Fault type detection method and device for overhead ground wire, and readable storage medium

Abstract

The invention discloses a fault type detection method and device for an overhead ground wire and a readable storage medium. The method comprises the following steps: acquiring a vibration response signal of the overhead ground wire; extracting fault features of the overhead ground wire from the vibration response signal; the fault features are input into a target detection model for analysis, multiple fault distribution probabilities of the overhead ground wire are obtained, the target detection model is obtained by training a fault feature sample corresponding to a vibration response signal sample of the overhead ground wire, and the fault distribution probabilities of the overhead ground wire are obtained; the fault distribution probability is used for indicating the probability of occurrence of faults of different fault types of the overhead ground wire; and determining a target fault type of the overhead ground wire based on the plurality of fault distribution probabilities. According to the invention, the technical problem of low efficiency of fault detection of the overhead ground wire is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of structural state identification of overhead power transmission lines, and in particular to a method, a device and a readable storage medium for detecting the fault type of an overhead ground wire. Background Art

[0002] At present, the power system generally uses high-voltage or ultra-high-voltage overhead lines for long-distance power distribution. This distribution method will expose the transmission lines to high-altitude environments for a long time, and they are easily affected by the natural environment, such as electric shocks, acid rain, etc., which will cause corrosion, wear and even breakage of the lines. As an important protection line in the transmission line, the overhead ground wire will also break due to environmental factors. If it is not handled in time, it is easy to cause large-scale power outages, resulting in huge economic losses.

[0003] In the related technologies, a large amount of manpower and material resources are invested in regular and frequent inspections of overhead ground wires to ensure the safety of the power system. However, the inspection cycle is long and the accuracy is poor, making it difficult to troubleshoot faults in a timely manner, especially in some areas with harsh environments, where the structural health status of the overhead ground wire cannot be obtained in a timely and effective manner. Minor local damage to the overhead ground wire will eventually develop into a serious accident due to long-term breeze vibration, causing many problems in people's lives and the economy. Therefore, there is a technical problem of low efficiency in fault detection of overhead ground wires.

[0004] With respect to the above-mentioned technical problem of low efficiency in fault detection of overhead ground wires, no effective solution has been proposed so far. Summary of the invention

[0005] The embodiments of the present invention provide a method, a device and a readable storage medium for detecting the fault type of an overhead ground wire, so as to at least solve the technical problem of low efficiency in fault detection of the overhead ground wire.

[0006] According to one aspect of an embodiment of the present invention, a method for detecting the fault type of an overhead ground wire is provided. The method may include: obtaining a vibration response signal of the overhead ground wire; extracting the fault characteristics of the overhead ground wire from the vibration response signal; inputting the fault characteristics into a target detection model for analysis to obtain multiple fault distribution probabilities of the overhead ground wire, wherein the target detection model is trained using fault feature samples corresponding to the vibration response signal samples of the overhead ground wire, and the fault distribution probability is used to indicate the probability of occurrence of different fault types of the overhead ground wire; based on the multiple fault distribution probabilities, determining the target fault type of the overhead ground wire.

[0007] Optionally, extracting fault characteristics of the overhead ground wire from the vibration response signal includes: decomposing the vibration response signal into multiple modal components, wherein the modal components are used to indicate the contribution of different frequencies and vibration modes to the vibration response signal; and determining the fault characteristics of the overhead ground wire based on the multiple modal components.

[0008] Optionally, based on multiple modal components, the fault characteristics of the overhead ground wire are determined, including: segmenting each modal component to determine the semantic relationship between the multiple modal components; based on the semantic relationship, determining the key mode, wherein the key mode is used to indicate representative information in different modal components; based on the key mode, determining the fault characteristics of the overhead ground wire.

[0009] Optionally, the method for detecting the fault type of the overhead ground wire further includes: training an initial detection model using fault feature samples of the overhead ground wire to obtain a target detection model.

[0010] Optionally, the initial detection model is trained using fault feature samples of the overhead ground wire to obtain a target detection model, including: multiple modal component samples extracted from the fault feature samples; multiple center frequencies are extracted from the multiple modal component samples, wherein the center frequency is used to indicate the dominant frequency of the modal component samples; based on the center frequency, the initial detection model is trained to obtain a target detection model.

[0011] Optionally, determining the fault type of the overhead ground wire based on multiple fault distribution probabilities includes: determining the fault type corresponding to the maximum fault distribution probability among the multiple fault distribution probabilities as the target fault type of the overhead ground wire.

[0012] According to another aspect of an embodiment of the present invention, a fault type detection device for an overhead ground wire is also provided. The device may include: an acquisition unit for acquiring a vibration response signal of the overhead ground wire; an extraction unit for extracting the fault characteristics of the overhead ground wire from the vibration response signal; an input unit for inputting the fault characteristics into a target detection model for analysis to obtain multiple fault distribution probabilities of the overhead ground wire, wherein the target detection model is trained using fault feature samples corresponding to the vibration response signal samples of the overhead ground wire, and the fault distribution probability is used to indicate the probability of occurrence of different fault types corresponding to the fault of the overhead ground wire; a determination unit for determining the target fault type of the overhead ground wire based on multiple fault distribution probabilities.

[0013] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is also provided, which includes a stored program, wherein when the program is executed by a processor, the device where the storage medium is located is controlled to execute the fault type detection method of the overhead ground line in an embodiment of the present invention.

[0014] According to another aspect of an embodiment of the present invention, a processor is provided, wherein the processor is used to run a program, wherein when the program is run, the method for detecting the fault type of an overhead ground wire in an embodiment of the present invention is executed.

[0015] According to another aspect of the embodiments of the present invention, a computer program product is provided, which includes computer instructions, and when the computer instructions are executed by a processor, the method for detecting the fault type of an overhead ground wire in the embodiments of the present invention is implemented.

