Tree line discharge fault identification method based on fault traveling wave current characteristics

Abstract

The invention relates to the technical field of distribution line fault detection, in particular to a tree line discharge fault identification method based on fault traveling wave current characteristics, which comprises the following steps of: based on a fault traveling wave current signal, performing high-frequency sampling on a current waveform on a distribution line, dividing the signal into a plurality of time slices, the current fluctuation characteristic of each time period is analyzed by calculating the time and frequency component change trend of each fragment, time-frequency changes are matched step by step, and a time-frequency fragment set is generated. The current waveform is sampled at high frequency, and the time-frequency change of the current waveform is analyzed in a segmented manner, so that the current fluctuation characteristics are careful to capture, and the influence of external interference on signals is avoided. By gradually reconstructing the waveform, the smoothness of waveform transition and the integrity of the signal are ensured, and signal distortion caused by waveform discontinuity is avoided. The feature points are screened and corrected to ensure that the points can accurately reflect the actual fault condition, and the possibility of false alarm or missing alarm is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution line fault detection, and in particular to a tree-line discharge fault identification method based on fault traveling wave current characteristics. Background Art

[0002] Overhead transmission channels are an important part of the power system. When 10kV overhead lines pass through mountainous areas, they are easily affected by weather factors such as strong winds and thunderstorms, causing the lines to come into contact with taller tree trunks, inducing power circuit grounding faults in the form of insulation breakdown, wire discharge to trees, and wire breakage and falling. If the tree-line discharge fault is not discovered and eliminated in time, it may evolve into an arc grounding fault, becoming a potential threat to causing wildfires. If the risk is discovered and stopped as early as possible in the early or even embryonic stage of tree-line discharge, it will play a more active preventive role.

[0003] However, when facing complex environments, the current waveform of the existing technology is easily disturbed by external factors, resulting in a decrease in the accuracy of feature point recognition. Due to the lack of continuity of the waveform, the reconstructed current signal may be incomplete or distorted, which increases the difficulty of fault identification and may lead to misjudgment or missed judgment. These deficiencies may result in the failure to detect tree-line discharge faults in a timely manner at an early stage, thereby increasing the risk of accidents. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings existing in the prior art and to propose a tree-line discharge fault identification method based on the fault traveling wave current characteristics.

[0005] In order to achieve the above object, the present invention adopts the following technical solution, a tree-line discharge fault identification method based on fault traveling wave current characteristics, comprising the following steps:

[0006] Based on the fault traveling wave current signal, the current waveform on the distribution line is sampled at high frequency, and the signal is divided into multiple time segments. By calculating the change trend of the time and frequency components of each segment, the current fluctuation characteristics of each period are analyzed, and the time-frequency changes are gradually matched to generate a set of time-frequency segments.

[0007] Based on the set of time-frequency segments, interpolation processing is performed on the transient waveform in each segment, a path with the minimum waveform difference between the segments is selected, a smooth transition of the segments is completed by fitting the waveform, and the complete waveform signal is gradually merged and reconstructed, and a synthetic signal sequence is obtained by reconstructing the segment signals;

[0008] Based on the synthetic signal sequence, the waveform features in the current waveform are screened, the time position and amplitude of the characteristic points in the waveform are extracted, the time difference and amplitude difference between each characteristic point and the adjacent points are calculated one by one, the validity and accuracy of the characteristic points in the waveform are screened, and a key signal characteristic point set is obtained;

[0009] Based on the key signal feature point set, the distribution of feature points in the distribution line current is analyzed, and the positions of the feature points are corrected in sequence by calculating the time difference and amplitude difference of each feature point, the time series distribution of the feature points is optimized, the spatial distribution of the feature points is adjusted, and the optimized signal feature distribution is generated;

[0010] Based on the optimized signal feature distribution, the time domain and frequency domain features of the distribution line fault signal are analyzed, and by comparing the time-frequency characteristics of the feature points one by one, the signals that meet the fault features are classified and labeled to generate a fault identification result.

