Power transmission line fault location system and method based on acoustic magnetic signal feature analysis

Abstract

The application belongs to the technical field of electromagnetic fault positioning, and discloses a power transmission line fault positioning system and method based on acoustic magnetic signal feature analysis. The method comprises the following steps: linear correction processing is performed on the magnetic field signal propagation speed of a magnetic field sensor and the ultrasonic signal propagation speed of an ultrasonic sensor through a dynamic time delay compensation mechanism; linearly corrected magnetic field signals and ultrasonic signals after linear correction are obtained through synchronous acquisition; strong electromagnetic interference suppression and transient pulse suppression are performed on the linearly corrected magnetic field signals to obtain first magnetic field signals; transient pulse suppression is performed on the linearly corrected ultrasonic signals to obtain first ultrasonic signals; and enhancement processing is performed on the first ultrasonic signals and the first magnetic field signals to obtain second magnetic field signals and second ultrasonic signals. The application can effectively shorten the fault repair time, reduce the operation and maintenance cost, and ensure the safe and stable operation of the power system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electromagnetic fault location, and more particularly, to a power transmission line fault location system and method based on acoustic magnetic signal feature analysis. BACKGROUND

[0002] The traditional manual inspection method is inefficient in locating faults on power transmission lines, and it is difficult to quickly find and locate the fault point, which seriously affects the rapid recovery of the power grid after a fault.

[0003] Chinese patent application No. CN119044676A discloses a method for locating underground cable faults based on acoustic magnetic feature matching analysis and a ground penetrating sensor system. The steps include: step 1: generating and collecting acoustic magnetic signals; step 2: signal preprocessing; step 3: fault point distance calculation; step 4: repeating steps 1 to 4 multiple times, and calculating the position coordinates of the fault point based on the calculation results of multiple sets of fault point distances. The invention automatically calculates the position of the underground cable fault point by collecting acoustic magnetic signals, reduces the labor cost of locating the fault point, and greatly improves the efficiency and accuracy of cable fault location.

[0004] Although the above method can meet most scenarios, research and practical application of the above method and prior art have found that the above method and prior art at least have the following defects:

[0005] There are various electromagnetic devices and signals in underground cables, and the electromagnetic environment is more complex, making it difficult to extract effective fault signals from the complex electromagnetic background. There are also various pipelines and other cables and facilities that interfere with the collection of magnetic field signals and sound signals by the sensor system, resulting in inaccurate fault location.

[0006] In view of this, the present application provides a power transmission line fault location system and method based on acoustic magnetic signal feature analysis to solve the above problems. SUMMARY

[0007] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned purpose, the present application provides the following technical solution: a power transmission line fault location method based on acoustic magnetic signal feature analysis, comprising the following steps:

[0008] The dynamic time delay compensation mechanism is used to linearly correct the propagation speed of the magnetic field signal of the magnetic field sensor and the propagation speed of the ultrasonic signal of the ultrasonic sensor, and the linearly corrected magnetic field signal and ultrasonic signal after linear correction are synchronously collected.

[0009] The linearly corrected magnetic field signal is subjected to strong electromagnetic interference suppression and transient pulse suppression to obtain a first magnetic field signal, the linearly corrected ultrasonic signal is subjected to transient pulse suppression to obtain a first ultrasonic signal, and the first ultrasonic signal and the first magnetic field signal are subjected to enhancement processing to obtain a second magnetic field signal and a second ultrasonic signal;

[0010] The second magnetic field signal and the second ultrasonic signal are subjected to feature extraction to obtain magnetic field features and ultrasonic features;

[0011] The ultrasonic features are taken as inputs of a fault recognition model to obtain mechanical vibration modes and corresponding fault types;

[0012] The magnetic field features and the ultrasonic features are taken as inputs of a dual-flow residual attention network to obtain fused joint feature mappings, the joint feature mappings of three adjacent monitoring terminals are fused through D-S evidence theory, and three-dimensional coordinates of a fault point are calculated and obtained.

[0013] Further, the method for linearly correcting the propagation speed of the magnetic field signal of the magnetic field sensor and the propagation speed of the ultrasonic signal of the ultrasonic sensor includes:

[0014] The propagation speed of the magnetic field signal is calculated based on the relative magnetic permeability of the wire material and the relative node constant of the wire material;

[0015] The propagation speed of the ultrasonic signal is calculated based on the wire bulk modulus and the wire material density;

[0016] For different wire types, the propagation speed of the magnetic field signal and the propagation speed of the ultrasonic signal are linearly corrected in combination with a temperature correction coefficient and a humidity correction coefficient of the wire.

[0017] Further, the training method of the dual-flow residual attention network includes:

[0018] R sets of training data are pre-collected, the training data include acoustic-magnetic features and corresponding joint feature mappings, and the acoustic-magnetic features include magnetic field features and ultrasonic features;

[0019] The acoustic magnetic feature is taken as an input of the dual-flow residual attention network, the joint feature mapping is taken as an output of the dual-flow residual attention network, and a natural heuristic optimization algorithm is used to optimize network parameters of the dual-flow residual attention network to obtain network parameters corresponding to the minimum error between the joint feature mapping output by the dual-flow residual attention network and the actual joint feature mapping, and the dual-flow residual attention network constructed by the corresponding network parameters is taken as the trained dual-flow residual attention network; the dual-flow residual attention network includes a magnetic field branch and an ultrasonic wave branch, the magnetic field branch includes an NG-layer ResNet and a channel attention module, the ultrasonic wave branch includes an NM-layer CNN+LSTM and a spatial attention module, and outputs of the magnetic field branch and the ultrasonic wave branch are fused through a cross-attention mechanism to obtain the output joint feature mapping; a loss function of the dual-flow residual attention network includes a classification loss composed of a cross entropy and a positioning loss composed of a mean square error;

[0020] The method for obtaining the three-dimensional coordinates of the fault point comprises:

[0021] The distance inversion is performed based on the joint feature mapping combined with the fault positioning model to obtain the fault distance.

[0022] The fault distances obtained by the three adjacent monitoring terminals are fused through the D-S evidence theory to obtain the three-dimensional coordinates of the fault point.

[0023] Further, the method for obtaining the first magnetic field signal comprises:

[0024] The power frequency harmonic filtering processing is performed based on the transfer function of the adaptive notch filter to obtain the filtered magnetic field signal.

[0025] The wavelet packet reconstruction is performed on the filtered magnetic field signal using the wavelet packet coefficients processed through the threshold shrinkage to obtain the first magnetic field signal.

[0026] Further, the method for obtaining the wavelet packet coefficients processed through the threshold shrinkage comprises:

[0027] The wavelet basis function and the corresponding decomposition layer number are selected;

[0028] The wavelet packet decomposition is performed on the filtered magnetic field signal according to the selected wavelet basis function and the decomposition layer number to obtain the wavelet packet coefficients;

[0029] The wavelet threshold is obtained based on the natural heuristic optimization algorithm with the minimum weighted sum of the signal-to-noise ratio and the mean square error as the target;

[0030] The wavelet packet coefficients with the absolute value less than the wavelet threshold are set to zero, and the wavelet packet coefficients with the absolute value greater than or equal to the wavelet threshold are shrunk to zero to obtain the wavelet packet coefficients processed through the threshold shrinkage.

