A Method, Device and Medium for Identifying and Compensating Multipath Interference of Cable Partial Discharge Signals

Abstract

The present invention relates to a method, device and medium for identifying and compensating multi-path interference of cable partial discharge signals. The method includes the following steps: using a plurality of high-frequency current sensors to synchronously collect multi-channel signals on different propagation paths at different sampling times; calculating the propagation paths based on the collected signals using time differences, and extracting the time-frequency characteristics of the signals; constructing input features of a deep learning model based on the collected signals, time differences, propagation paths and time-frequency characteristics, and using the deep learning model to identify multi-path interference and output the classification result of the signals; constructing a plurality of reflected wave signal models based on the classification result of the signals; and removing the reflected wave signals from the direct wave signals based on the reflected wave signal models to obtain compensated signals. Compared with the prior art, the present invention has the advantages of being able to accurately distinguish direct waves and reflected waves, accurately restoring direct wave signals, etc.

Description

Technical Field

[0001] The present invention relates to the technical field of cable partial discharge signal processing, and in particular to a method, device and medium for identifying and compensating multipath interference of cable partial discharge signals. Background Art

[0002] Partial discharge (PD) is an important diagnostic basis for cable insulation defects, and its signal characteristics directly reflect the insulation state of the cable. However, the PD signals propagating in the cable will form a multipath interference effect due to reasons such as medium discontinuity, port reflection, and joint scattering, resulting in signal superposition, amplitude attenuation, and spectrum distortion at the receiving end. This multipath interference effect increases the difficulty of extracting PD signals, may cause PD source localization deviation, feature recognition errors, and inaccurate diagnostic results. Therefore, there is an urgent need for an effective method to accurately identify the direct wave and the reflected wave and compensate for the interference to restore the true PD signal.

[0003] At present, there have been some technical progresses in the field of PD signal detection at home and abroad. For example, high-frequency current sensors (HFCTs) and ultrasonic sensors are widely used at home and abroad to collect PD signals, and signal processing methods such as short-time Fourier transform (STFT) or wavelet transform are used to extract the time-frequency characteristics of PD signals. However, most of the existing technologies mainly focus on the filtering of power frequency noise and lack the ability to accurately analyze and compensate the multipath interference effect. Internationally, some studies have tried to introduce machine learning algorithms for PD signal classification, but the distinction between the direct wave and the reflected wave in a complex multipath environment and the signal compensation effect are still not ideal. In addition, the domestic and foreign technologies generally lack the comprehensive utilization of multi-channel data and fail to fully utilize the signal time difference information of sensors at different positions to improve the restoration accuracy of multipath signals.

[0004] The deficiencies of the existing technologies are mainly reflected in the following three aspects: First, the characteristics of the direct wave and the reflected wave are severely aliased, and traditional time-domain or frequency-domain processing methods are difficult to accurately separate the two; Second, in a complex multipath scenario, the superimposed interference of the reflected wave has a great impact on PD source localization and feature extraction, and the existing compensation technologies cannot meet the high-precision requirements; Third, it is difficult to dynamically adapt to the complex environments of different cable lengths and structures. Summary of the Invention

[0005] The purpose of the present invention is to provide a method, device and medium for identifying and compensating multipath interference of cable partial discharge signals, which are mainly applied to PD detection, defect localization and condition assessment in power cables, submarine cables and other long-distance transmission media, and are widely applicable to on-line monitoring, fault diagnosis and condition assessment of power equipment, especially for accurate detection and analysis of PD signals in a complex multipath interference environment.

[0006] The object of the present invention can be achieved by the following technical solutions:

[0007] A method for identifying and compensating multipath interference of cable partial discharge signals, comprising the following steps:

[0008] S1, using a plurality of high-frequency current sensors to synchronously collect multi-channel signals on different propagation paths at different sampling times;

[0009] S2, based on the collected signals, calculating the propagation path using the time difference and extracting the time-frequency characteristics of the signals;

[0010] S3, constructing input features of a deep learning model based on the collected signals, time difference, propagation path and time-frequency characteristics, and using the deep learning model to identify multipath interference and output the classification result of the signal;

[0011] S4, constructing a plurality of reflected wave signal models based on the classification result of the signal;

[0012] S5, based on the reflected wave signal model, removing the reflected wave signal from the direct wave signal to obtain a compensated signal.

