Fault Location Method and System for Power Cable

Abstract

The present invention belongs to the technical field of fault location, and specifically relates to a fault location method and system for power cables. The steps include obtaining real-time cable current pulse waveform data, fault current pulse waveform data, and historical cable operation data; through Haar wavelet and Symlets wavelet analysis, summarizing signal mutation points into a signal mutation point feature set; constructing a threshold intelligent optimization model to calculate the time difference of the current pulse wavefront; using an RNN-CNN architecture to identify the wavefront time and fitting to obtain the wavefront time change value per unit distance of the entire cable; calculating and correcting the fault distance, and combining the corrected fault distance with the cable starting end position information to determine the position of the fault point. Through the deep combination of multi-stage wavelet analysis, a threshold intelligent optimization model, and an RNN-CNN architecture, the present invention realizes the fast and accurate location of power cable faults, significantly improving the fault diagnosis efficiency and accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault location, and particularly relates to a method and system for fault location of power cables. Background Art

[0002] Power cables play a crucial role in the power transmission system. However, due to various factors during long-term operation, such as aging, external force damage, and insulation moisture absorption, power cables are prone to failures. Accurately and quickly locating the fault points of power cables is of extremely crucial significance for ensuring the stable operation of the power system, reducing power outage time, and maintenance costs.

[0003] Traditional power cable fault location methods mainly rely on impedance methods and traveling wave methods. However, impedance methods usually judge the fault location based on the impedance change before and after the cable fault point, with low positioning accuracy and being easily affected by line parameter changes. Traveling wave methods often have difficulty accurately identifying the arrival time of the traveling wave head, resulting in large positioning errors. Therefore, it is difficult to meet the strict requirements of modern power cables for fast and accurate fault location. Summary of the Invention

[0004] To overcome the above defects of the prior art, the present invention provides a method and system for fault location of power cables. Through the deep combination of multi-stage wavelet analysis, threshold intelligent optimization model, and RNN-CNN architecture, fast and accurate location of power cable faults is achieved, significantly improving the efficiency and accuracy of fault diagnosis, especially performing well in complex cable networks.

[0005] The present invention provides the following technical solutions: A method for fault location of power cables, comprising the following steps:

[0006] S1: Obtain real-time cable current pulse waveform data, fault current pulse waveform data, and historical cable operation data;

[0007] S2: Initially decompose the real-time cable current pulse waveform data by Haar wavelet to obtain sub-signals of different frequency segments, analyze the sub-signals of different frequency segments by Symlets wavelet and locate the signal mutation points, and summarize the signal mutation points into a signal mutation point feature set;

[0008] S3: Construct a threshold intelligent optimization model, determine the optimal threshold through the threshold intelligent optimization model in combination with the signal mutation point feature set, accurately measure the current pulse wave head times TR1 and TR2, and calculate the current pulse wave head time difference;

[0009] S4: Segment the cable and set up test points at each segment. Inject multi-feature pulse signals into the test points. At the receiving end, use the RNN-CNN architecture to identify the wavefront time, combine the signal mutation point feature set to accurately calculate the change value of the propagation time per unit distance at each test point, and fit to obtain the change value of the wavefront time per unit distance of the entire cable segment;

[0010] S5: Calculate the fault distance through the time difference of the current pulse wavefront and the change value of the wavefront time per unit distance of the entire cable segment, and correct the fault distance through dynamic fuzzy logic. Combine the corrected fault distance with the cable starting end position information to determine the position of the fault point.

[0011] Preferably, in S1, obtain the real-time cable current pulse waveform data and the fault current pulse waveform data. The obtaining process is as follows:

[0012] Set the sampling frequency to Perform discrete sampling on the continuous current signal, where:

[0013] ;

[0014] In the formula, represents twice the highest frequency of the original signal .

[0015] Preferably, in S2, initially decompose the real-time cable current pulse waveform data through Haar wavelets to obtain sub-signals of different frequency bands. The decomposition process is as follows:

[0016] In the case of discrete acquisition of the real-time cable current pulse waveform data, set the given discrete current pulse waveform data sequence as x , where n = 1, 2, 3,..., N, n represents the discrete time series sample point serial number, N is the number of data points. For scale j, j = 1, 2, 3,..., J, and , J is the maximum decomposition layer, represents the upper limit of the total number of sub-bands obtained by decomposing the signal at the maximum decomposition layer J;

[0017] Based on scale j and translation parameter k, define the corresponding discrete wavelet function and the scaling function as:

[0018] ;

[0019] ;

[0020] In the formula, represents the scaling factor, represents the scale transformation factor, and represent the wavelet function and the scaling function after scale and translation transformations respectively;

[0021] Set the real-time cable current pulse waveform data as ;

[0022] Calculate the approximation coefficient at scale j , and the calculation method is:

[0023] ;

[0024] Calculate the detail coefficient at scale j , and the calculation method is:

[0025] ;

[0026] In discrete data, for scale j and translation parameter k, the corresponding discrete Symlets wavelet function is defined as:

[0027] ;

[0028] In the formula, represents the discrete Symlets wavelet function of the unscaled wavelet function at scale j and translation parameter k;

[0029] Use Symlets wavelet to analyze sub-signals in different frequency bands. Through discrete wavelet transform, calculate the detail coefficient, and the calculation method is:

[0030] ;

[0031] In the formula, represents the detail coefficient of the Symlets wavelet at scale j and translation parameter k, represents the sub-signal sequence after Haar wavelet decomposition, represents the discrete Symlets wavelet function after scaling and translation.

[0032] Preferably, in S2, use Symlets wavelet to analyze sub-signals in different frequency bands and locate the signal mutation point. The location process is as follows:

[0033] Set the detail coefficient , and the total number is M;

[0034] Calculate the mean value and the standard deviation of the detail coefficient , and the calculation method is:

[0035] ;

[0036] In the formula, i represents the index used to traverse the detail coefficients, and represents the scale j and translation parameter k corresponding to the i-th detail coefficient, represents at a given and after performing discrete wavelet transform using the Symlets wavelet, the obtained detail coefficients;

[0037] ;

[0038] Calculate the threshold T of the detail coefficients, and the calculation method is:

[0039] ;

[0040] In the formula, represents the empirical coefficient, ;

[0041] For each scale j and translation parameter k, traverse all the detail coefficients ;

[0042] When set the adjacent points as and , which are the detail coefficients of the Symlets wavelet at scales j and translation parameters k + 1, k - 1;

[0043] When and then is considered a modulus maximum point and is marked as a signal mutation point;

[0044] When then no marking is performed;

[0045] The process of marking the signal mutation points is as follows:

[0046] Create a marking sequence with the same length as the original signal data , initialized to zero. When a mutation point is detected at the position corresponding to then , represents any detected mutation point position, set the signal mutation point feature set as fare, , mc represents the total number of signal mutation points, represents the position of the mc-th mutation point in the signal.