[0031] Further, the method for obtaining the second magnetic field signal and the second ultrasonic signal comprises:

[0032] The collected acoustic magnetic signals are divided into a training set, a validation set and a test set according to a preset ratio; the acoustic magnetic signals include the first ultrasonic signal and the first magnetic field signal;

[0033] The U-Net structure is selected as the generator, a random noise vector is taken as the input of the generator, and a generated acoustic magnetic signal with the same size and channel number as the acoustic magnetic signal is obtained; the convolutional neural network is selected as the discriminator architecture, and the acoustic magnetic signal and the generated acoustic magnetic signal are taken as the input of the discriminator, and the probability that the input data is the acoustic magnetic signal is obtained;

[0034] M groups of acoustic magnetic signals are randomly extracted from the training set, and marked as 1, representing the real acoustic magnetic signal; the generator generates the corresponding generated acoustic magnetic signal according to the input random noise vector, and marked as 0, representing the generated acoustic magnetic signal; the acoustic magnetic signal and the generated acoustic magnetic signal are input into the discriminator respectively, and the discriminator outputs the discrimination result; according to the output of the discriminator, and the corresponding labels of the acoustic magnetic signal and the generated acoustic magnetic signal, the parameters of the generator and the discriminator are optimized by minimizing the value of the loss function; wherein the loss function is the sum of the generator loss and the discriminator loss, the generator loss fuses the difference term of the generated acoustic magnetic signal and the real acoustic magnetic signal, the Maxwell equation constraint and the acoustic wave equation constraint; the discriminator loss is the cross entropy loss; the steps of selecting acoustic magnetic signals for discrimination are repeated for multiple rounds of iterative training until the generative adversarial network reaches a convergent state;

[0035] The acoustic magnetic signal is input into the generator of the convergent state corresponding to the generative adversarial network to generate an enhanced acoustic magnetic signal, and the enhanced acoustic magnetic signal is taken as the acoustic magnetic signal after enhancement processing, that is, the second magnetic field signal and the second ultrasonic signal.

[0036] Further, the method for obtaining the magnetic field feature comprises:

[0037] The running wave mutation point of the second magnetic field signal is calculated as a mutation point feature;

[0038] The second magnetic field signal is detected based on a magnetic field mutation point detection threshold; the second magnetic field signal is traversed, and when the magnetic field signal difference of adjacent time points exceeds the magnetic field mutation point detection threshold, the corresponding magnetic field signal difference is extracted as a difference feature, and the corresponding two adjacent time points are spliced as a time point feature;

[0039] The harmonic distortion rate is extracted as a distortion rate feature based on the voltage amplitude of the second magnetic field signal;

[0040] The spectral gravity center is extracted as a gravity center feature based on the spectral value;

[0041] The S mutation point features, S difference features, S time point features, S distortion rate features and S center of gravity features are spliced as magnetic field features, wherein the features less than S are zero-padded.

[0042] Further, the method for obtaining the ultrasonic wave features comprises:

[0043] Step 1, obtaining a second ultrasonic signal g, decomposing g into k modal components, each modal component having a corresponding center frequency; constructing a constrained variational model;

[0044] Step 2, introducing a Lagrange multiplier and a quadratic penalty term, converting the constrained variational problem into an unconstrained augmented Lagrange function;

[0045] Step 3, using an alternating direction multiplier method to iteratively solve the augmented Lagrange function;

[0046] Step 4, updating the modal function;

[0047] Step 5, updating the center frequency;

[0048] Step 6, updating the Lagrange multiplier;

[0049] Step 7, repeating steps 4-6 until the change of each modal function in two consecutive iterations is less than the convergence tolerance, and stopping the iteration;

[0050] Step 8, based on steps 1-7, obtaining K modal components, extracting time domain features and frequency domain features of each modal component, the time domain features including amplitude features and time features, the amplitude features being amplitude mean, amplitude variance, amplitude maximum and amplitude minimum of the modal component, the time features being rise time, fall time and pulse width of the modal component; the frequency domain features including frequency features and energy features, the frequency features being peak frequency, center frequency and bandwidth of the modal component, the energy features being energy distribution of the modal component in different frequency bands; splicing the time domain features and the frequency domain features to obtain the ultrasonic wave features.

[0051] Further, the method for obtaining the time features comprises:

[0052] presetting an upper threshold and a lower threshold, traversing the corresponding modal component to obtain a first signal point greater than the lower threshold, continuing to traverse the corresponding modal component from the first signal point to obtain a second signal point greater than the upper threshold, and calculating the product of the time difference between the second signal point and the first signal point and the sampling period as the rise time;

[0053] Reverse traversal is performed from the end position of the corresponding modal component, a first third signal point less than the upper threshold is obtained, the corresponding modal component is continuously traversed from the third signal point, a first fourth signal point less than the lower threshold is obtained, and a product of a time difference between the fourth signal point and the third signal point and a sampling period is calculated as a falling time;

[0054] The maximum value and the minimum value of the modal component are obtained, the intermediate value of the maximum value and the minimum value is calculated, the first fifth signal point greater than the intermediate value is obtained by traversing the corresponding modal component, the first sixth signal point less than the upper threshold is obtained by continuously traversing the corresponding modal component from the fifth signal point, and the product of the difference between the sixth signal point and the fifth signal point and the sampling period is calculated as the pulse width.

[0055] The method for obtaining the energy feature comprises:

[0056] The sum of the energies corresponding to all frequency domain indexes in the preset frequency range of the modal component is calculated, and the energy is the square of the amplitude.

[0057] The power line fault positioning system based on the acoustic magnetic signal feature analysis and the power line fault positioning method based on the acoustic magnetic signal feature analysis are implemented, and the power line fault positioning system based on the acoustic magnetic signal feature analysis comprises:

[0058] The correction acquisition module performs linear correction processing on the magnetic field signal propagation speed of the magnetic field sensor and the ultrasonic signal propagation speed of the ultrasonic sensor through a dynamic time delay compensation mechanism, and synchronously acquires the linearly corrected magnetic field signal and the ultrasonic signal after linear correction processing;

[0059] The signal processing module performs strong electromagnetic interference suppression and transient pulse suppression on the linearly corrected magnetic field signal to obtain a first magnetic field signal, performs transient pulse suppression on the linearly corrected ultrasonic signal to obtain a first ultrasonic signal, and performs enhancement processing on the first ultrasonic signal and the first magnetic field signal to obtain a second magnetic field signal and a second ultrasonic signal.

[0060] The feature extraction module extracts features from the second magnetic field signal and the second ultrasonic signal to obtain magnetic field features and ultrasonic features.

[0061] The fault recognition module takes the ultrasonic features as the input of a fault recognition model to obtain a mechanical vibration mode and a corresponding fault type.

[0062] The fault positioning module takes the magnetic field features and the ultrasonic features as the input of a double-flow residual attention network to obtain a fused joint feature mapping, fuses the joint feature mappings of three adjacent monitoring terminals through D-S evidence theory, and calculates to obtain a fault point three-dimensional coordinate.

[0063] The power line fault positioning system and method based on the acoustic magnetic signal feature analysis have the following technical effects and advantages:

[0064] The application combines physical driving and data driving technologies to realize high-precision fault positioning; the complementary characteristics of acoustic magnetic signals are utilized, wherein the magnetic field signal reflects the sudden change of fault current, and the ultrasonic signal embodies the mechanical vibration characteristics; effective magnetic field signals are extracted through strong electromagnetic interference suppression and transient pulse suppression, and the dynamic time delay compensation mechanism is used to correct the propagation speed difference, effectively eliminating environmental interference errors; the Maxwell equation and acoustic wave equation are introduced to constrain the generator loss function, ensuring that the synthesized signal conforms to the physical law and improving the reliability of feature extraction; the energy weighted fusion and attention mechanism are used to dynamically allocate weights, a double-flow residual attention network is constructed, and the attenuation coefficient dynamic calibration and distance calculation formula are combined to significantly improve the positioning accuracy under complex working conditions; the application breaks through the limitations of low efficiency of traditional manual inspection and single signal interference, realizes rapid positioning and accurate identification of fault points, provides key technical support for intelligent operation and maintenance of power grids, can effectively shorten the fault repair time, reduce the operation and maintenance cost, and ensure the safe and stable operation of the power system. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 A flowchart of the power transmission line fault positioning method based on acoustic magnetic signal feature analysis of the application is shown in the figure.

[0066] Figure 2 A schematic diagram of the double-flow residual attention network structure of the application is shown in the figure.

[0067] Figure 3 A structure diagram of the power transmission line fault positioning system based on acoustic magnetic signal feature analysis of the application is shown in the figure.

[0068] Figure 4 A schematic diagram of the correction collection module interface of the power transmission line fault positioning system based on acoustic magnetic signal feature analysis of the application is shown in the figure.

[0069] Figure 5 A schematic diagram of the signal processing module interface of the power transmission line fault positioning system based on acoustic magnetic signal feature analysis of the application is shown in the figure. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0071] Embodiment 1

[0072] Please refer to Figure 1As shown, the power line fault positioning method based on the characteristic analysis of the acoustic-magnetic signals in the embodiment includes the following steps:

[0073] The dynamic time delay compensation mechanism is used to linearly correct the magnetic field signal propagation speed of the magnetic field sensor and the ultrasonic signal propagation speed of the ultrasonic sensor, and linearly corrected magnetic field signals and ultrasonic signals are synchronously collected.

