Cable fault accurate point-fixing method and system based on acoustic-magnetic synchronization and deep learning

Abstract

The invention discloses a cable fault accurate positioning method and system based on acoustic-magnetic synchronization and deep learning, and relates to the technical field of cable fault localization, and the method comprises the steps: synchronously collecting an acoustic signal and a magnetic signal generated in a cable fault region, and generating an original acoustic-magnetic synchronization signal pair; in combination with the current signal interference parameter, de-noising by using a deep convolutional neural network to obtain an enhanced acoustic-magnetic synchronization signal pair, calculating an acoustic-magnetic propagation speed difference according to the enhanced acoustic-magnetic synchronization signal pair, and determining first fault point positioning; inputting the regional cable distribution topology of the cable fault region, the current fault scene parameter and the enhanced acoustic-magnetic synchronization signal pair into a fault positioning identification plug-in, and outputting second fault point positioning; and determining a second positioning prediction confidence coefficient, carrying out positioning fitting on the first fault point positioning and the second fault point positioning, and outputting a cable fault fixed point result. According to the invention, the technical problem of insufficient cable fault fixed point accuracy in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cable fault location, in particular to a cable fault accurate pinpointing method and system based on acoustic-magnetic synchronization and deep learning. BACKGROUND

[0002] As a key carrier of power transmission, the operation reliability of the cable is directly related to the safety and stability of the power grid. With the acceleration of urbanization, the underground cable network is increasingly dense, and the frequency of cable faults also increases. Faults may cause large-scale power outages, resulting in significant economic losses and social impact. Therefore, quickly and accurately locating the fault point is the primary task of power system fault repair. In the prior art, the characteristics of acoustic and electromagnetic waves generated simultaneously when the fault point discharges can be used to locate the fault point by synchronously collecting the two signals and calculating the time difference of their arrival at the sensor. However, the on-site environment is complex, and the background noise interference is serious. The cable is usually laid underground, and there are various sources of noise around, such as traffic vibration, construction activities, and the operation of other electrical equipment. These noises can mask the weak fault sound signals, making signal detection difficult. Although the magnetic signal propagates stably, it is also susceptible to electromagnetic interference from adjacent cables or metal objects, causing signal distortion. Secondly, the acoustic signal attenuates, reflects, and refracts when propagating in the soil, and the propagation speed is affected by the medium properties. However, the magnetic signal propagation speed is fast but difficult to accurately measure the time difference, which introduces positioning errors. The traditional acoustic-magnetic synchronization method has limited noise reduction effect and lacks adaptive ability, and cannot adjust the processing parameters according to the specific environment. In areas with high noise levels, the traditional method may not be able to effectively extract the fault signal; in complex terrain, the inaccuracy of the acoustic wave propagation model can also cause pinpointing deviation. In addition, the existing method has poor generalization ability, and as the diversity of cable fault types, such as high-resistance faults and flashover faults, the traditional method is difficult to adapt to different fault characteristics, and has insufficient robustness, which is prone to misjudgment. SUMMARY

[0003] The present application provides a cable fault accurate pinpointing method and system based on acoustic-magnetic synchronization and deep learning, which solves the technical problem of insufficient accuracy of cable fault pinpointing in the prior art.

[0004] In view of the above problems, the present application provides a cable fault accurate pinpointing method and system based on acoustic-magnetic synchronization and deep learning.

[0005] In the first aspect, the present application provides a cable fault accurate pinpointing method based on acoustic-magnetic synchronization and deep learning, which comprises: synchronously collecting acoustic signals and magnetic signals generated in the cable fault area, and preprocessing to generate an original acoustic-magnetic synchronization signal pair; In combination with the current signal interference parameter, the original acoustic-magnetic synchronous signal pair is adaptively denoised by using a deep convolutional neural network to obtain an enhanced acoustic-magnetic synchronous signal pair, and the acoustic-magnetic propagation velocity difference is calculated according to the enhanced acoustic-magnetic synchronous signal pair to determine the first fault point positioning; The regional cable distribution topology of the cable fault area, the current fault scene parameter and the enhanced acoustic-magnetic synchronous signal pair are input into a fault positioning recognition plug-in constructed based on a long short-term memory network to output the second fault point positioning; The second positioning prediction confidence is determined in combination with the regional cable distribution topology and the current fault scene parameter, and the first fault point positioning and the second fault point positioning are fitted according to the adaptive fusion strategy formulated according to the second positioning prediction confidence to output the cable fault pinpointing result.

[0006] In a second aspect, the present application provides a cable fault pinpointing system based on acoustic-magnetic synchronization and deep learning, comprising: An acoustic-magnetic signal acquisition module is configured to synchronously acquire acoustic signals and magnetic signals generated in a cable fault area and to pre-process the acoustic signals and the magnetic signals to generate an original acoustic-magnetic synchronous signal pair; A first fault point acquisition module is configured to, in combination with a current signal interference parameter, adaptively denoise the original acoustic-magnetic synchronous signal pair by using a deep convolutional neural network to obtain an enhanced acoustic-magnetic synchronous signal pair, and to calculate an acoustic-magnetic propagation velocity difference according to the enhanced acoustic-magnetic synchronous signal pair to determine the first fault point positioning; A second fault point acquisition module is configured to input the regional cable distribution topology of the cable fault area, the current fault scene parameter and the enhanced acoustic-magnetic synchronous signal pair into a fault positioning recognition plug-in constructed based on a long short-term memory network to output the second fault point positioning; A cable fault pinpointing module is configured to determine a second positioning prediction confidence in combination with the regional cable distribution topology and the current fault scene parameter, and to fit the first fault point positioning and the second fault point positioning according to an adaptive fusion strategy formulated according to the second positioning prediction confidence to output the cable fault pinpointing result.

[0007] The one or more technical solutions provided in the present application have at least the following technical effects or advantages: The application provides a cable fault accurate pinpointing method and system based on acoustic-magnetic synchronization and deep learning, which significantly improves the accuracy, robustness and scene adaptability of fault pinpointing results by deeply integrating physical signal processing and data-driven modeling and introducing a dynamic self-adaptive fusion decision mechanism. Compared with the traditional method, the technical scheme provided by the application effectively overcomes the problems of inaccurate positioning caused by large signal noise interference, inaccurate propagation model and various fault types in a complex field environment, improves the anti-interference ability, self-adaptability and final result reliability of cable fault pinpointing detection, and provides effective technical support for realizing rapid and accurate repair of power cable faults. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0009] Figure 1 The flowchart of the cable fault accurate pinpointing method based on acoustic-magnetic synchronization and deep learning provided by the embodiments of the present application is shown.