[0013] In the step S1, high-frequency current sensors are respectively arranged at the head, middle and terminal of the cable, and signals are synchronously collected by the high-frequency current sensors at different sampling times, and the signal waveforms are stored.

[0014] The calculating the propagation path using the time difference based on the collected signals specifically includes the following steps:

[0015] Denote t the signals collected by the high-frequency current sensors at time as respectively, and calculate the arrival time

[0016] of the direct wave:

[0017] Based on the arrival time of the direct wave, calculate the time delay of the signal collected by each sensor relative to the direct wave:

[0018] where i =1, 2, 3,

[0019] wherein, represents the time difference of the signal collected by the i th sensor relative to the direct wave;

[0020] Calculate the propagation path based on the time difference:

[0021] where

[0022] wherein,v Indicates the propagation speed of the signal in the cable medium. d i Indicates the i propagation path of the signal collected by the

[0023] The extraction of the time-frequency features of the signal specifically includes the following steps:

[0024] Perform a short-time Fourier transform on the signal collected by the high-frequency current sensor to obtain its time-frequency distribution :

[0025] ,

[0026] where is the sliding window function, is the time variable, f is the frequency point, j is the imaginary unit;

[0027] Find the time points with a significant increase in intensity in the time-frequency distribution to obtain the signal arrival time :

[0028] ,

[0029] where the threshold is set according to the noise level;

[0030] Calculate the main frequency of the signal based on the time-frequency distribution :

[0031] ,

[0032] Calculate the global main frequency :

[0033] ,

[0034] Perform a wavelet transform on the signal to obtain the time-frequency distribution at different scales and times :

[0035] ,

[0036] where is the mother wavelet, used to generate wavelets at different scales and time positions, a is the scale factor, b is the translation factor;

[0037] For each scale factor a and translation factor b, calculate the energy density :

[0038] ,

[0039] Calculate the total energy :

[0040] .

[0041] In the step S3, the input feature X of the deep learning model is expressed as:

[0042]

[0043] where, represents the signal collected by the sensor, represents the time difference, represents the propagation path, represents the time-frequency distribution of the signal, represents the time of arrival of the signal, respectively represent the main frequency and the global main frequency of the signal, represents the total energy of the signal;

[0044] The deep learning model uses a convolutional neural network to extract the time-frequency map feature F: , and uses a long short-term memory network to process the time series characteristics: , obtain the classification feature H, and use the softmax function to output the classification result P: , where, CNN represents a convolutional neural network, LSTM represents a long short-term memory network, and P represents the probability that the signal belongs to the direct wave or the reflected wave.

[0045] The step S4 is specifically:

[0046] Construct the general form of the reflected wave signal model:

[0047] ,

[0048] where, represents the reflected wave signal, represents the direct wave signal, r is the attenuation factor of the reflected wave, is the time delay of the reflected wave, 、 are the parameters to be fitted;

[0049] Denote the number of high-frequency current sensors used to collect signals as N, the number of sampling times as M, and the high-frequency current sensor i at the sampling moment j The collected signal is , a total of N*M signals are collected, and the classification results of the N*M signals are output by using the deep learning model. Based on each direct wave signal and all the identified reflected wave signals, the parameters to be fitted are respectively fitted. 、 , and multiple reflected wave signal models corresponding to the direct wave are obtained. The number of the reflected wave signal models is the same as the number of the identified reflected wave signals.

[0050] The specific steps of step S5 are as follows:

[0051] For each direct wave signal identified by the deep learning model, the peak corresponding to the frequency component is detected through the spectrogram of the direct wave signal. Based on the frequency corresponding to the peak, the required reflected wave signal model is determined. Using the direct wave signal as the model input, the reflected wave signal is calculated based on the reflected wave signal model, and the calculated reflected wave signal is removed from the direct wave signal to obtain a compensation signal.