[0016] In an embodiment of the present invention, a vibration response signal of an overhead ground wire is obtained; from the vibration response signal, a fault feature of the overhead ground wire is extracted; the fault feature is input into a target detection model for analysis to obtain multiple fault distribution probabilities of the overhead ground wire, wherein the target detection model is trained using fault feature samples corresponding to the vibration response signal samples of the overhead ground wire, and the fault distribution probability is used to indicate the probability of occurrence of different fault types of the overhead ground wire; based on the multiple fault distribution probabilities, the target fault type of the overhead ground wire is determined. That is to say, the present invention uses a target detection model to analyze the vibration response signal of the overhead ground wire, determines multiple fault distribution probabilities of the overhead ground wire, and thereby determines the target fault type of the overhead ground wire according to the fault distribution probability, thereby solving the technical problem of low efficiency in fault detection of the overhead ground wire and achieving the technical effect of improving the efficiency of fault detection of the overhead ground wire. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0018] Figure 1 is a flow chart of a method for detecting a fault type of an overhead ground wire according to an embodiment of the present invention;

[0019] Figure 2 is a flow chart of a method for detecting an overhead ground wire fault based on deep learning according to an embodiment of the present invention;

[0020] Figure 3 is a flow chart of a method for optimizing parameters of variational mode decomposition according to an embodiment of the present invention;

[0021] Figure 4 is a schematic diagram of an overhead ground wire fault detection model based on deep learning according to an embodiment of the present invention;

[0022] Figure 5 is a schematic diagram of an overhead ground wire fault detection classification result based on an improved modal attention Transforme deep learning model according to an embodiment of the present invention;

[0023] Figure 6 is a schematic diagram of a fault type detection device for an overhead ground wire according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, functional component or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, functional components or devices.

[0026] According to an embodiment of the present invention, an embodiment of a method for detecting the fault type of an overhead ground wire is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0027] Figure 1 FIG. 1 is a flow chart of a method for detecting a fault type of an overhead ground wire according to an embodiment of the present invention. Figure 1 As shown, the method may include the following steps:

[0028] Step S101, obtaining a vibration response signal of the overhead ground wire.

[0029] In the technical solution provided in the above step S101 of the present invention, the vibration response signal may include frequency, amplitude, waveform and spectrum.

[0030] In this embodiment, the vibration response signal of the overhead ground wire is obtained. For example, the vibration response signal of the overhead ground wire is obtained by a sensor. This is only an illustrative example and does not limit the specific method of obtaining the vibration response signal of the overhead ground wire.

[0031] Optionally, the vibration response signal of the overhead ground wire usually includes the following information: Frequency: The frequency of the vibration signal reflects the speed and frequency of the ground wire vibration, and vibration signals of different frequencies may have different degrees of impact on the ground wire.

[0032] Optionally, amplitude: the amplitude of the vibration signal can reflect the intensity of the ground wire vibration. The larger the amplitude, the stronger the vibration force on the ground wire.

[0033] Optionally, waveform: The waveform of the vibration signal will show the shape and characteristics of the ground wire vibration, and the type and cause of the ground wire vibration can be analyzed through the waveform.

[0034] Optionally, spectrum: The spectrum distribution of the vibration signal can provide more detailed frequency information to help analyze the frequency components and frequency distribution characteristics of ground wire vibration.

[0035] Step S102: extracting fault characteristics of the overhead ground wire from the vibration response signal.

[0036] In the technical solution provided in the above step S102 of the present invention, extracting the fault characteristics of the overhead ground wire can also be called decoupling fault characteristics.

[0037] In this embodiment, after the vibration response signal of the overhead ground wire is obtained in step S101, the fault characteristics of the overhead ground wire are extracted from the vibration response signal. For example, a preprocessing method is used to decouple the fault characteristics, which is only an illustrative example and does not limit the specific method of extracting the fault characteristics of the overhead ground wire.

[0038] For example, the fault characteristics are decoupled by the vibration response preprocessing method of the overhead ground wire based on variational mode decomposition. The parameters of the variational mode decomposition are optimized by the Sparrow Intelligent Search and Optimization Algorithm (SCSSA) combined with sine and cosine and Cauchy mutation. The vibration response signal of the overhead ground wire is adaptively decomposed by the variational mode decomposition (VMD) optimized by the Sparrow Intelligent Search and Optimization Algorithm (SCSSA), and the signal is decomposed into a set of modal components. According to the set of modal components obtained by decomposition, each modal component in this group is segmented, and a VMD-based modal encoder is used to capture the temporal characteristics of these modal components and map them to a new feature space using a learnable linear transformation. The modal internal Transformer encoder captures the contextual information of the input sequence through a multi-head self-attention mechanism, and further processes this information through a feedforward neural network. To extract deeper feature information, that is, fault characteristics.

[0039] Step S103: input the fault characteristics into the target detection model for analysis to obtain multiple fault distribution probabilities of the overhead ground wire.

[0040] In the technical solution provided in the above step S103 of the present invention, the target detection model is trained using fault feature samples corresponding to the vibration response signal samples of the overhead ground wire, and the fault distribution probability is used to indicate the probability of occurrence of different fault types of the overhead ground wire.

[0041] In this embodiment, after extracting the fault features of the overhead ground wire in step S102, the fault features are input into the target detection model for analysis to obtain multiple fault distribution probabilities of the overhead ground wire. For example, multiple fault distribution probabilities of the overhead ground wire are obtained by calculating the activation function in the target detection model. This is only an illustrative example and does not limit the specific method for obtaining multiple fault distribution probabilities of the overhead ground wire.