[0047] Preferably, in the step S3, a threshold intelligent optimization model is constructed, and the construction process is as follows:

[0048] Set the set of real-time cable current pulse waveform data as , where , v is the total number of real-time cable current pulse waveform data points, is the current amplitude of the vth real-time cable current pulse waveform data point, is the time corresponding to the vth real-time cable current pulse waveform data point;

[0049] Set the set of fault current pulse waveform data as , where , g is the total number of fault current pulse waveform data points, is the current amplitude of the gth fault current pulse waveform data point, is the time corresponding to the gth fault current pulse waveform data point;

[0050] Merge the real-time cable current pulse waveform data set B and the fault current pulse waveform data set E into the total sample set S. The merging process is as follows:

[0051] ;

[0052] In the formula, represents the (v + g)th data point in the total sample set; , represents the data point 's current amplitude, represents the data point 's time;

[0053] Set the current amplitude threshold as LO;

[0054] If , then mark as , used to represent the wavefront;

[0055] If , then mark as , used to represent non-wavefront;

[0056] In the formula, represents a marking variable used to mark whether it is recognized as a wavefront;

[0057] Obtain the marked sample set ;

[0058] Perform normalization processing on . The processing process is as follows:

[0059] ;

[0060] Wherein, and are respectively the minimum and maximum values of the current amplitude in the entire sample set, represents the current amplitude after normalization processing;

[0061] Construct a threshold intelligent optimization model, and the construction process is as follows:

[0062] Set the input feature vector as , and the threshold intelligent optimization model is as follows:

[0063] ;

[0064] Wherein, represents the probability value that the sample XE belongs to the wavefront, and the value range is between ; represents the model parameter, and eh represents the natural constant;

[0065] Train the threshold intelligent optimization model through the cross-entropy cost function to determine the model parameter , and the training process is as follows:

[0066] ;

[0067] Wherein, represents the value of the cost function, represents the true class information of each sample, represents the probability value that the model predicts that the iq-th sample belongs to the wavefront, and log represents the logarithmic function;

[0068] Adopt the gradient descent algorithm to minimize , and the formula for updating the parameter is:

[0069] ;

[0070] Wherein, represents the jc-th parameter in the model parameter, represents the learning rate, represents the partial derivative of the cost function with respect to the parameter ;

[0071] ;

[0072] Wherein, represents the jc-th parameter in the feature vector ;

[0073] Set the number of iterations as BH, when the cost function When iterating to the BH-th time, stop the iteration and obtain the trained model parameters ;

[0074] Set the threshold of the model parameter to be ;

[0075] Calculate the accuracy of the threshold , and the calculation method is:

[0076] ;

[0077] In the formula, represents the number of true positive examples when the threshold is , that is, the number of validation set samples that are actually wave fronts and are correctly judged as wave fronts,

[0078]

[0079]

[0080]

[0081]

[0082]

[0083]

[0084]

[0085]

[0086]

[0087] ​​​​​​​​​​​​​​​​​​​​​​​​​Record the time points of the transition of the current pulse waveform data from non-wavefront to wavefront, denoted as TR1 and TR2 respectively;

[0088] Calculate the time difference of the current pulse wavefront , and the calculation method is:

[0089] .

[0090] Preferably, in S4, a multi-feature pulse signal is injected into the test point, and the wavefront time is identified by the receiving end using the RNN-CNN architecture. The identification process is as follows:

[0091] Set the pulse signal sequence input to the RNN-CNN architecture as , indicating the feature vector at time t;

[0092] Update the hidden state of the RNN layer. The update process is as follows:

[0093] ;

[0094] In the formula, represents the hidden state of the RNN layer at time t, represents the weight matrix input to the hidden layer, represents the weight matrix from the hidden layer to the hidden layer, represents the bias vector of the hidden layer, represents the hyperbolic tangent activation function, represents the hidden state of the RNN layer at time t-1;

[0095] Set the CNN layer to use kf convolutional kernels, and the convolutional kernel is , where mf is the size in the time dimension and nf is the size in the feature dimension. For the ie-th convolutional kernel, its convolution operation formula is:

[0096] ;

[0097] In the formula, represents the eigenvalue obtained at time t after the convolution operation of the ie-th convolutional kernel, represents the weight of the ie-th convolutional kernel at position , represents the bias vector of the ie-th convolutional kernel, represents the hidden state of the RNN layer at time t at position ;

[0098] After the convolution operation, a series of feature maps , represents the kj-th feature map;

[0099] Then perform a pooling operation, set the pooling window size, and the feature map after pooling , represents the km-th feature map, where:

[0100] ;

[0101] In the formula, represents the feature value of the km-th feature map at position after the pooling operation, represents the feature value of the kj-th feature map at position after the pooling operation, represents the area corresponding to the ja-th pooling window;

[0102] Flatten the feature map after pooling to obtain a one-dimensional vector FBB, and then perform mapping through a fully connected layer. Assume the fully connected layer has LBB neurons, and its calculation formula is:

[0103] ;

[0104] In the formula, represents the output of the jb-th neuron after the activation function of the fully connected layer, represents the activation function, represents the weight of the fully connected layer, NF represents the length of the one-dimensional vector FBB, represents the bias vector, represents the lb-th neuron in the one-dimensional vector FBB;

[0105] Determine the position information of the wavefront time through an output layer, set the output as yhz, and the calculation method is:

[0106] ;

[0107] In the formula, represents the weight of the output layer, represents the bias vector of the output layer;

[0108] Set the range of XVB from 0 to t, and yhz falls into , and calculate the wavefront time at the iyec-th test point through linear mapping. The calculation process is:

[0109] ;

[0110] In the formula, MLE represents the maximum value of yhz and is used for normalization processing.

[0111] Preferably, in S4, the change value of the propagation time per unit distance of each test point is accurately calculated in combination with the mutation point feature set, and the wavefront time transformation value per unit distance of the entire cable is obtained by fitting. The fitting process is as follows:

[0112] Calculate the change value of the propagation time per unit distance of each test point , and the calculation method is:

[0113] ;

[0114] In the formula, represents the distance between the (iyec - 1)-th test point and the iyec-th test point, represents the wavefront time at the (iyec - 1)-th test point;

[0115] Obtain the wavefront time transformation value per unit distance of the entire cable by fitting , and the fitting process is as follows:

[0116] ;

[0117] In the formula, ikkq represents the index variable, and nw represents the total number of test points.

