the modal center frequency, and the noise tolerance. The using the optimal decomposition parameters to perform multi-scale variational mode decomposition on the first traveling wave signal to obtain a number of initial modes is specifically: Perform signal preprocessing on the first traveling wave signal to obtain a third traveling wave signal; In each iteration process, perform iterative optimization decomposition on the third traveling wave signal based on the number of modes, the bandwidth penalty parameter, and the modal center frequency, and adjust the noise suppression ability of the decomposition process according to the noise tolerance to obtain a first decomposition result. Until the update amount of the iterative optimization decomposition is less than a preset threshold, then determine the corresponding initial mode based on the obtained first decomposition result.

3. The traveling wave head fault location method based on modal decomposition and neural network according to claim 2, characterized in that, The optimal waveform parameters are determined by performing parameter optimization on a traveling wave waveform dataset, specifically: Perform multi-scale variational mode decomposition on the traveling wave waveform dataset to obtain a second decomposition result, and determine the corresponding number of modes based on the traveling wave waveform dataset and the second decomposition result; Based on the number of modes and the second decomposition result, perform iterative optimization with the goal of minimizing the bandwidth objective function to determine the corresponding bandwidth penalty parameter; Calculate the main frequency peak value of the traveling wave waveform dataset, use the main frequency peak value as the initial center frequency, and determine the spectral energy density of each decomposition based on the initial center frequency and the bandwidth penalty parameter, and perform iterative optimization based on the spectral energy density to determine the modal center frequency; Determine the corresponding bandwidth based on the modal center frequency, and determine the frequency band overlap degree of adjacent modes in the variational mode decomposition process based on the bandwidth, and perform iterative optimization based on the frequency band overlap degree to determine the noise tolerance.

4. The traveling wave head fault location method based on modal decomposition and neural network according to claim 3, characterized in that The formula for performing iterative optimization based on the spectral energy density to determine the modal center frequency is specifically: where ω k is the modal center frequency, U k (ω) is the spectral energy density, F(ω) is the spectral function obtained by Fourier transform or wavelet transform of the traveling wave waveform dataset, Λ(ω) is the Lagrange multiplier term, α is the bandwidth penalty parameter, i is the iteration number, k is the k-th mode, and ω is the initial center frequency.

5. The traveling wave head fault location method based on modal decomposition and neural network according to claim 1, characterized in that The training process of the parameter optimization model is specifically: Input the traveling wave waveform dataset into a preset neural network model, and calculate the predicted modal parameter values through forward propagation; Calculate the loss value between the predicted parameter value and the optimal waveform parameter, and perform backpropagation through the Adam algorithm to update the weights of the neural network model according to the loss value until the loss value meets the preset condition. Then, determine the optimal weights and determine the parameter optimization model based on the optimal weights.

6. The traveling wave head fault location method based on modal decomposition and neural network according to claim 1, characterized in that The determination of the corresponding key mode based on each of the initial modes is specifically as follows: Calculate the high-frequency energy ratio, time-domain variance, and sample entropy of each of the initial modes; Based on the high-frequency energy ratio, the time-domain variance, and the sample entropy, score each of the initial modes, and use the initial mode with the highest score as the key mode.

7. The traveling wave head fault location method based on modal decomposition and neural network according to claim 6, characterized in that, The relevant formula for determining the corresponding key mode based on each of the initial modes is specifically as follows: Score(IMF k ) = ω1·E k + ω2·V k - ω3·H k ; In the formula, is the score of the k-th initial mode, ω1, ω2, and ω3 are the weight coefficients of each characteristic, and E k is the high-frequency energy ratio of the k-th initial mode, and V k is the time-domain variance of the k-th initial mode, and H k is the sample entropy of the k-th initial mode, ω max and ω min are the upper and lower limits of the frequency range respectively, ω high is the starting point of the high-frequency range, U k (ω) is the spectral energy density, T is the duration of the initial mode, μ k is the average value of the initial mode, M is the total number of sampling points of the initial mode, and P i is the probability distribution of the i-th sample.

8. The traveling wave head fault location method based on modal decomposition and neural network according to claim 1, characterized in that Input the key mode into a preset distortion rejection model to determine the distortion results of each sampling point in the key mode, specifically as follows: Input the key mode into the first convolutional layer and the first pooling layer of the distortion rejection model to obtain a first output feature; Input the first output feature into the second convolutional layer and the second pooling layer to obtain a second output feature; Input the second output feature into the third convolutional layer and the third pooling layer to obtain a third output feature, and input the third output feature into the global pooling layer to obtain a global feature vector; Map the global feature vector to a high-dimensional feature space through a first fully connected layer and a second fully connected layer to obtain the discrimination boundary between the wavefront and distortion, and output the distortion results of each sampling point in the key mode based on the output layer.

9. The traveling wave head fault location method based on modal decomposition and neural network according to claim 1, characterized in that The determination of the traveling wavefront mutation point based on the second traveling wave signal is specifically as follows: Perform a first-order difference operation on the second traveling wave signal to obtain a first difference sequence, and smooth the first difference sequence to obtain a second difference sequence; Screen the second difference sequence based on a dynamic threshold to obtain candidate wavefront mutation points, and further screen the candidate wavefront mutation points based on a local threshold to determine the traveling wavefront mutation point, where the dynamic threshold is determined according to the mean and standard deviation of the second difference sequence, and the local threshold is determined according to the maximum value in the neighborhood of the second difference sequence.

10. A traveling wave head fault location system based on modal decomposition and neural network, characterized in that, It includes: An acquisition module, a decomposition module, a rejection module, and a positioning module; The acquisition module is used to acquire the first traveling wave signal on the transmission line and extract the waveform features corresponding to the first traveling wave signal; The decomposition module is used to input the waveform features into a preset parameter optimization model to determine the corresponding optimal decomposition parameters, and perform multi-scale variational mode decomposition on the first traveling wave signal using the optimal decomposition parameters to obtain a plurality of initial modes, and determine the corresponding key mode based on each of the initial modes, where the parameter optimization model is trained using optimal waveform parameters, and the optimal waveform parameters are determined by performing parameter optimization on a traveling wave waveform dataset; The rejection module is used to input the key mode into a preset distortion rejection model to determine the distortion results of each sampling point in the key mode, and filter out the distorted sections of the first traveling wave signal based on the distortion results and the key mode to obtain a second traveling wave signal; The positioning module is configured to determine the traveling wave head mutation point based on the second traveling wave signal, and determine the corresponding fault location based on the traveling wave head mutation point.

Images