[0010] Figure 2 The structure diagram of the cable fault accurate pinpointing system based on acoustic-magnetic synchronization and deep learning provided by the embodiments of the present application is shown.

[0011] In the drawings, the components represented by the numbers are described as follows: The acoustic-magnetic signal acquisition module 100, the first fault point acquisition module 200, the second fault point acquisition module 300, and the cable fault pinpointing module 400. DETAILED DESCRIPTION

[0012] The application provides a cable fault accurate pinpointing method and system based on acoustic-magnetic synchronization and deep learning, which is used to solve the technical problem of insufficient accuracy of cable fault pinpointing in the prior art.

[0013] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0014] It is to be understood that the terms "including", "comprising", "having" and "with" are meant to be interpreted open-ended, for example, a process, method, system, product or apparatus that includes a series of steps or units is not necessarily limited to those listed, and can include other steps or modules that are not expressly listed or inherent to such processes, methods, products or apparatus.

[0015] Embodiment one, as shown in the present application provides a cable fault accurate pinpointing method based on acoustic-magnetic synchronization and deep learning, wherein the method comprises: Figure 1 S10: synchronously collecting acoustic signals and magnetic signals generated in a cable fault area, and performing preprocessing to generate a pair of original acoustic-magnetic synchronization signals.

[0016] The step S10 in the method provided in the present application comprises: performing continuous wavelet transform on the acoustic signals and the magnetic signals, extracting joint features of the signals in the time and frequency dimensions, and generating acoustic signal time-frequency diagrams and magnetic signal time-frequency diagrams as the pair of original acoustic-magnetic synchronization signals.

[0017] In the present application, an acoustic sensor and a magnetic induction intensity sensor are used to synchronously collect acoustic signals and magnetic signals generated in a cable fault area. And preprocessing is performed. Specifically, the acoustic signals and the magnetic signals are respectively processed by continuous wavelet transform to convert one-dimensional waveform signals into two-dimensional time-frequency diagrams. The time-frequency diagrams can represent the distribution of signal energy at different time and frequency points. The obtained acoustic signal time-frequency diagrams and magnetic signal time-frequency diagrams are used as the pair of original acoustic-magnetic synchronization signals.

[0018] Through high-precision synchronous acquisition technology and targeted preprocessing, a reliable data basis is constructed for the entire fault pinpointing process. The synchronous acquisition mechanism ensures the accurate alignment of acoustic signals and magnetic signals on the time axis, providing time-accurate signal sources for subsequent physical positioning methods based on time difference calculation. At the same time, the preprocessing operation can improve the signal quality at the preliminary level and provide an accurate basis for subsequent deep processing steps.

[0019] S20: combining current signal interference parameters, using a deep convolutional neural network to adaptively denoise the pair of original acoustic-magnetic synchronization signals to obtain an enhanced acoustic-magnetic synchronization signal pair, and calculating the acoustic-magnetic propagation velocity difference according to the enhanced acoustic-magnetic synchronization signal pair to determine the first fault point positioning.

[0020] ​The cable fault site environment is changing, the noise interference has non-stationary characteristics, and its strength and characteristics vary with specific location and time. Traditional denoising methods such as fixed threshold filtering algorithm lack adaptability, and in strong noise environment, the signal may be excessively smoothed, resulting in loss of fault characteristics, while in low noise environment, the denoising may be insufficient, and the optimal signal-to-noise ratio cannot be improved, which is difficult to cope with complex and variable actual working conditions, resulting in unstable quality of extracted acoustic and magnetic signals, and further causing large error of positioning results based on fixed acoustic and magnetic propagation speed difference calculation.

[0021] The step S20 in the method provided by the embodiments of the present application comprises: monitoring and acquiring a current signal interference parameter sequence in the cable fault area, wherein the signal interference parameters at least include full-band signal-to-noise ratio, key frequency band noise power spectral density, pulse interference density, pulse interference amplitude and peak signal-to-noise ratio of acoustic-magnetic cross-correlation sequence; performing signal interference strength evaluation and signal interference volatility analysis according to the current signal interference parameter sequence respectively, and outputting current signal interference strength and current signal interference volatility; performing denoising complexity evaluation based on the current signal interference strength and current signal interference volatility, and outputting current denoising complexity; wherein the denoising complexity evaluation based on the current signal interference strength and current signal interference volatility and outputting the current denoising complexity comprises: setting the ratio of the current signal interference strength to a preset standard signal interference strength as a first denoising complexity coefficient; setting the ratio of the current signal interference volatility to a preset standard signal interference volatility as a second denoising complexity coefficient; after non-dimensional processing of the first denoising complexity coefficient and the second denoising complexity coefficient, weighting and summing according to a preset weight proportion to obtain the current denoising complexity; constructing an acoustic signal denoising channel and a magnetic signal denoising channel based on a convolutional neural network, and calling the acoustic signal denoising channel and the magnetic signal denoising channel according to the current denoising complexity, to perform signal denoising on the acoustic signal time-frequency graph and the magnetic signal time-frequency graph respectively, and output an enhanced acoustic-magnetic synchronous signal pair; wherein constructing an acoustic signal denoising channel and a magnetic signal denoising channel based on a convolutional neural network, and calling the acoustic signal denoising channel and the magnetic signal denoising channel according to the current denoising complexity, to perform signal denoising on the acoustic signal time-frequency graph and the magnetic signal time-frequency graph respectively, comprises: The sample sound signal time-frequency graph set and the sample enhanced sound signal time-frequency graph set are collected as training data, and are equally divided into K parts, K times are selected from the K data sets with replacement to obtain a first training set, and K times are iteratively selected to obtain K training sets, wherein the sample enhanced sound signal time-frequency graph is obtained by denoising the sample sound signal time-frequency graph; The K training sets are used to train the convolutional neural network to convergence respectively, and K sound signal denoising units are obtained, and the sound signal denoising channel is integrated and constructed according to the mean fusion strategy, wherein K is an integer greater than 10; The ratio of the current denoising complexity to the preset maximum denoising complexity is multiplied by K to obtain the number P of selected adaptive units, wherein the preset maximum denoising complexity is the historical maximum denoising complexity in the historical time range, and P is greater than or equal to 3; P denoising units are randomly selected from the K sound signal denoising units of the sound signal denoising channel, the sound signal time-frequency graph is denoised, and the enhanced sound signal time-frequency graph is obtained after mean fitting of the P denoising results; The sample magnetic signal time-frequency graph set and the sample enhanced magnetic signal time-frequency graph set are collected as training data, and the convolutional neural network is trained to convergence to obtain K magnetic signal denoising units; P denoising units are randomly selected from the K magnetic signal denoising units, the magnetic signal time-frequency graph is denoised, and the enhanced magnetic signal time-frequency graph is obtained after mean fitting of the P denoising results; The enhanced sound signal time-frequency graph and the enhanced magnetic signal time-frequency graph are used as an enhanced acoustic-magnetic synchronous signal pair.