[0118] Preferably, in S5, the fault distance is calculated by the time difference of the current pulse wavefront and the wavefront time change value per unit distance of the entire cable. The calculation method is:

[0119] ;

[0120] In the formula, LJNT represents the fault distance;

[0121] And the fault distance is corrected by dynamic fuzzy logic. The correction process is as follows:

[0122] ;

[0123] In the formula, represents the revised fault distance, , and are the membership functions of the cable temperature, humidity, and aging degree respectively, , and are the weight coefficients corresponding to , and respectively.

[0124] Preferably, in S5, the corrected fault distance is combined with the cable starting end position information to determine the position of the fault point. The determination process is as follows:

[0125] Set the cable starting end position coordinates as , calculate the position coordinates of the fault point , and the calculation method is as follows:

[0126] ;

[0127] Set the actual laying length of the cable as LCDP;

[0128] If , determine that the position coordinates of the fault point are accurate;

[0129] If , determine that the position coordinates of the fault point are inaccurate, then recalculate through S1 to S4 until it reaches.

[0130] The fault location system of the power cable is used to implement the above-mentioned power cable fault location method, including: an acquisition module, an analysis module, a construction module, a calculation module and a location module;

[0131] The acquisition module is used to acquire real-time cable current pulse waveform data, fault current pulse waveform data and historical cable operation data;

[0132] The analysis module is used to initially decompose the real-time cable current pulse waveform data through Haar wavelet to obtain sub-signals of different frequency bands, analyze the sub-signals of different frequency bands by Symlets wavelet and locate the signal mutation points, and summarize the signal mutation points into a signal mutation point feature set;

[0133] The construction module is used to construct a threshold intelligent optimization model, determine the optimal threshold through the threshold intelligent optimization model combined with the signal mutation point feature set, accurately measure the current pulse wavefront times TR1 and TR2, and calculate the current pulse wavefront time difference;

[0134] The calculation module is used to segment the cable and set test points in each segment, inject multi-feature pulse signals at the test points, the receiving end uses the RNN-CNN architecture to identify the wavefront time, and accurately calculate the unit distance propagation time change value of each test point in combination with the signal mutation point feature set and fit to obtain the unit distance wavefront time change value of the entire cable segment;

[0135] The location module is used to calculate the fault distance through the current pulse wavefront time difference and the unit distance wavefront time change value of the entire cable segment, and correct the fault distance through dynamic fuzzy logic, and combine the corrected fault distance with the cable starting end position information to determine the position of the fault point.

[0136] The technical effects and advantages of the present invention:

[0137] The present invention adopts a multi-stage wavelet analysis strategy. First, the Haar wavelet is used to preliminarily decompose the real-time cable current pulse waveform data, quickly separating sub-signals in different frequency bands. Then, the Symlets wavelet is used to deeply analyze these sub-signals to accurately locate the signal mutation points. This combined wavelet analysis method from coarser to finer comprehensively and deeply excavates the potential feature information in the current pulse waveform data, significantly enhancing the accuracy and integrity of data feature extraction, providing a solid, reliable, and high-resolution data support for the subsequent fault diagnosis process, and greatly improving the quality of the initial data for fault diagnosis.

[0138] The threshold intelligent optimization model constructed by the present invention has a high degree of self-adaptability. By deeply learning the real-time cable current pulse waveform data and fault current pulse waveform data, the model can accurately adapt to the current operating state and data characteristics of the cable, thereby accurately determining the current pulse wavefront time. This data-driven and feature-fusion method for determining the threshold effectively avoids the subjectivity and limitations of traditional threshold setting methods, significantly improving the accuracy and reliability of measuring the current pulse wavefront time. Especially in complex cable networks with long distances and multiple branches, it can effectively reduce the fault location deviation.

[0139] The present invention introduces the RNN-CNN architecture into the cable fault detection process, giving full play to the advantages of the recurrent neural network in processing time-series data and the convolutional neural network in feature extraction. After injecting multi-feature pulse signals at the cable section test points, the RNN-CNN architecture can deeply analyze and identify complex wavefront time signals. Combining with the mutation point feature set obtained from the previous wavelet analysis, it accurately calculates the change value of the unit distance propagation time at each test point and fits to obtain the change value of the unit distance wavefront time for the entire cable section. This signal processing method based on deep learning can quickly process a large amount of data, significantly shortening the fault location time and improving the accuracy of fault distance calculation.

[0140] The present invention corrects the fault distance through dynamic fuzzy logic and determines the exact location of the fault point in combination with the cable starting end position information. This method comprehensively considers the influence of factors such as cable temperature, humidity, and aging degree, further improving the accuracy and reliability of fault location. Overall, the present invention performs excellently in improving the efficiency and accuracy of fault location, especially suitable for the fault detection requirements of complex cable networks in modern power systems. Brief Description of the Drawings

[0141] Figure 1 is the step flow chart of the fault location method for the power cable of the present invention;

[0142] Figure 2 is the block diagram of the fault location system for the power cable of the present invention. Detailed Embodiments

[0143] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0144] Embodiment 1:

[0145] As Figure 1 shown, the fault location method for power cables includes the following steps:

[0146] S1: Obtain real-time cable current pulse waveform data, fault current pulse waveform data, and historical cable operation data;

[0147] S2: Initially decompose the real-time cable current pulse waveform data by Haar wavelet to obtain sub-signals of different frequency bands, analyze the sub-signals of different frequency bands using Symlets wavelet to locate the signal mutation points, and summarize the signal mutation points into a signal mutation point feature set;

[0148] S3: Construct a threshold intelligent optimization model, determine the optimal threshold through the threshold intelligent optimization model in combination with the signal mutation point feature set, accurately measure the current pulse wavefront times TR1 and TR2, and calculate the current pulse wavefront time difference;

[0149] S4: Segment the cable and set test points in each segment, inject multi-feature pulse signals at the test points, use the RNN-CNN architecture at the receiving end to identify the wavefront time, accurately calculate the unit distance propagation time change value of each test point in combination with the signal mutation point feature set, and fit to obtain the unit distance wavefront time transformation value of the entire cable segment;

[0150] S5: Calculate the fault distance through the current pulse wavefront time difference and the unit distance wavefront time change value of the entire cable segment, correct the fault distance through dynamic fuzzy logic, and combine the corrected fault distance with the cable starting end position information to determine the position of the fault point.

