A method and system for locating building cable faults

Abstract

The present application provides a method and system for locating building cable faults, which relates to the technical field of fault location. First, the present application applies a detection signal to the target cable and obtains parameters such as feedback and transmission speed, and collects training data of different cables. The data processing stage includes calculating the product of the time difference and the transmission speed to obtain the initial distance, establishing and training a prediction model to predict the distance. When locating a fault, the initial distance is compared with the predicted distance. If the difference is less than a preset threshold, the fault location is output according to the distance information; otherwise, the data acquisition step is repeated. The present application can use the training data to train the prediction model, so that the prediction model can better adapt to different cable types and conditions, thereby improving the accuracy of predicting the distance. The present application can better adapt to complex and changeable actual situations than traditional methods.

Description

Technical Field

[0001] This application relates to the technical field of fault location, and particularly to a method and system for fault location of building cables. Background Art

[0002] In the fields of construction and infrastructure, cables, as key components for power transmission and signal communication, their reliability and stability are crucial for the normal operation of various systems. However, due to various factors such as environmental factors, construction quality, material aging, etc., cables may experience various faults during use, such as short circuits, open circuits, poor grounding, etc. These faults not only cause power or signal transmission interruptions but may also lead to serious consequences such as fires and equipment damage.

[0003] Traditional cable fault location methods usually rely on manual inspections and empirical judgments. This method is time-consuming and laborious, and the location accuracy is often limited by the experience and skill levels of the inspection personnel. Moreover, with the expansion of the cable network scale and the increase in complexity, traditional methods have been difficult to meet the requirements of quickly and accurately locating faults. Summary of the Invention

[0004] In order to improve the accuracy of building cable fault location, this application provides a method and system for building cable fault location.

[0005] In a first aspect, this application provides a method for building cable fault location, adopting the following technical solution:

[0006] A method for building cable fault location includes the following steps:

[0007] Data acquisition: including first acquisition and second acquisition;

[0008] First acquisition: Apply a detection signal to the target cable, obtain a first feedback signal and the transmission speed of the detection signal in the target cable, record the transmission speed as the first data; collect the physical parameters of the target cable;

[0009] Second acquisition: Collect training data, where the training data includes: the physical parameters of various cables, the second feedback signals after applying detection signals to various cables, and the distances corresponding to the second feedback signals;

[0010] Data processing: including first calculation, first modeling, and first prediction;

[0011] First calculation: Obtain the time difference between applying a detection signal to the target cable and receiving the first feedback signal, record it as the second data, perform a multiplication operation on the first data and the second data to obtain an initial distance;

[0012] First Modeling: Establish a prediction model, and train the prediction model using training data to obtain a trained prediction model;

[0013] First Prediction: Input the first feedback signal and the physical parameters of the target cable into the trained prediction model to obtain a predicted distance;

[0014] Fault Location: Includes a first judgment and an output;

[0015] First Judgment: Judge whether the difference between the initial distance and the predicted distance is less than a preset difference threshold. If so, execute the output step; if not, execute the data acquisition step;

[0016] Output: Determine the fault location based on the initial distance and / or the predicted distance, and output the fault location.

[0017] In this application, by applying a detection signal to the target cable and obtaining a feedback signal, the transmission speed and physical parameters of the cable can be understood in real time, and relevant data can be directly obtained for the target cable, providing accurate information about the current state of the target cable. Subsequently, the physical parameters, feedback signals, and their corresponding fault distances of multiple cables are collected, thereby covering more cable types and fault conditions and improving the generalization ability of the prediction model. Then, the training data is used to train the prediction model, enabling the prediction model to learn the complex relationship between the cable physical parameters, feedback signals, and distance. Then, by calculating the product of the time difference (second data) and the transmission speed, an initial distance can be quickly obtained. The predicted distance obtained by the prediction model in this application, together with the initial distance, constitutes a double verification of fault location, improving the reliability of location. Then, by comparing whether the difference between the initial distance and the predicted distance is less than the preset difference threshold, it can be judged whether the currently collected data is accurate and reliable enough for fault location, thereby screening out abnormal data that may be caused by factors such as noise and interference. If the difference is less than the preset difference threshold, the fault location is determined based on the initial distance and / or the predicted distance and output to relevant personnel or systems. Otherwise, it proves that there is abnormal data and data acquisition needs to be performed again.

[0018] Optionally, after executing the data acquisition step and before executing the data processing step, it further includes:

[0019] Feature Extraction: Perform spectral analysis on the second feedback signal using the Fourier algorithm to obtain an analysis result, and extract the features of the second feedback signal based on the analysis result, denoted as the first feature;

[0020] Construct a Graph Structure: Use the cables as nodes and the connection relationships between the cables as edges to construct a graph structure, and the attributes of the nodes include the physical parameters of the cables and the first feature;

[0021] First calculation: Calculate the Laplacian matrix of the graph structure, perform eigenvalue decomposition on the Laplacian matrix to obtain eigenvalues and eigenvectors;

[0022] First acquisition: Acquire the eigenvector corresponding to the largest eigenvalue, denoted as the first vector;

[0023] First mapping: Perform a multiplication operation on the matrix composed of the first features and the first vector to obtain an operation result, and use the operation result as the new second feedback signal.