[0022] In the embodiment of the application, the current signal interference parameter sequence in the cable fault area is monitored and acquired, wherein the signal interference parameters at least include the full-band signal-to-noise ratio, the key frequency band noise power spectral density, the pulse interference density, the pulse interference amplitude and the peak signal-to-noise ratio of the acoustic-magnetic cross-correlation sequence. Specifically, the full-band signal-to-noise ratio represents the relative size of the useful signal strength and the total background noise intensity in the entire frequency range, with the unit of dB, and the lower the value, the higher the overall noise level. The key frequency band noise power spectral density represents the power distribution intensity of the noise in the specific frequency interval where the fault signal characteristics are most concentrated, with the unit of dB / Hz, and the higher the value, the stronger the interference on the key frequency. The pulse interference density represents the number of sudden, high-amplitude pulse interferences appearing per unit time, with the unit of pieces / minute, and the higher the value, the more serious the signal pollution by random pulses. The pulse interference amplitude represents the average intensity of the appearing pulse interferences, with the unit of dB, and the higher the value, the greater the instantaneous distortion of the signal caused by the pulse interference. The peak signal-to-noise ratio of the acoustic-magnetic cross-correlation sequence represents the reliability of the synchronization between the acoustic signal and the magnetic signal, with the unit of dB, and the higher the value, the better the consistency of the acoustic-magnetic signal, and the smaller the time alignment error caused by noise.

[0023] According to the current signal interference parameter sequence, signal interference strength evaluation and signal interference fluctuation analysis are respectively performed, and the current signal interference strength and the current signal interference fluctuation degree are output. Specifically, each parameter in the current signal interference parameter sequence is divided by a preset standard value to obtain a parameter interference strength evaluation value, and the mean value of the parameter interference strength evaluation value is taken as the signal interference strength. The standard deviation of each parameter in the current signal interference parameter sequence is divided by the mean value to obtain a parameter interference fluctuation degree, and the mean value of the parameter interference fluctuation degree is taken as the signal interference fluctuation degree.

[0024] Based on the current signal interference strength and the current signal interference fluctuation degree, the denoising complexity evaluation is performed, and the current denoising complexity is output. Specifically, the ratio of the current signal interference strength to the preset standard signal interference strength is set as the first denoising complexity coefficient. The preset standard signal interference strength refers to the typical interference strength for denoising under general working conditions, which is exemplarily obtained by taking the median of the signal interference strength in the historical denoising log.

[0025] The ratio of the current signal interference fluctuation degree to the preset standard signal interference fluctuation degree is set as the second denoising complexity coefficient. The preset standard signal interference fluctuation degree refers to the typical interference fluctuation degree for denoising under general working conditions, which is exemplarily obtained by taking the median of the signal interference fluctuation degree in the historical denoising log.

[0026] After the first denoising complexity coefficient and the second denoising complexity coefficient are dimensionless processed, they are weighted and summed according to the preset weight proportion to obtain the current denoising complexity. Current denoising complexity = first weight x first denoising complexity coefficient + second weight x second denoising complexity coefficient, first weight + second weight = 1. The first weight and the second weight are obtained by expert evaluation of the influence of signal interference strength and signal interference fluctuation degree on denoising complexity. For example, when the signal interference fluctuation degree is large, denoising is more difficult, and a larger second weight is given, for example, 0.6, and the first weight is 0.4. The greater the obtained denoising complexity, the more complex the current noise mode, and the more computing resources are called for denoising.

[0027] Based on the convolutional neural network, the acoustic signal denoising channel and the magnetic signal denoising channel are constructed, and according to the current denoising complexity, the acoustic signal denoising channel and the magnetic signal denoising channel are called to perform signal denoising on the acoustic signal time-frequency graph and the magnetic signal time-frequency graph respectively, and the enhanced acoustic-magnetic synchronous signal pair is output.

[0028] Specifically, a sample sound signal time-frequency graph set and a sample enhanced sound signal time-frequency graph set are collected as training data, and the training data is equally divided into K parts, K times are selected from the K data sets with replacement to obtain a first training set, and K training sets are obtained by iterative selection of K times. Wherein, the sample enhanced sound signal time-frequency graph is obtained by denoising the sample sound signal time-frequency graph.

[0029] A sound signal denoising unit is constructed. Specifically, it is constructed based on a 4-layer convolutional neural network, wherein the input layer is used to receive the sound signal time-frequency graph, the first convolutional layer uses 16 3x3 convolutional kernels, the second convolutional layer uses 8 3x3 convolutional layers, and the output layer is used to output the denoised enhanced sound signal time-frequency graph.

[0030] The K training sets are used to train the convolutional neural network to convergence, respectively, to obtain K sound signal denoising units. According to the mean fusion strategy, a sound signal denoising channel is integrated and constructed, wherein K is an integer greater than 10, and the output result of the sound signal denoising channel is the mean fusion of the sound signal time-frequency graphs output by the multiple sound signal denoising units. For example, the sound signal time-frequency graphs output by the multiple sound signal denoising units are averaged at the pixel level, that is, the value of each pixel point is the arithmetic mean of the corresponding position values of the sound signal time-frequency graphs output by the multiple sound signal denoising units, and finally an enhanced sound signal time-frequency graph is synthesized.

[0031] The ratio of the current denoising complexity to the preset maximum denoising complexity is multiplied by K to obtain the number P of selected units, wherein the preset maximum denoising complexity is the historical maximum denoising complexity within the historical time range, and P is greater than or equal to 3. When the calculated P is less than or equal to 3, P is 3, and when the calculated P is greater than or equal to K, P = K.

[0032] Randomly select P denoising units from the K sound signal denoising units in the sound signal denoising channel, perform signal denoising on the sound signal time-frequency graph, and perform mean fitting on the P denoising results to obtain an enhanced sound signal time-frequency graph.