[0052] The method further includes:

[0053] S6, restoring the compensation signal, which specifically includes the following steps:

[0054] S61, normalizing the compensation signal to ensure the unity of the signal amplitude range:

[0055] ,

[0056] S62, extracting the frequency characteristics of the normalized signal by short-time Fourier transform:

[0057]

[0058] S63, smoothing the frequency characteristics to remove the noise frequency components:

[0059] ,

[0060] S64, performing an inverse transform on the smoothed frequency characteristics to reconstruct the time-domain signal:

[0061]

[0062] S65, performing weighted average fusion on the multi-channel reconstructed time-domain signals to obtain a restored signal :

[0063]

[0064] Among them, represents the channeli The reconstructed signal, indicating the channel weights, which are dynamically adjusted according to the signal quality, N is the number of sensors.

[0065] An electronic device includes a memory and a processor. A computer program is stored on the memory. When the processor executes the program, the method described above is implemented.

[0066] A computer-readable storage medium stores a computer program. When the program is executed by a processor, the method described above is implemented.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] 1. Combination of multi-channel signal acquisition and time difference calculation

[0069] In the present invention, multi-channel high-frequency current sensors (HFCTs) are arranged at the head, middle, and end of the cable. By synchronously acquiring partial discharge signals and extracting the time characteristics of the direct wave and the reflected wave, the time difference method (TDOA) is used to accurately calculate the signal propagation path difference. Compared with the existing single-point acquisition technology, multi-channel signal acquisition can comprehensively capture signal propagation information, providing a solid foundation for the identification and compensation of reflected waves.

[0070] 2. Multipath interference identification and compensation algorithm

[0071] Aiming at the multipath interference effect existing in the cable, the present invention introduces a reflected wave identification method based on time-frequency feature extraction. By using the short-time Fourier transform (STFT) and wavelet transform (WT) to extract features such as the main frequency, energy density, and arrival time of the signal, the direct wave and the reflected wave are accurately distinguished. In addition, combined with the dynamic compensation algorithm, the reflected wave is removed through the path difference and amplitude scaling model, and the original direct wave signal is restored, significantly improving the accuracy of partial discharge signal analysis.

[0072] 3. Adaptive reflected wave removal algorithm

[0073] The present invention can dynamically adjust the removal of the reflected wave according to the actual spectral characteristics of the signal, making the removal of the reflected wave more accurate and efficient, and adapting to different types of signal environments. Brief Description of the Drawings

[0074] Figure 1 is a flowchart of the method of the present invention. Detailed Embodiments

[0075] The present invention will be described in detail below with reference to the drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and the detailed implementation manners and specific operation processes are given, but the protection scope of the present invention is not limited to the following embodiments.

[0076] Example 1

[0077] This example provides a method for identifying and compensating for multipath interference in cable partial discharge signals, as Figure 1 shown, including the following steps:

[0078] S1. Use multiple high-frequency current sensors to synchronously collect multi-channel signals on different propagation paths at different sampling times.

[0079] In this example, high-frequency current sensors (HFCT) are arranged at the head, middle, and terminal of the cable respectively, and the signals are synchronously collected by the high-frequency current sensors at different sampling times, and the signal waveforms are stored. At each sampling time, the waveforms collected by the three sensors may have the following main differences due to their different positions and being affected by direct waves, multipath reflected waves, noise, etc.:

[0080] The head sensor mainly receives direct waves, and the waveform is relatively stable.

[0081] The middle sensor may receive some reflected waves, resulting in a slightly distorted waveform.

[0082] The terminal sensor may have obvious waveform distortion due to multipath interference, and there may be signal oscillation or periodic amplification / attenuation phenomena.

[0083] Denote t the signals collected by the high-frequency current sensors at time as respectively, and the sampling frequency is set to 500 MHz to ensure the complete capture of high-frequency partial discharge signals. In order to obtain multiple sets of test data, multiple samplings can be performed.