[0024] This application uses the Fourier algorithm to perform spectral analysis on the second feedback signal, and extracts features (i.e., the first features) from the analysis results. The first features can reflect key information about the cable state, such as fault type, location, or severity, etc. A graph structure constructed with cables as nodes and the connection relationships between cables as edges can intuitively represent the topological structure of the cable network. Then, a Laplacian matrix is established, and eigenvalue decomposition is performed on the Laplacian matrix to obtain eigenvalues and eigenvectors, thereby revealing the internal characteristics and structural patterns of the graph structure. The eigenvector with the largest eigenvalue (the first vector) reflects the most significant or important pattern in the graph structure. Performing a multiplication operation on the matrix composed of the first features and the first vector can obtain a new signal (the new second feedback signal). This new signal integrates the feature information of the original signal and the internal characteristics of the graph structure, providing richer and more accurate information for subsequent data processing. This application realizes the conversion from the original signal (referring to the original second feedback signal) to the enhanced signal (referring to the new second feedback signal). In this process, information is effectively integrated and enhanced, providing a more accurate and reliable information basis for subsequent data processing and fault location. And advanced algorithms such as graph structure and eigenvalue decomposition are introduced, which can more accurately identify and analyze fault patterns in the cable, thereby improving the accuracy and reliability of fault location.

[0025] Optionally, after performing the step of the first acquisition and before performing the step of the first mapping, it further includes:

[0026] First convolution: Integrate all the first features into a second vector, perform a convolution operation on the second vector with a preset first filter to obtain a filtered second vector, denoted as the third vector;

[0027] Second calculation: Perform a non-linear transformation on the third vector through the ReLU function, and use the non-linearly transformed third vector as the new first vector.

[0028] This application integrates all the first features into a second vector, achieving the conversion from multiple independent features to a unified vector. Then, the second vector is subjected to a convolution operation with a preset first filter to extract local feature information in the second vector. The filtered second vector (denoted as the third vector) obtained through the convolution operation contains the feature information processed by the filter, and these feature information are more focused on specific patterns or structures. The third vector is subjected to a non-linear transformation through the ReLU (Rectified Linear Unit) function. The third vector after being processed by the ReLU function (as the new first vector) not only retains the original feature information but also introduces non-linear factors, enabling subsequent steps to capture more complex feature relationships. The first convolution step extracts local feature information in the second vector through the convolution operation, enhancing the feature extraction ability. The second calculation step realizes the non-linear transformation processing of the third vector by introducing the ReLU function. The new first vector after the first convolution and the second calculation processing contains richer and more valuable feature information, which helps to improve the accuracy and reliability of fault location.

[0029] Optionally, after performing the step of constructing the graph structure and before performing the step of the first calculation, it further includes:

[0030] Third calculation: Based on the attributes of each node, calculate the correlation degree between each node using the Pearson correlation coefficient;

[0031] Fourth calculation: Perform a non-linear change on the correlation degree using the ReLU function to obtain the third data;

[0032] Set weights: Use the third data as the weights of the corresponding edges in the graph structure.

[0033] This application calculates the correlation degree between each node using the Pearson correlation coefficient based on the attributes of each node, thereby quantifying the correlation between nodes in the graph structure. Then, the ReLU function is used to perform a non-linear change on the correlation degree, mapping the correlation degree between each node to the non-negative interval and introducing non-linear factors to strengthen the edges with positive correlation and eliminate the edges with negative correlation. Strengthening the edges with positive correlation helps to highlight those node pairs with close connections in the graph structure. And eliminating the edges with negative correlation is to remove those connections that may cause interference or noise in the graph structure. Then, the third data is used as the weights of the corresponding edges in the graph structure, integrating the correlation degree information after non-linear change into the graph structure, so that the edges in the graph structure not only represent the connection relationship between nodes but also represent the correlation degree between nodes. By setting weights, the information in the graph structure is further enriched and refined. The step of setting weights integrates the correlation degree information after non-linear change into the graph structure, making the information in the graph structure richer and more complete.

[0034] Optionally, after the step of performing the second calculation and before the step of performing the first mapping, it further includes:

[0035] Third acquisition: Acquire the electrical parameters of various cables and integrate all the electrical parameters into a fourth vector;

[0036] Second judgment: Judge whether the dimension of the fourth vector is equal to the dimension of the new first vector. If so, perform the step of second convolution; if not, perform the step of unifying dimensions;

[0037] Unifying dimensions: Use the principal component analysis algorithm to unify the dimensions of the fourth vector and the new first vector;

[0038] Second convolution: Perform a convolution operation on the fourth vector and a preset second filter to obtain a filtered fourth vector, denoted as the fifth vector;

[0039] Fifth calculation: Perform a non-linear transformation on the fifth vector through the ReLU function, and use the non-linearly transformed fifth vector as the new fourth vector;

[0040] Second mapping: Use the electrical parameters corresponding to the new fourth vector as the new electrical parameters.

