Method, system and equipment for quickly diagnosing lightning stroke short-circuit fault of power transmission line and medium

Abstract

The invention discloses a power transmission line lightning stroke short circuit fault rapid diagnosis method, system and device and a medium, and the method comprises the steps: obtaining single-end fault recording data of a power transmission line, carrying out the denoising processing of the recording data, and carrying out the detection of a fault starting moment, and obtaining an effective fault transient signal with the fault starting moment as a reference; extracting a plurality of basic electrical features, and constructing composite features based on the basic electrical features to form an initial feature set; performing chi-square test and recursive feature elimination on the initial feature set in sequence, and performing two-stage feature selection to obtain an optimized feature subset; and taking the optimal feature subset as input, sending the optimal feature subset into a pre-trained multi-stage cascade classifier, and outputting a comprehensive diagnosis result containing a fault phase, a fault property, an impedance grade and a fault reason by the classifier. According to the method, high-precision, high-robustness, rapid and explainable integrated fault comprehensive judgment is realized.

Description

Technical Field

[0001] This invention relates to the field of power system fault monitoring and protection technology, and in particular to a method, system, equipment and medium for rapid diagnosis of lightning short-circuit faults in transmission lines. Background Technology

[0002] As the backbone of the power system, the safe and stable operation of transmission lines is of paramount importance. Because these lines are mostly erected in complex outdoor environments and are often exposed for extended periods, they are highly susceptible to damage from natural elements such as lightning strikes. Lightning strikes are frequently one of the leading causes of short-circuit faults in transmission lines, potentially damaging equipment, causing line tripping and power outages, and in severe cases, even threatening the stable operation of the entire power grid.

[0003] Currently, diagnostic methods for transmission line faults mainly fall into the following categories: The first category is based on classical electrical quantity calculations, such as analyzing the changes in voltage and current before and after a fault to determine the fault. While these methods are relatively simple to calculate, their accuracy drops significantly when encountering high-impedance faults or signals with high noise levels. Furthermore, they typically require data synchronization between both ends of the line, which is often difficult to achieve in practical applications. The second category is based on expert systems. Their diagnostic logic comes from expert experience, but the construction of the knowledge base is extremely complex, and its adaptability to new situations is limited. The third category comprises methods based on machine learning models that have emerged in recent years, such as support vector machines and neural networks. While these methods have shown some potential, they also have several problems—most research focuses only on fault location or the identification of a single fault type, failing to integrate key information such as fault nature, impedance level, and especially the cause of lightning strikes for diagnosis; model feature engineering often relies on manual design, lacking systematic optimization; in addition, many models are not robust enough to noise and asynchronous data, and are often "black box" models, with opaque decision-making processes that hinder understanding and adoption by on-site maintenance personnel. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method, system, equipment, and medium for rapid diagnosis of lightning short-circuit faults in transmission lines, addressing the problems of insufficient diagnostic comprehensiveness, weak anti-interference capability, lack of versatility, and poor interpretability in existing technologies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for rapid diagnosis of lightning short-circuit faults in transmission lines, comprising: Acquire single-end fault waveform data of the transmission line, and perform noise reduction processing and fault initiation time detection on the waveform data to obtain an effective fault transient signal based on the fault initiation time. Based on the effective fault transient signal, multiple basic electrical features are extracted, and composite features are constructed based on the basic electrical features to form an initial feature set; The initial feature set is subjected to chi-square test and recursive feature elimination in sequence to perform two-stage feature selection and obtain an optimized feature subset. The optimal feature subset is used as input and fed into a pre-trained multi-level cascaded classifier. The classifier outputs a comprehensive diagnostic result that includes fault phase, fault nature, impedance level and fault cause.

[0007] As a preferred embodiment of the rapid diagnosis method for lightning short-circuit faults in transmission lines according to the present invention, the noise reduction processing and fault initiation time detection include: A low-pass filter is used to suppress noise in the recorded waveform data; The initial change times of current and voltage mutations are calculated based on the least squares estimation algorithm. Centered on the initial change moment, a time window containing the brief period before the fault and the transient process after the fault is extracted as the effective fault transient signal.

[0008] As a preferred embodiment of the rapid diagnosis method for lightning short-circuit faults in transmission lines according to the present invention, the two-stage feature selection includes: The first stage uses chi-square test to screen candidate features that are more correlated with lightning strike fault categories than a preset threshold. In the second stage, the candidate features are input into the recursive feature elimination model, sorted according to the contribution of each feature to the classification performance, redundant features are eliminated one by one, and the resulting optimized feature subset is retained.