[0011] Preferably, the time-frequency segment set includes the frequency distribution, phase information and amplitude change of the time segment, the synthetic signal sequence includes the overall trend of the reconstructed waveform, the continuity of the time series and the smoothed amplitude information, the key signal feature point set includes the target time point, amplitude and the difference between the feature points, the optimized signal feature distribution includes the adjusted feature point position, the corrected time interval and the optimized spatial layout, and the fault identification result includes the identified fault type, the time mark and the spatial position of the fault.

[0012] Preferably, the steps of obtaining the time-frequency segment set are specifically as follows:

[0013] Based on the fault traveling wave current signal, the current waveform is sampled at high frequency, the signal is divided into multiple time segments, and a fixed time window is used to extract the time and frequency components of each segment one by one, calculate the time-frequency characteristics, and generate segmented time-frequency characteristics by analyzing the waveform change trend;

[0014] Based on the segmented time-frequency characteristics, the current fluctuation characteristics in each time segment are analyzed, and the frequency relationship between the segments is established by gradually matching the frequency and time between the segments, and the frequency changes within the time period are associated, so as to generate time-frequency association data;

[0015] Based on the time-frequency association data, the frequency and time information of each time segment are integrated, and the association relationship between the segments is merged in an inductive and organized manner. Through the unified integration of the time-frequency information between the segments, a time-frequency segment set is generated.

[0016] Preferably, the steps of obtaining the synthetic signal sequence are specifically as follows:

[0017] Based on the set of time-frequency segments, interpolation processing is performed on the transient waveform in each segment, a waveform difference path between the segments is selected, and the waveform between each segment is gradually and smoothly connected by calculating the minimum difference path, so as to minimize the difference between the segments and generate a waveform interpolation result;

[0018] Based on the waveform interpolation result, each segment waveform after interpolation is smoothed, by gradually merging the waveform differences of adjacent segments, using a weighted smoothing method to process the waveform transition at the segment connection, and arranging the merged segment waveforms to generate a smoothed waveform sequence;

[0019] Based on the smoothed waveform sequence, the reconstructed waveform signals are integrated, and by gradually merging the time series within the segments, the integrated time series are matched with the waveform, and the reconstructed segment signals are synthesized into a continuous waveform signal to generate a synthetic signal sequence.

[0020] Preferably, the steps of acquiring the key signal feature point set are specifically as follows:

[0021] Based on the synthetic signal sequence, key feature points in the reconstructed waveform are screened, the time position and amplitude information of the feature points are extracted one by one, the time position of each feature point is compared with the adjacent points, and the difference between adjacent feature points is calculated to generate a preliminary feature point set;

[0022] Based on the preliminary feature point set, the time difference and amplitude difference between the feature point and the adjacent points are calculated one by one, the valid feature point is determined by cumulative calculation, the valid feature point is matched with the adjacent points, and the difference values ​​between the feature points are screened to generate a valid feature point set;

[0023] Based on the effective feature point set, the effectiveness and accuracy of the feature points are judged by analyzing the time difference and amplitude change between the feature points, and the effective feature points are screened and classified to generate a key signal feature point set.

[0024] Preferably, the steps of obtaining the optimized signal feature distribution are specifically as follows:

[0025] Based on the key signal feature point set, the distribution of feature points in the distribution line current is analyzed, the time distribution data of the feature points are classified and sorted by calculating the time difference of each feature point, and the time difference information of adjacent feature points is integrated to generate a feature time series;

[0026] Based on the feature time series, the positions of the feature points are corrected in sequence, the time distribution and spatial layout of the feature points are gradually adjusted, the corrected feature points are sorted, and the adjusted feature point positions are matched with the time distribution to generate a corrected feature distribution;

[0027] Based on the corrected feature distribution, the time series distribution of the feature points is optimized, the spatial layout of the feature points is analyzed, the time series information of the feature points is gradually adjusted, the optimized feature points and the time series information are unified and integrated to generate an optimized signal feature distribution.