[0033] A sample magnetic signal time-frequency graph set and a sample enhanced magnetic signal time-frequency graph set are collected as training data, and the convolutional neural network is trained to convergence to obtain K magnetic signal denoising units. The same method as the sound signal denoising channel is used for construction and training, for example, a 4-layer convolutional neural network is used to construct the magnetic signal denoising unit, wherein the input layer is used to receive the magnetic signal time-frequency graph, the first convolutional layer uses 16 3x3 convolutional kernels, the second convolutional layer uses 8 3x3 convolutional layers, and the output layer is used to output the denoised enhanced magnetic signal time-frequency graph. The K magnetic signal denoising units are trained to convergence using the sample magnetic signal time-frequency graph set and the sample enhanced magnetic signal time-frequency graph set.

[0034] In the K magnetic signal denoising units, P denoising units are randomly selected, signal denoising is performed on the magnetic signal time-frequency diagram, and after mean fitting is performed on the P denoising results, an enhanced magnetic signal time-frequency diagram is obtained. For example, pixel-level average value calculation is performed on the magnetic signal time-frequency diagrams output by the plurality of magnetic signal denoising units, that is, the value of each pixel point is the arithmetic mean of the values of the corresponding positions on the magnetic signal time-frequency diagrams output by the plurality of magnetic signal denoising units, and finally, an enhanced acoustic signal time-frequency diagram is synthesized.

[0035] The enhanced acoustic signal time-frequency diagram and the enhanced magnetic signal time-frequency diagram are taken as an enhanced acoustic-magnetic synchronous signal pair.

[0036] According to the enhanced acoustic-magnetic synchronous signal pair, an acoustic-magnetic propagation speed difference is calculated, and a first fault point positioning is determined. The acoustic-magnetic synchronous method is a method for quantifying the distance of a fault point by using the difference between acoustic and magnetic signal propagation speeds. When the detection device is exactly above the fault point, the acoustic-magnetic signal time difference is minimum and close to zero, thereby accurately positioning the fault. For example, when the fault grounding resistance is small and the breakdown discharge sound is weak, the audio induction method can be used to determine the first fault point positioning. A 1 kHz audio current signal is injected into the fault cable to induce an audio magnetic field. After the magnetic field signal is collected by a coil, it is amplified and converted into sound. The position with the maximum audio sound is determined as the first fault point positioning.

[0037] By introducing a deep convolutional neural network and combining real-time signal interference parameters for adaptive denoising, the adaptive intelligence of the signal enhancement link is realized, and the quality of the fault feature signal and the reliability of the subsequent physical positioning calculation are significantly improved. The deep convolutional neural network can learn the difference between noise and effective signal in features, has strong feature extraction and nonlinear fitting capability, can more accurately separate noise and retain weak fault features, can dynamically adjust the denoising behavior according to the actual noise condition, and can ensure that a relatively optimal denoising effect is achieved in different interference environments, thereby further obtaining a relatively accurate fault point positioning.

[0038] S30: inputting the regional cable distribution topology of the cable fault area, the current fault scene parameter, and the enhanced acoustic-magnetic synchronous signal pair into a fault positioning recognition plug-in constructed based on a long short-term memory network, and outputting a second fault point positioning.

[0039] The physical positioning method based on acoustic-magnetic time difference is intuitive and reliable, but its accuracy depends to a great extent on the accurate estimation of the propagation speed of acoustic waves in the complex medium underground. The non-uniformity of the actual soil structure, the influence of the cable topology structure on the signal propagation path, and the difference in signal characteristics exhibited by different fault types all introduce errors. The traditional method often cannot effectively integrate these complex scene information, resulting in insufficient generalization ability of the positioning model when facing non-ideal working conditions or special fault types, and causing positioning deviation.

[0040] The step S30 in the method provided in the embodiments of the present application comprises: Based on historical cable fault detection records, a sample area cable distribution topology set, a sample fault scene parameter set and a sample enhanced acoustic-magnetic synchronous signal pair set are collected, and historical fault point positioning under different sample area cable distribution topologies, sample fault scene parameters and sample enhanced acoustic-magnetic synchronous signal pairs is collected as sample fault point positioning, to obtain a sample fault point positioning set, wherein the fault scene parameters include fault type, fault impedance and fault starting angle; The sample area cable distribution topology set, the sample fault scene parameter set and the sample enhanced acoustic-magnetic synchronous signal pair set are used as input, and the sample fault point positioning set is used as supervision to train the long short-term memory network to convergence, to generate a fault positioning recognition plug-in.

[0041] In the embodiments of the application, historical cable fault detection records are obtained, and based on the historical cable fault detection records, a sample area cable distribution topology set, a sample fault scene parameter set and a sample enhanced acoustic-magnetic synchronous signal pair set are collected. The cable distribution topology refers to the distribution topology information of the cable, including, for example, cable length, branch point, joint position, burial depth and the like. The fault scene parameters include fault type, fault impedance and fault starting angle. The fault type includes single-phase grounding, phase-to-phase short circuit and the like, the fault impedance includes high-impedance fault, low-impedance fault and the like, and the fault starting angle includes the phase angle of the power frequency alternating voltage waveform. Historical fault point positioning under different sample area cable distribution topologies, sample fault scene parameters and sample enhanced acoustic-magnetic synchronous signal pairs is collected as sample fault point positioning, to obtain a sample fault point positioning set.

[0042] Based on the long short-term memory network, a fault positioning recognition plug-in is constructed. The input layer is used to receive the area cable distribution topology, the fault scene parameter and the enhanced acoustic-magnetic synchronous signal, the LSTM layer contains 128 nodes to learn the dynamic characteristics of the input parameters, and the output layer is used to output the recognized fault positioning point. The sample area cable distribution topology set, the sample fault scene parameter set and the sample enhanced acoustic-magnetic synchronous signal pair set are used as input, and the sample fault point positioning set is used as supervision to train the long short-term memory network to convergence, for example, the accuracy rate of the output fault positioning point is more than 90%, that is, the fault positioning recognition plug-in training is completed.

[0043] By constructing a fault location recognition plug-in based on a long short-term memory network, an intelligent positioning approach capable of learning and utilizing historical experience and the correlation of complex scenarios is provided. The long short-term memory network is good at processing sequence data and capturing long-term dependencies, and can effectively model the deep nonlinear mapping between the dynamic characteristics of acoustic-magnetic signals and the cable topology and fault scene parameters. By inputting the regional cable distribution topology and the current fault scene parameters, specific environmental information can be integrated into the positioning judgment, such as considering the influence of cable bends and branches on signal propagation, or distinguishing signal patterns of different fault types. The output of the second fault point positioning result is an intelligent inference based on a large amount of historical data, which has stronger inclusiveness and understanding of complex factors, and this positioning result can effectively complement the physical positioning result of the first step.