[0041] In this application, by acquiring the electrical parameters of the cables and integrating them into a fourth vector, it provides basic data for subsequent processing. Then, by judging whether the dimension of the fourth vector is equal to the dimension of the new first vector, the number of features of the training data and the second feedback signal in the graph structure is made consistent with the number of electrical parameters. If the dimensions do not match, it is necessary to perform dimension unification processing to reduce calculation errors or data loss. When the dimensions of the fourth vector and the new first vector do not match, the principal component analysis (PCA) algorithm is used for dimension unification. The PCA algorithm can effectively reduce the dimension of the data while retaining the main features of the original data as much as possible. Then, perform a convolution operation on the fourth vector and a preset second filter to obtain a filtered fourth vector (i.e., the fifth vector). Then, perform a non-linear transformation on the fifth vector through the ReLU (Rectified Linear Unit) function to obtain a new fourth vector. Then, use the electrical parameters corresponding to the new fourth vector as the new electrical parameters.

[0042] Optionally, the attributes of the training data and the graph structure further include: the electrical parameters of various cables.

[0043] By adopting the above technical solution, the electrical parameters incorporated into the cable can significantly increase the dimension and richness of the training data. These electrical parameters include voltage, current, resistance, capacitance, etc., which directly reflect the physical characteristics and working status of the cable. Abundant training data helps the model learn more refined and accurate feature representations, thereby improving the accuracy of prediction or classification. Especially when dealing with tasks such as cable fault prediction and performance evaluation, the electrical parameters can provide key information. In the graph structure, nodes represent cables, while edges represent the connection relationships and similarities between them. Taking the electrical parameters as the attributes of the graph structure can further enhance the information content of nodes and edges, helping the graph structure better understand the internal connections and differences between cables. Especially when performing tasks such as fault propagation analysis and network optimization, the training data containing electrical parameters enables the model to learn a wider range of cable characteristics, thereby enhancing its generalization ability. This means that the model can still make reasonable predictions or judgments when faced with unseen cable types or fault patterns.

[0044] Optionally, after the step of performing the first judgment and before the step of performing data collection, it further includes:

[0045] Sixth calculation: Calculate the Euclidean distance between the first feedback signal and each second feedback signal, denoted as the fourth data;

[0046] Third judgment: Judge whether the fourth data is less than a preset value. If so, perform the step of the first prediction; if not, perform the step of model expansion;

[0047] Model expansion: Add a new hidden layer to the trained prediction model to obtain an expanded prediction model;

[0048] Retraining: Fix the original hidden layers in the expanded prediction model, and use the physical parameters of the target cable, the first feedback signal, and the initial distance to train the new hidden layer in the expanded prediction model to obtain a retrained prediction model;

[0049] Model update: Replace the trained prediction model with the retrained prediction model to perform the step of the first prediction.

[0050] In this application, the Euclidean distance between the first feedback signal and each second feedback signal is calculated to obtain the fourth data, and the degree of difference between the first feedback signal and the second feedback signal is judged based on the fourth data. The smaller the Euclidean distance, the closer the first feedback signal and the second feedback signal are. If the fourth data is greater than the preset value, it indicates that the performance of the prediction model may not be ideal enough, and it is necessary to make it have higher adaptability through incremental learning so that the prediction model always maintains a high performance level. Then, a new hidden layer is added to the prediction model, aiming to improve the complexity and learning ability of the model. After that, the original hidden layer is fixed, and only the newly added hidden layer is trained, which can not only reduce the risk of overfitting but also improve the speed of retraining. Then, by using the physical parameters of the target cable, the first feedback signal, and the initial distance for training, the new hidden layer can better adapt to the new task. Replace the original prediction model with the retrained prediction model, thereby improving the prediction accuracy and the generalization ability of the prediction model.

[0051] Optionally, after performing the step of model expansion and before performing the step of retraining, it further includes:

[0052] Establish connections: Establish connection relationships between each new hidden layer and each original hidden layer respectively;

[0053] Seventh calculation: Calculate the correlation degree between the first feedback signal and each second feedback signal respectively by using the Pearson correlation coefficient algorithm;

[0054] Fourth judgment: Judge whether all the correlation degrees are greater than the preset correlation degree threshold. If so, perform the step of retraining; if not, perform the pruning step;

[0055] Pruning: Remove the connection relationships between the original hidden layer and the new hidden layer corresponding to the correlation degree threshold lower than the preset correlation degree threshold.

[0056] In this application, by establishing the connection relationships between each new hidden layer and each original hidden layer in the retrained prediction model, the complexity and learning ability of the model are enhanced. This connection allows information to flow between different parts of the model, which helps the prediction model better capture the complex features in the data. Then, the Pearson correlation coefficient algorithm is used to calculate the correlation degree between the first feedback signal and each second feedback signal, and based on the comparison between the correlation degree and the preset correlation degree threshold, it is decided whether to perform the retraining or pruning step, which helps the model maintain complexity while not overfitting the data or ignoring important features. The pruned model usually has lower computational complexity and faster inference speed, and may maintain or improve the prediction accuracy at the same time.