[0074] The method for linearly correcting the magnetic field signal propagation speed of the magnetic field sensor and the ultrasonic signal propagation speed of the ultrasonic sensor includes:

[0075] The magnetic field signal propagation speed is calculated based on the relative magnetic permeability of the wire material and the relative permittivity of the wire material, such as the magnetic field signal propagation speed where c is the speed of light; μ r is the relative magnetic permeability of the wire material; ∈ r is the relative permittivity of the wire material.

[0076] The ultrasonic signal propagation speed is calculated based on the wire bulk modulus and the wire material density, such as the ultrasonic signal propagation speed where VW is the wire bulk modulus; and ρ is the wire material density.

[0077] For different wire types, the magnetic field signal propagation speed and the ultrasonic signal propagation speed are linearly corrected in combination with the temperature correction coefficient and the humidity correction coefficient of the wire, such as the magnetic field signal propagation speed adj = v base ×(1+β×ΔWD+γ×ΔSD); where v adj is the dynamically adjusted magnetic field signal or ultrasonic signal standard propagation speed; β is the temperature correction coefficient; ΔWD is the temperature change value; γ is the humidity correction coefficient; ΔSD is the humidity change value; and v base is the magnetic field signal propagation speed or the ultrasonic signal propagation speed.

[0078] The linear correction of the magnetic field and ultrasonic signal propagation speed by the dynamic time delay compensation mechanism can effectively compensate for the deviation caused by the influence of the environment (such as temperature, humidity, etc.) on the signal propagation speed, ensure the time synchronization of the signals collected by different sensors, avoid the fault time judgment error caused by the speed difference, improve the fault positioning accuracy, and based on the accurately corrected propagation speed, in combination with the time difference of the acoustic-magnetic signals reaching different sensors, the fault point position can be more accurately calculated, the positioning error range is reduced, and reliable protection is provided for the rapid repair and stable operation of the power transmission line.

[0079] The magnetic field signal can sensitively capture the magnetic field change caused by the current mutation at the fault moment, and the mutation moment, high-frequency harmonic distortion rate and other characteristics can be used to judge the fault type and provide key parameters for the positioning model, such as the attenuation coefficient. The ultrasonic signal is closely related to the mechanical vibration caused by the fault, and by analyzing the vibration energy integral, the main frequency offset and other characteristics, the fault location can be determined. By combining the two, the differences in their propagation characteristics can be used for time and space correction, and by using energy weighted fusion and other methods in the later stage, the fault point distance can be calculated, which can greatly improve the accuracy and reliability of fault location, and provide strong support for the rapid repair and safe operation of the transmission line. The linearly corrected magnetic field signal and ultrasonic signal can effectively compensate for the time delay caused by the difference in propagation speed, making the two types of signals more accurately correspond in time, and thus more accurately determining the fault occurrence time; The corrected signal can reduce the distance calculation deviation caused by speed error, and based on the accurate signal characteristics and propagation time difference, the fault point position can be more accurately calculated. In addition, these corrected signals can also enhance the recognition of fault features, which can help to extract effective fault information from complex environmental interference and provide more reliable basis for fault type discrimination, and improve the overall accuracy and reliability of transmission line fault location.

[0080] The linearly corrected magnetic field signal is subjected to strong electromagnetic interference suppression and transient pulse suppression to obtain a first magnetic field signal, and the linearly corrected ultrasonic signal is subjected to transient pulse suppression to obtain a first ultrasonic signal. The first ultrasonic signal and the first magnetic field signal are subjected to enhancement processing to obtain a second magnetic field signal and a second ultrasonic signal.

[0081] The method for obtaining the first magnetic field signal comprises:

[0082] The power frequency harmonic filtering processing is performed based on the adaptive notch filter transfer function H(s) to obtain the filtered magnetic field signal; for example, the transfer function is wherein z is a complex variable; Y(z) is an output signal; X(z) is an input signal; μ is a filter parameter, which can be obtained by natural heuristic optimization algorithm; ω0 is a notch frequency, ω0=2π×50Hz;

[0083] In the discrete time system, z -1 is a delay operator, z -1 Z(z)=x(n-1), z -2 X(z)=x(n-2); n is the sampling time; x(n-1) is the input signal corresponding to the sampling time n-1; x(n-2) is the input signal corresponding to the sampling time n-2;

[0084] Y(z)(1-2μcos(ω0)z -1 +μ 2 z -2)=X(z)(1-2cos(ω0)z -1 +z -2 Transforming the above equation into the time domain, we obtain the difference equation y(n)=2μcos(ω0)y(n-1)-μ 2 y(n-2)+x(n)-2cos(ω0)x(n-1)+x(n-2); y(n) is the output signal corresponding to sampling time n; y(n-1) is the output signal corresponding to sampling time n-1; x(n) is the input signal corresponding to sampling time n.

[0085] The error between the filter output and the desired output, e(n) = d(n) - y(n), is calculated using the least mean square algorithm; where d(n) is the desired output.

[0086] Update μ(n)+1) = μ(n) + α × e(n) × x(n) according to the LMS algorithm; where μ(n+1) is the filter parameter corresponding to sampling time n+1; μ(n) is the filter parameter corresponding to sampling time n; α is the step size factor, which is used to control the speed of adaptive adjustment and can be obtained by natural heuristic optimization algorithm;

[0087] The obtained output signal y(n) is used as the filtered magnetic field signal;

[0088] The first magnetic field signal is obtained by reconstructing the filtered magnetic field signal using wavelet packet coefficients that have undergone threshold shrinkage.

[0089] Further suppression of strong electromagnetic interference and transient pulses in the linearly corrected magnetic field signal can make the first magnetic field signal purer. Suppression of strong electromagnetic interference avoids interference from the complex surrounding electromagnetic environment, ensuring the signal accurately reflects the magnetic field characteristics generated by transmission line faults, such as accurately capturing the moment of fault abrupt change and high-frequency harmonic distortion rate. Transient pulse suppression eliminates brief abnormal fluctuations in the signal, making the magnetic field signal characteristics more stable. The resulting first magnetic field signal provides more accurate parameters for fault location, improves the accuracy of fault type identification, and thus enhances the precision and reliability of the fault location model.

[0090] Methods for obtaining wavelet packet coefficients after threshold shrinkage include:

[0091] Select the wavelet basis functions and the corresponding number of decomposition levels;

[0092] Based on the selected wavelet basis function and the number of decomposition levels, wavelet packet decomposition is performed on the filtered magnetic field signal to obtain the wavelet packet coefficients.

[0093] With the goal of minimizing the weighted sum of signal-to-noise ratio and mean square error, a wavelet threshold is obtained by optimization based on a natural heuristic algorithm.

[0094] The wavelet packet coefficients with absolute values less than the wavelet threshold are set to zero, and the wavelet packet coefficients with absolute values greater than or equal to the wavelet threshold are shrunk to zero, to obtain the wavelet packet coefficients after threshold shrinkage processing.

[0095] The method for obtaining the first ultrasonic signal comprises:

[0096] The wavelet packet coefficients after threshold shrinkage processing are used to perform wavelet packet reconstruction on the linearly corrected ultrasonic signal, to obtain the first ultrasonic signal. The transient pulse suppression on the linearly corrected ultrasonic signal can remove abnormal pulses in the signal caused by external impact or internal transient changes of the line. This makes the first ultrasonic signal more accurately reflect the mechanical vibration situation caused by the fault of the power transmission line, such as more reliable key features of vibration energy integral and main frequency offset. Based on the purer first ultrasonic signal, the related information of the sound wave energy peak can be more accurately determined, and when the fault is located in combination with the magnetic field signal, the time and space correction can be better realized, the energy weighted fusion result is more reasonable, and thus the accuracy and stability of the fault location of the power transmission line are improved.