[0044] S40: Analyze and determine a second positioning prediction confidence in combination with the regional cable distribution topology and the current fault scene parameters, and formulate an adaptive fusion strategy according to the second positioning prediction confidence to perform positioning fitting on the first fault point positioning and the second fault point positioning, and output a cable fault positioning result.

[0045] After obtaining the first fault point positioning based on physical principles and the second fault point positioning based on the deep learning model, effectively integrating these two results that may differ is crucial. Simply taking the average or choosing one cannot adapt to the dynamic changes of the field conditions, and cannot obtain the optimal and most reliable final positioning result.

[0046] The step S40 in the method provided by the embodiments of the present application includes: high-frequency feature extraction is performed on the sample regional cable distribution topology set and the sample fault scene parameter set to obtain a high-frequency regional cable distribution topology set and a high-frequency fault scene parameter set; feature extraction is respectively performed on the high-frequency regional cable distribution topology set and the high-frequency fault scene parameter set to construct a high-frequency distribution topology feature vector library and a high-frequency scene parameter feature vector library; feature extraction is performed on the regional cable distribution topology and the current fault scene parameters to construct a current distribution topology feature vector and a current scene parameter feature vector; the current distribution topology feature vector is compared with the high-frequency distribution topology feature vector library through similar traversal to determine a distribution topology similarity; the current scene parameter feature vector is compared with the high-frequency scene parameter feature vector library through similar traversal to determine a scene parameter similarity; a comprehensive data similarity is calculated based on the distribution topology similarity and the scene parameter similarity, and a second positioning prediction confidence is determined according to the comprehensive data similarity matching, wherein the second positioning prediction confidence and the comprehensive data similarity are positively correlated. a ratio of the second positioning prediction confidence and a preset standard prediction confidence is set as a second weight compensation coefficient; a product of the second weight compensation coefficient and a second initial weight is set as a second adaptive weight, and a first adaptive weight is obtained by subtracting the second adaptive weight from 1; wherein the second initial weight is 0.4, and the second adaptive weight is not less than 0.3 and not more than 0.6; based on the second adaptive weight and the first adaptive weight, the first fault point positioning and the second fault point positioning are positioned and fitted, and a weighted center point of the two is selected as a cable fault positioning result.

[0047] In the embodiments of the application, high-frequency feature extraction is performed on the sample regional cable distribution topology set and the sample fault scene parameter set to obtain a high-frequency regional cable distribution topology set and a high-frequency fault scene parameter set. Specifically, high-frequency feature extraction is performed on the sample regional cable distribution topology set, for example, the features are extracted as a length of 1500 meters, a branch point number of 2, and an average burial depth of 1.2 meters, and high-frequency fault scene parameters are extracted, for example, the fault type is single-phase grounding, the fault impedance is high-impedance fault, the fault impedance is greater than or equal to 10 kΩ and less than or equal to 50 kΩ, and the fault starting angle is greater than or equal to 60 degrees and less than or equal to 80 degrees, to obtain the high-frequency regional cable distribution topology set and the high-frequency fault scene parameter set.

[0048] Feature extraction is performed on the regional cable distribution topology and the current fault scene parameter to construct a current distribution topology feature vector and a current scene parameter feature vector.

[0049] The current distribution topology feature vector is compared with the high-frequency distribution topology feature vector library through similar traversal to determine a distribution topology similarity. For example, multiple cosine similarities of the current distribution topology feature vector and the high-frequency distribution topology feature vector library are calculated through traversal comparison, and the maximum value is selected as the distribution topology similarity.

[0050] The current scene parameter feature vector is compared with the high-frequency scene parameter feature vector library through similar traversal in the same way as the distribution topology similarity is obtained to determine a scene parameter similarity.

[0051] The comprehensive data similarity is weightedly calculated based on the distribution topology similarity and the scene parameter similarity. The comprehensive data similarity = (topology weight x distribution topology similarity) + (scene weight x scene parameter similarity), wherein the topology weight + the scene weight = 1. The topology weight and the scene weight are obtained based on expert evaluation. For example, the distribution topology has a greater impact on fault positioning, and a greater topology weight is given. For example, the topology weight is 0.6, and the scene weight is 0.4. The second positioning prediction confidence is determined according to the comprehensive data similarity matching, wherein the second positioning prediction confidence is positively correlated with the comprehensive data similarity. Exemplarily, the second positioning prediction confidence = comprehensive data similarity = comprehensive data similarity x 1.05, so as to improve the influence of the prediction confidence in the case of high similarity.

[0052] The ratio of the second positioning prediction confidence to the preset standard prediction confidence is set as the second weight compensation coefficient. The preset standard prediction confidence is a standard prediction confidence set in advance, which can be set to 0.85 as an example to serve as a basis for evaluating the size of the prediction confidence. The second weight compensation coefficient = the second positioning prediction confidence ÷ the preset standard prediction confidence.

[0053] The product of the second weight compensation coefficient and the second initial weight is set as the second adaptive weight, and the first adaptive weight is obtained by subtracting 1 from the second adaptive weight, wherein the second initial weight is 0.4, and the second adaptive weight is not less than 0.3 and not greater than 0.6. Specifically, the second adaptive weight = the second initial weight x the second weight compensation coefficient. When the calculated second adaptive weight is less than or equal to 0.3, 0.3 is taken, and when the second adaptive weight is greater than or equal to 0.6, 0.6 is taken, so as to prevent excessive weight from distorting the fitting result. The first adaptive weight = 1 - the second adaptive weight.

[0054] Based on the second adaptive weight and the first adaptive weight, the first fault point positioning and the second fault point positioning are positioned and fitted, and the weighted center point of the two is selected as the cable fault positioning result. For example, the first fault positioning point and the second fault positioning point are connected as a straight line, for example, the straight line length is 1 meter. When the first adaptive weight is 0.6 and the second adaptive weight is 0.4, the point closer to the fault positioning point with greater weight is selected, for example, the point with a distance from the first fault positioning point = 1 - (1 x 0.6) is taken as the cable fault positioning result.

[0055] By introducing the prediction confidence evaluation and adaptive fusion strategy, the decision-making link is refined and intelligentized, ensuring the optimality and reliability of the final result. The second positioning prediction confidence is determined by combining the regional cable distribution topology and the current fault scene parameter analysis to evaluate the matching degree of the current working condition and the deep learning model training experience library. Based on this confidence, the fusion strategy can dynamically adjust the contribution weight of the first positioning and the second positioning in the final result. When the second positioning confidence is high, the fusion strategy will give it more weight, making full use of the advantages of data model in complex pattern recognition; when the confidence is low, it will rely more on the physical positioning result with stronger explainability. This dynamic weighted fusion method effectively avoids the limitations of a single method, realizing the complementary advantages between the stability of the physical model and the intelligence of the data model. The final output of the cable fault positioning result is the optimized solution obtained after accurately evaluating the reliability of each method under the current scene, thereby significantly improving the adaptability, robustness and final positioning accuracy of the cable fault positioning method in various complex and unknown environments.