[0057] Optionally, after performing the pruning step and before performing the retraining step, it further includes:

[0058] Fifth judgment: Determine whether there is a connection relationship between the i-th new hidden layer and the remaining hidden layers. If so, perform the iterative steps; if not, perform the deletion steps.

[0059] Iteration: Take the (i + 1)-th new hidden layer as the new i-th hidden layer, and perform the steps of the fifth judgment until the preset stop condition is met.

[0060] Deletion: Delete the i-th new hidden layer, and perform the iterative steps.

[0061] This application checks whether there is a connection relationship between each new hidden layer and the remaining hidden layers in the prediction model after pruning. If there is a connection relationship between the i-th new hidden layer and the remaining hidden layers, the iterative step takes the (i + 1)-th new hidden layer as the new object to be checked and repeats the fifth judgment. This process continues until the preset stop condition is met. If there is no connection relationship between the i-th new hidden layer and the remaining hidden layers, the deletion steps are performed. Deleting these isolated new hidden layers helps to simplify the model structure, reduce unnecessary computational overhead, and may improve the generalization ability of the model.

[0062] In a second aspect, this application provides a building cable fault location system, adopting the following technical solution:

[0063] A building cable fault location system includes:

[0064] A processor and a memory,

[0065] Program code is stored in the memory;

[0066] When the processor calls the program code in the memory, it executes the steps of the method.

[0067] In summary, this application includes at least one of the following beneficial technical effects:

[0068] 1. The predicted distance obtained by this application through the prediction model, together with the initial distance, constitutes a double verification for fault location, improving the reliability of location. Then, by comparing whether the difference between the initial distance and the predicted distance is less than the preset difference threshold, it can be determined whether the currently collected data is accurate and reliable enough for fault location, so as to screen out abnormal data that may be caused by factors such as noise and interference. If the difference is less than the preset difference threshold, the fault location is determined according to the initial distance and / or the predicted distance and output to relevant personnel or systems. Otherwise, it proves that there is abnormal data and data collection needs to be performed again.

[0069] 2. Based on the attributes of each node, this application calculates the correlation degree between each node using the Pearson correlation coefficient, thereby quantifying the correlation between nodes in the graph structure. Then, the ReLU function is used to perform a non-linear transformation on the correlation degree, mapping the correlation degree between each node to the non-negative interval, and introducing non-linear factors to strengthen the edges with positive correlation and eliminate the edges with negative correlation. Strengthening the edges with positive correlation helps to highlight those node pairs with close connections in the graph structure. And eliminating the edges with negative correlation is to remove those connections that may cause interference or noise in the graph structure. Then, the third data is used as the weight of the corresponding edge in the graph structure, and the correlation degree information after non-linear transformation is incorporated into the graph structure, so that the edges in the graph structure not only represent the connection relationship between nodes, but also represent the degree of correlation between nodes. By setting the weights, the information in the graph structure is further enriched and refined. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 is the flowchart of Embodiment 1 of this application;

[0071] Figure 2 is the flowchart from S41 Feature Extraction to S47 First Mapping of Embodiment 2 of this application;

[0072] Figure 3 is the flowchart from S50 Third Calculation to S58 Second Mapping of Embodiment 2 of this application;

[0073] Figure 4 is the flowchart of Embodiment 3 of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0074] The following is combined with Figures 1 to 4 to further elaborate on this application in detail.

[0075] Embodiment 1: This embodiment discloses a method for locating building cable faults. Referring to Figure 1 , the method includes: S1 Data Acquisition, S2 Data Processing, and S3 Fault Location. First, various data of the target cable and training data are collected. Then, the initial distance of the fault point is calculated using the various data of the target cable. Then, a prediction model is established, and the prediction model is trained using the training data. Then, the various data of the target cable are input into the trained prediction model to obtain the predicted distance. By comparing the initial distance and the predicted distance, it is determined whether the fault location can be determined based on the initial distance and / or the predicted distance, and the fault location is output. The process of this embodiment is as follows:

[0076] S1 Data Acquisition includes S11 First Acquisition and S12 Second Acquisition.

[0077] S11 First acquisition: Apply a detection signal to the target cable, obtain the first feedback signal and the transmission speed of the detection signal in the target cable, and record the transmission speed as the first data. Collect the physical parameters of the target cable, such as length, diameter, conductor material, insulation material, etc., through the stored data information.

[0078] Apply a detection signal to the target cable. The detection signal can be a pulse signal, a continuous wave signal, or other forms of signals. When the detection signal propagates in the cable, it will encounter various impedance changes (i.e., fault points), resulting in reflection or scattering and returning the first feedback signal.

[0079] Collect the transmission speed of the detection signal in the target cable. The collection methods can include many types. Taking the frequency domain analysis method and the direct measurement method as examples, the process is as follows:

[0080] Frequency domain analysis method: Send a series of signals with different frequencies to the target cable and analyze the propagation characteristics of these signals in the cable. By measuring parameters such as the phase delay or group delay of the signal, the transmission speed of the signal can be deduced.

