High and low voltage switch cabinet feeder line fault positioning method and system based on transient traveling wave

Abstract

The invention discloses a high-low voltage switch cabinet feeder fault positioning method and system based on transient traveling waves, and belongs to the technical field of power system fault detection and positioning, and the method comprises the steps: synchronously collecting electric and acoustic multi-mode signals, and generating a weighted transient synchronization feature matrix; performing time-frequency transformation and feedback optimization on the matrix, and outputting a multi-scale time-frequency feature set; analyzing the feature set by using an integrated learning network, and outputting a layered preliminary positioning result; convergence to an accurate fault section is carried out through iteration calibration; and finally, fusing multi-model calculation and outputting a comprehensive fault positioning report. According to the method, a technical path of combining multi-modal signal fusion and a physical model is adopted, convergence from fuzzy region division to an accurate position can be realized in stages, and the accuracy, the speed and the anti-interference capability of switch cabinet feeder line fault positioning are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system fault detection and positioning, and in particular to a method and system for positioning high and low voltage switchgear feeder faults based on transient traveling waves. Background Art

[0002] High- and low-voltage switchgear is a key hub in power systems, connecting the power grid and users. Its feeder lines, or feeders, are crucial components of the distribution network. When a feeder experiences a fault, such as a short circuit or ground fault, it generates a transient traveling wave signal containing rich fault information. This is an electromagnetic wave that propagates along the line at nearly the speed of light. Using traveling waves for fault location is an important electrical measurement technique. It determines the physical location of the fault by detecting the arrival time of the traveling wave signal at a monitoring point or analyzing its characteristics. This technique is crucial for ensuring safe and stable power grid operation and rapidly restoring power.

[0003] Existing methods for locating feeder faults based on transient traveling waves primarily rely on analyzing voltage or current traveling wave signals. Some methods install monitoring devices at both ends or multiple points on the feeder and use the time difference between the arrival of the traveling waves at different devices for distance measurement. Other methods utilize single-ended traveling wave signals and analyze the time interval between the initial wave head and the secondary wave head reflected from the fault point for location. At the signal processing level, time-frequency analysis tools such as wavelet transforms are typically used to identify the traveling wave head and calculate the fault distance.

[0004] However, the above-mentioned existing technologies have obvious technical defects in practical applications. The electromagnetic environment inside and near high and low voltage switchgear is complex, and there are a large number of metal structure reflections and coupling interferences. This makes the single current traveling wave or voltage traveling wave signal very easy to be distorted, and the wave head characteristics are submerged or false wave heads are generated, resulting in low positioning accuracy. In addition, when faced with non-ideal fault forms such as high-resistance grounding and multi-point faults, the traveling wave characteristics are weak and atypical. Traditional methods based on fixed models and algorithms have poor adaptability, and the positioning results are unstable or even fail. It is difficult to meet the high requirements of modern distribution networks for the speed and accuracy of fault location. Summary of the Invention

[0005] To solve the above problems, the present invention provides a high- and low-voltage switchgear feeder fault location method and system based on transient traveling waves. It adopts a technical path that combines multimodal signal fusion with physical models, and can achieve convergence from fuzzy area division to precise location in stages, significantly improving the accuracy, speed and anti-interference ability of switchgear feeder fault location.

[0006] The above objectives can be achieved through the following solutions: The method for locating the feeder fault of a high- and low-voltage switchgear based on transient traveling waves includes synchronously collecting broadband voltage and current traveling wave signals of the fault feeder, acoustic emission and ultrasonic signals inside the switchgear, constructing a signal cross matrix, performing signal correlation analysis on the signal cross matrix to adjust the modal weight, and generating a weighted transient synchronization feature matrix; performing a two-dimensional time-frequency transformation on the weighted transient synchronization feature matrix to obtain a transient spectrum time sequence diagram, and performing real-time feature scoring on the transient spectrum time sequence diagram, adjusting the weighted transient synchronization feature matrix according to the real-time feature scoring, and outputting a multi-scale time-frequency feature set; inputting the multi-scale time-frequency feature set into a preset integrated learning network to obtain The preliminary positioning result is obtained, and the confidence distribution analysis is used to divide the preliminary positioning result into a credible area and an area to be verified; for each segment of the area to be verified, waveform edge detection and data-driven detection are performed, and the traveling wave head moment is calibrated. The calibration result is fed back to the integrated learning network to update the internal parameters. The positioning result is converged to the fault segment through iteration, and an accurate wave head moment pair is output; based on the accurate wave head moment pair, combined with energy distribution analysis and fault propagation constraints, the multi-dimensional positioning result is calculated, and the result consistency test is performed. The verification result is fed back to update the parameter configuration of the integrated learning network, and a comprehensive fault location report is output.

[0007] Optionally, generating a weighted transient synchronization characteristic matrix includes: taking the broadband voltage and current traveling wave signal as the first signal group, taking the acoustic emission and ultrasonic signals as the second signal group, performing correlation analysis on the first signal group and the second signal group, and calculating a canonical correlation coefficient; based on the canonical correlation coefficient, iteratively correcting each modal weight, and monitoring the cross-correlation changes of the signal cross matrix to generate an iterative weight adjustment sequence; applying the iterative weight adjustment sequence to perform weighted fusion on the signal cross matrix, and outputting a weighted transient synchronization characteristic matrix.

[0008] Optionally, the output multi-scale time-frequency feature set includes: adaptively adjusting the time-frequency resolution of the weighted transient synchronization feature matrix, optimizing the selection of transformation parameters through a dynamic window function, and performing two-dimensional time-frequency transformation processing to generate a transient spectrum time series diagram; performing information entropy evaluation on the transient spectrum time series diagram to calculate a real-time feature score, and quantifying the spectrum distribution uniformity and time series consistency, and outputting a feature score vector; based on the feature score vector, performing parameter feedback correction on the weighted transient synchronization feature matrix, and fusing the transient spectrum time series diagram to generate a multi-scale time-frequency feature set.

[0009] Optionally, the output feature score vector includes: performing multi-scale entropy spectrum analysis on the transient spectrum time series diagram, calculating the local information entropy distribution and global spectrum complexity, and filtering out noise interference to generate a preliminary entropy spectrum quantization set; for the preliminary entropy spectrum quantization set, calculating the spectrum distribution uniformity and time series consistency, and generating a real-time feature score sequence through weighted aggregation; performing vector normalization processing based on the real-time feature score sequence, statistically correcting the confidence interval to optimize the score distribution, and outputting a feature score vector.