[0097] The method for obtaining the second magnetic field signal and the second ultrasonic signal comprises:

[0098] Step 1, data division: the collected acoustic and magnetic signals are divided into a training set, a validation set and a test set according to a preset proportion; the acoustic and magnetic signals include the first ultrasonic signal and the first magnetic field signal;

[0099] Step 2, model construction: selecting a U-Net structure as a generator, obtaining a generated acoustic and magnetic signal with the same size and channel number as the acoustic and magnetic signal by taking a random noise vector as the input of the generator; selecting a convolutional neural network as a discriminator architecture, obtaining a probability that the input data is the acoustic and magnetic signal by taking the acoustic and magnetic signal and the generated acoustic and magnetic signal as the input of the discriminator;

[0100] Step 3, network training: randomly extracting M acoustic and magnetic signals from the training set, and marking them as 1, indicating real acoustic and magnetic signals; the generator generates corresponding generated acoustic and magnetic signals according to the input random noise vector, and marks them as 0, indicating generated acoustic and magnetic signals; the acoustic and magnetic signals and the generated acoustic and magnetic signals are input into the discriminator, and the discriminator outputs a discrimination result; according to the output of the discriminator and the corresponding labels of the acoustic and magnetic signals and the generated acoustic and magnetic signals, the parameters of the generator and the discriminator are optimized by the back propagation algorithm with the minimum value of the loss function as the target; wherein, the loss function is the sum of the generator loss and the discriminator loss, the generator loss f combines the Maxwell equation constraint and the acoustic wave equation constraint; the discriminator loss is the cross-entropy loss; for example, the generator loss f wherein, is the generated acoustic and magnetic signal; f real is the real acoustic and magnetic signal; ‖·‖2 is the L2 norm; ξ is a regularization parameter; Let B be the curl of the magnetic field strength B in the first magnetic field signal; ∈ is the loss constant; Let E be the partial derivative of the electric field strength E with respect to time t; Let p be the Laplace operator for the sound pressure p of the first ultrasonic signal; The second partial derivative of sound pressure p with respect to time t; discriminator loss. in, D is the mathematical expectation; D(·) is the discriminator function; These are weight parameters; The loss is spectral correlation; repeat the above steps for multiple rounds of iterative training until the generative adversarial network reaches convergence.

[0101] Step 4: Acoustomagnetic signal enhancement: The acoustic magnetic signal is input into the generator of the generative adversarial network corresponding to the convergence state to generate an enhanced acoustic magnetic signal. The enhanced acoustic magnetic signal is used as the enhanced acoustic magnetic signal, namely the second magnetic field signal and the second ultrasonic signal.

[0102] The second magnetic field signal and the second ultrasonic signal, obtained by enhancing the first magnetic field signal and the first ultrasonic signal, have improved amplitude and signal-to-noise ratio, making fault characteristics more prominent. For example, features such as the fault abrupt change time and high-frequency harmonic distortion rate are more obvious in the second magnetic field signal, which helps to more accurately identify the fault type; features such as vibration energy integral and dominant frequency offset of the second ultrasonic signal are also easier to identify, which can more accurately determine the mechanical vibration caused by the fault. In addition, the enhanced signal can reduce false positives and false negatives caused by weak signals, and provide more reliable basic data when calculating the distance to the fault point by multimodal feature fusion, thereby significantly improving the accuracy and reliability of transmission line fault location.

[0103] Feature extraction is performed on the second magnetic field signal and the second ultrasonic signal to obtain magnetic field features and ultrasonic features;

[0104] Methods for obtaining magnetic field characteristics include:

[0105] The abrupt change point of the traveling wave in the second magnetic field signal is calculated as a characteristic of the abrupt change point, such as the traveling wave abrupt change point. Among them, A loop t represents the effective area of ​​the magnetic field sensor; t represents time; B(t) represents the magnetic field strength value at time t.

[0106] Threshold for detecting sudden magnetic field points The second magnetic field signal is detected, where s is the detection threshold coefficient for magnetic field abrupt change points; N is the number of samples for the statistical characteristics of the magnetic field signal; B c This represents the magnetic field strength value at the c-th sample point; is the average value of the magnetic field strength of N sample points; traversing the second magnetic field signal, when the difference value of the magnetic field signals of adjacent time points exceeds the magnetic field mutation point detection threshold, extracting the corresponding magnetic field signal difference value as a difference value feature, and extracting the corresponding two adjacent time points for splicing as a time point feature;

[0107] Based on the voltage amplitude of the second magnetic field signal, the harmonic distortion rate is extracted as a distortion rate feature, such as the harmonic distortion rate wherein, V h is the voltage amplitude of the hth time; V1 is the voltage amplitude of the 1st time; N h is the highest harmonic order;

[0108] Based on the frequency spectrum value, the frequency spectrum center of gravity is extracted as a center of gravity feature, such as the frequency spectrum center of gravity wherein, f is the frequency; F is the maximum frequency; X(f) is the frequency spectrum value of the second magnetic field signal at the frequency f;

[0109] Splicing S mutation point features, S difference value features, S time point features, S distortion rate features and S center of gravity features as magnetic field features, wherein the features less than S are zero-padded.

[0110] The magnetic field features such as the traveling wave mutation point and the high-frequency harmonic distortion rate can effectively reflect the electromagnetic change at the fault moment, and can be used for accurate discrimination of the fault type (such as lightning stroke, broken strand, etc.) in the later stage, and then select appropriate attenuation coefficient and propagation speed correction rule for the positioning model, and provide important basis for fault point distance calculation.

[0111] The method for obtaining the ultrasonic wave feature comprises:

[0112] Obtaining a second ultrasonic wave signal g, decomposing g into k modal components, each modal component having a corresponding center frequency ω k ; constructing a constrained variational model; such as the constrained variational model The constraint condition is wherein, u k is the kth modal component; ω k is the center frequency of the kth modal component u k ; K is the total number of modal components decomposed from the second ultrasonic wave signal g; k is the number of modal components decomposed from the second ultrasonic wave signal g; is the derivation operation with respect to time t; δ(t) is the Dirac function; j is the imaginary unit; * is the convolution operation; e is a mathematical constant; t is time;

[0113] Introducing a Lagrange multiplier and a quadratic penalty term, the constrained variational problem is converted into an unconstrained augmented Lagrange function; such as the augmented Lagrange function λ is the Lagrange multiplier;

[0114] The augmented Lagrange function is iteratively solved by using the alternating direction multiplier method;

[0115] Update the modal function: fix ω k and λ, for each k, by solving Update u k ; is the kth modal component after the hth update; i is the number of modal components decomposed from the ultrasonic signal g;

[0116] Update the center frequency: fix u k and λ, by solving Update ω k ; is the center frequency of the kth modal component after the hth update ;

[0117] The Lagrange multiplier is updated according to ; where λ h+1 is the Lagrange multiplier after the hth update; λ h is the Lagrange multiplier before the hth update; τ is the update step size;

[0118] Repeat the above update steps until the changes in the modal functions in two consecutive iterations satisfy ; where tol is the convergence tolerance; is the kth modal component before the hth update;

[0119] Based on the above steps, K modal components u k are obtained, the time domain features and frequency domain features of each modal component are extracted, the time domain features include amplitude features and time features, the amplitude features are the mean, variance, maximum and minimum of the amplitude of the modal component, the time features are the rise time, fall time and pulse width of the modal component; the frequency domain features include frequency features and energy features, the frequency features are the peak frequency, center frequency and bandwidth of the modal component, the energy features are the energy distribution of the modal component in different frequency bands; the time domain features and frequency domain features are spliced to obtain the ultrasonic features.

[0120] The ultrasonic features include time domain features and frequency domain features of different modal components, which are closely related to the mechanical vibration caused by the fault and help to determine the fault location and nature from another dimension. By analyzing these features, the direction and energy distribution of the fault point can be determined, combined with the magnetic field features, multi-modal information fusion can be realized, the distance of the fault point can be calculated more accurately, the accuracy and reliability of fault location can be improved, and the severity of the fault can also be evaluated to provide reference for subsequent maintenance decision.