[0151] The Haar wavelet is a simple basis function based on piecewise constant functions, with symmetry, compact support, and orthogonality. The Symlets wavelet is a symmetric Daubechies wavelet basis with better symmetry. The RNN-CNN architecture is the architecture in which the recurrent neural network RNN and the convolutional neural network CNN are used in combination.

[0152] In S1, to obtain the real-time cable current pulse waveform data and the fault current pulse waveform data, the obtaining process is as follows:

[0153] Set the sampling frequency to Perform discrete sampling on the continuous current signal, where:

[0154] ;

[0155] In the formula, Represents twice the highest frequency of the original signal times.

[0156] In S2, the real-time cable current pulse waveform data is initially decomposed by Haar wavelet to obtain sub-signals in different frequency bands. The decomposition process is as follows:

[0157] In the case of discrete acquisition of real-time cable current pulse waveform data, the given discrete current pulse waveform data sequence is set as x , where n = 1, 2, 3,..., N, n represents the serial number of discrete time series sample points, N is the number of data points. For scale j, j = 1, 2, 3,..., J, and , J is the maximum decomposition level, represents the upper limit of the total number of sub-bands obtained by decomposing the signal at the maximum decomposition level J;

[0158] Based on scale j and translation parameter k, the corresponding discrete wavelet function and scaling function are defined as:

[0159] ;

[0160] ;

[0161] In the formula, represents the scaling factor, represents the scale transformation factor, and respectively represent the wavelet function and scaling function after scale and translation transformations;

[0162] Set the real-time cable current pulse waveform data as ;

[0163] Calculate the approximation coefficient at scale j, and the calculation method is:

[0164] ;

[0165] Calculate the detail coefficient at scale j, and the calculation method is:

[0166] ;

[0167] In discrete data, for scale j and translation parameter k, the corresponding discrete Symlets wavelet function is defined as:

[0168] ;

[0169] In the formula, Denote the discrete Symlets wavelet function of the unscaled wavelet function at scale \(j\) and translation parameter \(k\);

[0170] Use Symlets wavelet to analyze the sub-signals of different frequency bands. Through discrete wavelet transform, calculate the detail coefficients, and the calculation method is:

[0171] ;

[0172] In the formula, Denote the detail coefficients of the Symlets wavelet at scale \(j\) and translation parameter \(k\), Denote the sub-signal sequence after Haar wavelet decomposition, Denote the discrete Symlets wavelet function after scaling and translation.

[0173] In S2, use Symlets wavelet to analyze the sub-signals of different frequency bands and locate the signal mutation points. The location process is as follows:

[0174] Set the detail coefficients , and the total number is \(M\);

[0175] Calculate the mean value and standard deviation of the detail coefficients , and the calculation method is:

[0176] ;

[0177] In the formula, \(i\) represents the index used to traverse the detail coefficients, and Denote the scale \(j\) and translation parameter \(k\) corresponding to the \(i\)-th detail coefficient, Denote the detail coefficients obtained after discrete wavelet transform using Symlets wavelet under the given and ;

[0178] ;

[0179] Calculate the threshold \(T\) of the detail coefficients, and the calculation method is:

[0180] ;

[0181] In the formula, Denote the empirical coefficient, ;

[0182] For each scale \(j\) and translation parameter \(k\), traverse all the detail coefficients ;

[0183] When , set the adjacent points as and are the detail coefficients of the Symlets wavelet at scale j and translation parameters k + 1 and k - 1;

[0184] When and then it is considered that is a modulus maximum point and is marked as a signal mutation point;

[0185] When then no marking is performed;

[0186] The process of marking signal mutation points is as follows:

[0187] Create a marking sequence with the same length as the original signal data , initialized to zero. When a mutation point is detected at the corresponding position of then , represents the position of any detected mutation point. Set the signal mutation point feature set as fare, , mc represents the total number of signal mutation points, represents the position of the mc-th mutation point in the signal in the signal.

[0188] In this embodiment, a multi-stage wavelet analysis strategy is adopted. First, the Haar wavelet is used to preliminarily decompose the real-time cable current pulse waveform data, which can quickly separate the sub-signals of different frequency bands and preliminarily outline the general frequency feature framework of the data. On this basis, the Symlets wavelet is further used to analyze each frequency sub-signal. Relying on the excellent performance of the Symlets wavelet in time-frequency localization characteristics, the signal mutation points are accurately located. This combination of wavelet analysis methods from coarse to fine comprehensively and deeply excavates the potential feature information in the current pulse waveform data, significantly enhancing the accuracy and integrity of data feature extraction. Compared with traditional single wavelet analysis methods, it can more effectively capture weak signal changes and complex waveform details, providing solid, reliable and high-resolution data support for the subsequent fault diagnosis link, greatly improving the quality of the initial data for fault diagnosis, and thus ensuring the accuracy and stability of the entire fault detection process from the source.

[0189] In S3, a threshold intelligent optimization model is constructed, and the construction process is as follows:

[0190] Set the set of real-time cable current pulse waveform data as , where , v is the total number of real-time cable current pulse waveform data points, $I_{v}$ is the current amplitude of the $v$-th real-time cable current pulse waveform data point, $t_{v}$ is the time corresponding to the $v$-th real-time cable current pulse waveform data point;

[0191] Set the set of fault current pulse waveform data as , where , $g$ is the total number of fault current pulse waveform data points, $I_{g}$ is the current amplitude of the $g$-th fault current pulse waveform data point, $t_{g}$ is the time corresponding to the $g$-th fault current pulse waveform data point;

[0192] Merge the real-time cable current pulse waveform data set $B$ and the fault current pulse waveform data set $E$ into the total sample set $S$. The merging process is as follows:

[0193] ;

[0194] In the formula, represents the $(v + g)$-th data point in the total sample set; , represents the current amplitude of the data point , represents the time of the data point ;

[0195] Set the current amplitude threshold as $LO$;

[0196] If , then mark as , which is used to represent the wavefront;

[0197] If , then mark as , which is used to represent non-wavefront;

[0198] In the formula, represents a marking variable used to mark whether is recognized as the wavefront;

[0199] Obtain the marked sample set ;

[0200] Perform normalization processing on . The processing process is as follows:

[0201] ;

[0202] In the formula, and are respectively the minimum and maximum values of the current amplitude in the entire sample set, represents the current amplitude after normalization;

[0203] Construct a threshold intelligent optimization model, and the construction process is as follows:

[0204] Set the input feature vector as , and the threshold intelligent optimization model is as follows:

[0205] ;

[0206] In the formula, represents the probability value that the sample XE belongs to the wavefront, and the value range is between ; represents the model parameter, and eh represents the natural constant;