[0010] Optionally, the method also includes: performing multi-layer association mapping on the iterative weight adjustment sequence and the feature score vector to generate a weight score joint matrix; calculating the dynamic optimization factor based on the weight score joint matrix, and analyzing the signal modality and time-frequency features for vector-level enhancement, and outputting an enhanced weight vector set; injecting the enhanced weight vector set into the integrated learning network parameters, optimizing the internal weights and bias terms through gradient guidance, and generating a fusion feature optimization map.

[0011] Optionally, dividing the preliminary positioning result into a credible area and an area to be verified includes: inputting the multi-scale time-frequency feature set into the integrated learning network to calculate a preliminary positioning result; analyzing the confidence probability distribution of the preliminary positioning result, performing distribution entropy calculation to quantify the uncertainty level, and outputting a confidence distribution map; based on the confidence distribution map, assigning a dynamic verification priority to each area, and dividing the area into a credible area and an area to be verified.

[0012] Optionally, the calculation to obtain the preliminary positioning result includes: obtaining historical fault recording data and historical power grid physical simulation data of high and low voltage switchgear to generate a historical fault data set; using multi-scale time-frequency features as input and actual position as output, and utilizing the historical fault data set to establish and train a neural network model to obtain an integrated learning network; inputting the multi-scale time-frequency feature set into the integrated learning network to output a preliminary positioning result.

[0013] Optionally, the output of the precise wave front moment pair includes: performing dual-path preliminary calibration for each section of the area to be verified, executing traveling wave edge detection and data-driven detection in parallel, and cross-validating the calibration results of the two paths, and outputting the initial wave front moment; for the initial wave front moment, calculating the intermediate positioning result, and evaluating the confidence gradient of the integrated learning network for the intermediate positioning result; using the confidence gradient, feedback updating the internal parameters of the integrated learning network; iteratively performing the preliminary positioning and the confidence gradient evaluation until the drift of the positioning result and the confidence gain of the integrated learning network are both lower than a preset convergence threshold, and outputting the wave front moment of the positioning result in the converged state as a precise wave front moment pair.

[0014] Optionally, the output of the comprehensive fault location report includes: performing multi-terminal ranging and energy distribution analysis based on the precise wave head moment, integrating the fusion feature optimization map to perform propagation path constraint correction, and generating a preliminary multi-dimensional positioning result; applying a cross-consistency test to the preliminary multi-dimensional positioning result, and feeding the test result back to the integrated learning network to update the parameter configuration, and outputting an optimized positioning result set; performing weighted fusion on the positioning results in the optimized positioning result set, calculating the fault location coordinates and the posterior probability, and encapsulating the fault location coordinates and the posterior probability into the comprehensive fault location report for output.

[0015] Based on the same inventive concept, the present invention also provides a high- and low-voltage switchgear feeder fault location system based on transient traveling waves, the system comprising: an adaptive signal fusion module for synchronously collecting broadband voltage and current traveling wave signals of the fault feeder, acoustic emission and ultrasonic signals inside the switchgear, constructing a signal cross matrix, performing signal correlation analysis on the signal cross matrix to adjust the modal weights, and generating a weighted transient synchronization feature matrix; a time-frequency feature optimization module for performing a two-dimensional time-frequency transformation on the weighted transient synchronization feature matrix to obtain a transient spectrum time sequence diagram, and performing real-time feature scoring on the transient spectrum time sequence diagram, adjusting the weighted transient synchronization feature matrix according to the real-time feature score, and outputting a multi-scale time-frequency feature set; an intelligent pre-location and area division module for converting the multi-scale The time-frequency feature set is input into a preset integrated learning network to obtain a preliminary positioning result, and the preliminary positioning result is divided into a credible area and an area to be verified through confidence distribution analysis; the iterative convergence positioning module is used to perform waveform edge detection and data-driven detection on each segment of the area to be verified, calibrate the traveling wave head moment, and feed the calibration result back to the integrated learning network to update the internal parameters, and iterate until the positioning result converges to the fault segment, and output an accurate wave head moment pair; the multi-model fusion and verification module is used to calculate the multi-dimensional positioning result based on the accurate wave head moment pair, combined with energy distribution analysis and fault propagation constraints, and perform result consistency check, and feed back the verification result to update the parameter configuration of the integrated learning network, and output a comprehensive fault location report.

[0016] Compared with the prior art, the present invention has the following advantages: 1. This invention dynamically and collaboratively integrates multi-source signals and adaptively adjusts modal weights based on the inherent correlation between signals. This effectively overcomes the limitations of a single signal source, such as susceptibility to interference and insufficient information dimension in complex electromagnetic environments. It improves the quality of input features and the signal-to-noise ratio from the source, laying a solid and reliable data foundation for subsequent precise positioning. 2. This invention builds a hierarchical processing framework from intelligent prediction to iterative precision positioning. Through intelligent region division through confidence analysis, high-cost precision computing resources are focused on the most uncertain sections. Through closed-loop feedback iteration between the model and physical calibration results, high-precision and high-efficiency positioning of complex and difficult faults is achieved, improving the robustness and response speed of the method. 3. This invention establishes a multi-layered, full-process closed-loop optimization system, from signal processing to model parameters, from physical calculations to model configuration. By feeding back the results of feature quality assessment and cross-consistency testing to update the core intelligent network, it possesses the ability of continuous learning and self-evolution, continuously optimizing its performance during use and maintaining a high level of positioning accuracy over the long term.

[0017] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 The present invention is a flowchart of a method for locating a high- and low-voltage switchgear feeder fault based on transient traveling waves according to an embodiment of the present invention.

[0020] Figure 2 4 is a cross-correlation and distribution matrix diagram of an embodiment of the present invention.

[0021] Figure 3 This is a confidence distribution and area division cloud and rain map according to an embodiment of the present invention.

[0022] Figure 4 It is a composite timing diagram of the closed-loop intervention and model adaptive optimization process in an embodiment of the present invention.

[0023] Figure 5 1 is a schematic structural diagram of a high- and low-voltage switchgear feeder fault location system based on transient traveling waves according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0025] Reference Figure 1 One embodiment of the present invention proposes a method for locating high and low voltage switchgear feeder faults based on transient traveling waves. By adopting a technical path that combines multimodal signal fusion with physical models, it can achieve convergence from fuzzy region division to precise location in stages, significantly improving the accuracy, speed, and anti-interference capability of switchgear feeder fault location.