[0121] The method for obtaining the time feature comprises:

[0122] presetting an upper threshold and a lower threshold, traversing the corresponding modal component to obtain a first signal point greater than the lower threshold, continuing to traverse the corresponding modal component from the first signal point to obtain a second signal point greater than the upper threshold, and calculating the product of the time difference between the second signal point and the first signal point and the sampling period as the rise time;

[0123] traversing the corresponding modal component in reverse from the end position to obtain a third signal point less than the upper threshold, continuing to traverse the corresponding modal component from the third signal point to obtain a fourth signal point less than the lower threshold, and calculating the product of the time difference between the fourth signal point and the third signal point and the sampling period as the fall time;

[0124] obtaining the maximum value and the minimum value of the modal component, calculating the intermediate value of the maximum value and the minimum value, traversing the corresponding modal component to obtain a fifth signal point greater than the intermediate value, continuing to traverse the corresponding modal component from the fifth signal point to obtain a sixth signal point less than the upper threshold, and calculating the product of the difference between the sixth signal point and the fifth signal point and the sampling period as the pulse width.

[0125] The method for obtaining the energy feature comprises:

[0126] calculating the sum of the energies corresponding to all frequency domain indexes in the preset frequency range of the modal component, and the energy is the square of the amplitude.

[0127] taking the ultrasonic wave feature as the input of the fault recognition model, obtaining the mechanical vibration mode and the corresponding fault type;

[0128] The training method of the fault recognition model comprises:

[0129] pre-acquiring Q sets of type recognition data, and the type recognition data comprises the ultrasonic wave feature, the mechanical vibration mode and the corresponding fault type.

[0130] taking the ultrasonic wave feature as the input of the fault recognition model, taking the mechanical vibration mode and the corresponding fault type as the output of the fault recognition model, taking the minimum error between the output mechanical vibration mode and the corresponding fault type and the actual mechanical vibration mode and the corresponding fault type as the target, optimizing the network parameters of the fault recognition model through the natural heuristic optimization algorithm, obtaining the network parameters corresponding to the minimum error between the output mechanical vibration mode and the corresponding fault type of the fault recognition model and the actual mechanical vibration mode and the corresponding fault type, and constructing the fault recognition model with the corresponding network parameters as the trained fault recognition model.

[0131] By accurately identifying the unique mechanical vibration patterns caused by different faults such as loosening and breaking through ultrasonic characteristics, the fault type can be determined, which can help to narrow down the location. At the same time, the propagation characteristics of different vibration patterns can assist in judging the approximate location of the fault, and combined with other positioning methods, it can further accurately locate the fault and provide a basis for subsequent targeted maintenance, improving the efficiency and safety of power transmission line maintenance.

[0132] The magnetic field characteristics and ultrasonic characteristics are taken as inputs of the dual-flow residual attention network to obtain a fused joint feature mapping. Based on the joint feature mapping of adjacent monitoring terminals, the three-dimensional coordinates of the fault point are calculated by D-S evidence theory fusion. The joint feature mapping includes magnetic field characteristics, acoustic wave characteristics, and cross-modal correlation characteristics. The magnetic field characteristics include the row wave mutation time, high-frequency harmonic distortion rate, and magnetic field polarity reversal times. The acoustic wave characteristics include vibration energy integration, main frequency offset, and waveguide direction. The cross-modal correlation characteristics include the time sequence alignment deviation of magnetic field mutation and acoustic wave energy peak and the inter-modal weight distribution coefficient.

[0133] The training method of the dual-flow residual attention network includes:

[0134] R sets of training data are collected in advance, including acoustic-magnetic characteristics and corresponding joint feature mappings. The acoustic-magnetic characteristics include magnetic field characteristics and ultrasonic characteristics.

[0135] The acoustic-magnetic characteristics are taken as inputs of the dual-flow residual attention network, and the joint feature mapping is taken as the output of the dual-flow residual attention network. The error of the loss function is minimized as the goal. The network parameters of the dual-flow residual attention network are optimized by a natural heuristic optimization algorithm to obtain the network parameters corresponding to the minimum error of the joint feature mapping output by the dual-flow residual attention network and the actual joint feature mapping. The dual-flow residual attention network constructed by the corresponding network parameters is taken as the trained dual-flow residual attention network. Referring to Figure 2 , the dual-flow residual attention network includes a magnetic field branch and an ultrasonic wave branch. The magnetic field branch includes NG layers (such as 5 layers) of ResNet and a channel attention module. The ultrasonic wave branch includes NM layers (such as 3 layers) of CNN+LSTM and a spatial attention module. The outputs of the magnetic field branch and the ultrasonic wave branch are fused through a cross-attention mechanism to obtain the output joint feature mapping. The loss function of the dual-flow residual attention network includes a classification loss composed of cross-entropy and a positioning loss composed of mean square error. For example, the total loss wherein, is the classification loss; is a weight balancing coefficient, which can be obtained by optimization through a natural heuristic optimization algorithm; is the positioning loss.

[0136] The double-flow residual attention network can significantly improve the fault location accuracy by integrating the magnetic field features and ultrasonic wave features, combining the cross-modal correlation features (such as modal weight coefficients). The network uses the attention mechanism to dynamically allocate modal weights, captures the spatio-temporal correlation between acoustic and magnetic signals (such as the timing alignment of magnetic field mutations and acoustic energy peaks), generates a joint feature mapping containing multi-modal complementary information, effectively suppresses single-modal noise interference, enhances the recognition of fault features under complex working conditions, and provides more comprehensive and reliable input basis for the subsequent positioning model.

[0137] The method for obtaining the three-dimensional coordinates of the fault point comprises:

[0138] Based on the joint feature mapping, a fault distance is obtained by combining a fault location model; for example, the fault distance Wherein, ; is the attenuation coefficient, f THD is the total harmonic distortion rate; k material is the material correction factor; E source is the energy weighted fusion value, E mag is the magnetic field energy, E sonic is the acoustic energy, and ζ is the inter-modal weight distribution coefficient; E received is the received signal energy; v sonic is the acoustic signal propagation speed; Δt align is the timing deviation compensation value;

[0139] The fault distances calculated by the three adjacent monitoring terminals are fused by the D-S evidence theory to obtain the three-dimensional coordinates of the fault point; for example, the three-dimensional coordinates of the fault point Wherein, d r is the fault point distance calculated by the rth monitoring terminal; (x r , y r , z r ) is the three-dimensional coordinates of the rth monitoring terminal in space.

[0140] The D-S evidence theory fuses the spatio-temporal information of the acoustic and magnetic signals of multiple terminals to construct a multi-source evidence body, reduces the single-point positioning error, uses the uncertainty reasoning ability of the evidence theory to comprehensively consider the confidence of different terminals on the position of the fault point, combines the signal propagation speed correction coefficient, and accurately calculates the three-dimensional coordinates of the fault point; through the cooperation of multiple terminals, the influence of environmental noise and signal attenuation is eliminated, which is especially suitable for fault location under long-distance power transmission lines or complex terrain, and can significantly improve the reliability and robustness of the positioning result.

[0141] Embodiment 2

[0142] The embodiment provides a self-adaptive dynamic time delay compensation mechanism, which introduces an LSTM network to predict temperature and humidity correction coefficients corresponding to a magnetic field signal propagation speed or an ultrasonic signal propagation speed.

[0143] The training method of the LSTM network comprises the following steps:

[0144] L groups of type identification data are collected in advance, the type identification data comprising input training data and output training data; the input training data comprising a first data group or a second data group, the first data group comprising temperature, humidity and a magnetic field signal propagation speed, and the second data group comprising temperature, humidity and an ultrasonic signal propagation speed; and the output training data comprising corresponding temperature correction coefficients and humidity correction coefficients.

[0145] The input training data is taken as the input of the LSTM network, the output training data is taken as the output of the LSTM network, and the error between the output training data and the actual output training data is minimized as the target; the network parameters of the LSTM network are optimized through a natural heuristic optimization algorithm, the network parameters corresponding to the minimum error between the output training data and the actual output training data are obtained, and the LSTM network constructed by the corresponding network parameters is taken as the trained LSTM network.