[0009] The beneficial effects of this preferred technical solution are that, through a two-stage feature selection combining chi-square test and recursive feature elimination, redundant and weakly correlated features are effectively eliminated, thereby reducing model complexity while improving the accuracy and generalization ability of fault diagnosis.

[0010] As a preferred embodiment of the rapid diagnosis method for lightning short-circuit faults in transmission lines according to the present invention, the multi-level cascaded classifier includes: The first-level classifier distinguishes between single-phase, two-phase, and three-phase faults. The second-level classifier determines whether a fault is a ground fault based on the fault phase output result. The third-level classifier classifies low-impedance faults into high-impedance faults based on the grounding type. The fourth-level classifier combines the impedance level, the fault phase, and external environmental information to determine whether the fault was caused by lightning.

[0011] As a preferred embodiment of the rapid diagnosis method for lightning-induced short-circuit faults in transmission lines according to the present invention, the external environmental information includes geographical area information and seasonal information, wherein: The seasonal information is divided into spring, summer, autumn, and winter based on the month corresponding to the time of the fault occurrence; The geographical area information is determined based on the geographical location of the line towers; The seasonal information and geographical region information are input as additional input features to the fourth-level classifier.

[0012] The beneficial effect of this preferred technical solution is that by introducing geographical area information and seasonal information as additional input features into the lightning strike discrimination process, the spatiotemporal distribution pattern of lightning strike faults is fully integrated, significantly improving the accuracy and reliability of fault cause identification.

[0013] As a preferred embodiment of the rapid diagnosis method for lightning short-circuit faults in transmission lines according to the present invention, the extraction of multiple basic electrical features includes: Based on the original single-end fault recording data of the transmission line, through noise reduction processing and fault initiation time detection, an effective fault transient signal covering the time period before and after the fault is obtained with the fault initiation time as the reference. Based on the effective fault transient signal, the power frequency component is extracted by Fourier transform to obtain the fundamental amplitude and fundamental phase angle of the three-phase current and voltage. Based on the fundamental amplitude and phase angle of the three-phase current and voltage, the zero-sequence component is synthesized by the symmetrical component method, and the root mean square value of the zero-sequence component is calculated to obtain the effective values ​​of the zero-sequence current and zero-sequence voltage. Based on the effective fault transient signal, which is divided into a pre-fault sub-window and a post-fault sub-window according to the fault initiation time, the change rate of the power frequency component before and after the fault is obtained by extracting the fundamental frequency and calculating the relative change rate of the pre-fault sub-window and the post-fault sub-window respectively.

[0014] As a preferred embodiment of the rapid diagnosis method for lightning short-circuit faults in transmission lines according to the present invention, the composite structural features include: Based on the fundamental amplitudes of the three-phase current and the three-phase voltage in the basic electrical characteristics, the ratio characteristics between the current and voltage of each phase are obtained by calculating the ratio between the amplitude of the current in the same phase or across phases. Based on the fundamental amplitudes of the three-phase current, the fundamental amplitudes of the three-phase voltage, the fundamental phase angles of the three-phase current and the fundamental phase angles of the three-phase voltage, positive-sequence components, negative-sequence components and zero-sequence components are synthesized through symmetrical component transformation. The information entropy of the energy distribution of the reconstructed waveforms of the positive-sequence components, negative-sequence components and zero-sequence components within the effective fault transient signal time window is calculated to obtain the energy entropy of the positive-sequence components, the energy entropy of the negative-sequence components and the energy entropy of the zero-sequence components. Based on the fault phase information indicated by the effective value of the zero-sequence current and the ratio characteristics between the current and voltage of each phase, the absolute value of the fault phase current in the time domain is obtained by extracting the original current waveform of the corresponding fault phase from the effective fault transient signal and performing absolute value integration over a time period after the fault initiation time. Based on the original current waveform or original voltage waveform in the effective fault transient signal, detail coefficients of multiple frequency bands are obtained through multi-scale wavelet decomposition, and the ratio of the energy of the detail coefficient of each frequency band to the total energy of the detail coefficients of all frequency bands is calculated to obtain the energy proportion of wavelet coefficients in different frequency bands.