[0026] The method of this embodiment specifically includes: Synchronously collect broadband voltage and current traveling wave signals of the fault feeder, acoustic emission and ultrasonic signals inside the switchgear, construct a signal cross matrix, perform signal correlation analysis on the signal cross matrix, adjust modal weights, and generate a weighted transient synchronization feature matrix; Performing a two-dimensional time-frequency transformation on the weighted transient synchronization feature matrix to obtain a transient spectrum time sequence diagram, performing real-time feature scoring on the transient spectrum time sequence diagram, adjusting the weighted transient synchronization feature matrix according to the real-time feature score, and outputting a multi-scale time-frequency feature set; Inputting the multi-scale time-frequency feature set into a preset integrated learning network to obtain a preliminary positioning result, and dividing the preliminary positioning result into a credible area and an area to be verified through confidence distribution analysis; For each section of the area to be verified, waveform edge detection and data-driven detection are performed to calibrate the traveling wave front moment. The calibration results are fed back to the integrated learning network to update internal parameters. The positioning result is converged to the fault section through iteration, and an accurate wave front moment pair is output; Based on the precise wavefront time pair, combined with energy distribution analysis and fault propagation constraints, multi-dimensional positioning results are calculated, and the result consistency is checked. The verification results are fed back to update the parameter configuration of the integrated learning network and output a comprehensive fault location report.

[0027] The technical approach of combining multimodal signal fusion with physical models can achieve convergence from fuzzy area division to precise location in stages, significantly improving the accuracy, speed and anti-interference ability of switchgear feeder fault location.

[0028] Optionally, generating a weighted transient synchronization characteristic matrix includes: Taking the broadband voltage and current traveling wave signals as a first signal group and the acoustic emission and ultrasonic signals as a second signal group, performing correlation analysis on the first signal group and the second signal group to calculate a canonical correlation coefficient; Specifically, this step aims to quantify the intrinsic correlation strength between the two heterogeneous signal groups. The synchronously collected broadband voltage and current traveling wave signals are divided into the first signal group, and the acoustic emission and ultrasonic signals are divided into the second signal group. To maximize the correlation between the two sets of variables, a canonical correlation analysis (CCA) is performed on the first and second signal groups. This analysis maximizes the correlation coefficient between the two signal groups by finding a linear combination of the two signal groups, namely the canonical variable. This maximized correlation coefficient is the first canonical correlation coefficient.

[0029] Iteratively correcting each modal weight based on the canonical correlation coefficient, and monitoring the cross-correlation change of the signal cross matrix to generate an iterative weight adjustment sequence; Specifically, the canonical correlation coefficient directly reflects the degree of consistency between the fault information represented by the electrical and acoustic signal groups under the current weight distribution. An iterative optimization process is initiated, with maximizing the canonical correlation coefficient as the objective function. Using gradient ascent or a similar optimization algorithm, the weights of each signal mode are adjusted at each iteration. Simultaneously, changes in the internal correlation of the signal cross-matrix are monitored in real time until the canonical correlation coefficient reaches stability or converges. The weight changes during this iterative process are recorded, forming an iterative weight adjustment sequence.

[0030] The iterative weight adjustment sequence is applied to perform weighted fusion on the signal cross matrix, and a weighted transient synchronization feature matrix is ​​output.

[0031] Specifically, the optimized weights that finally converge in the iterative weight adjustment sequence are applied to perform weighted fusion on the original signal cross matrix. The weighted fusion process can be expressed by the following matrix operation formula: , in, The weighted transient synchronization characteristic matrix representing the final output; It is a signal cross matrix composed of original multimodal signals; is a weight vector whose elements are the final optimized weights of each signal mode derived from the iterative weight adjustment sequence; is an operator that converts the weight vector into a diagonal matrix. This fusion operation is achieved by multiplying each column signal in the signal cross matrix by its corresponding final optimized weight, thereby highlighting the components that are most relevant to the fault and verified by multiple source signals, while suppressing the noise or artifacts that only exist in a single signal source, such as Figure 2 As shown in the figure, the diagonal line shows the probability density distribution of the three core signal modes: broadband traveling waves, acoustic emissions, and ultrasonic signals. The off-diagonal scatter plot reveals the correlation between any two signal modes. This figure is an important basis for signal correlation analysis and modal weight adjustment before generating the weighted transient synchronization feature matrix.

[0032] Optionally, the output multi-scale time-frequency feature set includes: Adaptively adjusting the time-frequency resolution of the weighted transient synchronization feature matrix, optimizing and selecting transformation parameters through a dynamic window function, and performing a two-dimensional time-frequency transformation process to generate a transient spectrum time sequence diagram; Specifically, this step aims to dynamically optimize the parameters of the time-frequency transform based on the instantaneous characteristics of the weighted transient synchronization feature matrix. For example, when performing a short-time Fourier transform (STFT), the transform parameters are selected through dynamic window function optimization. Specifically, a shorter analysis window is used to achieve high time resolution for the transient portions of the signal with dramatic changes, while a longer analysis window is used to achieve high frequency resolution for the portions with gentle changes. After determining the optimized transform parameters, a two-dimensional time-frequency transform is performed on the weighted transient synchronization feature matrix, converting it from a one-dimensional time-domain signal to a two-dimensional time-frequency domain representation, thereby generating a transient spectrum time series diagram.

[0033] Performing information entropy evaluation on the transient spectrum time sequence diagram to calculate a real-time feature score, quantifying spectrum distribution uniformity and time sequence consistency, and outputting a feature score vector; Specifically, in order to quantify the quality and information content of the transient spectrum time series diagram, the information entropy is evaluated to calculate the real-time feature score. This step aims to evaluate the clarity and reliability of the time-frequency features. The real-time feature score of a local area of ​​a transient spectrum time series diagram is It can be calculated using a compound formula that includes multiple metrics: , in, It is the information entropy after calculating and normalizing the energy distribution of the region. A region with concentrated energy and clear features corresponds to a lower information entropy, which makes The value of is higher; It is an index of spectrum distribution uniformity obtained by calculating the Gini coefficient of spectrum energy. The more concentrated the energy, the higher the index value. It is a temporal consistency index obtained by calculating the autocorrelation coefficient of the feature vectors in adjacent time windows. The more stable the feature, the higher the index value. 、 、 These are weight coefficients derived from a feature evaluation rule base, used to balance the contributions of information entropy, uniformity, and consistency to the overall score. This calculation ultimately outputs a feature score vector, where each element corresponds to the feature quality score for a different region or time on the transient spectrum time series diagram.

[0034] Based on the feature scoring vector, parameter feedback correction is performed on the weighted transient synchronization feature matrix, and the transient spectrum time series diagram is fused to generate a multi-scale time-frequency feature set.