[0056] In one embodiment, as shown in FIG. 2, the cable fault precise positioning system based on acoustic-magnetic synchronization and deep learning provided by the present application comprises: Figure 2 An acoustic-magnetic signal acquisition module 100 is configured to synchronously acquire acoustic signals and magnetic signals generated in a cable fault area and perform preprocessing to generate a pair of original acoustic-magnetic synchronous signals. A first fault point acquisition module 200 is configured to combine current signal interference parameters, use a deep convolutional neural network to adaptively denoise the pair of original acoustic-magnetic synchronous signals to obtain a pair of enhanced acoustic-magnetic synchronous signals, and calculate an acoustic-magnetic propagation velocity difference based on the pair of enhanced acoustic-magnetic synchronous signals to determine a first fault point positioning. A second fault point acquisition module 300 is configured to input a regional cable distribution topology of the cable fault area, current fault scene parameters and the pair of enhanced acoustic-magnetic synchronous signals into a fault positioning recognition plug-in constructed based on a long short-term memory network to output a second fault point positioning. A cable fault positioning module 400 is configured to combine the regional cable distribution topology and the current fault scene parameter analysis to determine a second positioning prediction confidence, and formulate an adaptive fusion strategy based on the second positioning prediction confidence to perform positioning fitting on the first fault point positioning and the second fault point positioning to output a cable fault positioning result.

[0057] In one embodiment, the acoustic-magnetic signal acquisition module 100 is further configured to: ​The acoustic signal and the magnetic signal are continuously wavelet transformed to extract joint features of the signals in time and frequency dimensions, and generate an acoustic signal time-frequency graph and a magnetic signal time-frequency graph as a pair of original acoustic-magnetic synchronous signals.

[0058] In one embodiment, the first fault point acquisition module 200 is further configured to: monitor and acquire a current signal interference parameter sequence in the cable fault area, wherein the signal interference parameters at least include a full-band signal-to-noise ratio, a key frequency band noise power spectral density, a pulse interference density, a pulse interference amplitude, and a peak signal-to-noise ratio of an acoustic-magnetic cross-correlation sequence; perform signal interference intensity evaluation and signal interference fluctuation analysis according to the current signal interference parameter sequence respectively, and output a current signal interference intensity and a current signal interference fluctuation degree; perform denoising complexity evaluation based on the current signal interference intensity and the current signal interference fluctuation degree, and output a current denoising complexity; wherein the denoising complexity evaluation based on the current signal interference intensity and the current signal interference fluctuation degree, and outputting the current denoising complexity, comprises: setting a ratio of the current signal interference intensity to a preset standard signal interference intensity as a first denoising complexity coefficient; setting a ratio of the current signal interference fluctuation degree to a preset standard signal interference fluctuation degree as a second denoising complexity coefficient; after the first denoising complexity coefficient and the second denoising complexity coefficient are dimensionless processed, performing weighted summation according to a preset weight proportion to obtain the current denoising complexity; constructing an acoustic signal denoising channel and a magnetic signal denoising channel based on a convolutional neural network, and calling the acoustic signal denoising channel and the magnetic signal denoising channel according to the current denoising complexity to perform signal denoising on the acoustic signal time-frequency graph and the magnetic signal time-frequency graph respectively, and output an enhanced acoustic-magnetic synchronous signal pair; wherein the acoustic signal denoising channel and the magnetic signal denoising channel are constructed based on a convolutional neural network, and the acoustic signal denoising channel and the magnetic signal denoising channel are called according to the current denoising complexity to perform signal denoising on the acoustic signal time-frequency graph and the magnetic signal time-frequency graph respectively, comprising: collecting a sample acoustic signal time-frequency graph set and a sample enhanced acoustic signal time-frequency graph set as training data, and equally dividing them into K parts, selecting K times from the K parts of data with replacement to obtain a first training set, and iteratively selecting K times to obtain K training sets, wherein the sample enhanced acoustic signal time-frequency graph is obtained by denoising processing of the sample acoustic signal time-frequency graph; training the convolutional neural network using the K training sets respectively until convergence to obtain K acoustic signal denoising units, and integrating and constructing an acoustic signal denoising channel according to a mean fusion strategy, wherein K is an integer greater than 10; Adapt the current denoising complexity to the preset maximum denoising complexity ratio, and take the integer part of K to obtain the adaptive unit selection number P, wherein the preset maximum denoising complexity is the historical maximum denoising complexity in the historical time range, and P is greater than or equal to 3; Randomly select P denoising units in the K acoustic signal denoising units of the acoustic signal denoising channel, perform signal denoising on the acoustic signal time-frequency graph, and perform mean fitting on the P denoising results to obtain an enhanced acoustic signal time-frequency graph; Collect sample magnetic signal time-frequency graph sets and sample enhanced magnetic signal time-frequency graph sets as training data, train the convolutional neural network to convergence, and obtain K magnetic signal denoising units; Randomly select P denoising units in the K magnetic signal denoising units, perform signal denoising on the magnetic signal time-frequency graph, and perform mean fitting on the P denoising results to obtain an enhanced magnetic signal time-frequency graph; Take the enhanced acoustic signal time-frequency graph and the enhanced magnetic signal time-frequency graph as an enhanced acoustic-magnetic synchronous signal pair.

[0059] In one embodiment, the second fault point acquisition module 300 is also used for: Based on the historical cable fault detection record, collect sample regional cable distribution topology sets, sample fault scene parameter sets, and sample enhanced acoustic-magnetic synchronous signal pair sets, and collect historical fault point positioning under different sample regional cable distribution topologies, sample fault scene parameters, and sample enhanced acoustic-magnetic synchronous signal pairs as sample fault point positioning, and acquire a sample fault point positioning set, wherein the fault scene parameters include fault type, fault impedance, and fault start angle; Take the sample regional cable distribution topology set, the sample fault scene parameter set, and the sample enhanced acoustic-magnetic synchronous signal pair set as input, take the sample fault point positioning set as supervision, train the long short-term memory network to convergence, and generate a fault positioning recognition plug-in.