[0146] Embodiment 3

[0147] Please refer to Figure 3 The power transmission line fault positioning system based on the acoustic-magnetic signal feature analysis comprises:

[0148] The correction acquisition module performs linear correction processing on the magnetic field signal propagation speed of the magnetic field sensor and the ultrasonic signal propagation speed of the ultrasonic sensor through a dynamic time delay compensation mechanism, and synchronously acquires linearly corrected magnetic field signals and ultrasonic signals;

[0149] The linear correction processing on the magnetic field signal propagation speed of the magnetic field sensor and the ultrasonic signal propagation speed of the ultrasonic sensor comprises the following steps:

[0150] The magnetic field signal propagation speed is calculated based on the relative magnetic permeability of the wire material and the relative node constant of the wire material;

[0151] The ultrasonic signal propagation speed is calculated based on the wire volume modulus and the wire material density;

[0152] The magnetic field signal propagation speed and the ultrasonic signal propagation speed are linearly corrected in combination with the temperature correction coefficients and the humidity correction coefficients of the wire for different wire types.

[0153] Since the dynamic time delay compensation mechanism in Embodiment 1 is difficult to adapt to changes in environmental factors such as temperature and humidity in real time, the adaptive dynamic time delay compensation mechanism in Embodiment 2 can be combined to optimize the temperature correction coefficient and humidity correction coefficient corresponding to the magnetic field signal propagation speed or the ultrasonic wave signal propagation speed, thereby improving the accuracy of the correction results of the magnetic field signal propagation speed and the ultrasonic wave signal propagation speed. Referring to Figure 4 , the LSTM network is used to optimize the temperature correction coefficient and humidity correction coefficient corresponding to the magnetic field signal propagation speed and the ultrasonic wave signal propagation speed, respectively. The temperature correction coefficient corresponding to the magnetic field signal propagation speed is 1.15, and the humidity correction coefficient is 0.88. The temperature correction coefficient corresponding to the ultrasonic wave signal propagation speed is 0.87, and the humidity correction coefficient is 1.05. The training frequency is 20,000 times, and the training accuracy is 98.5%, indicating that the model has been trained a large number of times and has high reliability.

[0154] The signal processing module: the linearly corrected magnetic field signal is subjected to strong electromagnetic interference suppression and transient pulse suppression to obtain a first magnetic field signal, the linearly corrected ultrasonic wave signal is subjected to transient pulse suppression to obtain a first ultrasonic wave signal, and the first ultrasonic wave signal and the first magnetic field signal are subjected to enhancement processing to obtain a second magnetic field signal and a second ultrasonic wave signal.

[0155] The steps of obtaining the first magnetic field signal are as follows:

[0156] The power frequency harmonic filtering processing is performed based on the adaptive notch filter transfer function to obtain the filtered magnetic field signal.

[0157] The wavelet packet coefficients subjected to threshold shrinkage processing are used to reconstruct the filtered magnetic field signal to obtain the first magnetic field signal.

[0158] The steps of obtaining the wavelet packet coefficients subjected to threshold shrinkage processing are as follows:

[0159] The wavelet basis function and the corresponding decomposition layer number are selected;

[0160] The filtered magnetic field signal is subjected to wavelet packet decomposition based on the selected wavelet basis function and the decomposition layer number to obtain the wavelet packet coefficients.

[0161] The wavelet threshold is obtained based on the natural heuristic optimization algorithm with the weighted sum of the signal-to-noise ratio and the mean square error being the minimum as the target.

[0162] The wavelet packet coefficients with absolute values less than the wavelet threshold are set to zero, and the wavelet packet coefficients with absolute values greater than or equal to the wavelet threshold are shrunk to zero to obtain the wavelet packet coefficients subjected to threshold shrinkage processing. Referring to Figure 5The initial threshold value of the wavelet packet coefficient is set to 0.25, and the iteration number is set to 100; the preset wavelet base function is db4, the decomposition layer number is 4 layers, and the optimization state has converged, in the wavelet packet coefficient optimization process, the population size of the natural heuristic optimization algorithm is 50, the maximum evolution generation number is 200, the obtained optimization threshold value is 0.487, the optimal fitness is 0.956, and the optimization converges at the 47th generation, and the convergence time is 8.5 seconds, through the analysis of the wavelet packet coefficient distribution, a series of operations and results of threshold optimization. Real-time data monitoring ensures dynamic tracking of the signal, and the setting of the optimization state and algorithm configuration parameters ensures the effectiveness and efficiency of the optimization process, and the final optimization result provides an important parameter basis for subsequent signal processing and fault analysis.

[0163] The method for obtaining the first ultrasonic signal comprises:

[0164] The wavelet packet reconstruction is performed on the linearly corrected ultrasonic signal by using the wavelet packet coefficient subjected to the threshold shrinkage processing, and the first ultrasonic signal is obtained.

[0165] The method for obtaining the second magnetic field signal and the second ultrasonic signal comprises:

[0166] The collected acoustic magnetic signals are divided into a training set, a verification set and a test set according to a preset ratio; the acoustic magnetic signals comprise the first ultrasonic signal and the first magnetic field signal;

[0167] The U-Net structure is selected as the generator, the random noise vector is selected as the input of the generator, the generated acoustic magnetic signal with the same size and channel number as the acoustic magnetic signal is obtained, the convolutional neural network is selected as the discriminator architecture, the acoustic magnetic signal and the generated acoustic magnetic signal are selected as the input of the discriminator, and the probability that the input data is the acoustic magnetic signal is obtained;

[0168] M groups of acoustic magnetic signals are randomly extracted from the training set and marked as 1, representing the real acoustic magnetic signal; the corresponding generated acoustic magnetic signal is generated by the generator according to the input random noise vector, and is marked as 0, representing the generated acoustic magnetic signal; the acoustic magnetic signal and the generated acoustic magnetic signal are input into the discriminator, and the discriminator outputs the discrimination result; according to the output of the discriminator and the corresponding labels of the acoustic magnetic signal and the generated acoustic magnetic signal, the parameters of the generator and the discriminator are optimized by the natural heuristic optimization algorithm, with the value of the loss function being minimized as the target; wherein the loss function is the sum of the generator loss and the discriminator loss, the generator loss fuses the difference term of the generated acoustic magnetic signal and the real acoustic magnetic signal, the Maxwell equation constraint and the acoustic wave equation constraint; the discriminator loss is the cross-entropy loss; the step of selecting the acoustic magnetic signal for discrimination is repeated, and the multi-round iteration training is performed until the generative adversarial network reaches the convergence state;

[0169] The acoustic-magnetic signal is input into the generator of the generative adversarial network corresponding to the convergence state to generate an enhanced acoustic-magnetic signal, and the enhanced acoustic-magnetic signal is taken as the acoustic-magnetic signal after enhancement processing, that is, a second magnetic field signal and a second ultrasonic signal.

[0170] The feature extraction module: performing feature extraction on the second magnetic field signal and the second ultrasonic signal to obtain magnetic field features and ultrasonic features;

[0171] The method for obtaining the magnetic field features comprises:

[0172] The running wave mutation point of the second magnetic field signal is calculated as a mutation point feature;

[0173] The second magnetic field signal is detected based on a magnetic field mutation point detection threshold; the second magnetic field signal is traversed, and when the magnetic field signal difference of adjacent time points exceeds the magnetic field mutation point detection threshold, the corresponding magnetic field signal difference is extracted as a difference value feature, and the corresponding two adjacent time points are spliced as a time point feature;

[0174] The harmonic distortion rate is extracted based on the voltage amplitude of the second magnetic field signal as a distortion rate feature;

[0175] The spectral barycenter is extracted based on the spectral value as a barycenter feature;

[0176] The S mutation point features, S difference value features, S time point features, S distortion rate features, and S barycenter features are spliced as the magnetic field features, wherein the features less than S are zero-padded.

[0177] The method for obtaining the ultrasonic features comprises:

[0178] Step 1, obtaining a second ultrasonic signal g, decomposing g into k modal components, each modal component having a corresponding center frequency; constructing a constrained variational model;

[0179] Step 2, introducing a Lagrange multiplier and a quadratic penalty term, and converting the constrained variational problem into an unconstrained augmented Lagrange function;

[0180] Step 3, using the alternating direction multiplier method to iteratively solve the augmented Lagrange function;

[0181] Step 4, updating the modal function;

[0182] Step 5, updating the center frequency;

[0183] Step 6, updating the Lagrange multiplier;

[0184] Step 7, repeating steps 4-6 until the change of each modal function in two consecutive iterations is less than a convergence tolerance, and stopping the iteration;

[0185] Step 8, obtaining K modal components based on steps 1-7, extracting time domain features and frequency domain features of each modal component, the time domain features including amplitude features and time features, the amplitude features being amplitude mean, amplitude variance, amplitude maximum and amplitude minimum of the modal component, the time features being rise time, fall time and pulse width of the modal component, the frequency domain features including frequency features and energy features, the frequency features being peak frequency, center frequency and bandwidth of the modal component, the energy features being energy distribution of the modal component in different frequency bands; and splicing the time domain features and the frequency domain features to obtain ultrasonic wave features.