[0084] S2. Based on the collected signals, use the time difference to calculate the propagation path and extract the time-frequency characteristics of the signals.

[0085] S21. Time-domain feature extraction, specifically including the following steps:

[0086] S211. Calculate the arrival time of the direct wave:

[0087]

[0088] S212. Based on the arrival time of the direct wave, calculate the time delay of the signal collected by each sensor relative to the direct wave:

[0089] , i = 1, 2, 3,

[0090] where represents thei The time difference between the signals collected by a sensor relative to the direct wave. >0 may be a reflected wave, =0 may be a direct wave.

[0091] S213. Calculate the propagation path based on the time difference:

[0092] ,

[0093] where, v represents the propagation speed of the signal in the cable medium, usually taking v ≈2×10 8 m / s, d i represents the i th propagation path of the signal collected by the sensor.

[0094] Through step S21, the signal waveforms received by each sensor S ( t ), time differences and propagation paths d are obtained.

[0095] S22. Extract the time-frequency features of the signal.

[0096] The purpose of time-frequency feature extraction is to obtain time, frequency, and amplitude features from the partial discharge signal through short-time Fourier transform (STFT) or wavelet transform (WT) to distinguish between direct waves and reflected waves. The specific steps are as follows:

[0097] S221. Extract the time-frequency distribution.

[0098] Perform short-time Fourier transform on the signal collected by the high-frequency current sensor to obtain its time-frequency distribution :

[0099] ,

[0100] where, is the sliding window function, is the time variable, f is the frequency point, j is the imaginary unit.

[0101] S222. Calculate the signal arrival time.

[0102] The signal arrival time reflects the first occurrence time of the partial discharge signal at different sensors. The arrival time of the direct wave is earlier than that of the reflected wave.

[0103] In the time-frequency distribution Find the time point with a significant increase in intensity in the signal to obtain the signal arrival time :

[0104] ,

[0105] Among them, the threshold is set according to the noise level.

[0106] S223, calculate the main frequency.

[0107] The main frequency of the signal reflects the frequency band range where its main energy is located. The direct wave usually has a higher main frequency, while the main frequency of the reflected wave is lower or has a spectral drift.

[0108] Based on the time-frequency distribution Calculate the main frequency of the signal :

[0109]

[0110] Calculate the global main frequency :

[0111]

[0112] S224, calculate the energy density.

[0113] The signal energy density reflects the intensity distribution of the signal. The energy density of the direct wave is usually higher than that of the reflected wave. Through wavelet transform, the time-frequency energy of the signal can be accurately calculated.

[0114] For the signal Perform wavelet transform to obtain the time-frequency distribution at different scales and times :

[0115] ,

[0116] Among them, is the mother wavelet, which is used to generate wavelets at different scales and time positions, a is the scale factor, b is the translation factor.

[0117] For each scale factor a and translation factor b , calculate the energy density :

[0118]

[0119] Calculate the total energy :

[0120]

[0121] S3. Based on the collected signals, time differences, propagation paths, and time-frequency characteristics, construct the input features of the deep learning model, and use the deep learning model for multipath interference recognition to output the classification results of the signals.

[0122] Through the deep learning model, use time-frequency characteristics to distinguish direct waves and reflected waves. The amplitude of the direct wave is usually higher and the time delay is smaller, while the amplitude of the reflected wave is lower and there is a significant time delay.

[0123] 1) Input features

[0124] The input feature X of the deep learning model is expressed as:

[0125]

[0126] Among them, represents the signal collected by the sensor, represents the time difference, represents the propagation path, represents the time-frequency distribution of the signal, represents the signal arrival time, respectively represent the main frequency and the global main frequency of the signal, represents the total energy of the signal.

[0127] 2) Model structure

[0128] Use a convolutional neural network to extract the time-frequency map features F:

[0129] ,

[0130] Use a long short-term memory network to process the time series characteristics:

[0131] ,

[0132] Obtain the classification feature H, and use the softmax function to output the classification result P:

[0133] ,

[0134] Among them, CNN represents the convolutional neural network, LSTM represents the long short-term memory network, P represents the probability that the signal belongs to the direct wave or the reflected wave, represents the probability that the signal belongs to the direct wave, represents the probability that the signal belongs to the reflected wave.