[0015] Secondly, the present invention provides a rapid diagnostic system for lightning short-circuit faults in power transmission lines, comprising: The data preprocessing module is used to acquire single-end fault waveform data of the transmission line, and to perform noise reduction processing and fault initiation time detection on the waveform data to obtain an effective fault transient signal based on the fault initiation time. The feature construction module is used to extract multiple basic electrical features based on the effective fault transient signal, and construct composite features based on the basic electrical features to form an initial feature set; The feature optimization module is used to sequentially perform chi-square test and recursive feature elimination on the initial feature set, and perform two-stage feature selection to obtain an optimized feature subset; The intelligent diagnostic module is used to take the optimal feature subset as input and feed it into a pre-trained multi-level cascaded classifier. The classifier outputs a comprehensive diagnostic result that includes fault phase, fault nature, impedance level and fault cause.

[0016] Thirdly, the present invention provides an electronic device, comprising: Memory, used to store programs; A processor is configured to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the rapid diagnosis method for lightning short-circuit faults in transmission lines.

[0017] Fourthly, the present invention provides a computer-readable storage medium, comprising: when the program is executed by a processor, the steps of implementing the rapid diagnosis method for lightning short-circuit faults in transmission lines.

[0018] The beneficial effects of this invention are as follows: By employing single-ended fault recording data combined with denoising processing and precise fault initiation time detection techniques, this invention achieves effective purification and temporal alignment of the original signal, providing a high signal-to-noise ratio transient data foundation for subsequent feature extraction; by constructing a high-dimensional initial feature set combining basic electrical features and composite features, and introducing a two-stage joint feature selection technique of chi-square test and recursive feature elimination, it achieves efficient removal of redundant features and accurate retention of discriminative features, significantly improving the model's generalization ability and computational efficiency; by designing a multi-level cascaded classifier structure and integrating geographical area information and seasonal information as external environmental features in the lightning strike discrimination stage, it achieves step-by-step interpretable diagnosis of fault phase, grounding type, impedance level, and fault cause (especially lightning strike), effectively improving the accuracy and engineering practicality of lightning short-circuit fault identification. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a basic flowchart illustrating a rapid diagnosis method for lightning-induced short-circuit faults in power transmission lines, provided as an embodiment of the present invention. Detailed Implementation

[0020] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0021] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for rapid diagnosis of lightning short-circuit faults in transmission lines is provided, comprising: S100: Acquire single-end fault waveform data of the transmission line, and perform noise reduction processing and fault start time detection on the waveform data to obtain an effective fault transient signal based on the fault start time. S200: Based on the effective fault transient signal, extract multiple basic electrical features, and construct composite features based on the basic electrical features to form an initial feature set; S300: Perform chi-square test and recursive feature elimination sequentially on the initial feature set to conduct two-stage feature selection and obtain an optimized feature subset; S400: The optimal feature subset is taken as input and fed into a pre-trained multi-level cascaded classifier. The classifier outputs a comprehensive diagnostic result that includes fault phase, fault nature, impedance level and fault cause.

[0022] It should be noted that existing technologies face a series of challenges in the operation of lightning short-circuit fault diagnosis for transmission lines. These include: traditional methods based on two-terminal electrical quantities rely on high-precision synchronous sampling, which is easily affected by communication delays or equipment out-of-synchronization in actual engineering, resulting in large positioning errors; single machine learning models often only focus on a certain link in fault classification or ranging, lacking the ability to systematically and jointly distinguish fault phase, grounding type, impedance level, and cause; feature engineering relies heavily on human experience, without scientifically screening high-dimensional features, easily introducing redundancy and noise, reducing model robustness; in addition, most methods treat lightning strikes as other short-circuit faults, failing to integrate the unique spatiotemporal distribution patterns of lightning strikes (such as regional thunderstorm characteristics and seasonal correlation), resulting in insufficient accuracy in fault cause identification; finally, existing solutions generally have poor interpretability, making it difficult to meet the requirements of power system operation and maintenance for transparency in the diagnostic process and credibility of results.

[0023] Therefore, in response to the problems of insufficient diagnostic comprehensiveness, weak anti-interference ability, weak versatility and poor interpretability of the existing technologies, the S100-S400 steps are used to achieve high-precision, fast, robust and interpretable integrated diagnosis of lightning short-circuit faults in transmission lines. This is achieved by integrating single-end waveform signal processing, two-stage feature optimization and multi-level interpretable classification architecture, and introducing geographical area and seasonal information to assist in lightning strike identification.