[0035] Specifically, this step uses the feature scoring vector generated in the previous step to perform parameter feedback correction on the weighted transient synchronization feature matrix. For example, for time periods with low scores, the corresponding data points in the original weighted transient synchronization feature matrix are suppressed or filtered. Finally, the weighted transient synchronization feature matrix, which has undergone parameter feedback correction, is fused with the transient spectrum time series diagram that clearly demonstrates the time-frequency dynamics. Together, they form a multi-scale time-frequency feature set with richer information dimensions and more reliable features, which is then output to the subsequent processing module.

[0036] Optionally, the output feature score vector includes: Performing multi-scale entropy spectrum analysis on the transient spectrum time sequence diagram, calculating the local information entropy distribution and global spectrum complexity, filtering and removing noise interference, and generating a preliminary entropy spectrum quantization set; Specifically, this step aims to comprehensively evaluate the uncertainty of the signal at different time resolutions. Multi-scale entropy spectrum analysis is performed on the transient spectrum time series diagram, for example, using a sample entropy algorithm, at different time scales, that is, after the signal is subjected to different degrees of coarse-graining processing, its information entropy is calculated. Through this analysis, the local information entropy distribution of the transient spectrum time series diagram can be calculated, that is, the degree of disorder in each tiny area of ​​the diagram, as well as the global spectrum complexity that reflects the complexity of the overall signal. High entropy areas usually correspond to irregular noise, while low entropy areas correspond to structured fault features. Then, based on the results of the above entropy spectrum analysis, an entropy value threshold is set, and areas with entropy values ​​higher than the threshold are judged as noise and suppressed, thereby filtering and eliminating noise interference to generate a preliminary entropy spectrum quantization set.

[0037] For the preliminary entropy spectrum quantization set, the spectrum distribution uniformity and temporal consistency are calculated, and a real-time feature score sequence is generated through weighted aggregation; Specifically, after obtaining a preliminary entropy spectrum quantization set that has been cleaned by the entropy dimension, its spectrum distribution uniformity and time series consistency are further calculated. Spectrum distribution uniformity is used to measure the distribution of energy in the frequency domain and can be quantified by calculating the Gini coefficient or spectrum flatness. Time series consistency is measured by evaluating the autocorrelation or stability of features in adjacent time slices. Subsequently, the indicators of the three dimensions of entropy value, uniformity, and consistency are fused, and the weight coefficients are derived from a feature evaluation rule base to generate a real-time feature score sequence. Each value in this sequence represents the comprehensive feature quality of the transient spectrum time series diagram at the corresponding moment.

[0038] Vector normalization is performed based on the real-time feature score sequence, and the score distribution is optimized by statistical confidence interval correction to output a feature score vector.

[0039] Specifically, to facilitate subsequent processing, the real-time feature score sequence is first vector-normalized, for example, by using min-max normalization to scale its values ​​to a range between 0 and 1. Finally, to enhance the robustness of the score, robust statistical methods are used, such as calculating the interquartile range to construct a confidence interval. This confidence interval is then used to correct for extreme scores that are far from the main distribution. This process of optimizing the score distribution ultimately outputs a stable feature score vector that accurately reflects feature quality.

[0040] Optionally, the method further includes: Performing multi-layer association mapping on the iterative weight adjustment sequence and the feature score vector to generate a weight score joint matrix; Specifically, this step aims to establish a deep connection between front-end signal processing and mid-end feature evaluation. The iterative weight adjustment sequence reflects the evolving importance of different signal modalities during the fusion process, while the feature score vector quantifies the quality of the extracted time-frequency features at different moments. A two-dimensional joint weight-score matrix is ​​constructed, with rows corresponding to signal modalities and columns corresponding to time or feature dimensions. The matrix elements are calculated using nonlinear functions from the weight values ​​in the iterative weight adjustment sequence and the score values ​​in the feature score vector. This matrix reveals which signal modalities contribute the most at which high-quality feature moments.

[0041] Calculating a dynamic optimization factor based on the weight score joint matrix, analyzing the signal modality and time-frequency characteristics for vector-level enhancement, and outputting an enhanced weight vector set; Specifically, this step aims to extract key optimization guidance information from the weight score joint matrix. First, by applying principal component analysis (PCA) or similar dimensionality reduction techniques to the weight score joint matrix, its main change trends are extracted and a dynamic optimization factor is calculated. Subsequently, the dynamic optimization factor is used to perform vector-level enhancement on the original iterative weight adjustment sequence. An enhanced weight vector It can be calculated by the following formula: , in, is the original weight vector derived from the iterative weight adjustment sequence; represents the Hadamard product (element-wise product); It is a hyperbolic tangent function, which is used for nonlinear mapping and smoothly limits the enhancement effect to the range of -1 to 1; is a hyperparameter that controls the magnitude of reinforcement; is the dynamic optimization factor obtained from the above calculation; is the original weight vector and the feature score derived from the feature score vector The mutual information between them is used to quantify the nonlinear correlation between the two. This formula integrates the dynamic optimization factor and mutual information to achieve intelligent reinforcement of those weight components that are both important and reliable, and ultimately outputs an enhanced weight vector set.

[0042] The enhanced weight vector set is injected into the integrated learning network parameters, and the internal weights and bias terms are optimized through gradient guidance to generate a fusion feature optimization map.

[0043] Specifically, this step applies the enhanced weight vector set generated in the previous step to optimize the internal parameters of the ensemble learning network. This process does not simply replace the original parameters, but is carried out in a gradient-guided manner. For example, the enhanced weight vector set is used as an attention mask and multiplied element-by-element on the feature map of a specific layer of the ensemble learning network. During the backpropagation process of network training or fine-tuning, this operation naturally modulates the gradient, making the update of the network parameters more biased towards the important features identified by the enhanced weight vector set. The final product of this series of operations is a fused feature optimization map, which is both a graphical representation of the entire optimization process and represents the new state of the ensemble learning network after this round of optimization.

[0044] Optionally, dividing the preliminary positioning result into a trusted area and an area to be verified includes: Inputting the multi-scale time-frequency feature set into the integrated learning network to calculate and obtain a preliminary positioning result; Specifically, this step feeds the rich, multi-scale time-frequency feature set generated in the previous step into a trained ensemble learning network. The ensemble learning network, through its complex internal nonlinear mapping relationships, performs in-depth analysis and reasoning on the input features, ultimately calculating and outputting a preliminary location result. This preliminary location result may be one or more potential fault sections and their corresponding initial confidence probabilities.