[0207] Train the threshold intelligent optimization model through the cross-entropy cost function to determine the model parameter , and the training process is as follows:

[0208] ;

[0209] In the formula, represents the value of the cost function, represents the true class information of each sample, represents the probability value that the model predicts the iq-th sample to belong to the wavefront, and log represents the logarithmic function;

[0210] Adopt the gradient descent algorithm to minimize , and the formula for updating the parameter is:

[0211] ;

[0212] In the formula, represents the jc-th parameter in the model parameter, represents the learning rate, represents the cost function with respect to the parameter partial derivative;

[0213] ;

[0214] In the formula, represents the jc-th parameter in the feature vector ;

[0215] Set the number of iterations to BH. When the cost function iterates to BH times, stop the iteration and obtain the trained model parameter ;

[0216] Set the threshold of the model parameter as ;

[0217] Calculate the accuracy of the threshold as follows: :

[0218] ;

[0219] In the formula, represents the number of true positive examples when the threshold is , that is, the number of validation set samples that are actually wavefronts and are correctly judged as wavefronts, represents the number of true negative examples, that is, the number of validation set samples that are actually not wavefronts and are correctly judged as not wavefronts, represents the number of false positive examples, that is, the number of validation set samples that are actually not wavefronts but are wrongly judged as wavefronts, is the number of false negative examples, that is, the number of validation set samples that are actually wavefronts but are wrongly judged as not wavefronts;

[0220] Calculate the recall rate of the threshold as follows: :

[0221] ;

[0222] Calculate the precision of the threshold as follows: :

[0223] ;

[0224] Calculate the F1 score of the threshold as follows: :

[0225] ;

[0226] Set the threshold of the F1 score to FDC;

[0227] When , determine that is the optimal threshold ;

[0228] When , then do not determine that is the optimal threshold ;

[0229] And record the time points of the transition of the current pulse waveform data from non-wavefront to wavefront, denoted as TR1 and TR2 respectively;

[0230] Calculate the time difference of the current pulse wavefront as follows:

[0231] 。

[0232] The threshold intelligent optimization model constructed in this embodiment has high innovation and self - adaptability. Using rich real - time cable current pulse waveform data and fault current pulse waveform data as samples for in - depth training, the model can fully learn the internal laws and characteristic differences of current pulse waveforms under different working conditions. During the actual operation process, the model closely cooperates with the signal mutation point feature set obtained through wavelet analysis. Based on the multi - dimensional information contained in the feature set, an optimal threshold is determined through complex intelligent algorithms. This optimal threshold can accurately adapt to the current cable operation state and data characteristics, and then accurately measure the current pulse wavefront times TR1 and TR2 of the two - end monitoring devices. Through this data - driven and feature - fusion threshold determination method, it not only effectively avoids the subjectivity and limitations of traditional threshold setting methods, but also greatly improves the accuracy and reliability of current pulse wavefront time measurement. In a complex cable network with long distances and multiple branches, it can significantly reduce the fault location deviation caused by wavefront time measurement errors, and effectively improve the stability and accuracy of the entire fault location in a complex environment.

[0233] In S4, a multi - feature pulse signal is injected into the test point, and the receiving end uses the RNN - CNN architecture to identify the wavefront time. The identification process is as follows:

[0234] Set the pulse signal sequence input to the RNN - CNN architecture as , where represents the feature vector at time t;

[0235] Update the hidden state of the RNN layer. The update process is as follows:

[0236] ;

[0237] In the formula, represents the hidden state of the RNN layer at time t, represents the weight matrix input to the hidden layer, represents the weight matrix from the hidden layer to the hidden layer, represents the bias vector of the hidden layer, represents the hyperbolic tangent activation function, represents the hidden state of the RNN layer at time t - 1;

[0238] Set the CNN layer to use kf convolutional kernels, and the convolutional kernel is where mf is the size in the time dimension and nf is the size in the feature dimension. For the ie - th convolutional kernel, its convolution operation formula is:

[0239] ;

[0240] In the formula, represents the eigenvalue obtained at time t after the convolution operation by the ie-th convolution kernel, represents the weight of the ie-th convolution kernel at position ; represents the bias vector of the ie-th convolution kernel, represents the hidden state of the RNN layer at time t at position ;

[0241] After the convolution operation, a series of feature maps are obtained, represents the kj-th feature map;

[0242] Then, a pooling operation is performed. Set the pooling window size (which can be set according to requirements). The pooled feature map is obtained, represents the km-th feature map, where:

[0243] ;

[0244] In the formula, represents the eigenvalue of the km-th feature map after the pooling operation at position ; represents the eigenvalue of the kj-th feature map after the pooling operation at position ; represents the region corresponding to the ja-th pooling window;

[0245] The pooled feature map is flattened to obtain a one-dimensional vector FBB, and then it is mapped through a fully connected layer. Assume that the fully connected layer has LBB neurons, and its calculation formula is:

[0246] ;

[0247] In the formula, represents the output of the jb-th neuron of the fully connected layer after passing through the activation function, represents the activation function, represents the weight of the fully connected layer, NF represents the length of the one-dimensional vector FBB, represents the bias vector, represents the lb-th neuron in the one-dimensional vector FBB;

[0248] The position information of the wavefront time is determined through an output layer. Set the output as yhz, and the calculation method is:

[0249] ;

[0250] In the formula, denotes the weights of the output layer, denotes the bias vector of the output layer;

[0251] Set the range of XVB from 0 to t, and yhz falls into Through linear mapping, calculate the wavefront time at the iyec-th test point , and the calculation process is:

[0252] ;

[0253] In the formula, MLE represents the maximum value of yhz and is used for normalization processing.

[0254] In S4, combined with the mutation point feature set, accurately calculate the change value of the propagation time per unit distance at each test point and fit to obtain the change value of the wavefront time per unit distance of the entire cable. The fitting process is as follows:

[0255] Calculate the change value of the propagation time per unit distance at each test point , and the calculation method is:

[0256] ;

[0257] In the formula, represents the distance between the (iyec - 1)-th test point and the iyec-th test point, represents the wavefront time at the (iyec - 1)-th test point;

[0258] Fit to obtain the change value of the wavefront time per unit distance of the entire cable , and the fitting process is as follows:

[0259] ;

[0260] In the formula, ikkq represents the index variable, and nw represents the total number of test points.