[0024] Example 2, this is an embodiment of the present invention, which provides a rapid diagnosis method for lightning short-circuit faults in transmission lines based on the previous embodiment, including: In this embodiment of the application, the single-end fault recording data of the transmission line in step S100 comes from a CSV format file recorded by the relay protection device, which includes timestamps, sampling sequences of three-phase voltage and three-phase current; this data format is a commonly used standard in field engineering, ensuring that the method has good compatibility and deployability.

[0025] In this embodiment of the application, the noise reduction filter in step S100 is a second-order Butterworth low-pass filter with a cutoff frequency of 100 Hz to filter the three-phase voltage and current signals in the recorded waveform data, so as to suppress high-frequency noise and retain the key transient frequency components of the lightning strike fault.

[0026] In one optional implementation, the denoising filter type in step S100 can also perform multi-layer wavelet decomposition on the three-phase voltage and current signals in the recorded waveform data, apply soft thresholding to the high-frequency wavelet coefficients to suppress noise, and then obtain the denoised signal through wavelet reconstruction.

[0027] In an optional implementation, the noise reduction filter type in step S100 can also use the voltage or current signal of the steady-state section before the fault as a reference input, and use the LMS or Kalman adaptive algorithm to estimate and filter out the noise components in the waveform data in real time, and output the noise-reduced three-phase electrical quantities.

[0028] In this embodiment, the normalization process for the denoised waveform signal includes discarding the data of the first power frequency cycle before the fault, and using the data of the subsequent two complete cycles to calculate the root mean square values ​​of the three-phase voltage and current, respectively, and using these as the normalization reference. This process makes the method applicable to transmission lines of different voltage levels, significantly improving its versatility.

[0029] In this embodiment of the application, when fault waveform data from both ends of the line are acquired simultaneously, the end with the larger change in current amplitude is preferentially selected as the single-ended analysis object to ensure that the selected signal has more significant fault characteristics and improve diagnostic reliability.

[0030] In this embodiment of the application, the noise reduction processing and fault start time detection in step S100 include: A low-pass filter is used to suppress noise in the recorded waveform data; The initial change times of current and voltage mutations are calculated based on the least squares estimation algorithm. Centered on the initial change moment, a time window containing the brief period before the fault and the transient process after the fault is extracted as the effective fault transient signal.

[0031] In this embodiment, the fault initiation time detection in step S100 is based on a least squares estimation algorithm. This involves fitting the current signal within a sliding window of one power frequency cycle length and calculating the residual sequence between the reconstructed signal and the measured signal. When the residual first exceeds a dynamic threshold... The time is determined as the start time of the fault, where It is a residual sequence. This strategy is highly robust to noise, averaging the absolute deviation.

[0032] In an optional implementation, the fault initiation time detection in step S100 can also perform multi-scale wavelet decomposition on the denoised current signal, extract the modulus maxima sequence in the high-frequency subband, and determine the time point corresponding to the first modulus maxima with significant amplitude and continuous existence as the fault initiation time.

[0033] In an optional implementation, the fault start time detection in step S100 can also calculate the output of the differential energy operator (such as the Teager-Kaiser energy operator) of the denoised current signal. When the energy value exceeds the preset threshold for the first time, the corresponding time is determined as the fault start time.

[0034] In this embodiment of the application, the fault initiation time is determined. Then, capture the window before the failure. After the failure window The data is used for phasor estimation, where For the estimated time of failure, The timeframe is approximately two cycles, or 40 ms. This allows for the estimation of voltage and current phasors before and after the fault. This window length effectively captures the transient characteristics of the fault.

[0035] Before the malfunction: After the malfunction: in, For the estimated time of failure, This is the phasor evaluation window.

[0036] In this embodiment of the application, step S200 extracts several basic electrical features, including: Based on the original single-end fault recording data of the transmission line, through noise reduction processing and fault initiation time detection, an effective fault transient signal covering the time period before and after the fault is obtained with the fault initiation time as the reference. Based on the effective fault transient signal, the power frequency component is extracted by Fourier transform to obtain the fundamental amplitude and fundamental phase angle of the three-phase current and voltage. Based on the fundamental amplitude and phase angle of the three-phase current and voltage, the zero-sequence component is synthesized by the symmetrical component method, and the root mean square value of the zero-sequence component is calculated to obtain the effective values ​​of the zero-sequence current and zero-sequence voltage. Based on the effective fault transient signal, which is divided into a pre-fault sub-window and a post-fault sub-window according to the fault initiation time, the change rate of the power frequency component before and after the fault is obtained by extracting the fundamental frequency and calculating the relative change rate of the pre-fault sub-window and the post-fault sub-window respectively.