[0042] For example, after extracting features through the modal encoder and Transformer encoder, the classifier composed of a fully connected layer and a softmax function associates the extracted features with the fault type. Subsequently, the softmax function calculates the probability distribution of each fault type.

[0043] Step S104: determining a target fault type of the overhead ground wire based on a plurality of fault distribution probabilities.

[0044] In the technical solution provided in the above step S104 of the present invention, after the multiple fault distribution probabilities are determined in step S103, the target fault type of the overhead ground wire is determined according to the multiple fault distribution probabilities.

[0045] In this embodiment, a plurality of fault distribution probabilities are compared to determine the target fault type of the overhead ground wire.

[0046] For example, assuming that the fault distribution probability of fault type A is 40%, the fault distribution probability of fault type B is 20%, the fault distribution probability of fault type C is 30%, and the fault distribution probability of fault type D is 10%, the fault distribution probabilities of fault type A, fault type B, fault type C and fault type D are compared. Since 40%>30%>20%>10%, that is, the fault distribution probability of fault type A is the largest, based on this, the target fault type of the overhead ground wire is determined to be fault type A.

[0047] It should be noted that the above embodiment can be executed by a fault type detection device for an overhead ground wire.

[0048] In the above steps S101 to S104 of the present invention, the vibration response signal of the overhead ground wire is obtained; the fault characteristics of the overhead ground wire are extracted from the vibration response signal; the fault characteristics are input into the target detection model for analysis to obtain multiple fault distribution probabilities of the overhead ground wire, wherein the target detection model is trained using the fault characteristic samples corresponding to the vibration response signal samples of the overhead ground wire, and the fault distribution probability is used to indicate the probability of occurrence of different fault types of the overhead ground wire; based on the multiple fault distribution probabilities, the target fault type of the overhead ground wire is determined. In other words, the present invention uses the target detection model to analyze the vibration response signal of the overhead ground wire, determines the multiple fault distribution probabilities of the overhead ground wire, and thus determines the target fault type of the overhead ground wire according to the fault distribution probability, thereby solving the technical problem of low efficiency of fault detection of the overhead ground wire and achieving the technical effect of improving the efficiency of fault detection of the overhead ground wire.

[0049] The above method of this embodiment is further introduced below.

[0050] As an optional implementation method, the fault characteristics of the overhead ground wire are extracted from the vibration response signal, including: decomposing the vibration response signal into multiple modal components, wherein the modal components are used to indicate the contribution of different frequencies and vibration modes to the vibration response signal; based on the multiple modal components, determining the fault characteristics of the overhead ground wire.

[0051] In this embodiment, the vibration response signal is decomposed into multiple modal components. For example, the vibration response signal is decomposed into multiple modal components by using an overhead ground wire vibration response preprocessing method based on variational modal decomposition.

[0052] For example, the sparrow intelligent search algorithm combining sine and cosine and Cauchy mutation is used to optimize the parameters of variational mode decomposition. First, the sparrow population initialization is optimized, and the important parameters of the SCSSA algorithm are set, including the number of sparrow populations as θ, the current number of iterations as t, and the maximum number of iterations as t. max The warning threshold issued by individual sparrows when they discover predators is R2, the preset safety threshold is ST, etc., and the sparrow population is initialized according to the refraction reverse learning strategy to reduce the probability of premature convergence of the optimization algorithm in the later stage, and the refraction reverse population It is shown in the following formula (1):

[0053]

[0054] Among them, x m,n is the n-dimensional position of the mth sparrow in the sparrow population, m = 1, 2, ..., M, n = 1, 2, ..., N; k is a scaling factor to adjust the search range, For x m,n The refraction reverse position; α n , β n are the minimum and maximum values ​​of the n-dimensional position respectively; then the fitness value of each sparrow is calculated and sorted, the sparrow individuals with the best and worst fitness values ​​are found, and the current position is recorded; the sine and cosine search strategy is introduced at the seeker position of the sparrow algorithm SSA to update the seeker position, and the new seeker position update formula becomes the following formula (2):

[0055]

[0056] Where ω is the nonlinear weight factor; is the n-dimensional position of the mth sparrow under the current iteration number; r1 is the improved step search factor, which is related to the population iteration number and the maximum iteration number; r2 and r3 are both random values ​​of 0 to 2π; X best is the current optimal position of the sparrow; Cauchy variation is introduced into the follower position update formula, and the new follower position update formula becomes the following formula (3):

[0057]

[0058] Among them, X best (t) is the optimal position of the sparrows under the current iteration number; cauchy(0,1) is the standard Cauchy distribution function; then the position of the sentinels in the sparrow population is updated, and the update formula of the sentinel position is the following formula (4):

[0059]

[0060] Among them, ζ is the step size correction control coefficient; μ is a random value between -1 and 1; f i is the target constraint value of the current sparrow individual; f a , f b are the optimal and worst fitness values ​​of the current sparrow respectively; ε is the minimum constant, is the overall worst position of the sparrows under the current number of iterations; finally, determine whether the end condition can be met under the current maximum number of iterations. If so, output the optimal fitness value and the best position of the population at this time, and the algorithm process ends; otherwise, repeat the algorithm process until the end condition is met.