[0081] Direct measurement method: It is possible to measure the time required for a signal to propagate on a cable with a known length and divide this length by the time to obtain an approximate value of the transmission speed.

[0082] S12 Second acquisition: Obtain training data by calling historical data. The training data includes: the physical parameters of various cables, the second feedback signals of various cables after applying the detection signal, and the distances corresponding to the second feedback signals.

[0083] In the historical data, the second feedback signals of various cables applied with different detection signals and the distances corresponding to the second feedback signals are recorded. These second feedback signals and the distances corresponding to the second feedback signals are obtained through data calling.

[0084] The physical parameters of various cables are stored in the relevant database during the construction stage / planning stage and can be obtained through calling.

[0085] S2 Data processing, including S21 First calculation, S22 First modeling, and S23 First prediction.

[0086] S21 First calculation: Obtain the timestamp when the detection signal is applied to the target cable, denoted as the first timestamp, and obtain the timestamp when the first feedback signal is received, denoted as the second timestamp.

[0087] Calculate the absolute value of the difference between the first timestamp and the second timestamp, denoted as the second data.

[0088] Perform a multiplication operation on the first data and the second data, and record the operation result as the initial distance.

[0089] S22 First modeling: Establish a prediction model and train the prediction model using training data. During the training process, the physical parameters of various cables and the second feedback signals after applying detection signals to various cables are used as the data during training, and the distance corresponding to the second feedback signal is used as the true label of the data during training. After that, the trained prediction model is obtained.

[0090] The prediction model can be any one of neural network models such as CNN model, RNN model, LSTM model, BI-LSTM model, etc.

[0091] S23 First prediction: Input the first feedback signal and the physical parameters of the target cable into the trained prediction model, and output the predicted label corresponding to the first feedback signal, that is, the predicted distance.

[0092] S3 Fault location, including S31 First judgment and S32 Output.

[0093] S31 First judgment: Judge whether the difference between the initial distance and the predicted distance is less than the preset difference threshold. If so, it means that the prediction result is reliable, and then execute S32 Output; if not, it means that the prediction result is unreliable, and there may be noise or incorrect data, and it is necessary to execute S1 Data acquisition to re-collect data.

[0094] S32 Output: Define the position with a length of the initial distance and / or the predicted distance from the end of the target cable where the detection signal is input as the fault position, and output the fault position.

[0095] In this embodiment, through the S1 Data acquisition stage, the physical parameters, transmission speed, and historical feedback signal data of the cable are accurately obtained. Combining the accurate time difference calculation and advanced prediction model in the S2 Data processing stage for distance prediction improves the accuracy and reliability of the fault location result. At the same time, the dual verification mechanism and data re-sampling process improve the robustness of the fault location and enhance the overall processing efficiency.

[0096] Embodiment 2: Refer to Figure 2 , the difference between this embodiment and Embodiment 1 is that after executing S1 Data acquisition and before executing S2 Data processing, it further includes:

[0097] S41 Feature extraction: By performing spectral analysis on the second feedback signal through the Fourier algorithm, the representation of the second feedback signal in the frequency domain, that is, the spectrum, can be obtained. The spectrum shows the intensity or amplitude of the second feedback signal at different frequencies.

[0098] Extract the features of the second feedback signal based on the spectrum, denoted as the first features. These first features can be certain statistics of the spectrum (such as mean, variance, maximum value, minimum value, etc.), or certain specific frequency components of the spectrum (such as fundamental frequency, harmonics, etc.). In this embodiment, the form of the first features is not specifically limited. As long as the combination of the first features contained in the second feedback signal forms a vector, this vector can uniquely represent the second feedback signal.

[0099] S42 Construct a graph structure with cables as nodes and the connection relationships between cables as edges. The attributes of the nodes include the physical parameters of the cables and the first features.

[0100] If there is a connection relationship between two nodes, there is an undirected edge between the two nodes; otherwise, there is no undirected edge between the two nodes.

[0101] S43 First calculation: Calculate the Laplacian matrix of the graph structure, perform eigenvalue decomposition on the Laplacian matrix to obtain eigenvalues and eigenvectors.

[0102] The Laplacian matrix is a special form of symmetric matrix composed of the adjacency matrix and the degree matrix of the graph. It reflects the structural characteristics of the graph. The calculation process of the Laplacian matrix is as follows:

[0103] 1. Construct the adjacency matrix and the degree matrix:

[0104] The adjacency matrix represents the connection relationships between nodes in the graph. If there is an edge connecting node i and node j, the value at the corresponding position in the adjacency matrix is 1 (for an unweighted graph); otherwise, it is 0.

[0105] The degree matrix is a diagonal matrix, and the elements on the diagonal represent the degrees of the corresponding nodes (i.e., the number of edges connected to the nodes).

[0106] 2. Calculate the Laplacian matrix: The Laplacian matrix L is defined as the degree matrix D minus the adjacency matrix A, that is, L = D - A. This matrix reflects the structural characteristics of the graph, which contains the connection relationships between nodes and the degree information of nodes.