[0186] The method for obtaining the time features includes:

[0187] presetting an upper threshold and a lower threshold, obtaining a first signal point greater than the lower threshold from the corresponding modal component, obtaining a second signal point greater than the upper threshold from the corresponding modal component starting from the first signal point, and calculating the product of the time difference between the second signal point and the first signal point and the sampling period as the rise time;

[0188] reversely traversing the corresponding modal component from the end position to obtain a third signal point less than the upper threshold, obtaining a fourth signal point less than the lower threshold from the corresponding modal component starting from the third signal point, and calculating the product of the time difference between the fourth signal point and the third signal point and the sampling period as the fall time;

[0189] obtaining the maximum value and the minimum value of the modal component, calculating the intermediate value of the maximum value and the minimum value, obtaining a fifth signal point greater than the intermediate value from the corresponding modal component, obtaining a sixth signal point less than the upper threshold from the corresponding modal component starting from the fifth signal point, and calculating the product of the difference between the sixth signal point and the fifth signal point and the sampling period as the pulse width.

[0190] The method for obtaining the energy features includes:

[0191] calculating the sum of the energies corresponding to all frequency domain indexes in the preset frequency band of the modal component, the energy being the square of the amplitude.

[0192] The fault identification module: taking the ultrasonic wave features as the input of the fault identification model to obtain the mechanical vibration mode and the corresponding fault type.

[0193] The training method of the fault identification model includes:

[0194] pre-collecting Q sets of type identification data, the type identification data including ultrasonic wave features, mechanical vibration modes and corresponding fault types.

[0195] The ultrasonic wave features are taken as inputs of the fault recognition model, the mechanical vibration mode and the corresponding fault type are taken as outputs of the fault recognition model, and the error between the mechanical vibration mode and the corresponding fault type output by the fault recognition model and the actual mechanical vibration mode and the corresponding fault type is minimized as the target.

[0196] The magnetic field features and the ultrasonic wave features are taken as inputs of the double-flow residual attention network, a fused joint feature map is obtained, the joint feature maps of three adjacent monitoring terminals are fused through D-S evidence theory, and a three-dimensional coordinate of the fault point is calculated.

[0197] The training method of the double-flow residual attention network comprises:

[0198] R groups of training data are collected in advance, the training data comprise acoustic-magnetic features and corresponding joint feature maps, the acoustic-magnetic features comprise magnetic field features and ultrasonic wave features;

[0199] The acoustic-magnetic features are taken as inputs of the double-flow residual attention network, the joint feature maps are taken as outputs of the double-flow residual attention network, the error between the joint feature maps output by the double-flow residual attention network and the actual joint feature maps is minimized as the target, the network parameters of the double-flow residual attention network are optimized through a natural heuristic optimization algorithm, the network parameters corresponding to the minimum error between the joint feature maps output by the double-flow residual attention network and the actual joint feature maps are obtained, and the double-flow residual attention network constructed by the corresponding network parameters is taken as the trained double-flow residual attention network; the double-flow residual attention network comprises a magnetic field branch and an ultrasonic wave branch, the magnetic field branch comprises NG layers of ResNet and one channel attention module, the ultrasonic wave branch comprises NM layers of CNN+LSTM and one spatial attention module, the outputs of the magnetic field branch and the ultrasonic wave branch are fused through a cross-attention mechanism to obtain the output joint feature maps; and a loss function of the double-flow residual attention network comprises a classification loss composed of cross-entropy and a positioning loss composed of mean square error.

[0200] The method for obtaining the three-dimensional coordinate of the fault point comprises:

[0201] The joint feature map is combined with the fault positioning model to perform distance inversion, and the fault distance is obtained.

[0202] The fault distances calculated by the three adjacent monitoring terminals are fused through D-S evidence theory, and the three-dimensional coordinate of the fault point is calculated.

[0203] The above merely describes specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any modification or substitution within the technical range disclosed by the present application can be easily thought of by those skilled in the art, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0204] Finally, the above merely describes preferred embodiments of the present application, and is not used to limit the present application, any modification, equivalent substitution, improvement, etc. within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for locating a fault on a power transmission line based on analysis of characteristics of magneto-acoustic signals, characterized in that, The method comprises the following steps: The magnetic field signal propagation speed of the magnetic field sensor and the ultrasonic signal propagation speed of the ultrasonic sensor are linearly corrected by a dynamic time delay compensation mechanism, and linearly corrected magnetic field signals and ultrasonic signals are synchronously collected; The linearly corrected magnetic field signals are subjected to strong electromagnetic interference suppression and transient pulse suppression to obtain first magnetic field signals, the linearly corrected ultrasonic signals are subjected to transient pulse suppression to obtain first ultrasonic signals, and the first ultrasonic signals and the first magnetic field signals are subjected to enhancement processing to obtain second magnetic field signals and second ultrasonic signals; The second magnetic field signals and the second ultrasonic signals are subjected to feature extraction to obtain magnetic field features and ultrasonic features; The ultrasonic features are taken as inputs of a fault recognition model to obtain mechanical vibration modes and corresponding fault types; The magnetic field features and the ultrasonic features are taken as inputs of a double-flow residual attention network to obtain fused joint feature mappings, and the joint feature mappings of three adjacent monitoring terminals are fused by D-S evidence theory to calculate and obtain three-dimensional coordinates of a fault point.

2. The power transmission line fault location method based on the analysis of the signature of the acoustomagnetic signal according to claim 1, characterized in that, The method for linearly correcting the magnetic field signal propagation speed of the magnetic field sensor and the ultrasonic signal propagation speed of the ultrasonic sensor comprises: The magnetic field signal propagation speed is calculated based on the relative magnetic permeability of the wire material and the relative node constant of the wire material; The ultrasonic signal propagation speed is calculated based on the wire volume modulus and the wire material density; The magnetic field signal propagation speed and the ultrasonic signal propagation speed are linearly corrected in combination with the temperature correction coefficient and the humidity correction coefficient of the wire for different wire types.

3. The method for power transmission line fault location based on the analysis of the signature of the acoustic-magnetic signal according to claim 1, characterized in that, The training method of the double-flow residual attention network comprises: R groups of training data are collected in advance, the training data comprising acoustic-magnetic features and corresponding joint feature mappings, the acoustic-magnetic features comprising magnetic field features and ultrasonic features; The acoustic-magnetic features are taken as inputs of the double-flow residual attention network, the joint feature mappings are taken as outputs of the double-flow residual attention network, and the error between the output joint feature mappings and the actual joint feature mappings is minimized as the target, the network parameters of the double-flow residual attention network are optimized by a natural heuristic optimization algorithm, the network parameters corresponding to the minimum error between the output joint feature mappings and the actual joint feature mappings of the double-flow residual attention network are obtained, and the double-flow residual attention network constructed by the corresponding network parameters is taken as the trained double-flow residual attention network; the double-flow residual attention network comprises a magnetic field branch and an ultrasonic branch, the magnetic field branch comprises NG layers of ResNet and one channel attention module, the ultrasonic branch comprises NM layers of CNN+LSTM and one spatial attention module, the outputs of the magnetic field branch and the ultrasonic branch are fused by a cross-attention mechanism to obtain the output joint feature mappings; the loss function of the double-flow residual attention network comprises a classification loss composed of cross-entropy and a positioning loss composed of mean square error; The method for obtaining the three-dimensional coordinates of the fault point comprises: The joint feature mappings are combined with a fault positioning model to perform distance inversion and obtain fault distances; The fault distances calculated by the three adjacent monitoring terminals are fused by D-S evidence theory to calculate and obtain the three-dimensional coordinates of the fault point.