[0061] For another example, the VMD optimized by SCSSA parameters is used to adaptively decompose the vibration response signal of the overhead ground wire and decompose the signal into a set of modal components. The constrained variational problem solved by variational modal decomposition can be described by the following formula (5):

[0062]

[0063] Among them, u k and ω k represent the decomposed mode and the corresponding center frequency respectively, and f represents the signal to be decomposed, that is, the input signal. j is an imaginary number, is the partial derivative of the function with respect to time t; δ is the Dirac distribution, and * represents the convolution operation. In order to transform the constrained variational problem into an unconstrained variational problem, a quadratic penalty term is introduced and the Lagrangian multiplication is used. The augmented Lagrangian function can be expressed as the following formula (6):

[0064]

[0065] Among them, α is the quadratic penalty factor, and λ represents the Lagrange multiplier. L is the Lagrange function; λ(t) is the Lagrange multiplier. By finding the extreme value of the function, the modal component u in the spectral domain k and the corresponding center frequency ω k Finally, it can be expressed as the following formula (7) and formula (8):

[0066]

[0067] Where: "^" is Fourier transform; i is an integer value from 0 to K. The input signal f is decomposed into modal components {u1,u2,...,u K} and the corresponding center frequencies {ω1,ω2,...,ω K}. SCSSA is used to adaptively optimize the parameters of VMD, including the penalty factor α and the number of decomposition layers K; the minimum envelope entropy E is selected p As the objective constraint function. The envelope entropy E of the original signal x(i)p It is the following formula (9):

[0068]

[0069] Where: x(i) is the original signal; p(i) is the signal after envelope analysis; Y(i) is the normalized form of p(i).

[0070] Optionally, after decomposing the vibration response signal into a plurality of modal components, the fault characteristics of the overhead ground wire are determined.

[0071] As an optional implementation method, the fault characteristics of the overhead ground wire are determined based on multiple modal components, including: segmenting each modal component to determine the semantic relationship between the multiple modal components; determining the key mode based on the semantic relationship, wherein the key mode is used to indicate representative information in different modal components; based on the key mode, determining the fault characteristics of the overhead ground wire.

[0072] In this embodiment, each modal component is segmented, and the semantic relationship between each modal component is determined; based on the semantic relationship, the key modality is determined. For example, a VMD-based modal encoder is used to capture the temporal characteristics of these modal components and map them to a new feature space. At the same time, the multi-head self-attention mechanism and feedforward neural network of the modal internal transformer encoder are used to learn the relationship between modalities and calculate the key modality.

[0073] For example, according to a set of modal components obtained by decomposition, each modal component in this group is segmented, and a VMD-based modal encoder is used to capture the temporal characteristics of these modal components and map them to a new feature space using a learnable linear transformation. Consider the modal component u obtained by modal decomposition k , whose length is L m , with a specific center frequency and finite bandwidth. First, in the time domain, u k Divide into length L t N segments, that is These fragments are then mapped to patches using a learnable linear transformation where d p is the length of a single patch, as shown in the following formulas (10) and (11):

[0074]

[0075] Among them, the parameters is a linear transformation matrix, which is trainable. In order to improve the model’s sensitivity to the position of the time series in the modal component, a learnable position encoding E is added posIn addition, a learnable class label is appended to the beginning of the modal sequence and used as output. The final output can be expressed as the following formula (12):

[0076]

[0077] The modal internal Transformer encoder captures the contextual information of the input sequence through a multi-head self-attention mechanism, and further processes this information through a feedforward neural network to extract deeper feature information. The Transformer encoder is mainly composed of a multi-head attention mechanism and a feedforward neural network. The normalization layer and residual connection are used to accelerate convergence and learning. The self-attention mechanism is a special variant of the attention mechanism. It uses a scaled dot product attention mechanism to filter the target information between multiple features. The weighted probability is obtained by performing a dot product on the query matrix Q and the key matrix K, and then passing Softmax. After that, the attention vector can be obtained by weighted summing the matrix V through the weighted probability, which can be specifically expressed as the following formula (13):

[0078]

[0079] Among them, Q, K, V are query matrix, key matrix and value matrix respectively; T represents the transpose of the matrix, d k is the matrix dimension of Q, K, and V. In order to obtain more comprehensive information, Transformer uses a multi-head self-attention mechanism to divide the model into multiple heads. In this way, the model can pay attention to different aspects of information and integrate multiple aspects of information. The calculation formula corresponding to the multi-head self-attention mechanism is as follows (14):

[0080]

[0081] in, represents the input of the modal encoder, MHSA refers to multi-head self-attention, W i Q , W i K , W i V are the weight matrices of the i-th attention head Q, K, and V respectively; h is the number of feature subspaces; W O is the multi-head self-attention weight matrix. The output values ​​of each attention head are concatenated through the concat function and finally combined with the weight matrix W O The multi-head self-attention value is calculated. Then, a multi-layer perceptron with one hidden layer is used to implement the feedforward neural network. It can be expressed as the following formula (15):

[0082] MLP(X)=GeLU(XW 1+b1)W 2 +b2 (15)

[0083] Among them, MLP refers to a feedforward neural network, W 1 , W 2 , b1 and b2 represent the weights and biases of the two layers, and GeLU is the activation function which can be expressed as the following formula (16):

[0084]

[0085] The activation function is mapped through the tanh(·) function.

[0086] Optionally, after determining the key modes, the fault characteristics of the overhead ground wire are determined according to the key modes. For example, the center frequency corresponding to each modal segment is embedded into the modal segment through a position encoding method based on the center frequency, the sensitivity of the model to different modal frequencies is enhanced, and the input is input into the Transformer encoder for analysis to determine the fault characteristics of the overhead ground wire.

[0087] As an optional implementation manner, the device for detecting the fault type of an overhead ground wire further includes: using fault feature samples of the overhead ground wire to train an initial detection model to obtain a target detection model.

[0088] In this embodiment, the initial detection model is trained using the fault feature samples of the overhead ground wire to obtain the target detection model. The fault feature samples may also be the historical fault features of the overhead ground wire. The target detection model may be a Transformer encoder.

[0089] Optionally, the initial detection model is trained by fault feature samples to obtain a target detection model, thereby ensuring the accuracy of the target detection model and improving the efficiency of fault detection for overhead ground wires.