[0107] 3. Perform eigenvalue decomposition: Perform eigenvalue decomposition on the Laplacian matrix L, that is, decompose it into the product of the eigenvector matrix and the eigenvalue matrix. The process of eigenvalue decomposition can be regarded as the process of diagonalizing the Laplacian matrix. The elements on the diagonal matrix are the eigenvalues of the Laplacian matrix. After completing the eigenvalue decomposition of the Laplacian matrix, a set of eigenvalues and corresponding eigenvectors can be obtained.

[0108] S44 First acquisition: Obtain the eigenvector corresponding to the maximum eigenvalue, denoted as the first vector.

[0109] S45 First Convolution: Integrate the first features of the same second feedback signal into a second vector, perform a convolution operation on the second vector with a preset first filter to obtain a filtered second vector, denoted as the third vector.

[0110] During the convolution process, a feature map or output is generated on the input data (i.e., the second vector) through the first filter (or called the convolution kernel).

[0111] In this process, each element of the first filter will perform a weighted sum with the corresponding element of the second vector to generate an element of the output data. This weighted sum process is actually extracting features from the second vector, and the design of the first filter determines the type of features to be extracted.

[0112] The result of the convolution operation is a new vector, that is, the filtered second vector. Denote the filtered second vector as the third vector. The third vector contains the feature information of the input data (i.e., the second vector) after being processed by the filter.

[0113] S46 Second Calculation: Perform a non - linear transformation on the third vector through the ReLU function, and use the non - linearly transformed third vector as the new first vector.

[0114] S47 First Mapping: Multiply the matrix composed of the first features by the new first vector in the second calculation of S46 to obtain an operation result, and use the operation result as the new second feedback signal.

[0115] In this embodiment, Fourier spectrum analysis is used to extract the first features of the second feedback signal, a graph structure with cables as nodes is constructed and its attributes are assigned, the eigenvectors are obtained by calculating the eigen - decomposition of the Laplacian matrix, the first features are integrated into a vector for convolution operation to obtain a filtered vector, then through ReLU non - linear transformation, and finally multiplied by the feature matrix to obtain a new second feedback signal. This embodiment comprehensively applies techniques such as Fourier spectrum analysis, graph structure construction, Laplacian matrix eigen - decomposition, convolution operation, and ReLU non - linear transformation, realizing in - depth mining and characterization enhancement of the features of the second feedback signal. By constructing a graph structure with cables as nodes and integrating physical parameters and feature information, further refining key features through convolution and non - linear transformation, and finally generating a new second feedback signal with higher characterization ability and application value.

[0116] Refer to Figure 3 , in other embodiments, the attributes of the training data and the graph structure further include: the electrical parameters of various cables. After performing S42 to construct the graph structure and before performing S43 first calculation, it further includes:

[0117] S50 The third calculation uses the Pearson correlation coefficient algorithm to calculate the correlation between nodes based on the attributes of each node in the graph structure (these attributes include the physical parameters of the cable and the first feature).

[0118] S51 is a fourth calculation, in which a ReLU function is used to perform a nonlinear change on the correlation described in the third calculation of S51 to obtain third data.

[0119] S52 sets the weight, and uses the third data as the weight of the edge in the graph structure. In this embodiment, the nonlinearly transformed correlation is used as the weight of the edge in the graph structure, which reflects the strength and importance of the connection between nodes.

[0120] Then, the first calculation in S43 to the second calculation in S46 are performed. After the second calculation in S46 is performed and before the first mapping in S47 is performed, the method further includes:

[0121] S53 is a third collection, collecting electrical parameters of various cables and integrating all electrical parameters into a fourth vector, wherein the electrical parameters include: voltage, current, resistance, capacitance, inductance, etc.

[0122] S54 is a second judgment, judging whether the dimension of the fourth vector is equal to the dimension of the new first vector. If so, executing S56 a second convolution; if not, executing S55 a unified dimension.

[0123] S55 unifies the dimensions and uses the principal component analysis algorithm to unify the dimensions of the fourth vector and the new first vector.

[0124] Principal component analysis (PCA) is a commonly used dimensionality reduction technique that can reduce the dimension of data by retaining the main components (i.e. the most important features) in the data while retaining as much information as possible from the original data. In this step, PCA is used to adjust the dimensions of the fourth vector and the new first vector so that they have the same dimensions.

[0125] S56 second convolution, performing a convolution operation on the fourth vector and a preset second filter to obtain a filtered fourth vector, which is recorded as a fifth vector.

[0126] S57 is a fifth calculation, in which a nonlinear transformation is performed on the fifth vector by using a ReLU function, and the fifth vector after the nonlinear transformation is used as a new fourth vector.

[0127] S58 second mapping, taking the electrical parameters corresponding to the new fourth vector as new electrical parameters, and updating the new electrical parameters to the attributes of the training data and the graph structure.