[0045] Analyze the confidence probability distribution based on the preliminary positioning results, perform distribution entropy calculation to quantify the uncertainty level, and output a confidence distribution map; Specifically, this step aims to quantify the degree of uncertainty of the model in its own prediction results. First, the confidence probability distribution in the preliminary positioning results is analyzed. Then, a generalized information entropy, namely Rényi Entropy, is used to calculate the distribution entropy. A probability distribution Rainey entropy It can be calculated by the following formula: , in, It is the first The confidence probability of a possible fault section; represents the sum of all possible segments; is an order parameter greater than or equal to 0 and not equal to 1. By adjusting the order parameter, the sensitivity to different parts of the probability distribution can be changed, thereby achieving multi-dimensional quantification of uncertainty. For example, when When it approaches 1, the formula converges to the traditional Shannon entropy; when When is larger, more attention is paid to the peak part in the probability distribution.,Through this calculation, a confidence distribution map is output, which intuitively,shows the positioning uncertainty level of each section along the,feeder.

[0046] Based on the confidence distribution map, a dynamic verification priority is assigned to each area, and the trusted area and the area to be verified are divided.

[0047] Specifically, based on the confidence distribution map generated in the previous step, a dynamic verification priority is assigned to each potential fault area. Areas with high entropy values ​​or low confidence values ​​will be given a higher verification priority. Finally, based on the dynamic verification priority, a confidence threshold or entropy threshold is set, and areas with a priority higher than this threshold are divided into areas to be verified. These are the parts that require more computing resources to be invested in for detailed verification; and areas with a priority lower than this threshold are divided into trusted areas. The positioning results of these areas are considered to be preliminarily reliable, such as Figure 3The figure shows the confidence distribution for four potential fault zones. The shape of the violin reflects the confidence probability density, the boxplot inside shows the statistical summary, and the scattered points represent the raw confidence scores from multiple evaluations. For example, the violins in "Segments A" and "Segments C" are tall and thin, with concentrated data points, indicating that the model is very confident in its location results and belongs to the "trusted region." In contrast, the violins in "Segments B" and "Segments D" are short and fat, with scattered data points, indicating high uncertainty and belonging to the "unverified region."

[0048] Optionally, the calculating to obtain a preliminary positioning result includes: Obtain historical fault recording data of high and low voltage switchgear and historical power grid physical simulation data to generate a historical fault data set; Specifically, this step aims to provide sufficient and diverse samples for training the ensemble learning network. Historical fault recording data is derived from waveform files recorded by field equipment during real-world fault events. Historical power grid physical simulation data is generated by using electromagnetic transient simulation software to simulate various boundary conditions, fault types, and fault locations based on the actual switchgear topology and parameters. After integrating, cleaning, and uniformly formatting these two types of data, a historical fault dataset containing a rich set of fault modes is generated.

[0049] Taking multi-scale time-frequency features as input and actual positions as output, a neural network model is established and trained using the historical fault data set to obtain an integrated learning network; Specifically, this step involves building the core intelligent positioning model. A neural network model is constructed, whose architecture adopts the stacked generalization strategy used in ensemble learning. This strategy consists of two layers. The first layer is the base learner, which parallelizes a convolutional neural network (CNN) unit to extract spatial features from a multi-scale time-frequency feature set and a long short-term memory (LSTM) unit to analyze its temporal evolution pattern. The CNN unit can consist of three convolutional layers, each followed by a rectified linear unit (ReLU) activation function and a max pooling layer. The LSTM unit can consist of two stacked LSTM layers, each with 128 hidden units. The second layer is the meta-learner, which concatenates the output feature vectors of the CNN and LSTM units in the first layer and feeds this concatenation into a gradient boosting decision tree (GBDT) model, which outputs the final preliminary positioning result. The training process of the integrated learning network is as follows: the multi-scale time-frequency feature set extracted from the historical fault dataset is used as the input of the model, and the actual location label of the fault and the corresponding confidence distribution are used as the output. The Adam optimizer is used in training, and the learning rate can be dynamically adjusted during the training process. Its initial value can be set in the range of 1e-4 to 1e-3. In order to enhance the generalization ability of the model, the historical fault dataset is enhanced by adding Gaussian white noise with different signal-to-noise ratios, random time offset and amplitude scaling. To prevent overfitting, the Dropout mechanism and L2 regularization are used in the training of CNN and LSTM units. The training goal of the model is to optimize a composite loss function To achieve: , in, is the positioning loss term, for example, the mean squared error (MSE) is used to calculate the deviation between the model's predicted position and the actual position label; It is the confidence loss term, which penalizes prediction results with high uncertainty by calculating the information entropy of the confidence distribution output by the model, and guides the model to produce more confident positioning; is a regularization term used to prevent the model from overfitting; and are two hyperparameters derived from the model training configuration, used to balance the contributions of the confidence loss and regularization term to the total loss. By minimizing this composite loss function using optimization algorithms such as gradient descent, a convergent and stable ensemble learning network is ultimately trained.

[0050] The multi-scale time-frequency feature set is input into the integrated learning network, and a preliminary positioning result is output.

[0051] Specifically, after receiving the multi-scale time-frequency feature set generated by a new fault event, it is input into the previously trained ensemble learning network. The network performs forward propagation calculations using its internal weights and activation functions, deeply analyzing and reasoning on the input features, and ultimately calculates and outputs a preliminary location result. This preliminary location result includes predictions for one or more potential fault sections, as well as the model's confidence probability distribution for each predicted section.

[0052] Optionally, the outputting of the precise wave front moment pair includes: Perform preliminary dual-path calibration for each section of the area to be verified, perform traveling wave edge detection and data-driven detection in parallel, cross-validate the calibration results of the two paths, and output the initial wave front moment; Specifically, this step aims to perform high-precision wavefront calibration in areas to be verified where the ensemble learning network is uncertain. Two independent detection paths are initiated in parallel: the first is based on physical model-based traveling wave edge detection, for example, applying an image-based line detection algorithm to find steep edges in the two-dimensional representation of the original traveling wave signal; the second is based on data-driven detection, for example, applying a pre-trained temporal pattern recognition network to search for patterns similar to historical fault wavefronts. The calibration results output by the two paths are cross-validated, and only when the two results are consistent within a time tolerance is it confirmed as a valid initial wavefront moment.

[0053] Calculating an intermediate positioning result for the initial wave front moment, and evaluating a confidence gradient of the integrated learning network on the intermediate positioning result; Specifically, the initial wavefront time output from the previous step is substituted into a two-terminal traveling wave ranging algorithm to calculate an intermediate positioning result. This intermediate positioning result is then fed back into the ensemble learning network as a query. Through a single forward and backpropagation pass, the model's confidence gradient at that positioning result is evaluated. The confidence gradient is a vector that indicates the direction in which its internal parameters should be adjusted to maximize the model's confidence in the current positioning result.