[0135] 3) Model training

[0136] In this embodiment, the model is trained using simulation data and labeled real partial discharge signals to construct a dataset, and the loss function used in the training process is cross-entropy:

[0137]

[0138] Among them, i is the category, and y i is the true probability of category i, is the predicted probability of category i.

[0139] S4. Construct multiple reflected wave signal models based on the classification results of the signals.

[0140] In this embodiment, it is assumed that the reflected wave signal is a scaled and translated form of the direct wave signal to construct the general form of the reflected wave signal model:

[0141] ,

[0142] Among them, r is the attenuation factor of the reflected wave, is the time delay of the reflected wave, , are the parameters to be fitted.

[0143] Let the number of high-frequency current sensors for collecting signals be N, the number of sampling times be M, and the high-frequency current sensor i collects the signal at the sampling moment j is . Then a total of N*M signals are collected. Using the deep learning model to output the classification results of the N*M signals, based on each direct wave signal and all the identified reflected wave signals, the parameters to be fitted , are respectively fitted to obtain multiple reflected wave signal models corresponding to the direct wave. The number of the reflected wave signal models is the same as the number of the identified reflected wave signals.

[0144] In this embodiment, N = 3. Assuming M = 3, a total of nine signals are collected. Due to the sensor arrangement method in step S1, 3 direct wave signals and 6 reflected wave signals are identified among the 9 signals. For the direct wave signal and the reflected wave signals ~ , the corresponding six reflected wave signal models for are respectively fitted. For the remaining two direct wave signals, six reflected wave signal models are respectively constructed in the same way, which will not be elaborated in this embodiment.

[0145] S5. Based on the reflected wave signal model, remove the reflected wave signal from the direct wave signal to obtain the compensated signal.

[0146] The core objective of reflected signal compensation is to eliminate the interference of multipath reflected waves and restore the original direct wave signal. Through the Time Difference of Arrival (TDOA) method and the signal compensation model, the influence of reflected waves can be effectively identified and eliminated.

[0147] For each direct wave signal identified by the deep learning model, detect the peak corresponding to the frequency component through the spectrogram of the direct wave signal. Based on the frequency corresponding to the peak, determine the reflected wave signal model to be adopted. Use the direct wave signal as the model input, calculate the reflected wave signal based on the reflected wave signal model, and subtract the calculated reflected wave signal from the direct wave signal to obtain the compensated signal.

[0148] The formula for removing the reflected wave signal can be expressed as:

[0149]

[0150] where, is the direct wave signal containing reflected wave interference collected by the sensor, is with as the input, and the reflected wave signal calculated through the corresponding reflected wave signal model, is the compensated signal.

[0151] If there are multiple reflected waves in the signal, repeat the above removal process to gradually restore the original direct wave signal.

[0152] There are differences in the spectral characteristics between the direct wave and multiple reflected waves in a signal, and the number and characteristics of reflected waves can be judged through spectral analysis.

[0153] Through the spectrogram, detect the peak corresponding to the frequency component of the direct wave signal. Assume that the spectrum S ( f ) has multiple peaks , and these peaks can represent different reflected waves.

[0154]

[0155] where, is the spectrum of s(t), is the amplitude of the spectrum.

[0156] According to the peaks in the spectrogram, determine the frequencies of the reflected waves to be removed, calculate the peak frequencies of the reflected wave signal models, perform frequency matching based on the frequencies of the reflected waves to be removed and the peak frequencies of the multiple reflected wave signal models corresponding to the direct wave signal, and determine one or more reflected wave signal models to be adopted. According to the selected reflected wave signal models and the direct wave signal, calculate one or more reflected wave signals to be removed, and subtract them from the direct wave signal one by one.

[0157] S6. Restore the compensation signal.