[0261] In S5, calculate the fault distance through the time difference of the current pulse wavefront and the change value of the wavefront time per unit distance of the entire cable. The calculation method is:

[0262] ;

[0263] In the formula, LJNT represents the fault distance;

[0264] And correct the fault distance through dynamic fuzzy logic. The correction process is as follows:

[0265] ;

[0266] In the formula, represents the revised fault distance, , and Membership functions for cable temperature, humidity, and aging degree respectively , and are weight coefficients corresponding to , and respectively.

[0267] In S5, the corrected fault distance is combined with the cable starting end position information to determine the position of the fault point. The determination process is as follows:

[0268] Set the position coordinates of the cable starting end as , and calculate the position coordinates of the fault point . The calculation method is:

[0269] ;

[0270] Set the actual laying length of the cable as LCDP;

[0271] If , it is determined that the position coordinates of the fault point are accurate;

[0272] If , it is determined that the position coordinates of the fault point are inaccurate, and recalculate through S1 to S4 until is reached.

[0273] In this embodiment, an optimization algorithm based on gradient descent is used to reversely adjust the relevant parameters in the threshold intelligent optimization model, realizing the adaptive dynamic matching of the threshold and the actual operating state of the cable, and improving the accuracy and stability of subsequent fault diagnosis.

[0274] At the same time, this embodiment cleverly introduces the RNN-CNN architecture into the cable fault detection process, giving full play to the unique advantages of the recurrent neural network in processing time series data and the powerful capabilities of the convolutional neural network in the field of image feature extraction and pattern recognition. After injecting multi-feature pulse signals at the cable segment test points, the RNN-CNN architecture at the receiving end can deeply analyze and identify complex wavefront time signals. Combining with the mutation point feature set obtained from the previous wavelet analysis, through rigorous mathematical calculations and data fitting, the propagation time change value per unit distance of each test point is accurately calculated, and further the wavefront time transformation value per unit distance of the entire cable is fitted;

[0275] This signal processing and analysis method based on deep learning architecture can automatically learn and adapt to various complex signal patterns and changing rules in the cable network, effectively overcoming the deficiencies of traditional signal processing methods when facing large-scale and highly complex data. In terms of improving the fault location efficiency, the parallel computing ability and efficient feature extraction mechanism of the RNN-CNN architecture can quickly process a large amount of test point data, significantly shortening the time required for fault location. In terms of accuracy, its powerful learning and generalization ability can accurately identify the subtle changes in the wavefront time, thus ensuring high-precision fault distance calculation, providing a strong technical guarantee for the rapid and accurate location of cable network faults, especially suitable for the increasingly complex cable network structure and working conditions in modern power systems, achieving the effect of quickly and accurately locating cable faults.

[0276] In this embodiment, when calculating each formula, dimensionless or dimension removal processing can be performed on each parameter as needed.

[0277] The verification process is as follows:

[0278] First, set the cable parameters. The total length of the cable is 10 km, and the preset fault point is located at 3.5 km (simulating a short-circuit fault). For test point segmentation, set a test point every 1 km (a total of 10 test points).

[0279] The sampling frequency is 40 kHz, which satisfies , and the real-time current pulse waveform data (normal state) and fault current pulse waveform data are generated by electromagnetic transient simulation software (such as PSCAD). Considering environmental factors, set the temperature to 25 °C, the humidity to 60%, the aging degree to 5%, and the membership function weights

[0280] , , , and are 0.2, 0.1, and 0.3 respectively.

[0281] Perform wavelet decomposition and mutation point detection. In Haar wavelet decomposition, the decomposition level J = 5, and 5 frequency sub-signals (low-frequency approximation signal + high-frequency detail signal) are obtained. The distribution of the sub-signal sequence is that the low-frequency band accounts for 85% and the high-frequency band accounts for 15%. In Symlets wavelet analysis, the total number of detected mutation points mc = 12, which are distributed near the fault point (in the 3.3 - 3.6 km interval).

[0282] Perform threshold intelligent optimization and wavefront time determination. Combine 5000 samples of real-time and fault data, divide them into a training set (80%) and a validation set (20%), and the optimal threshold (The corresponding F1 score is 0.92), the measured result of the wave head time is , , and is obtained.

[0283] Calculate the propagation time of the test points. The recognition result of the RNN-CNN is that the model accuracy is 98.5% (test set), the recognition error of the wave head time is less than 0.5%. Among the wave head times of each test point, for the 1st, 2nd, 6th, and 7th test points, their wave head times are 2.1, 4.3, 12.5, and 14.8 μs respectively.

[0284] Fit the propagation time per unit distance, and is obtained.

[0285] Calculate and correct the fault distance. The initial fault distance , perform dynamic fuzzy correction, and the membership values , and are 0.3, 0.2, and 0.1 respectively. Then , because the preset fault point of the simulation data is 3.5 km and the error is 0.3 km. When the starting end position coordinate is 0, the fault point coordinate , 3.47 km is less than the total length of 10 km. Therefore, it is determined that the fault point position is accurate.

[0286] Embodiment 2:

[0287] As Figure 2 shown, a fault location system for a power cable, which is used to implement the method in Embodiment 1, includes: an acquisition module, an analysis module, a construction module, a calculation module, and a location module;

[0288] The acquisition module is used to acquire real-time cable current pulse waveform data, fault current pulse waveform data, and historical cable operation data;

[0289] The analysis module is used to preliminarily decompose the real-time cable current pulse waveform data by Haar wavelet to obtain sub-signals of different frequency bands, analyze the sub-signals of different frequency bands by Symlets wavelet and locate the signal mutation points, and summarize the signal mutation points into a signal mutation point feature set;

[0290] The construction module is used to construct a threshold intelligent optimization model, determine the optimal threshold by combining the threshold intelligent optimization model with the signal mutation point feature set, accurately measure the current pulse wave head times TR1 and TR2, and calculate the current pulse wave head time difference;

[0291] A calculation module is used to segment the cable and set up test points on each segment, inject multi-feature pulse signals into the test points, and the receiving end uses the RNN-CNN architecture to identify the wavefront time, combines the signal mutation point feature set to accurately calculate the change value of the propagation time per unit distance of each test point, and fits to obtain the change value of the wavefront time per unit distance of the entire cable segment.

[0292] A positioning module is used to calculate the fault distance through the wavefront time difference of the current pulse and the change value of the wavefront time per unit distance of the entire cable segment, correct the fault distance through dynamic fuzzy logic, and combine the corrected fault distance with the cable starting end position information to determine the position of the fault point.

[0293] Embodiment 3:

[0294] A fault location device for a power cable includes: a memory on which a computer processing program is stored; a processor for executing the computer processing program in the memory to implement the system in Embodiment 2.