[0037] In this embodiment of the application, constructing the composite feature in step S200 includes: Based on the fundamental amplitudes of the three-phase current and the three-phase voltage in the basic electrical characteristics, the ratio characteristics between the current and voltage of each phase are obtained by calculating the ratio between the amplitude of the current in the same phase or across phases. Based on the fundamental amplitudes of the three-phase current, the fundamental amplitudes of the three-phase voltage, the fundamental phase angles of the three-phase current and the fundamental phase angles of the three-phase voltage, positive-sequence components, negative-sequence components and zero-sequence components are synthesized through symmetrical component transformation. The information entropy of the energy distribution of the reconstructed waveforms of the positive-sequence components, negative-sequence components and zero-sequence components within the effective fault transient signal time window is calculated to obtain the energy entropy of the positive-sequence components, the energy entropy of the negative-sequence components and the energy entropy of the zero-sequence components. Based on the fault phase information indicated by the effective value of the zero-sequence current and the ratio characteristics between the current and voltage of each phase, the absolute value of the fault phase current in the time domain is obtained by extracting the original current waveform of the corresponding fault phase from the effective fault transient signal and performing absolute value integration over a time period after the fault initiation time. Based on the original current waveform or original voltage waveform in the effective fault transient signal, detail coefficients of multiple frequency bands are obtained through multi-scale wavelet decomposition, and the ratio of the energy of the detail coefficient of each frequency band to the total energy of the detail coefficients of all frequency bands is calculated to obtain the energy proportion of wavelet coefficients in different frequency bands.

[0038] In this embodiment of the application, after the initial feature set is formed in step S200, all features are subjected to z-score standardization, that is, for each feature dimension, the mean and standard deviation in the training samples are centered and scaled to eliminate the difference in units and ensure the stability and fairness of subsequent classifier training.

[0039] In this embodiment of the application, z-score normalization is represented as follows: in, Let i be the standardized value of the i-th feature on the j-th sample. Let i be the feature value of the j-th instance. Features The sample mean, Let be the sample standard deviation of feature i.

[0040] In this embodiment of the application, the two-stage feature selection in step S300 includes: The first stage uses chi-square test to screen candidate features that are more correlated with lightning strike fault categories than a preset threshold. In the second stage, the candidate features are input into the recursive feature elimination model, sorted according to the contribution of each feature to the classification performance, redundant features are eliminated one by one, and the resulting optimized feature subset is retained.

[0041] In this embodiment of the application, the two-stage feature selection in step S300 uses classification accuracy as the core evaluation index. Accuracy is defined as the ratio of the sum of true positives and true negatives to the total number of samples. In the chi-square test stage, different numbers of features are traversed to select the candidate set that optimizes the accuracy. In the recursive feature elimination stage, the neighborhood of the candidate set is further searched, and the optimal feature subset is dynamically updated to achieve data-driven adaptive screening.

[0042] In an optional implementation, the two-stage feature selection in step S300 can also train a random forest or gradient boosting tree model on the initial feature set, sort each feature according to its contribution to the classification of lightning strike faults (such as reduction of average impurity or ranking importance), and select the top K features with the highest importance as the optimized feature subset.

[0043] In an optional implementation, the two-stage feature selection in step S300 can also calculate the mutual information between each feature and the lightning fault label to assess the nonlinear correlation, and then input all features into an L1 regularized logistic regression model to automatically filter out a subset of features that have discriminative power for classification using its sparsity.

[0044] In this embodiment of the application, the two-stage feature selection in step S300 adopts the following optimization strategy: In the first stage, for each integer, where is the initial total number of features, the highest-scoring features are selected through chi-square hypothesis testing. A classifier is built based on this subset of features, and its classification accuracy is calculated on the test set. in, For the actual number of cases, The number of true negative cases, The number of false positives The number of false negatives is used; if the current accuracy is better than the best historical value, the best feature set is updated. ; The second phase, within the neighborhood. Inside, for each Select using the Recursive Feature Elimination (RFE) method Find the optimal features, construct a classifier, and re-evaluate the accuracy. If the result is better than the current best, update the feature set. The final result This refers to optimizing the feature subset.