[0060] In one embodiment, the cable fault positioning module 400 is also used for: Perform high-frequency feature extraction on the sample regional cable distribution topology set and the sample fault scene parameter set to obtain a high-frequency regional cable distribution topology set and a high-frequency fault scene parameter set; Perform feature extraction on the high-frequency regional cable distribution topology set and the high-frequency fault scene parameter set, respectively, to construct a high-frequency distribution topology feature vector library and a high-frequency scene parameter feature vector library; Perform feature extraction on the regional cable distribution topology and the current fault scene parameter to construct a current distribution topology feature vector and a current scene parameter feature vector; Compare the current distribution topology feature vector with the high-frequency distribution topology feature vector library for similarity traversal to determine a distribution topology similarity; The current scene parameter feature vector is compared with the high-frequency scene parameter feature vector library for similarity traversal comparison to determine a scene parameter similarity; A comprehensive data similarity is calculated based on the distribution topology similarity and the scene parameter similarity, and a second positioning prediction confidence is determined according to the comprehensive data similarity, wherein the second positioning prediction confidence is positively correlated with the comprehensive data similarity; A second weight compensation coefficient is set as a ratio of the second positioning prediction confidence to a preset standard prediction confidence; A product of the second weight compensation coefficient and a second initial weight is set as a second adaptive weight, and a first adaptive weight is obtained by subtracting the second adaptive weight from 1; The second initial weight is 0.4, and the second adaptive weight is not less than 0.3 and not more than 0.6; The first fault point positioning and the second fault point positioning are fitted based on the second adaptive weight and the first adaptive weight, and a weighted center point of the two is set as a cable fault positioning result.

[0061] In summary, the embodiments of the present application have at least the following technical effects: The application provides a cable fault accurate pinpointing method and system based on acoustic-magnetic synchronization and deep learning. By deeply fusing physical signal processing and data-driven modeling and introducing a dynamic adaptive fusion decision mechanism, the accuracy, robustness and scene adaptability of fault pinpointing results are significantly improved. Specifically, a deep convolutional neural network is used to adaptively denoise the synchronously collected acoustic-magnetic signals, which can dynamically adjust the denoising strength according to the real-time monitored signal interference parameters, so as to effectively extract weak fault feature signals in a strong noise background and generate high-quality enhanced acoustic-magnetic synchronization signals, thereby laying a reliable data foundation for subsequent accurate pinpointing based on acoustic-magnetic time difference. Further, a fault location recognition plug-in based on a long short-term memory network is trained. The model can learn and remember the complex nonlinear mapping relationship between the cable distribution topology, fault scene parameters and fault point position, so as to directly output an intelligent inferred location result according to the input data, thereby making up for the defects of insufficient consideration of complex propagation paths and fault characteristics. Further, the similarity between the current scene data and the historical samples is analyzed to evaluate the confidence of the deep learning model output, and the weights of the acoustic-magnetic synchronization method result and the deep learning model result are dynamically and reasonably allocated for weighted fusion. The advantages of intelligent pinpointing of the model are fully utilized in the scene where the model is familiar and the data quality is high, while in the new or complex scene where the generalization ability of the model may be limited, the stability of the physical signal positioning result is relied on more, so as to realize the complementary advantages and improve the reliability and universality of the overall scheme under different working conditions. Compared with the traditional method, the technical scheme provided by the application effectively overcomes the problems of inaccurate positioning caused by large signal noise interference, inaccurate propagation model and various fault types in a complex field environment, and achieves the technical effects of improving the anti-interference ability, adaptive ability and final result credibility of cable fault pinpointing detection, thereby providing effective technical support for realizing the rapid and accurate repair of power cable faults.

[0062] It should be noted that the above sequence of the embodiments of the application is only for description, and does not represent the advantages and disadvantages of the embodiments. Moreover, the above describes specific embodiments of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0063] The above only describes the preferred embodiments of the application and does not limit the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.

[0064] The specification and drawings are, of course, to be regarded in an illustrative rather than a restrictive sense. It is to be understood that any such modifications, variations, combinations or equivalents that fall within the scope of the application are intended to be embraced herein.

Claims

1. A method for accurate pinpointing of cable faults based on simultaneous acousto-magnetic and deep learning, characterized in that, The method comprises: Synchronously collecting acoustic signals and magnetic signals generated in the cable fault area, and preprocessing to generate original acoustic-magnetic synchronous signal pairs; Adaptive denoising of the original acoustic-magnetic synchronous signal pairs is performed using a deep convolutional neural network to obtain enhanced acoustic-magnetic synchronous signal pairs, and the acoustic-magnetic propagation velocity difference is calculated based on the enhanced acoustic-magnetic synchronous signal pairs to determine the first fault point positioning; The regional cable distribution topology of the cable fault area, the current fault scene parameters and the enhanced acoustic-magnetic synchronous signal pairs are input into a fault positioning recognition plug-in constructed based on a long short-term memory network to output the second fault point positioning; The second positioning prediction confidence is determined by analyzing the regional cable distribution topology and the current fault scene parameters, and the first fault point positioning and the second fault point positioning are fitted according to the adaptive fusion strategy formulated based on the second positioning prediction confidence to output the cable fault pinpointing result.

2. The method of claim 1, wherein, Continuous wavelet transform is performed on the acoustic signals and the magnetic signals to extract joint features of the signals in the time and frequency dimensions, and acoustic signal time-frequency diagrams and magnetic signal time-frequency diagrams are generated as original acoustic-magnetic synchronous signal pairs.

3. The method of claim 2, wherein, Adaptive denoising of the original acoustic-magnetic synchronous signal pairs is performed using a deep convolutional neural network to obtain enhanced acoustic-magnetic synchronous signal pairs, including: The current signal interference parameter sequence in the cable fault area is monitored and acquired, wherein the signal interference parameters at least include full-band signal-to-noise ratio, key frequency band noise power spectral density, pulse interference density, pulse interference amplitude and peak signal-to-noise ratio of acoustic-magnetic cross-correlation sequence; Signal interference strength evaluation and signal interference fluctuation analysis are performed based on the current signal interference parameter sequence to output the current signal interference strength and the current signal interference fluctuation degree; Denoising complexity evaluation is performed based on the current signal interference strength and the current signal interference fluctuation degree to output the current denoising complexity; Convolutional neural network is used to construct acoustic signal denoising channels and magnetic signal denoising channels, and the acoustic signal denoising channels and the magnetic signal denoising channels are called according to the current denoising complexity to perform signal denoising on the acoustic signal time-frequency diagrams and the magnetic signal time-frequency diagrams respectively, and enhanced acoustic-magnetic synchronous signal pairs are output.