[0054] Using the confidence gradient, performing feedback update on internal parameters of the ensemble learning network; Specifically, this step uses the confidence gradient calculated in the previous step to perform a feedback update on the internal parameters of the ensemble learning network. This update process can use gradient ascent or a similar optimization algorithm to fine-tune the model to generate higher confidence in the wavefront moment and the calculated positioning result in the current iteration.

[0055] Iteratively perform the preliminary positioning and the confidence gradient evaluation until the drift of the positioning result and the confidence gain of the integrated learning network are both lower than a preset convergence threshold, and output the wave head moment of the positioning result in the converged state as an accurate wave head moment pair.

[0056] Specifically, the aforementioned preliminary positioning, confidence gradient evaluation and parameter update steps are iteratively performed. At the end of each iteration, a convergence criterion function To determine whether the iteration is terminated: , in, and They are and The positioning result vector calculated by the iteration; It is After iterations, the improvement in the model's confidence in the positioning results is the confidence gain of the integrated learning network; is a weight coefficient derived from the configuration, used to balance the importance of result stability and confidence gain; is the convergence threshold, which comes from the calibration experiment. When the value of is less than the threshold, the iteration terminates. The wave head moment corresponding to the positioning result in the final converged state is output as the precise wave head moment pair.

[0057] Optionally, the outputting of the comprehensive fault location report includes: Based on the precise wavefront moment, multi-terminal ranging and energy distribution analysis are performed, and the fusion feature optimization map is incorporated to perform propagation path constraint correction to generate a preliminary multi-dimensional positioning result; Specifically, this step aims to obtain a set of independent positioning solutions through parallel calculations based on multiple physical principles. Based on the precise wave head moment pair output in the previous step, at least two positioning algorithms are started in parallel. The first is the multi-terminal traveling wave ranging method, which directly uses the time difference for calculation. The second is the transient energy distribution analysis method, which infers the fault point by analyzing the distribution gradient of the fault energy in each section of the feeder. During the calculation process, the information derived from the fusion feature optimization map is used as a constraint condition for the propagation path to be corrected to ensure that the calculation meets the actual credibility of the signal characteristics, thereby generating a preliminary multi-dimensional positioning result containing multiple independent physical position solutions.

[0058] Applying a cross-consistency check to the preliminary multi-dimensional positioning results, and feeding the check results back to the integrated learning network to update parameter configuration and output an optimized positioning result set; Specifically, this step aims to verify and optimize the results obtained by parallel computing. A cross-consistency check is performed on each independent physical location solution in the preliminary multi-dimensional positioning results, for example, by calculating the Euclidean distance between each solution or the physical deviation on the Geographic Information System (GIS). The test results, that is, the deviation between each solution, are quantified and fed back to the integrated learning network to fine-tune the parameter configuration of its output layer or confidence assessment module. This feedback process is designed to allow the integrated learning network to learn and understand the performance differences of different physical positioning algorithms under specific working conditions. After the parameter update is completed, the positioning calculation is performed again to output a preliminarily optimized and more consistent optimized positioning result set.

[0059] The positioning results in the optimized positioning result set are weightedly fused to calculate the fault location coordinates and the posterior probability, and the fault location coordinates and the posterior probability are packaged into the comprehensive fault location report for output.

[0060] Specifically, in order to obtain a unique and most credible final positioning conclusion, this step performs weighted fusion on multiple positioning results in the optimized positioning result set. This fusion process uses the Bayesian Model Averaging (BMA) method. This method does not perform simple arithmetic averaging, but regards each positioning algorithm as an independent expert model, and calculates the posterior probability of each positioning result based on its historical performance under the current working conditions and the consistency of the output results this time. The final fault location coordinate is the weighted average of all positioning results with the posterior probability as the weight. Through this method, not only an accurate fault location coordinate is calculated, but also a statistical confidence evaluation is provided for it, namely the posterior probability. Finally, the coordinate and the posterior probability are packaged together as a comprehensive fault location report for output, such as Figure 4 As shown in the figure, the dynamic evolution of robust risk indicators and their uncertainty is illustrated, with time as the horizontal axis. When the risk indicator reaches the dynamically adjusted safety threshold, the system triggers a closed-loop intervention command. Each successful intervention not only effectively controls the risk but also learns from it. This is reflected in the step-by-step growth of the knowledge base of the reused intelligent template library on the right Y-axis, demonstrating the complete intelligent closed loop of decision-making, intervention, learning, and evolution of the entire system.

[0061] To verify the feasibility of the present invention, it was applied to the fault location of the switchgear feeder in an underground 10kV substation at a core transportation hub in a certain city. This switchgear is responsible for supplying power to critical loads and operates in a complex environment with strong electromagnetic interference and a compact internal structure. Traditional fault location methods have difficulty quickly and accurately identifying and locating early high-resistance grounding faults with subtle characteristics caused by slow insulation degradation. To improve operation and maintenance efficiency and prevent fault expansion, the project team adopted the method of the present invention to achieve advanced and precise positioning of feeder faults within the switchgear.

[0062] In this embodiment, the project manager first inputs the feeder topology, equipment models, and historical operating data for all high- and low-voltage switchgear within the underground substation into the diagnostic process corresponding to the present method. This analysis prepares corresponding models and parameters for each potential fault diagnosis task. When a fault occurs, the present method performs multimodal signal acquisition and fusion, time-frequency feature extraction and optimization, preliminary localization using an integrated learning network, and precise calibration through iterative convergence, ultimately producing a comprehensive fault location report containing the fault location and reliability.

[0063] In order to verify the effectiveness of the present invention, the diagnosis process of a real early high-resistance grounding fault that occurred on the night of July 10, 2024 is selected for illustration.

[0064] At 11:15 PM on July 10, 2024, a weak transient disturbance was detected on feeder A12 of switchgear #2 within the station. The proposed method was immediately activated, synchronously collecting broadband voltage and current traveling wave signals at the feeder port, as well as acoustic emission and ultrasonic signals from within the switchgear.

[0065] Canonical correlation analysis was performed on the collected electrical signals as the first signal group and the acoustic signals as the second signal group, resulting in a calculated canonical correlation coefficient of 0.78. Three rounds of iterative correction generated an optimal iterative weight adjustment sequence, which was then applied to weighted fusion of the signals, resulting in a weighted transient synchronization feature matrix with an optimized signal-to-noise ratio.