[0045] In this embodiment of the application, the multi-level cascaded classifier in step S400 includes: The first-level classifier distinguishes between single-phase, two-phase, and three-phase faults. The second-level classifier determines whether a fault is a ground fault based on the fault phase output result. The third-level classifier classifies low-impedance faults into high-impedance faults based on the grounding type. The fourth-level classifier combines the impedance level, the fault phase, and external environmental information to determine whether the fault was caused by lightning.

[0046] In this embodiment of the application, each classifier in the multi-level cascaded classifier in step S400 is constructed using a decision tree model. Four cascaded decision tree classifiers are constructed in sequence: the first level identifies the fault phase, the second level determines whether it is a ground fault, the third level classifies the impedance level, and the fourth level combines the output of the previous level with geographical and seasonal information to determine whether it is caused by lightning. The ground type classifier can effectively distinguish between line-to-line faults and line-to-ground faults. The impedance level classifier divides the fault impedance into three categories: low impedance (0–20Ω), medium impedance (20–40Ω), and high impedance (>40Ω).

[0047] In an optional implementation, the model type of the multi-level classifier in step S400 can also be four cascaded support vector machine classifiers to be constructed in sequence, which are used to distinguish the fault phase, grounding type, impedance level and lightning strike cause, respectively. Each level is trained and predicted based on the output results of the previous level and the optimized feature subset, and a radial basis function (RBF) kernel is used to handle the nonlinear decision boundary.

[0048] In an optional implementation, the model type of the multi-level classifier in step S400 can also be four cascaded support vector machine (SVM) classifiers constructed sequentially, which are used for fault phase identification, ground fault judgment, impedance level classification and lightning strike cause discrimination, respectively. Each SVM is trained based on the optimized feature subset and the output results of the previous stage, and the RBF kernel function is used to adapt to the nonlinear classification requirements.

[0049] In this embodiment of the application, the external environment information in step S400 includes geographical area information and seasonal information, wherein: Seasonal information is divided into spring, summer, autumn, and winter based on the month corresponding to the time of the fault. Geographic area information is determined based on the geographical location of the line towers; Seasonal information and geographical region information are input as additional input features to the fourth-level classifier.

[0050] In this embodiment, the training data for each classifier comes from the equivalent model of the transmission line established by the EMTP-ATP electromagnetic transient simulation platform. A total of 3,000 simulation cases covering different fault locations, types, impedances and phases are generated for training and verification. The fault cause classifier is further calibrated by combining 46 real lightning strike fault cases, and its lightning strike identification accuracy reaches 80%.

[0051] In this embodiment of the application, when a new fault event occurs, the system first reads and preprocesses the waveform data, extracts and standardizes the predefined optimized feature subset, and inputs it sequentially into the trained phase, grounding and impedance level classifiers. The system then integrates the output results with external environmental information, and finally generates a structured diagnostic report containing the fault location, type, impedance and cause by the fault cause classifier.

[0052] In this embodiment, the multi-level cascaded classifier adopts a hierarchically dependent dataflow architecture: the output of the previous level classifier not only serves as the final determination for the current diagnostic task, but also as a structured input feature passed to the next level classifier, forming the input vector of the next level together with the optimized feature subset. Specifically, the fault phase category (e.g., single-phase A, two-phase BC, etc.) output by the first level classifier is encoded as a discrete variable and input into the second level classifier to assist in determining whether it is a ground fault; the grounding type (grounded / ungrounded) output by the second level is further used as a feature in the third level impedance level classification; the impedance level category output by the third level, together with the original optimized features and external environmental information (geographical region, season), is input into the fourth level classifier to finally determine whether the fault is caused by lightning. This cascade mechanism ensures that each level of diagnostic task shares the underlying electrical features while fully utilizing the semantic information of the preceding diagnostic conclusions, realizing hierarchical and collaborative reasoning of fault attributes.

[0053] In this embodiment, energy entropy is used to characterize the concentration of energy distribution of a certain sequence component (positive sequence, negative sequence, or zero sequence) within the effective fault transient signal time window. The calculation method is as follows: First, the reconstructed time-domain waveform of the sequence component is divided into several continuous time segments, and the energy of each segment is calculated, which is the sum of the squares of the signal sample values ​​in the segment. Then, the energy of each segment is divided by the total energy to obtain the proportion of the energy of each segment to the total energy, which is used as the probability weight of the segment. Finally, based on the probability weights of all segments, according to the definition of information entropy, the negative value of the sum of the products of each weight and its logarithm is calculated. The result is the energy entropy of the sequence component. The smaller the entropy value, the more concentrated the energy; the larger the entropy value, the more dispersed the energy.