[0090] As an optional implementation method, the initial detection model is trained using fault feature samples of the overhead ground wire to obtain a target detection model, including: multiple modal component samples extracted from the fault feature samples; multiple center frequencies are extracted from the multiple modal component samples, wherein the center frequency is used to indicate the dominant frequency of the modal component samples; based on the center frequency, the initial detection model is trained to obtain a target detection model.

[0091] In this embodiment, multiple modal component samples are extracted from the vibration response signal samples; and multiple center frequencies are extracted from the multiple modal component samples. The center frequency can also be called the center frequency code E f .

[0092] For example, through the position encoding method based on the center frequency, the center frequency corresponding to each modal segment is embedded into the modal segment to enhance the model's sensitivity to different modal frequencies and input into the Transformer encoder. The structure of the Transformer encoder is similar to that of the modal encoder. The input of the Transformer encoder is represented as a set of modal patches. Corresponding to the original modes with different center frequencies where d m Represents the length of a single modality patch. Similar to the patch and position embeddings in the above modality encoder, the center frequency encoding E f is also incorporated into the modal embedding to improve the sensitivity of the target detection model to different modal frequencies, as shown in the following formula (17):

[0093]

[0094] Among them, m class Represents the category label, parameter W t is the linear transformation matrix, E f Represents the center frequency encoding. The output embedding is then input to the Transformer encoder. Each encoding layer includes a multi-head self-attention (MHSA) module and a feedforward neural network (MLP) module. This process can be expressed by the following formulas (17) and (18):

[0095]

[0096] Among them, LN(·) represents the normalization layer, and i represents the number of encoder layers. After extracting features through the modal encoder and the Transformer encoder, the classifier composed of the fully connected layer and the softmax function associates the extracted features with the fault type. Subsequently, the softmax function calculates the probability distribution of each fault type.

[0097] As an optional implementation manner, determining the fault type of the overhead ground wire based on multiple fault distribution probabilities includes: determining the fault type corresponding to the maximum fault distribution probability among the multiple fault distribution probabilities as the target fault type of the overhead ground wire.

[0098] In this embodiment, the fault type corresponding to the maximum fault distribution probability among the multiple fault distribution probabilities is determined as the target fault type of the overhead ground wire.

[0099] Optionally, by comparing various fault distribution probabilities and determining the fault type corresponding to the maximum fault distribution probability, targeted preventive measures can be taken to reduce the probability of occurrence of this type of fault and improve the reliability and stability of the system.

[0100] It should be noted that the above embodiment can be executed by a fault type detection device for an overhead ground wire.

[0101] In this embodiment, a vibration response signal of an overhead ground wire is obtained; from the vibration response signal, a fault feature of the overhead ground wire is extracted; the fault feature is input into a target detection model for analysis to obtain multiple fault distribution probabilities of the overhead ground wire, wherein the target detection model is trained using fault feature samples corresponding to the vibration response signal samples of the overhead ground wire, and the fault distribution probability is used to indicate the probability of occurrence of different fault types of the overhead ground wire; based on the multiple fault distribution probabilities, the target fault type of the overhead ground wire is determined. In other words, the present invention uses a target detection model to analyze the vibration response signal of the overhead ground wire, determines multiple fault distribution probabilities of the overhead ground wire, and thereby determines the target fault type of the overhead ground wire according to the fault distribution probability, thereby solving the technical problem of low efficiency in fault detection of the overhead ground wire and achieving the technical effect of improving the efficiency of fault detection of the overhead ground wire.

[0102] The technical solution of the embodiment of the present invention is illustrated below in conjunction with preferred implementation modes.

[0103] At present, the power system generally uses high-voltage or ultra-high-voltage overhead lines for long-distance power distribution. This distribution method will expose the transmission lines to high-altitude environments for a long time, and they are easily affected by the natural environment, such as electric shocks, acid rain, etc., which will cause corrosion, wear and even breakage of the lines. As an important protection line in the transmission line, the overhead ground wire will also break due to environmental factors. If it is not handled in time, it is easy to cause large-scale power outages, resulting in huge economic losses.

[0104] In the related technology, a large amount of manpower and material resources are invested in regular and frequent inspections of overhead ground wires to ensure the safety of the power system. However, the inspection cycle is long and the accuracy is poor, making it difficult to troubleshoot faults in a timely manner, especially in some areas with harsh environments, where the structural health status of the overhead ground wire cannot be obtained in a timely and effective manner. Minor local damage to the overhead ground wire will eventually develop into a serious accident due to long-term breeze vibration, causing many problems in people's lives and the economy. Therefore, there is a technical problem of low efficiency in fault detection of overhead ground wires. At present, no effective solution has been proposed to the above-mentioned technical problem of low efficiency in fault detection of overhead ground wires.

[0105] However, an embodiment of the present invention proposes a method for detecting overhead ground wire faults based on deep learning. The Sparrow Smart Search Algorithm (SCSSA) combining sine, cosine and Cauchy mutations is used to optimize the parameter penalty factor (α) and the number of modes (K) of the variational mode decomposition (VMD). The vibration response signal of the overhead ground wire can be adaptively decomposed. In view of the problem that the characteristics of multi-band coupled signals are difficult to identify, a VMD-based modal encoder is proposed to decompose the coupled modes and calculate the key modes. Secondly, in view of the problem that the existing position encoding methods do not pay enough attention to the signal frequency, a position encoding method based on the center frequency is proposed. Finally, through the proposed innovative deep learning model of Transformer for fault diagnosis based on VMD, the fault detection of the overhead ground wire is realized, thereby solving the technical problem of low efficiency of fault detection of the overhead ground wire and achieving the technical effect of improving the efficiency of fault detection of the overhead ground wire.

[0106] The embodiments of the present invention are further described below.