4. The method of claim 3, wherein, Denoising complexity evaluation is performed based on the current signal interference strength and the current signal interference fluctuation degree to output the current denoising complexity, including: The ratio of the current signal interference strength to the preset standard signal interference strength is set as a first denoising complexity coefficient; The ratio of the current signal interference fluctuation degree to the preset standard signal interference fluctuation degree is set as a second denoising complexity coefficient; After the first denoising complexity coefficient and the second denoising complexity coefficient are dimensionless processed, the current denoising complexity is obtained by weighted summation according to the preset weight proportion.

5. The method of claim 3, wherein, Convolutional neural network is used to construct acoustic signal denoising channels and magnetic signal denoising channels, and the acoustic signal denoising channels and the magnetic signal denoising channels are called according to the current denoising complexity to perform signal denoising on the acoustic signal time-frequency diagrams and the magnetic signal time-frequency diagrams respectively, including: The sample sound signal time-frequency graph set and the sample enhanced sound signal time-frequency graph set are collected as training data, and are equally divided into K parts, K times are selected from the K data sets with replacement to obtain a first training set, and K times are iteratively selected to obtain K training sets, wherein the sample enhanced sound signal time-frequency graph is obtained by denoising the sample sound signal time-frequency graph; The K training sets are used to train the convolutional neural network to convergence, respectively, to obtain K sound signal denoising units, which are integrated to construct a sound signal denoising channel according to a mean fusion strategy, wherein K is an integer greater than 10; The ratio of the current denoising complexity to the preset maximum denoising complexity is multiplied by K to obtain an adaptive unit selection quantity P, wherein the preset maximum denoising complexity is the historical maximum denoising complexity in a historical time range, and P is greater than or equal to 3; P denoising units are randomly selected from the K sound signal denoising units of the sound signal denoising channel, the sound signal time-frequency graph is denoised, and an enhanced sound signal time-frequency graph is obtained after mean fitting of the P denoising results; The sample magnetic signal time-frequency graph set and the sample enhanced magnetic signal time-frequency graph set are collected as training data, and a convolutional neural network is trained to convergence to obtain K magnetic signal denoising units; P denoising units are randomly selected from the K magnetic signal denoising units, the magnetic signal time-frequency graph is denoised, and an enhanced magnetic signal time-frequency graph is obtained after mean fitting of the P denoising results; The enhanced sound signal time-frequency graph and the enhanced magnetic signal time-frequency graph are used as an enhanced acoustic and magnetic synchronous signal pair.

6. The method of claim 1, wherein, The construction method of the fault location recognition plug-in includes: Based on historical cable fault detection records, sample regional cable distribution topology sets, sample fault scene parameter sets, and sample enhanced acoustic and magnetic synchronous signal pair sets are collected, and historical fault point locations under different sample regional cable distribution topologies, sample fault scene parameters, and sample enhanced acoustic and magnetic synchronous signal pairs are collected as sample fault point locations, and a sample fault point location set is obtained, wherein the fault scene parameters include fault type, fault impedance, and fault start angle; The sample regional cable distribution topology set, the sample fault scene parameter set, and the sample enhanced acoustic and magnetic synchronous signal pair set are used as input, and the sample fault point location set is used as supervised training of a long short-term memory network to convergence to generate a fault location recognition plug-in.

7. The method of claim 6, wherein, A second positioning prediction confidence is determined by combining the regional cable distribution topology and the current fault scene parameter, including: High-frequency feature extraction is performed on the sample regional cable distribution topology set and the sample fault scene parameter set to obtain a high-frequency regional cable distribution topology set and a high-frequency fault scene parameter set; Feature extraction is performed on the high-frequency regional cable distribution topology set and the high-frequency fault scene parameter set, respectively, to construct a high-frequency distribution topology feature vector library and a high-frequency scene parameter feature vector library; Feature extraction is performed on the regional cable distribution topology and the current fault scene parameter to construct a current distribution topology feature vector and a current scene parameter feature vector; The current distribution topology feature vector is compared with the high-frequency distribution topology feature vector library through similar traversal to determine a distribution topology similarity. The current scene parameter feature vector is compared with the high-frequency scene parameter feature vector library for similarity traversal comparison to determine a scene parameter similarity; A comprehensive data similarity is calculated based on the distribution topology similarity and the scene parameter similarity, and a second positioning prediction confidence is determined according to the comprehensive data similarity, wherein the second positioning prediction confidence is positively correlated with the comprehensive data similarity.

8. The method of claim 1, wherein, An adaptive fusion strategy is formulated according to the second positioning prediction confidence to perform positioning fitting on the first fault point positioning and the second fault point positioning, and a cable fault positioning result is output, including: A second weight compensation coefficient is set as a ratio of the second positioning prediction confidence to a preset standard prediction confidence; A second adaptive weight is set as a product of the second weight compensation coefficient and a second initial weight, and a first adaptive weight is obtained by subtracting the second adaptive weight from 1; The first fault point positioning and the second fault point positioning are subjected to positioning fitting based on the second adaptive weight and the first adaptive weight, and a weighted center point of the two is set as the cable fault positioning result.

9. The method of claim 8, wherein, The second initial weight is 0.4, and the second adaptive weight is not less than 0.3 and not more than 0.

6.

10. A cable fault pinpointing system based on acoustomagnetic synchronization and deep learning, characterized in that, The system for implementing the cable fault precise positioning method based on acoustic-magnetic synchronization and deep learning according to any one of claims 1-9, the system comprising: An acoustic-magnetic signal acquisition module configured to synchronously acquire acoustic signals and magnetic signals generated in a cable fault area, and to preprocess the acoustic signals and the magnetic signals to generate a pair of original acoustic-magnetic synchronous signals; A first fault point acquisition module configured to combine current signal interference parameters, use a deep convolutional neural network to perform adaptive denoising on the pair of original acoustic-magnetic synchronous signals to obtain a pair of enhanced acoustic-magnetic synchronous signals, and calculate an acoustic-magnetic propagation velocity difference based on the pair of enhanced acoustic-magnetic synchronous signals to determine a first fault point positioning; A second fault point acquisition module configured to input a regional cable distribution topology of the cable fault area, current fault scene parameters, and the pair of enhanced acoustic-magnetic synchronous signals into a fault positioning recognition plug-in constructed based on a long short-term memory network to output a second fault point positioning; A cable fault positioning module configured to combine the regional cable distribution topology and the current fault scene parameters to analyze and determine a second positioning prediction confidence, and to formulate an adaptive fusion strategy according to the second positioning prediction confidence to perform positioning fitting on the first fault point positioning and the second fault point positioning, and to output a cable fault positioning result.

Images