[0054] In the embodiments of this application, when performing multi-scale wavelet decomposition on the original current or voltage waveform, the mother wavelet types used include, but are not limited to, Daubechies series wavelets (such as db4), Symlets series wavelets (such as sym5), or Coiflets series wavelets. Among them, the db4 wavelet is preferred for transient feature extraction due to its good tight support and approximate symmetry. In practical applications, the optimal type can be selected from the above mother wavelets according to the signal characteristics and noise level to balance the time-frequency localization capability and reconstruction accuracy.

[0055] Example 3 is an embodiment of the present invention. This embodiment differs from the first embodiment in that it provides a rapid diagnostic system for lightning short-circuit faults in power transmission lines.

[0056] It should be noted that the technical solution of the rapid diagnosis system for lightning short-circuit faults in transmission lines is based on the same concept as the technical solution of the rapid diagnosis method for lightning short-circuit faults in transmission lines described above. For details not described in detail in the technical solution of the rapid diagnosis system for lightning short-circuit faults in transmission lines in this embodiment, please refer to the description of the technical solution of the rapid diagnosis method for lightning short-circuit faults in transmission lines described above.

[0057] This embodiment provides a rapid diagnostic system for lightning short-circuit faults in power transmission lines, comprising: The data preprocessing module is used to acquire single-end fault waveform data of the transmission line, and to perform noise reduction processing and fault initiation time detection on the waveform data to obtain an effective fault transient signal based on the fault initiation time. The feature construction module is used to extract multiple basic electrical features based on the effective fault transient signal, and construct composite features based on the basic electrical features to form an initial feature set; The feature optimization module is used to sequentially perform chi-square test and recursive feature elimination on the initial feature set, and perform two-stage feature selection to obtain an optimized feature subset; The intelligent diagnostic module is used to take the optimal feature subset as input and feed it into a pre-trained multi-level cascaded classifier. The classifier outputs a comprehensive diagnostic result that includes fault phase, fault nature, impedance level and fault cause.

[0058] This embodiment also provides an electronic device applicable to a rapid diagnosis method for lightning short-circuit faults in power transmission lines, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a rapid diagnosis method for lightning-induced short-circuit faults in power transmission lines, as proposed in the above embodiments.

[0059] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a method for rapid diagnosis of lightning short-circuit faults in transmission lines as proposed in the above embodiments.

[0060] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for rapid diagnosis of lightning short-circuit faults in transmission lines proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0061] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0062] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A rapid diagnostic method for lightning-induced short-circuit faults in transmission lines, characterized in that, include: Acquire single-end fault waveform data of the transmission line, and perform noise reduction processing and fault initiation time detection on the waveform data to obtain an effective fault transient signal based on the fault initiation time. Based on the effective fault transient signal, multiple basic electrical features are extracted, and composite features are constructed based on the basic electrical features to form an initial feature set; The initial feature set is subjected to chi-square test and recursive feature elimination in sequence to perform two-stage feature selection and obtain an optimized feature subset. The optimal feature subset is used as input and fed into a pre-trained multi-level cascaded classifier. The classifier outputs a comprehensive diagnostic result that includes fault phase, fault nature, impedance level and fault cause.

2. The rapid diagnosis method for lightning short-circuit faults in transmission lines as described in claim 1, characterized in that: The noise reduction process and fault initiation time detection include: A low-pass filter is used to suppress noise in the recorded waveform data; The initial change times of current and voltage mutations are calculated based on the least squares estimation algorithm. Centered on the initial change moment, a time window containing the brief period before the fault and the transient process after the fault is extracted as the effective fault transient signal.

3. The rapid diagnosis method for lightning short-circuit faults in transmission lines as described in claim 1 or 2, characterized in that: The two-stage feature selection includes: The first stage uses chi-square test to screen candidate features that are more correlated with lightning strike fault categories than a preset threshold. In the second stage, the candidate features are input into the recursive feature elimination model, sorted according to the contribution of each feature to the classification performance, redundant features are eliminated one by one, and the resulting optimized feature subset is retained.