Claims

1. A method for fault location of power cables, characterized in that, It includes the following steps: S1: Obtain real-time cable current pulse waveform data, fault current pulse waveform data, and historical cable operation data; S2: Initially decompose the real-time cable current pulse waveform data by Haar wavelet to obtain sub-signals of different frequency segments. Analyze the sub-signals of different frequency segments using Symlets wavelet to locate the signal mutation points, and summarize the signal mutation points into a signal mutation point feature set; S3: Construct a threshold intelligent optimization model. Determine the optimal threshold through the threshold intelligent optimization model combined with the signal mutation point feature set, accurately measure the current pulse wavefront times TR1 and TR2, and calculate the current pulse wavefront time difference; S4: Segment the cable and set test points in each segment. Inject multi-feature pulse signals at the test points. The receiving end uses the RNN-CNN architecture to identify the wavefront time, accurately calculate the change value of the propagation time per unit distance at each test point in combination with the signal mutation point feature set, and fit to obtain the change value of the wavefront time per unit distance of the entire cable segment; S5: Calculate the fault distance through the current pulse wavefront time difference and the change value of the wavefront time per unit distance of the entire cable segment, and correct the fault distance through dynamic fuzzy logic. Combine the corrected fault distance with the cable starting end position information to determine the position of the fault point.

2. The fault location method of the power cable according to claim 1, characterized in that, In the above S1, the process of obtaining the real-time cable current pulse waveform data and the fault current pulse waveform data is as follows: Set the sampling frequency to Discretely sample the continuous current signal, where: ； In the formula, represents twice the highest frequency of the original signal .

3. The fault location method of the power cable according to claim 1, characterized in that, In the above S2, the process of initially decomposing the real-time cable current pulse waveform data by Haar wavelet to obtain sub-signals of different frequency segments is as follows: In the case of discrete acquisition of real-time cable current pulse waveform data, set the given discrete current pulse waveform data sequence as x , where n = 1, 2, 3,..., N, n represents the serial number of discrete time series sample points, N is the number of data points, for scale j, j = 1, 2, 3,..., J, and , J is the maximum decomposition level, represents the upper limit of the total number of subbands obtained by decomposing the signal at the maximum decomposition level J; Based on the scale j and the translation parameter k, the corresponding discrete wavelet function and the scaling function are defined as: ； ； In the formula, represents the scaling factor, represents the scale transformation factor, and respectively represent the wavelet function and the scaling function after scale and translation transformations; Set the real-time cable current pulse waveform data as ; Calculate the approximation coefficient at scale j , and the calculation method is as follows: ； Calculate the detail coefficient at scale j , and the calculation method is as follows: ； Under discrete data, for scale j and translation parameter k, the corresponding discrete Symlets wavelet function is defined as: ； wherein, represents the discrete Symlets wavelet function of the unscaled wavelet function at scale j and translation parameter k; Use Symlets wavelet to analyze the sub-signals of different frequency segments. Through discrete wavelet transform, calculate the detail coefficients. The calculation method is: ； In the formula, represents the detail coefficient of the Symlets wavelet at scale j and translation parameter k, represents the sub-signal sequence after Haar wavelet decomposition, represents the discrete Symlets wavelet function after scaling and translation.

4. The fault location method of the power cable according to claim 3, characterized in that, In the above S2, the process of using Symlets wavelet to analyze the sub-signals of different frequency segments and locate the signal mutation points is as follows: Set the detail coefficient , and its total number is M; Calculate the mean value of the detail coefficient and the standard deviation , the calculation method is as follows: ； Where \(i\) represents the index used to traverse the detail coefficients, and represent the scale \(j\) and the translation parameter \(k\) corresponding to the \(i\)-th detail coefficient, represents the detail coefficient obtained after performing a discrete wavelet transform using the Symlets wavelet under the given and ; ； Calculate the threshold T of the detail coefficients. The calculation method is: ； In the formula, represents the empirical coefficient, ; For each scale j and translation parameter k, traverse all the detail coefficients ; When is satisfied, the adjacent points are set as and , which are the detail coefficients of the Symlets wavelet at scale j and translation parameters k + 1, k - 1; When and then it is considered that is a modulus maximum point and is marked as a signal mutation point; When there is no marking The process of marking the signal mutation points is: Create a marker sequence with the same length as the original signal data , Initialize to zero. When detecting the corresponding position is a mutation point, then , represents the position of any detected mutation point. Set the signal mutation point feature set as fare, , mc represents the total number of signal mutation points, represents the position of the mc-th mutation point in the signal in the signal.