4. The method for power transmission line fault location based on the analysis of the signature of the acoustic-magnetic signal according to claim 1, characterized in that, The method for obtaining the first magnetic field signal comprises: Performing power frequency harmonic filtering processing based on an adaptive notch filter transfer function to obtain a filtered magnetic field signal; Performing wavelet packet reconstruction on the filtered magnetic field signal using the wavelet packet coefficients after threshold shrinkage processing to obtain the first magnetic field signal.

5. The method for power transmission line fault location based on the analysis of the signature of the acoustic-magnetic signal according to claim 3, characterized in that, The method for obtaining the wavelet packet coefficients after threshold shrinkage processing comprises: Selecting a wavelet base function and a corresponding decomposition layer number; Performing wavelet packet decomposition on the filtered magnetic field signal according to the selected wavelet base function and the decomposition layer number to obtain the wavelet packet coefficients; Optimizing the wavelet threshold based on a natural heuristic optimization algorithm to minimize the weighted sum of the signal-to-noise ratio and the mean square error; Setting the wavelet packet coefficients with absolute values less than the wavelet threshold to zero and shrinking the wavelet packet coefficients with absolute values greater than or equal to the wavelet threshold to zero to obtain the wavelet packet coefficients after threshold shrinkage processing.

6. The power transmission line fault location method based on the analysis of the signature of the acoustomagnetic signal according to claim 4, characterized in that, The method for obtaining the second magnetic field signal and the second ultrasonic signal comprises: Dividing the collected acoustic-magnetic signals into a training set, a validation set and a test set according to a preset proportion; the acoustic-magnetic signals include the first ultrasonic signal and the first magnetic field signal; Selecting a U-Net structure as a generator, inputting a random noise vector into the generator to obtain a generated acoustic-magnetic signal with the same size and channel number as the acoustic-magnetic signal; selecting a convolutional neural network as a discriminator architecture, inputting the acoustic-magnetic signal and the generated acoustic-magnetic signal into the discriminator to obtain a probability that the input data is the acoustic-magnetic signal; Randomly selecting M groups of acoustic-magnetic signals from the training set, marking them as 1, indicating real acoustic-magnetic signals; the generator generates corresponding generated acoustic-magnetic signals according to the input random noise vector, marking them as 0, indicating generated acoustic-magnetic signals; inputting the acoustic-magnetic signal and the generated acoustic-magnetic signal into the discriminator, the discriminator outputs a discrimination result; optimizing the parameters of the generator and the discriminator through a natural heuristic optimization algorithm to minimize the value of a loss function, wherein the loss function is the sum of the generator loss and the discriminator loss, the generator loss combines the difference between the generated acoustic-magnetic signal and the real acoustic-magnetic signal, the Maxwell equation constraint and the acoustic wave equation constraint; the discriminator loss is the cross-entropy loss; repeating the steps of selecting acoustic-magnetic signals for discrimination to perform multiple rounds of iterative training until the generative adversarial network reaches a convergent state; Inputting the acoustic-magnetic signal into the generator of the convergent state corresponding generative adversarial network to generate an enhanced acoustic-magnetic signal, taking the enhanced acoustic-magnetic signal as the enhanced acoustic-magnetic signal, i.e., the second magnetic field signal and the second ultrasonic signal.

7. The method for power transmission line fault location based on analysis of signature of acoustic magnetic signals as claimed in claim 1 wherein, The method for obtaining the magnetic field features comprises: Calculating the traveling wave mutation point of the second magnetic field signal as a mutation point feature; Detecting the second magnetic field signal based on a magnetic field mutation point detection threshold; traversing the second magnetic field signal, when the magnetic field signal difference of adjacent time points exceeds the magnetic field mutation point detection threshold, extracting the corresponding magnetic field signal difference as a difference feature and extracting the corresponding two adjacent time points for splicing as a time point feature; Extracting the harmonic distortion rate based on the voltage amplitude of the second magnetic field signal as a distortion rate feature; Extracting the spectral barycenter based on the spectral value as a barycenter feature; The S mutation point features, S difference features, S time point features, S distortion rate features and S gravity center features are spliced as the magnetic field features, wherein the features less than S are zero-padded.

8. The method for power transmission line fault location based on analysis of signature of acoustic magnetic signals as claimed in claim 1 wherein, The method for obtaining the ultrasonic wave features comprises: Step 1, obtaining a second ultrasonic signal g, decomposing g into k modal components, each modal component having a corresponding center frequency, and constructing a constrained variational model; Step 2, introducing a Lagrange multiplier and a quadratic penalty term, and converting the constrained variational problem into an unconstrained augmented Lagrange function; Step 3, iteratively solving the augmented Lagrange function by using an alternating direction multiplier method; Step 4, updating the modal function; Step 5, updating the center frequency; Step 6, updating the Lagrange multiplier; Step 7, repeating steps 4-6 until the change of each modal function in two consecutive iterations is less than a convergence tolerance; Step 8, obtaining K modal components based on steps 1-7, extracting time domain features and frequency domain features of each modal component, the time domain features including amplitude features and time features, the amplitude features being amplitude mean, amplitude variance, amplitude maximum and amplitude minimum of the modal component, the time features being rise time, fall time and pulse width of the modal component, the frequency domain features including frequency features and energy features, the frequency features being peak frequency, center frequency and bandwidth of the modal component, the energy features being energy distribution of the modal component in different frequency bands; and splicing the time domain features and the frequency domain features to obtain the ultrasonic wave features.

9. The power transmission line fault location method based on the analysis of the signature of the acoustomagnetic signal according to claim 8, characterized in that, The method for obtaining the time features comprises: presetting an upper threshold and a lower threshold, traversing the corresponding modal component to obtain a first signal point greater than the lower threshold, continuing to traverse the corresponding modal component from the first signal point to obtain a second signal point greater than the upper threshold, and calculating the product of the time difference between the second signal point and the first signal point and the sampling period as the rise time; traversing the corresponding modal component in reverse from the end position to obtain a third signal point less than the upper threshold, continuing to traverse the corresponding modal component from the third signal point to obtain a fourth signal point less than the lower threshold, and calculating the product of the time difference between the fourth signal point and the third signal point and the sampling period as the fall time; obtaining the maximum value and the minimum value of the modal component, calculating the intermediate value of the maximum value and the minimum value, traversing the corresponding modal component to obtain a fifth signal point greater than the intermediate value, continuing to traverse the corresponding modal component from the fifth signal point to obtain a sixth signal point less than the upper threshold, and calculating the product of the difference between the sixth signal point and the fifth signal point and the sampling period as the pulse width; The method for obtaining the energy features comprises: calculating the sum of the energies corresponding to all frequency domain indexes in a preset frequency band of the modal component, the energy being the square of the amplitude.

10. A power line fault location system based on analysis of acoustomagnetic signal characteristics, implementing the method of any one of claims 1-9, characterized in that, It comprises: a correction acquisition module: linearly correcting the magnetic field signal propagation speed of the magnetic field sensor and the ultrasonic wave signal propagation speed of the ultrasonic wave sensor through a dynamic time delay compensation mechanism, and synchronously acquiring linearly corrected magnetic field signals and ultrasonic wave signals after linear correction; The signal processing module: the linearly corrected magnetic field signal is subjected to strong electromagnetic interference suppression and transient pulse suppression to obtain a first magnetic field signal, the linearly corrected ultrasonic signal is subjected to transient pulse suppression to obtain a first ultrasonic signal, and the first ultrasonic signal and the first magnetic field signal are subjected to enhancement processing to obtain a second magnetic field signal and a second ultrasonic signal; The feature extraction module: the second magnetic field signal and the second ultrasonic signal are subjected to feature extraction to obtain a magnetic field feature and an ultrasonic feature; The fault identification module: the ultrasonic feature is taken as an input of a fault identification model to obtain a mechanical vibration mode and a corresponding fault type; The fault positioning module: the magnetic field feature and the ultrasonic feature are taken as inputs of a double-flow residual attention network to obtain a fused joint feature mapping, the joint feature mappings of three adjacent monitoring terminals are fused through D-S evidence theory, and a three-dimensional coordinate of a fault point is calculated and obtained.

Images