[0066] An adaptive time-frequency transform is performed on the weighted transient synchronization feature matrix, and a feature scoring vector is calculated using metrics such as information entropy. This scoring vector reveals that the ultrasonic signal feature quality is low during the initial stages of the fault. Feedback is then used to correct the weighting process, appropriately reducing the ultrasonic signal weight, ultimately outputting a high-quality multi-scale time-frequency feature set.

[0067] The optimized multi-scale time-frequency feature set was fed into a pre-set ensemble learning network. The network's initial positioning result indicated a section approximately 1.5 meters near the A12 feeder cable terminal. However, the confidence distribution map indicated a high level of uncertainty. Therefore, this section was designated as a "pending verification area," and the next step of the precise calibration process was initiated.

[0068] For the 1.5-meter verification area, dual-path wavefront calibration was initiated, performing traveling wave edge detection and data-driven detection in parallel, followed by cross-validation, to output an initial wavefront moment. Four rounds of iterative feedback were performed based on the intermediate positioning results calculated at this moment and the confidence gradient of the ensemble learning network. During this iterative process, the drift of the positioning results converged from the initial 0.5 meter to 0.05 meter, and the confidence gain also stabilized, ultimately outputting a pair of accurate wavefront moment pairs.

[0069] Based on precise wavefront moment pairs, a dual-end ranging algorithm and energy distribution analysis were applied in parallel to obtain two independent positioning results. At this point, a fusion feature optimization map was incorporated for weighting, and a Bayesian consistency check was initiated. The fault location coordinates were ultimately calculated to be 0.73 meters from the cable terminal, with a posterior probability of 98.5%. The entire process took 1.2 seconds, generating and outputting a comprehensive fault location report. Based on this report, the operations and maintenance personnel accurately discovered early signs of insulation degradation at the cable terminal during a power outage maintenance the following morning, successfully eliminating a major hidden danger.

[0070] Table 1 Multimodal signal adaptive fusion data table

[0071] Table 2 AI pre-positioning and iterative convergence process data

[0072] Table 3 Final positioning results and model update data table

[0073] The above Tables 1-3 record the actual application data of the present invention in switch cabinet feeder fault location. Table 1 shows that through canonical correlation analysis, the adaptive weighting of multimodal signals is completed in just tens of milliseconds, highlighting its intelligent fusion capability. Table 2 clearly records the convergence process from AI fuzzy pre-positioning to iterative precise positioning, proving the efficiency of its phased and step-by-step approximation strategy. Table 3 quantifies the final positioning accuracy and speed, and demonstrates the self-learning capability, by feeding back the experience of this successful positioning to update and optimize the background model. This embodiment fully demonstrates the huge technical advantages of the present invention in solving the problem of precise positioning of weak faults in complex environments.

[0074] Based on the same inventive concept, the present invention also provides a high and low voltage switch cabinet feeder fault location system based on transient traveling waves, such as Figure 5 As shown, the system includes: An adaptive signal fusion module is used to synchronously collect broadband voltage and current traveling wave signals from the fault feeder, acoustic emission signals from the switchgear, and ultrasonic signals, construct a signal cross-matrix, perform signal correlation analysis on the signal cross-matrix, adjust modal weights, and generate a weighted transient synchronization feature matrix. a time-frequency feature optimization module, configured to perform a two-dimensional time-frequency transformation on the weighted transient synchronization feature matrix to obtain a transient spectrum time sequence diagram, perform real-time feature scoring on the transient spectrum time sequence diagram, adjust the weighted transient synchronization feature matrix according to the real-time feature scoring, and output a multi-scale time-frequency feature set; An intelligent pre-positioning and region division module is used to input the multi-scale time-frequency feature set into a preset integrated learning network to obtain a preliminary positioning result, and divide the preliminary positioning result into a credible area and a to-be-verified area through confidence distribution analysis; An iterative convergence positioning module is used to perform waveform edge detection and data-driven detection on each section of the area to be verified, calibrate the traveling wave front moment, and feed the calibration result back to the integrated learning network to update internal parameters. It iterates until the positioning result converges to the fault section and outputs an accurate wave front moment pair; The multi-model fusion and verification module is used to calculate the multi-dimensional positioning results based on the precise wave head moment pair, combined with energy distribution analysis and fault propagation constraints, and perform result consistency verification. The verification results are fed back to update the parameter configuration of the integrated learning network and output a comprehensive fault location report.

[0075] It should be noted that the functional division and information exchange between the above modules are logical. Physically, they can be integrated into the same software platform or deployed in a distributed manner. The connections between them represent data flow and control flow, and are intended to collaboratively achieve the objectives of the present invention. The above description is merely an exemplary embodiment of the present invention and is not intended to limit the scope of protection of the present invention.

Claims

1. A method for locating feeder faults in high and low voltage switchgear based on transient traveling waves, characterized in that: The method comprises: Synchronously collect broadband voltage and current traveling wave signals of the fault feeder, acoustic emission and ultrasonic signals inside the switchgear, construct a signal cross matrix, perform signal correlation analysis on the signal cross matrix, adjust modal weights, and generate a weighted transient synchronization feature matrix; Performing a two-dimensional time-frequency transformation on the weighted transient synchronization feature matrix to obtain a transient spectrum time sequence diagram, performing real-time feature scoring on the transient spectrum time sequence diagram, adjusting the weighted transient synchronization feature matrix according to the real-time feature score, and outputting a multi-scale time-frequency feature set; Inputting the multi-scale time-frequency feature set into a preset integrated learning network to obtain a preliminary positioning result, and dividing the preliminary positioning result into a credible area and an area to be verified through confidence distribution analysis; For each section of the area to be verified, waveform edge detection and data-driven detection are performed to calibrate the traveling wave front moment. The calibration results are fed back to the integrated learning network to update internal parameters. The positioning result is converged to the fault section through iteration, and an accurate wave front moment pair is output; Based on the precise wavefront time pair, combined with energy distribution analysis and fault propagation constraints, multi-dimensional positioning results are calculated, and the result consistency is checked. The verification results are fed back to update the parameter configuration of the integrated learning network and output a comprehensive fault location report.

2. The method for locating high and low voltage switchgear feeder faults based on transient traveling waves according to claim 1 is characterized in that: Generating a weighted transient synchronization characteristic matrix includes: Taking the broadband voltage and current traveling wave signals as a first signal group and the acoustic emission and ultrasonic signals as a second signal group, performing correlation analysis on the first signal group and the second signal group to calculate a canonical correlation coefficient; Iteratively correcting each modal weight based on the canonical correlation coefficient, and monitoring the cross-correlation change of the signal cross matrix to generate an iterative weight adjustment sequence; The iterative weight adjustment sequence is applied to perform weighted fusion on the signal cross matrix, and a weighted transient synchronization feature matrix is ​​output.