5. The fault location method for the power cable according to claim 1, characterized in that, In the above S3, the process of constructing the threshold intelligent optimization model is as follows: Set the real-time cable current pulse waveform data The set of is , where v is the total number of real-time cable current pulse waveform data points, is the current amplitude of the v-th real-time cable current pulse waveform data point, is the time corresponding to the v-th real-time cable current pulse waveform data point; Set the fault current pulse waveform data set , where , g is the total number of fault current pulse waveform data points, is the current amplitude of the g-th fault current pulse waveform data point, is the time corresponding to the g-th fault current pulse waveform data point; Merge the real-time cable current pulse waveform data set B and the fault current pulse waveform data set E into a total sample set S. The merging process is as follows: ； In the formula, represents the (v + g)-th data point in the total sample set; , represents the current amplitude of the data point represents the time of the data point. Set the current amplitude threshold as LO; If , then is marked as for indicating the wavefront; If , then is marked as , for indicating non-wavefront; In the formula, represents a flag variable used to flag whether it is recognized as a wavefront; Obtain a labeled sample set ; Pair is normalized as follows: ； Wherein, and are respectively the minimum and maximum values of the current amplitude in the entire sample set, represents the current amplitude after normalization; Construct the threshold intelligent optimization model. The construction process is as follows: Set the input feature vector as , and the threshold intelligent optimization model is as follows: ； In the formula, represents the probability value that the sample XE belongs to the wavefront, and its value range is between which represents the model parameter, and eh represents the natural constant; Train the threshold intelligent optimization model through the cross-entropy cost function to determine the model parameters , and the training process is as follows: ； In the formula, represents the value of the cost function, represents the true class information of each sample, represents the probability value that the model predicts the iq-th sample to belong to the wavefront, and log represents the logarithmic function; Use the gradient descent algorithm to minimize , and the formula for updating the parameter is as follows: ； In the formula, represents the jc-th parameter in the model parameters, represents the learning rate, represents the cost function partial derivative of with respect to the parameter; ； In the formula, represents the jc-th parameter in the eigenvector; Set the number of iterations to BH. When the cost function iterates to BH times, stop the iteration and obtain the trained model parameters ; Set the model parameters with a threshold of ; Calculation threshold Accuracy The calculation method is as follows: ； In the formula, represents the number of true positive examples at the threshold of , that is, the number of validation set samples that are actually wavefronts and are correctly judged as wavefronts, represents the number of true negative examples, that is, the number of validation set samples that are actually not wavefronts and are correctly judged as not wavefronts, represents the number of false positive examples, that is, the number of validation set samples that are actually not wavefronts but are misjudged as wavefronts, is the number of false negative examples, that is, the number of validation set samples that are actually wavefronts but are misjudged as not wavefronts; Calculation threshold Recall rate The calculation method is as follows: ； Calculation threshold Precision , and the calculation method is as follows: ； Calculation threshold F1 score , and the calculation method is as follows: ； Set the threshold of the F1 score as FDC; When , it is determined that is the optimal threshold ; When , it is not determined that is the optimal threshold ; And record the time points when the current pulse waveform data changes from non-wavefront to wavefront, denoted as TR1 and TR2 respectively; Calculate the time difference of the current pulse wavefront , and the calculation method is as follows: 。 6. The fault location method for the power cable according to claim 5, characterized in that, In the above S4, the process of injecting multi-feature pulse signals at the test points and the receiving end using the RNN-CNN architecture to identify the wavefront time is as follows: Set the pulse signal sequence input to the RNN-CNN architecture as , denote the feature vector at time t; Update the hidden state of the RNN layer. The update process is as follows: ； Wherein, represents the hidden state of the RNN layer at time t, represents the weight matrix input to the hidden layer, represents the weight matrix from the hidden layer to the hidden layer, represents the bias vector of the hidden layer, represents the hyperbolic tangent activation function, represents the hidden state of the RNN layer at time t-1; It is set that the CNN layer uses kf convolutional kernels, and the convolutional kernels are , where mf is the size in the time dimension and nf is the size in the feature dimension. For the ie-th convolutional kernel, its convolution operation formula is: ； In the formula, represents the eigenvalue obtained at time t after the convolution operation by the ie-th convolutional kernel, represents the weight of the ie-th convolutional kernel at position ; represents the bias vector of the ie-th convolutional kernel, represents the hidden state of the RNN layer at time t at position ; A series of feature maps are obtained after the convolution operation , denotes the kj-th feature map; Then perform a pooling operation, set the pooling window size, and the feature map after pooling , represents the km-th feature map, where: ； In the formula, represents the eigenvalue of the km-th feature map after the pooling operation at the position . represents the eigenvalue of the kj-th feature map after the pooling operation at the position . represents the region corresponding to the ja-th pooling window; Flatten the pooled feature map to obtain a one-dimensional vector FBB, and then map it through a fully connected layer. Assume the fully connected layer has LBB neurons. Its calculation formula is: ； In the formula, represents the output of the jb-th neuron after the fully connected layer passes through the activation function, represents the activation function, represents the weight of the fully connected layer, NF represents the length of the one-dimensional vector FBB, represents the bias vector, represents the lb-th neuron in the one-dimensional vector FBB; Determine the position information of the wavefront time through an output layer. Set the output as yhz. The calculation method is: ； In the formula, represents the weight of the output layer, represents the bias vector of the output layer; Set the range of XVB from 0 to t, and yhz falls into and calculate the wavefront time at the iyec-th test point through linear mapping , and the calculation process is as follows: ； In the formula, MLE represents the maximum value of yhz, which is used for normalization processing.

7. The fault location method of the power cable according to claim 6, wherein, In S4, the change value of the propagation time per unit distance of each test point is accurately calculated in combination with the mutation point feature set, and the wavefront time transformation value per unit distance of the entire cable is obtained by fitting. The fitting process is as follows: Calculate the change value of the propagation time per unit distance at each test point , and the calculation method is as follows: ； In the formula, represents the distance between the (iyec - 1)-th test point and the iyec-th test point, represents the wavefront time at the (iyec - 1)-th test point; The wavefront time transformation value per unit distance of the entire cable is obtained by fitting. , and the fitting process is as follows: ； In the formula, ikkq represents the index variable, and nw represents the total number of test points.

8. The fault location method of the power cable according to claim 7, characterized in that In S5, the fault distance is calculated through the time difference of the current pulse wavefront and the wavefront time change value per unit distance of the entire cable. The calculation method is: ； In the formula, LJNT represents the fault distance; And the fault distance is corrected by dynamic fuzzy logic. The correction process is as follows: ； In the formula, represents the revised fault distance, , and are the membership functions of cable temperature, humidity and aging degree respectively, , and are the corresponding , and weight coefficients.

9. The fault location method of the power cable according to claim 8, characterized in that, In S5, the corrected fault distance is combined with the position information of the cable starting end to determine the position of the fault point. The determination process is as follows: Set the starting position coordinates of the cable as , and calculate the position coordinates of the fault point . The calculation method is as follows: ； Set the actual laying length of the cable as LCDP; If , determine that the location coordinates of the fault point are accurate; If the position coordinates of the fault point are determined to be inaccurate, recalculate through S1 to S4 until it reaches 10. A fault location system for a power cable, which is used to implement the fault location method for the power cable according to any one of claims 1-9, characterized in that, Including: An acquisition module, an analysis module, a construction module, a calculation module, and a positioning module; The acquisition module is used to acquire real-time cable current pulse waveform data, fault current pulse waveform data, and historical cable operation data; The analysis module is used to preliminarily decompose the real-time cable current pulse waveform data by Haar wavelet to obtain sub-signals of different frequency bands, analyze the sub-signals of different frequency bands by Symlets wavelet and locate the signal mutation points, and summarize the signal mutation points into a signal mutation point feature set; The construction module is used to construct a threshold intelligent optimization model, determine the optimal threshold in combination with the signal mutation point feature set through the threshold intelligent optimization model, accurately measure the current pulse wavefront times TR1 and TR2, and calculate the current pulse wavefront time difference; The calculation module is used to segment the cable and set test points in each segment, inject multi-feature pulse signals into the test points, use the RNN-CNN architecture at the receiving end to identify the wavefront time, accurately calculate the change value of the propagation time per unit distance of each test point in combination with the signal mutation point feature set, and obtain the wavefront time transformation value per unit distance of the entire cable by fitting; The positioning module is used to calculate the fault distance through the current pulse wavefront time difference and the wavefront time change value per unit distance of the entire cable, correct the fault distance by dynamic fuzzy logic, and combine the corrected fault distance with the position information of the cable starting end to determine the position of the fault point.

Images