[0107] Figure 2 1 is a flowchart of a method for detecting an overhead ground wire fault based on deep learning according to an embodiment of the present invention, the method comprising the following steps:

[0108] Step S201 : decoupling fault characteristics by using an overhead ground wire vibration response preprocessing method based on variational mode decomposition.

[0109] In this embodiment, the sparrow intelligent search algorithm combining sine, cosine and Cauchy mutation is used to optimize the parameters of variational mode decomposition.

[0110] Optionally, the VMD optimized by SCSSA parameters is used to adaptively decompose the vibration response signal of the overhead ground wire, and the signal is decomposed into a group of modal components.

[0111] Optionally, the constrained variational problem solved by variational mode decomposition can be described by the aforementioned formula (5), which will not be described in detail here.

[0112] Optionally, in order to transform the constrained variational problem into an unconstrained variational problem, a quadratic penalty term is introduced and the Lagrangian multiplication is used. The augmented Lagrangian function can be expressed as the aforementioned formula (6), which will not be described here.

[0113] Optionally, by finding the extreme value of the function, the modal component u in the spectral domain k and the corresponding center frequency ω k Finally, it can be expressed as the above formula (7) and formula (8), which will not be described here.

[0114] Optionally, use SCSSA to perform adaptive optimization of VMD parameters, including penalty factor α and decomposition level K; select the minimum envelope entropy E p As the objective constraint function. The envelope entropy E of the original signal x(i) p It is the aforementioned formula (9), which will not be described again here.

[0115] Step S202, based on a set of modal components obtained by decomposition, the temporal characteristics in the modal components are captured and mapped to a new feature space, while the relationship between the modes is learned and the key modes are calculated.

[0116] In this embodiment, according to a set of modal components decomposed in step S201, a VMD-based modal encoder is used to capture the temporal characteristics of these modal components and map them to a new feature space. At the same time, the multi-head self-attention mechanism and feedforward neural network of the modal internal transformer encoder are used to learn the relationship between modalities and calculate the key modalities.

[0117] Optionally, according to a group of modal components decomposed in step S201, each modal component in this group is segmented, and the temporal characteristics of these modal components are captured by a VMD-based modal encoder and mapped to a new feature space using a learnable linear transformation.

[0118] Optionally, the modality-internal Transformer encoder captures the contextual information of the input sequence through a multi-head self-attention mechanism, and further processes this information through a feed-forward neural network to extract deeper feature information.

[0119] Optionally, according to a set of modal components obtained by decomposition, each modal component in the set is segmented, and a VMD-based modal encoder is used to capture the temporal characteristics of these modal components and map them to a new feature space using a learnable linear transformation. Consider the modal component u obtained by modal decomposition k , whose length is L m , with a specific center frequency and finite bandwidth. First, in the time domain, u k Divide into length L t N segments, that is These fragments are then mapped to patches using a learnable linear transformation where d p is the length of a single patch, as shown in the above formulas (10) and (11), which will not be repeated here.

[0120] Optionally, to increase the model’s sensitivity to the position of the time series in the modal component, a learnable positional encoding E is added posIn addition, a learnable class label is appended to the beginning of the modal sequence and used as the output. The final output can be expressed as the aforementioned formula (12), which will not be repeated here.

[0121] Optionally, the modality-internal Transformer encoder captures the contextual information of the input sequence through a multi-head self-attention mechanism, and further processes this information through a feedforward neural network. To extract deeper feature information. The Transformer encoder is mainly composed of a multi-head attention mechanism and a feedforward neural network. The normalization layer and residual connection are used to accelerate convergence and learning speed. The self-attention mechanism is a special variant of the attention mechanism. It uses a scaled dot product attention mechanism to filter the target information between multiple features. The weighted probability is obtained by performing a dot product on the query matrix Q and the key matrix K, and then passing through Softmax. After that, the attention vector can be obtained by weighted summing the matrix V through the weighted probability, which can be specifically expressed as the aforementioned formula (13), which will not be repeated here.

[0122] Optionally, in order to obtain more comprehensive information, Transformer uses a multi-head self-attention mechanism to divide the model into multiple heads. In this way, the model can pay attention to different aspects of information and integrate multiple aspects of information. The calculation formula corresponding to the multi-head self-attention mechanism is the aforementioned formula (14), which will not be repeated here.

[0123] Optionally, the output values ​​of each attention head are concatenated through the concat function and finally combined with the weight matrix W O The multi-head self-attention value is calculated. Then, a multi-layer perceptron with one hidden layer is used to implement a feedforward neural network. It can be expressed as the aforementioned formula (15), which will not be repeated here.

[0124] Optionally, GeLU is an activation function which can be expressed as the aforementioned formula (16), which will not be described in detail here. The activation function is mapped by a tanh(·) function.

[0125] Step S203: According to the output modal segments, the center frequency corresponding to each modal segment is embedded into the modal segment and input into the Transformer encoder.

[0126] In this embodiment, according to the modal segment outputted in step S202, the center frequency corresponding to each modal segment is embedded into the modal segment through a position encoding method based on the center frequency, the sensitivity of the model to different modal frequencies is enhanced, and the modal segment is inputted into the Transformer encoder.

[0127] Step S204: associate the extracted features with the fault types and calculate the probability distribution of each fault type.

[0128] In this embodiment, after extracting features through the modal encoder and the Transformer encoder, the classifier composed of a fully connected layer and a softmax function associates the extracted features with the fault type. Subsequently, the softmax function calculates the probability distribution of each fault type, and finally completes the fault diagnosis process of the overhead ground wire.

[0129] Figure 3 1 is a flow chart of a method for optimizing parameters of variational mode decomposition according to an embodiment of the present invention, the optimization method comprising the following steps:

[0130] Step S301, initializing SCSSA algorithm parameters.