[0158] After removing the reflected wave, the compensation signal may still have noise, amplitude distortion or other minor errors. Restore the compensation signal in the following way, and the compensated signal should conform to the PD characteristics for subsequent defect classification and location:

[0159] S61. Normalize the compensation signal to ensure a unified signal amplitude range:

[0160] ,

[0161] S62. Extract the frequency characteristics of the normalized signal by short-time Fourier transform :

[0162]

[0163] S63. Smooth the frequency characteristics to remove the noise frequency components:

[0164] ,

[0165] S64. Perform an inverse transform on the smoothed frequency characteristics to reconstruct the time-domain signal:

[0166]

[0167] S65. Perform weighted average fusion on the multi-channel reconstructed time-domain signals to obtain the restored signal :

[0168]

[0169] where represents the reconstructed signal of channel i , represents the channel weight, which is dynamically adjusted according to the signal quality, N and

[0170] In summary, the present invention proposes a method for identifying and compensating multipath interference of cable partial discharge signals, aiming to solve the interference problem of multipath interference effects on the accurate extraction of partial discharge signals in cables. By arranging high-frequency current sensors (HFCTs) at the head, middle, and end of the cable, combining time-frequency feature extraction, time difference of arrival (TDOA) method, multipath signal identification, and dynamic compensation algorithms, the influence of reflected waves is effectively eliminated, the direct wave signal is accurately restored, the accuracy of partial discharge source location and fault classification is significantly improved, and it has high robustness and wide applicability, making it suitable for partial discharge signal analysis in complex cable environments.

[0171] Embodiment 2

[0172] The electronic device of the present invention includes a central processing unit (CPU), which can execute various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or computer program instructions loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0173] Multiple components in the device are connected to the I / O interface, including: an input unit, such as a keyboard, mouse, etc.; an output unit, such as various types of displays, speakers, etc.; a storage unit, such as a disk, optical disc, etc.; and a communication unit, such as a network card, modem, wireless communication transceiver, etc. The communication unit allows the device to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0174] The processing unit executes the various methods and processes described above, such as methods S1 - S6. For example, in some embodiments, methods S1 - S6 can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device via the ROM and / or the communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 - S6 described above can be executed. Alternatively, in other embodiments, the CPU can be configured to execute methods S1 - S6 in any other suitable manner (e.g., by means of firmware).

[0175] The functions described above herein can be at least partially executed by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that can be used include: field programmable gate arrays (FPGA), application specific integrated circuits (ASIC), application specific standard products (ASSP), system on a chip (SOC), complex programmable logic devices (CPLD), and so on.

[0176] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing devices, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowchart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, executed partially on the machine and partially on a remote machine as an independent software package, or executed entirely on a remote machine or server.