3. The method for locating high and low voltage switchgear feeder faults based on transient traveling waves according to claim 2 is characterized in that: The output multi-scale time-frequency feature set includes: Adaptively adjusting the time-frequency resolution of the weighted transient synchronization feature matrix, optimizing and selecting transformation parameters through a dynamic window function, and performing a two-dimensional time-frequency transformation process to generate a transient spectrum time sequence diagram; Performing information entropy evaluation on the transient spectrum time sequence diagram to calculate a real-time feature score, quantifying spectrum distribution uniformity and time sequence consistency, and outputting a feature score vector; Based on the feature scoring vector, parameter feedback correction is performed on the weighted transient synchronization feature matrix, and the transient spectrum time series diagram is fused to generate a multi-scale time-frequency feature set.

4. The method for locating high and low voltage switchgear feeder faults based on transient traveling waves according to claim 3 is characterized in that: The output feature score vector includes: Performing multi-scale entropy spectrum analysis on the transient spectrum time sequence diagram, calculating the local information entropy distribution and global spectrum complexity, filtering and removing noise interference, and generating a preliminary entropy spectrum quantization set; For the preliminary entropy spectrum quantization set, the spectrum distribution uniformity and temporal consistency are calculated, and a real-time feature score sequence is generated through weighted aggregation; Vector normalization is performed based on the real-time feature score sequence, and the score distribution is optimized by statistical confidence interval correction to output a feature score vector.

5. The method for locating high and low voltage switchgear feeder faults based on transient traveling waves according to claim 3 is characterized in that: The method further comprises: Performing multi-layer association mapping on the iterative weight adjustment sequence and the feature score vector to generate a weight score joint matrix; Calculating a dynamic optimization factor based on the weight score joint matrix, analyzing the signal modality and time-frequency characteristics for vector-level enhancement, and outputting an enhanced weight vector set; The enhanced weight vector set is injected into the integrated learning network parameters, and the internal weights and bias terms are optimized through gradient guidance to generate a fusion feature optimization map.

6. The method for locating high and low voltage switchgear feeder faults based on transient traveling waves according to claim 1, characterized in that: The dividing the preliminary positioning result into a credible area and an area to be verified includes: Inputting the multi-scale time-frequency feature set into the integrated learning network to calculate and obtain a preliminary positioning result; Analyze the confidence probability distribution based on the preliminary positioning results, perform distribution entropy calculation to quantify the uncertainty level, and output a confidence distribution map; Based on the confidence distribution map, a dynamic verification priority is assigned to each area, and the trusted area and the area to be verified are divided.

7. The method for locating high and low voltage switchgear feeder faults based on transient traveling waves according to claim 1, characterized in that: The calculation to obtain the preliminary positioning result includes: Obtain historical fault recording data of high and low voltage switchgear and historical power grid physical simulation data to generate a historical fault data set; Taking multi-scale time-frequency features as input and actual positions as output, a neural network model is established and trained using the historical fault data set to obtain an integrated learning network; The multi-scale time-frequency feature set is input into the integrated learning network, and a preliminary positioning result is output.

8. The method for locating high and low voltage switchgear feeder faults based on transient traveling waves according to claim 1, characterized in that: The output precise wave front moment pair includes: Perform preliminary dual-path calibration for each section of the area to be verified, perform traveling wave edge detection and data-driven detection in parallel, cross-validate the calibration results of the two paths, and output the initial wave front moment; Calculating an intermediate positioning result for the initial wave front moment, and evaluating a confidence gradient of the integrated learning network on the intermediate positioning result; Using the confidence gradient, performing feedback update on internal parameters of the ensemble learning network; Iteratively perform the preliminary positioning and the confidence gradient evaluation until the drift of the positioning result and the confidence gain of the integrated learning network are both lower than a preset convergence threshold, and output the wave head moment of the positioning result in the converged state as an accurate wave head moment pair.

9. The method for locating feeder faults of high and low voltage switch cabinets based on transient traveling waves according to claim 5, characterized in that: The output comprehensive fault location report includes: Based on the precise wavefront moment, multi-terminal ranging and energy distribution analysis are performed, and the fusion feature optimization map is incorporated to perform propagation path constraint correction to generate a preliminary multi-dimensional positioning result; Applying a cross-consistency check to the preliminary multi-dimensional positioning results, and feeding the check results back to the integrated learning network to update parameter configuration and output an optimized positioning result set; The positioning results in the optimized positioning result set are weightedly fused to calculate the fault location coordinates and the posterior probability, and the fault location coordinates and the posterior probability are packaged into the comprehensive fault location report for output.

10. A system for locating a high- and low-voltage switchgear feeder fault based on transient traveling waves, applied to a method for locating a high- and low-voltage switchgear feeder fault based on transient traveling waves according to any one of claims 1 to 9, characterized in that: The system comprises: An adaptive signal fusion module is used to synchronously collect broadband voltage and current traveling wave signals from the fault feeder, acoustic emission signals from the switchgear, and ultrasonic signals, construct a signal cross-matrix, perform signal correlation analysis on the signal cross-matrix, adjust modal weights, and generate a weighted transient synchronization feature matrix. a time-frequency feature optimization module, configured to perform a two-dimensional time-frequency transformation on the weighted transient synchronization feature matrix to obtain a transient spectrum time sequence diagram, perform real-time feature scoring on the transient spectrum time sequence diagram, adjust the weighted transient synchronization feature matrix according to the real-time feature scoring, and output a multi-scale time-frequency feature set; An intelligent pre-positioning and region division module is used to input the multi-scale time-frequency feature set into a preset integrated learning network to obtain a preliminary positioning result, and divide the preliminary positioning result into a credible area and a to-be-verified area through confidence distribution analysis; An iterative convergence positioning module is used to perform waveform edge detection and data-driven detection on each section of the area to be verified, calibrate the traveling wave front moment, and feed the calibration result back to the integrated learning network to update internal parameters. It iterates until the positioning result converges to the fault section and outputs an accurate wave front moment pair; The multi-model fusion and verification module is used to calculate the multi-dimensional positioning results based on the precise wave head moment pair, combined with energy distribution analysis and fault propagation constraints, and perform result consistency verification. The verification results are fed back to update the parameter configuration of the integrated learning network and output a comprehensive fault location report.

Images