[0128] In this embodiment, the Pearson correlation coefficient algorithm is used to calculate the correlation degree between nodes, which is set as the graph edge weight after ReLU non-linear transformation. The electrical parameters of the cable are collected and integrated into the fourth vector. According to dimension matching, direct convolution or PCA is selected to unify the dimensions, and then the electrical parameters are updated through convolution and non-linear transformation and integrated into the training data and the graph structure to optimize data processing and subsequent analysis.

[0129] Embodiment 3: Refer to Figure 4 , the difference between this embodiment and Embodiment 1 is that after performing the first judgment in S31 and before performing data collection in S1, it further includes:

[0130] S60 Sixth calculation, encoding the first feedback signal and the second feedback signal respectively, and calculating the Euclidean distance between the encoded first feedback signal and each encoded second feedback signal, denoted as the fourth data.

[0131] S61 Third judgment, judging whether the fourth data is less than a preset value. If so, it indicates that the first feedback signal and the current second feedback signal have a high similarity, and then S32 first prediction is executed; if not, S62 model expansion is executed.

[0132] S62 Model expansion, adding a new hidden layer to the trained prediction model to obtain an expanded prediction model. By adding hidden layers, the model can learn more complex and abstract feature representations, thereby improving its prediction ability.

[0133] S63 Establish connections. For each newly added hidden layer node, establish its connection with each node in all previous hidden layers. By adding connections, the model can capture more feature interactions, which helps to improve the generalization ability and accuracy of the model.

[0134] S64 Seventh calculation, using the Pearson correlation coefficient algorithm to calculate the correlation degree between the first feedback signal and each second feedback signal respectively. By calculating the correlation degree, the contribution degree of different hidden layers to the final output and their mutual influence can be evaluated, which helps to identify which hidden layers are crucial for the model performance and which may be redundant.

[0135] S65 Fourth judgment, judging whether all the correlation degrees are greater than a preset correlation degree threshold. If so, it means that the current network structure is effective and the interaction between hidden layers is positive, and then S71 model update is executed; if not, it indicates that some connections do not contribute significantly to the model performance and may even introduce noise, and then S66 pruning is executed.

[0136] S66 Pruning, removing the connection relationship between the original hidden layer and the new hidden layer corresponding to the correlation degree threshold lower than the preset correlation degree threshold.

[0137] S67 Fifth judgment: Determine whether there is a connection relationship between the i-th new hidden layer and the remaining hidden layers in the expanded prediction model after pruning in S66. If so, execute S68 iteration; if not, it indicates that the i-th hidden layer has no special effect and S69 deletion needs to be executed.

[0138] S68 Iteration: Take the (i + 1)-th new hidden layer as the new i-th hidden layer and execute S67 fifth judgment until the preset stop condition is met. The preset stop condition includes: the performance of the prediction model has met the requirements or all hidden layers have been processed.

[0139] S69 Deletion: Delete the i-th new hidden layer and execute S68 iteration. After all hidden layers have been processed, execute S70 retraining.

[0140] S70 Retraining: Fix the original hidden layers in the expanded prediction model and train the new hidden layers in the expanded prediction model using the physical parameters of the target cable, the first feedback signal, and the initial distance.

[0141] During the training process, take the physical parameters of the target cable and the first feedback signal as the training data, take the initial distance as the true label of the training data, and take the training data and its true label as the input data of the expanded prediction model. These input data are used to train the newly added hidden layers so that they can learn more complex and accurate feature representations and obtain the retrained prediction model.

[0142] S71 Model update: Replace the trained prediction model with the retrained prediction model and execute S32 first prediction.

[0143] In this embodiment, the Euclidean distance and correlation degree are calculated to quantify the difference between the model output and the actual observation value and the interaction between hidden layers, and then decide whether to perform model expansion and pruning operations. By adding hidden layers and their connection relationships, removing redundant connections, and only retraining the newly added hidden layers, this process optimizes the model structure, reduces the risk of overfitting, and improves the training efficiency. The optimized model has been significantly improved in terms of accuracy and generalization ability and can better meet the actual application requirements.

[0144] Embodiment 4: This embodiment discloses a building cable fault location system, and the system includes: a processor and a memory,

[0145] Program code is stored in the memory;

[0146] When the processor calls the program code in the memory, it executes the steps of the method.

[0147] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present application shall be covered within the protection scope of the present application.