[0177] In the context of the present invention, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0178] As described above, it is only the specific implementation manners of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for identifying and compensating multipath interference of cable partial discharge signals based on multi-channel time-frequency analysis, characterized in that: The following steps are involved: S1, using multiple high-frequency current sensors to synchronously collect multi-channel signals on different propagation paths at different sampling times; S2, based on the collected signal, calculate the propagation path using the time difference and extract the time-frequency characteristics of the signal; S3, constructing the input features of the deep learning model based on the collected signals, time differences, propagation paths and time-frequency features, using the deep learning model to identify multipath interference, and outputting the classification results of the signals; S4, constructing multiple reflection wave signal models based on the signal classification results; S5, based on the reflected wave signal model, removing the reflected wave signal from the direct wave signal to obtain a compensation signal; The step of extracting the time-frequency characteristics of the signal specifically comprises the following steps: The signal collected by the high-frequency current sensor Perform short-time Fourier transform to obtain its time-frequency distribution : ， in, is a sliding window function, is the time variable, f is the frequency point, j is an imaginary unit; In time-frequency distribution Find the time point when the intensity increases significantly and get the signal arrival time : ， Among them, the threshold is set according to the noise level; Based on time-frequency distribution Calculate the dominant frequency of a signal : ， Calculate the global dominant frequency : ， Signal Perform wavelet transform to obtain time-frequency distribution of different scales and times : ， in, is the mother wavelet, used to generate wavelets of different scales and time positions. a is the scale factor, b is the translation factor; For each scale factor a and translation factor b , calculate the energy density : ， Calculate total energy : ， In step S3, the input feature X of the deep learning model is expressed as: ， in, Represents the signal collected by the sensor, Indicates the time difference, represents the propagation path, represents the time-frequency distribution of the signal, represents the signal arrival time, Represent the main frequency of the signal and the global main frequency respectively, Represents the total energy of the signal; The deep learning model uses a convolutional neural network to extract the time-frequency graph features F: , and use long short-term memory networks to process time series characteristics: , get the classification feature H, and use the softmax function to output the classification result P: , where CNN represents convolutional neural network, LSTM represents long short-term memory network, and P represents the probability that the signal belongs to direct wave or reflected wave. The step S4 is specifically as follows: The general form of constructing the reflected wave signal model is: ， in, Represents the reflected wave signal, represents the direct wave signal, r is the attenuation factor of the reflected wave, is the time delay of the reflected wave, , is the parameter to be fitted; The number of high-frequency current sensors used to collect signals is N, the number of sampling times is M, and the high-frequency current sensor i At sampling time j The collected signal is , a total of N*M signals are collected, and the classification results of N*M signals are output using the deep learning model. Based on each direct wave signal and all the identified reflected wave signals, the parameters to be fitted are fitted respectively. , , and obtain multiple reflected wave signal models corresponding to the direct wave, where the number of the reflected wave signal models is the same as the number of identified reflected wave signals.

2. According to claim 1, a cable partial discharge signal multipath interference identification and compensation method based on multi-channel time-frequency analysis is characterized in that: In the step S1, high-frequency current sensors are arranged at the head end, the middle and the terminal of the cable respectively, and the high-frequency current sensors synchronously collect signals at different sampling times and store signal waveforms.

3. According to claim 2, a cable partial discharge signal multipath interference identification and compensation method based on multi-channel time-frequency analysis is characterized in that: The method of calculating the propagation path based on the collected signal using the time difference specifically includes the following steps: remember t The signals collected by the high-frequency current sensor at each moment are , calculate the arrival time of the direct wave : ， Based on the arrival time of the direct wave, calculate the time delay of the signal collected by each sensor relative to the direct wave: ， i =1,2,3， in, Indicates i The time difference of the signal collected by each sensor relative to the direct wave; Calculate the propagation path based on the time difference: ， in, v Indicates the propagation speed of the signal in the cable medium. d i Indicates i The propagation path of the signal collected by the sensor.

4. According to claim 1, a cable partial discharge signal multipath interference identification and compensation method based on multi-channel time-frequency analysis is characterized in that: The step S5 is specifically as follows: For each direct wave signal identified by the deep learning model, the peak value corresponding to the frequency component is detected through the spectrum diagram of the direct wave signal. Based on the frequency corresponding to the peak value, the required reflected wave signal model is determined. The direct wave signal is used as the model input, and the reflected wave signal is calculated based on the reflected wave signal model. The calculated reflected wave signal is removed from the direct wave signal to obtain a compensation signal.

5. According to the method of claim 1, the cable partial discharge signal multipath interference identification and compensation method based on multi-channel time-frequency analysis is characterized in that: The method further comprises: S6, restoring the compensation signal, specifically comprising the following steps: S61, compensation signal Normalize to ensure that the signal amplitude range is uniform: ， S62, extract the normalized signal by short-time Fourier transform Frequency characteristics : ， S63, frequency characteristics Perform smoothing to remove noise frequency components: ， S64, frequency characteristics after smoothing Perform inverse transform and reconstruct the time domain signal: ， S65, weighted average fusion is performed on the multi-channel reconstructed time domain signal to obtain the restored signal : ， in, Indicates channel i The reconstructed signal, Indicates the channel weight, which is dynamically adjusted according to the signal quality. N is the number of sensors.

6. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 5 is implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images