4. The rapid diagnosis method for lightning short-circuit faults in transmission lines as described in claim 3, characterized in that: The multi-level cascaded classifier includes: The first-level classifier distinguishes between single-phase, two-phase, and three-phase faults. The second-level classifier determines whether a fault is a ground fault based on the fault phase output result. The third-level classifier classifies low-impedance faults into high-impedance faults based on the grounding type. The fourth-level classifier combines the impedance level, the fault phase, and external environmental information to determine whether the fault was caused by lightning.

5. The rapid diagnosis method for lightning short-circuit faults in transmission lines as described in claim 4, characterized in that: The external environmental information includes geographical region information and seasonal information, wherein: The seasonal information is divided into spring, summer, autumn, and winter based on the month corresponding to the time of the fault occurrence; The geographical area information is determined based on the geographical location of the line towers; The seasonal information and geographical region information are input as additional input features to the fourth-level classifier.

6. The rapid diagnosis method for lightning short-circuit faults in transmission lines as described in claim 5, characterized in that: The extraction of multiple basic electrical features includes: Based on the original single-end fault recording data of the transmission line, through noise reduction processing and fault initiation time detection, an effective fault transient signal covering the time period before and after the fault is obtained with the fault initiation time as the reference. Based on the effective fault transient signal, the power frequency component is extracted by Fourier transform to obtain the fundamental amplitude and fundamental phase angle of the three-phase current and voltage. Based on the fundamental amplitude and phase angle of the three-phase current and voltage, the zero-sequence component is synthesized by the symmetrical component method, and the root mean square value of the zero-sequence component is calculated to obtain the effective values ​​of the zero-sequence current and zero-sequence voltage. Based on the effective fault transient signal, which is divided into a pre-fault sub-window and a post-fault sub-window according to the fault initiation time, the change rate of the power frequency component before and after the fault is obtained by extracting the fundamental frequency and calculating the relative change rate of the pre-fault sub-window and the post-fault sub-window respectively.

7. The rapid diagnosis method for lightning short-circuit faults in transmission lines as described in claim 6, characterized in that: The constructed composite features include: Based on the fundamental amplitudes of the three-phase current and the three-phase voltage in the basic electrical characteristics, the ratio characteristics between the current and voltage of each phase are obtained by calculating the ratio between the amplitude of the current in the same phase or across phases. Based on the fundamental amplitudes of the three-phase current, the fundamental amplitudes of the three-phase voltage, the fundamental phase angles of the three-phase current and the fundamental phase angles of the three-phase voltage, positive-sequence components, negative-sequence components and zero-sequence components are synthesized through symmetrical component transformation. The information entropy of the energy distribution of the reconstructed waveforms of the positive-sequence components, negative-sequence components and zero-sequence components within the effective fault transient signal time window is calculated to obtain the energy entropy of the positive-sequence components, the energy entropy of the negative-sequence components and the energy entropy of the zero-sequence components. Based on the fault phase information indicated by the effective value of the zero-sequence current and the ratio characteristics between the current and voltage of each phase, the absolute value of the fault phase current in the time domain is obtained by extracting the original current waveform of the corresponding fault phase from the effective fault transient signal and performing absolute value integration over a time period after the fault initiation time. Based on the original current waveform or original voltage waveform in the effective fault transient signal, detail coefficients of multiple frequency bands are obtained through multi-scale wavelet decomposition, and the ratio of the energy of the detail coefficient of each frequency band to the total energy of the detail coefficients of all frequency bands is calculated to obtain the energy proportion of wavelet coefficients in different frequency bands.

8. A rapid diagnostic system for lightning-induced short-circuit faults in transmission lines, employing the method described in any one of claims 1-7, characterized in that, include: The data preprocessing module is used to acquire single-end fault waveform data of the transmission line, and to perform noise reduction processing and fault initiation time detection on the waveform data to obtain an effective fault transient signal based on the fault initiation time. The feature construction module is used to extract multiple basic electrical features based on the effective fault transient signal, and construct composite features based on the basic electrical features to form an initial feature set; The feature optimization module is used to sequentially perform chi-square test and recursive feature elimination on the initial feature set, and perform two-stage feature selection to obtain an optimized feature subset; The intelligent diagnostic module is used to take the optimal feature subset as input and feed it into a pre-trained multi-level cascaded classifier. The classifier outputs a comprehensive diagnostic result that includes fault phase, fault nature, impedance level and fault cause.

9. An electronic device, characterized in that, include: Memory, used to store programs; A processor for loading the program to perform the steps of the method as claimed in any one of claims 1-7.

10. A computer-readable storage medium storing a program, characterized in that, When the program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images