[0131] In this embodiment, the initialization of the sparrow population is optimized first, and the important parameters of the SCSSA algorithm are set, including the number of sparrow populations as θ, the current number of iterations as t, and the maximum number of iterations as t. max The warning threshold issued by individual sparrows when they discover predators is R2, the preset safety threshold is ST, etc., and the sparrow population is initialized according to the refraction reverse learning strategy to reduce the probability of premature convergence of the optimization algorithm in the later stage, and the refraction reverse population As shown in the above formula (1), it will not be repeated here.

[0132] Step S302, calculate the current fitness value of each sparrow and sort them.

[0133] In this embodiment, the fitness value of each sparrow is calculated and sorted, the sparrow individuals with the best and worst fitness values ​​are found, and the current positions are recorded.

[0134] Step S303, updating the seeker's position.

[0135] In this embodiment, a sine-cosine search strategy is introduced at the seeker position of the sparrow algorithm SSA to update the seeker position. The new seeker position update formula becomes as shown in the aforementioned formula (2), which will not be described again here.

[0136] Step S304: Update the follower position.

[0137] In this embodiment, the Cauchy mutation is introduced into the follower position update formula, and the new follower position update formula becomes as shown in the aforementioned formula (3), which will not be described again here.

[0138] Step S305, updating the position of the alerter.

[0139] In this embodiment, the position of the sentinel in the sparrow population is updated, and the updating formula of the sentinel position is as shown in the above formula (4), which will not be described again here.

[0140] Step S306, determining whether the maximum number of iterations has been reached.

[0141] In this embodiment, it is determined whether the maximum number of iterations has been reached. If so, step S307 is executed; if not, step S303 is executed.

[0142] Step S307, output the best parameter combination.

[0143] In this embodiment, the optimal fitness value and the best position of the population at this time are output.

[0144] Figure 4 It is a schematic diagram of an overhead ground wire fault detection model based on deep learning according to an embodiment of the present invention, and the overhead ground wire fault detection model includes a Transformer encoder.

[0145] In this embodiment, the structure of the Transformer encoder is similar to that of the modality encoder, and the input of the Transformer encoder is represented as a set of modality patches Corresponding to the original modes with different center frequencies where d m Represents the length of a single modality patch. Similar to the patch and position embeddings in the above modality encoder, the center frequency encoding E f is also incorporated into the modal embedding to improve the model's sensitivity to different modal frequencies, as shown in the following formula (17).

[0146]

[0147] Among them, m class Represents the category label, parameter W t is the linear transformation matrix, E f Indicates the center frequency code.

[0148] Optionally, the output embedding is input to the Transformer encoder, and each encoding layer includes a multi-head self-attention (MHSA) module and a feedforward neural network (MLP) module. This process can be expressed by the following formulas (18) and (19):

[0149]

[0150] Here, LN(·) represents the normalization layer and i represents the number of encoder layers.

[0151] Figure 5 is a schematic diagram of overhead ground wire fault detection classification results based on an improved modal attention Transformer deep learning model according to an embodiment of the present invention, such as Figure 5 As shown, it includes the probability of the target category and the output category.

[0152] In this embodiment, the Sparrow Smart Search Algorithm (SCSSA) combined with sine, cosine and Cauchy mutations is used to optimize the parameter penalty factor (α) and the number of modes (K) of the variational mode decomposition (VMD), so that the vibration response signal of the overhead ground wire can be adaptively decomposed. In view of the problem that the characteristics of multi-band coupled signals are difficult to identify, a modal encoder based on VMD is proposed to decompose the coupled modes and calculate the key modes. Secondly, in view of the problem that the existing position encoding method does not pay enough attention to the signal frequency, a position encoding method based on the center frequency is proposed. Finally, through the proposed innovative deep learning model of Transformer for fault diagnosis based on VMD, the fault detection of the overhead ground wire is realized, thereby solving the technical problem of low efficiency of fault detection of the overhead ground wire and achieving the technical effect of improving the efficiency of fault detection of the overhead ground wire.

[0153] According to an embodiment of the present invention, a device for detecting the fault type of an overhead ground wire is also provided. It should be noted that the device for detecting the fault type of an overhead ground wire can be used to execute the method for detecting the fault type of an overhead ground wire in the method embodiment.

[0154] Figure 6 FIG. 1 is a schematic diagram of a fault type detection device for an overhead ground wire according to an embodiment of the present invention. Figure 6 As shown, the overhead ground wire fault type detection device 600 may include: an acquisition unit 601 , an extraction unit 602 , an input unit 603 and a determination unit 604 .

[0155] The acquisition unit 601 is used to acquire the vibration response signal of the overhead ground wire.

[0156] The extraction unit 602 is used to extract the fault characteristics of the overhead ground wire from the vibration response signal.

[0157] The input unit 603 is used to input the fault characteristics into the target detection model for analysis to obtain multiple fault distribution probabilities of the overhead ground wire, wherein the target detection model is trained using fault feature samples corresponding to the vibration response signal samples of the overhead ground wire, and the fault distribution probability is used to indicate the probability of occurrence of different fault types corresponding to the fault of the overhead ground wire.

[0158] The determination unit 604 is configured to determine a target fault type of the overhead ground wire based on a plurality of fault distribution probabilities.

[0159] Optionally, the extraction unit 602 may include: a decomposition module, used to decompose the vibration response signal into multiple modal components, wherein the modal components are used to indicate the contribution of different frequencies and vibration modes to the vibration response signal; and a first determination module, used to determine the fault characteristics of the overhead ground wire based on the multiple modal components.

[0160] Optionally, the first determination module may include: a first determination submodule, used to segment each modal component and determine the semantic relationship between multiple modal components; a second determination submodule, used to determine the key mode based on the semantic relationship, wherein the key mode is used to indicate representative information in different modal components; and a third determination submodule, used to determine the fault characteristics of the overhead ground wire based on the key mode.