Claims

1. A method for locating a building cable fault, characterized in that: include: Data collection: including the first collection and the second collection; First acquisition: applying a detection signal to the target cable, obtaining a first feedback signal and a transmission speed of the detection signal in the target cable, and recording the transmission speed as first data; Collect the physical parameters of the target cable; Second collection: collecting training data, the training data including: physical parameters of various cables, second feedback signals of various cables after applying detection signals, and distances corresponding to the second feedback signals; Data processing: including first calculation, first modeling and first prediction; First calculation: obtaining the time difference between applying the detection signal to the target cable and receiving the first feedback signal, recording it as the second data, performing a product operation on the first data and the second data, and obtaining the initial distance; First modeling: establish a prediction model, use training data to train the prediction model, and obtain a trained prediction model; First prediction: input the first feedback signal and the physical parameters of the target cable into the trained prediction model to obtain the predicted distance; Fault location: including first judgment and output; First judgment: judging whether the difference between the initial distance and the predicted distance is less than a preset difference threshold, if so, executing the output step; if not, executing the data collection step; Output: determine the fault location according to the initial distance and / or predicted distance, and output the fault location; After executing the step of data collection and before executing the step of data processing, the method further includes: Feature extraction: using Fourier algorithm to perform spectrum analysis on the second feedback signal to obtain analysis results, and extracting features of the second feedback signal based on the analysis results, which are recorded as first features; Constructing a graph structure: constructing a graph structure with cables as nodes and connection relationships between cables as edges, wherein the attributes of the nodes include physical parameters and the first feature of the cables; First calculation: construct an adjacency matrix and a degree matrix based on the graph structure, calculate the Laplace matrix of the graph structure, the Laplace matrix is ​​equal to the difference between the adjacency matrix and the degree matrix, perform eigendecomposition on the Laplace matrix, and obtain eigenvalues ​​and eigenvectors; First acquisition: obtain the eigenvector corresponding to the maximum eigenvalue, recorded as the first vector; First mapping: performing a product operation on the matrix formed by the first features and the first vector to obtain an operation result, and using the operation result as a new second feedback signal.

2. The building cable fault location method according to claim 1, characterized in that: After executing the first acquisition step and before executing the first mapping step, the method further includes: First convolution: Integrate all the first features into a second vector, perform a convolution operation on the second vector and a preset first filter, and obtain a filtered second vector, which is recorded as a third vector; Second calculation: Perform nonlinear transformation on the third vector through the ReLU function, and use the third vector after the nonlinear transformation as the new first vector.

3. The building cable fault location method according to claim 1, characterized in that: After executing the step of constructing the graph structure and before executing the step of the first calculation, the method further includes: The third calculation: based on the attributes of each node, the Pearson correlation coefficient is used to calculate the correlation between each node; Fourth calculation: using a ReLU function to perform nonlinear change on the correlation to obtain third data; Set weight: Use the third data as the weight of the corresponding edge in the graph structure.

4. The building cable fault location method according to claim 2, characterized in that: After the step of performing the second calculation and before the step of performing the first mapping, the method further includes: The third acquisition: collect the electrical parameters of various cables and integrate all the electrical parameters into the fourth vector; Second judgment: judging whether the dimension of the fourth vector is equal to the dimension of the new first vector, if so, executing the second convolution step; if not, executing the dimension unification step; Unify dimensions: Use the principal component analysis algorithm to unify the dimensions of the fourth vector and the new first vector; Second convolution: performing a convolution operation on the fourth vector and a preset second filter to obtain a filtered fourth vector, which is recorded as a fifth vector; Fifth calculation: performing nonlinear transformation processing on the fifth vector through the ReLU function, and using the fifth vector after the nonlinear transformation as the new fourth vector; Second mapping: taking the electrical parameters corresponding to the new fourth vector as new electrical parameters.

5. The building cable fault location method according to claim 4, characterized in that: The attributes of the training data and graph structure also include: electrical parameters of various cables.

6. The building cable fault location method according to claim 1 or 2, characterized in that: After executing the first judgment step and before executing the data collection step, the method further includes: Sixth calculation: calculating the Euclidean distance between the first feedback signal and each second feedback signal, recorded as fourth data; Third judgment: judging whether the fourth data is less than a preset value, if so, executing the first prediction step; if not, executing the model expansion step; Model expansion: Add a new hidden layer to the trained prediction model to obtain an expanded prediction model; Retraining: fix the original hidden layer in the expanded prediction model, use the physical parameters of the target cable, the first feedback signal and the initial distance to train the new hidden layer in the expanded prediction model, and obtain the retrained prediction model; Model update: The step of performing the first prediction by replacing the trained prediction model with the retrained prediction model.

7. The building cable fault location method according to claim 6, characterized in that: After executing the model expansion step and before executing the retraining step, it also includes: Establish connections: Establish the connection relationship between each new hidden layer and each original hidden layer; Seventh calculation: using a Pearson correlation coefficient algorithm to respectively calculate the correlation between the first feedback signal and each second feedback signal; Fourth judgment: judging whether all the correlations are greater than the preset correlation threshold, if so, executing the retraining step; if not, executing the pruning step; Pruning: Remove the connection relationship between the original hidden layer and the new hidden layer corresponding to the correlation threshold lower than the preset correlation threshold.

8. The building cable fault location method according to claim 7, characterized in that: After the pruning step and before the retraining step, the following steps are also included: Fifth judgment: judge whether there is a connection relationship between the i-th new hidden layer and the remaining hidden layers. If so, execute the iteration step; if not, execute the deletion step; Iteration: taking the i+1th new hidden layer as the new ith hidden layer, and executing the fifth judgment step until the preset stop condition is met; Delete: Delete the i-th new hidden layer and perform the iterative steps.

9. A building cable fault location system, characterized in that: include: processor and memory, The memory stores program code; When the processor calls the program code in the memory, the steps of the method according to any one of claims 1 to 8 are executed.

Images