[0161] Optionally, the overhead ground wire fault type detection device 600 may further include: a training unit, configured to train an initial detection model using fault feature samples of the overhead ground wire to obtain a target detection model.

[0162] Optionally, the first training unit may include: a first extraction module, used to extract multiple modal component samples from the fault feature samples; a second extraction module, used to extract multiple center frequencies from the multiple modal component samples, wherein the center frequency is used to indicate the dominant frequency of the modal component samples; and a training module, used to train the initial detection model based on the center frequency to obtain a target detection model.

[0163] Optionally, the determination unit 604 may include: a comparison module, configured to determine the fault type corresponding to the maximum fault distribution probability among the multiple fault distribution probabilities as the target fault type of the overhead ground wire.

[0164] In this embodiment, a vibration response signal of an overhead ground wire is obtained; from the vibration response signal, a fault feature of the overhead ground wire is extracted; the fault feature is input into a target detection model for analysis to obtain multiple fault distribution probabilities of the overhead ground wire, wherein the target detection model is trained using fault feature samples corresponding to the vibration response signal samples of the overhead ground wire, and the fault distribution probability is used to indicate the probability of occurrence of different fault types of the overhead ground wire; based on the multiple fault distribution probabilities, the target fault type of the overhead ground wire is determined. In other words, the present invention uses a target detection model to analyze the vibration response signal of the overhead ground wire, determines multiple fault distribution probabilities of the overhead ground wire, and thereby determines the target fault type of the overhead ground wire according to the fault distribution probability, thereby solving the technical problem of low efficiency in fault detection of the overhead ground wire and achieving the technical effect of improving the efficiency of fault detection of the overhead ground wire.

[0165] According to an embodiment of the present invention, there is further provided a computer-readable storage medium, the storage medium comprising a stored program, wherein the program executes the method for detecting the fault type of an overhead ground wire in the method embodiment.

[0166] According to an embodiment of the present invention, a processor is further provided. The processor is used to run a program, wherein the method for detecting the fault type of an overhead ground line in the method embodiment is executed when the program is run.

[0167] According to an embodiment of the present invention, a computer program product is further provided. The computer program product includes computer instructions. When the computer instructions are executed by a processor, the method for detecting the fault type of an overhead ground line in the method embodiment is implemented.

[0168] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0169] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0170] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0171] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed over multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0172] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0173] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent functional component, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software functional component, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc., which can store program codes.

[0174] The above are only preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for detecting the fault type of an overhead ground wire, characterized in that: include: Obtaining vibration response signals of overhead ground wires; Extracting fault characteristics of the overhead ground wire from the vibration response signal; The fault feature is input into a target detection model for analysis to obtain multiple fault distribution probabilities of the overhead ground wire, wherein the target detection model is trained using fault feature samples corresponding to vibration response signal samples of the overhead ground wire, and the fault distribution probability is used to indicate the probability of occurrence of different fault types of the overhead ground wire; Based on the multiple fault distribution probabilities, a target fault type of the overhead ground wire is determined.

2. The method according to claim 1, characterized in that: Extracting the fault characteristics of the overhead ground wire from the vibration response signal includes: Decomposing the vibration response signal into a plurality of modal components, wherein the modal components are used to indicate the contribution of different frequencies and vibration modes to the vibration response signal; Based on the multiple modal components, the fault characteristics of the overhead ground wire are determined.

3. The method according to claim 2, characterized in that Determining the fault characteristics of the overhead ground wire based on the multiple modal components includes: Segmenting each of the modal components to determine a semantic relationship between the multiple modal components; Based on the semantic relationship, determining a key modality, wherein the key modality is used to indicate representative information in different modal components; Based on the key mode, the fault feature of the overhead ground wire is determined.

4. The method according to claim 1, characterized in that The method further comprises: The initial detection model is trained using the fault feature samples of the overhead ground wire to obtain the target detection model.

5. The method according to claim 4, characterized in that The initial detection model is trained using the fault feature sample of the overhead ground wire to obtain the target detection model, including: A plurality of modal component samples extracted from the fault feature samples; Extracting a plurality of center frequencies from a plurality of the modal component samples, wherein the center frequencies are used to indicate dominant frequencies of the modal component samples; Based on the center frequency, the initial detection model is trained to obtain the target detection model.

6. The method according to any one of claims 1 to 5, characterized in that: Determining the fault type of the overhead ground wire based on the multiple fault distribution probabilities includes: The fault type corresponding to the maximum fault distribution probability among the multiple fault distribution probabilities is determined as the target fault type of the overhead ground wire.

7. A fault type detection device for an overhead ground wire, characterized in that: include: An acquisition unit, used for acquiring a vibration response signal of the overhead ground wire; An extraction unit, used to extract the fault characteristics of the overhead ground wire from the vibration response signal; An input unit, used for inputting the fault feature into a target detection model for analysis to obtain a plurality of fault distribution probabilities of the overhead ground wire, wherein the target detection model is trained using fault feature samples corresponding to vibration response signal samples of the overhead ground wire, and the fault distribution probability is used to indicate the probability of occurrence of different fault types corresponding to the fault of the overhead ground wire; A determination unit is used to determine a target fault type of the overhead ground wire based on a plurality of the fault distribution probabilities.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed by a processor, the device where the storage medium is located is controlled to execute the method according to any one of claims 1 to 6.

9. A processor, characterized in that: The processor is used to run a program, wherein the program executes the method according to any one of claims 1 to 6 when running.

10. A computer program product, characterized in that The computer program product comprises computer instructions, which implement the method according to any one of claims 1 to 6 when executed by a processor.

Images