[0028] Preferably, the steps of obtaining the fault identification result are specifically as follows:

[0029] Based on the optimized signal feature distribution, a long short-term memory neural network is used to perform time series data analysis, gradually extract the time feature information of each time series segment, analyze the dynamic change trend of each time feature, and generate time domain fault features;

[0030] Based on the time domain fault characteristics, the frequency domain characteristics of the fault signal are analyzed, the frequency information of the characteristic points are compared section by section, the frequency domain data that meets the fault characteristics are classified and sorted by matching the frequency domain changes of the characteristic points, and the frequency domain characteristics are uniformly summarized to generate the frequency domain fault characteristics;

[0031] Based on the frequency domain fault characteristics, the signals that meet the fault characteristics are uniformly classified and labeled, each labeled fault signal is sorted and classified to generate a fault identification result.

[0032] Preferably, the long short-term memory neural network is according to the formula:

[0033]

[0034] Calculate the time domain fault characteristics, where h t is the hidden state at the current moment, W h is the weight matrix of hidden state and input, h t-1 is the hidden state of the previous moment, x t is the input data at the current moment, b h is the bias term, α t is the dynamically adjusted time delay parameter, δ t is the time series feature weight, γ is the weight coefficient related to the input, λ is the hidden state compensation coefficient, ρ t is the adjustment factor, β t is the environmental noise correction term, ξ is the noise amplification factor, and σ is the activation function.

[0035] Compared with the prior art, the advantages and positive effects of the present invention are:

[0036] The present invention samples the current waveform at high frequency and analyzes its time-frequency changes in segments, so that the current fluctuation characteristics are captured more meticulously and the influence of external interference on the signal is avoided. By gradually reconstructing the waveform, the smoothness of the waveform transition and the integrity of the signal are guaranteed, and signal distortion caused by waveform discontinuity is avoided. The screening and correction of characteristic points ensure that these points can accurately reflect the actual fault situation and reduce the possibility of false alarms or missed alarms. Finally, by accurately comparing and classifying the time-frequency characteristics of the characteristic points, the accuracy and reliability of fault identification are improved, so that early risks can be discovered and stopped in time, reducing potential dangers. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the steps of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0039] In the description of the present invention, it should be understood that the terms length, width, up, down, front, back, left, right, vertical, horizontal, top, bottom, inside, outside, etc. indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific position, be constructed and operated in a specific position, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of multiple is two or more, unless otherwise clearly and specifically defined.

[0040] See also Figure 1 The present invention provides a technical solution, a method for identifying tree-line discharge faults based on fault traveling wave current characteristics, comprising the following steps:

[0041] Based on the fault traveling wave current signal, the current waveform on the distribution line is sampled at high frequency, and the signal is divided into multiple time segments. By calculating the change trend of the time and frequency components of each segment, the current fluctuation characteristics of each period are analyzed, and the time-frequency changes are gradually matched to generate a set of time-frequency segments.

[0042] Based on the set of time-frequency segments, the transient waveform in each segment is interpolated, the path with the minimum waveform difference between the segments is selected, the smooth transition of the segments is completed by fitting the waveform, and the complete waveform signal is gradually merged and reconstructed. The synthetic signal sequence is obtained by reconstructing the segment signals;

[0043] Based on the synthetic signal sequence, the waveform features in the current waveform are screened, the time position and amplitude of the characteristic points in the waveform are extracted, the time difference and amplitude difference between each characteristic point and the adjacent points are calculated one by one, the effectiveness and accuracy of the characteristic points in the waveform are screened, and a set of key signal characteristic points is obtained;

[0044] Based on the key signal feature point set, the distribution of feature points in the distribution line current is analyzed. By calculating the time difference and amplitude difference of each feature point, the position of the feature points is corrected in turn, the time distribution of the feature points is optimized, the spatial distribution of the feature points is adjusted, and the optimized signal feature distribution is generated;

[0045] Based on the optimization of signal feature distribution, the time domain and frequency domain characteristics of the distribution line fault signal are analyzed. By comparing the time-frequency characteristics of the feature points one by one, the signals that meet the fault characteristics are classified and labeled to generate fault identification results.

[0046] The time-frequency segment set includes the frequency distribution, phase information and amplitude change of the time segment. The synthetic signal sequence includes the overall trend of the reconstructed waveform, the continuity of the time series and the smoothed amplitude information. The key signal feature point set includes the target time point, amplitude and the difference between the feature points. The optimized signal feature distribution includes the adjusted feature point position, the corrected time interval and the optimized spatial layout. The fault identification result includes the identified fault type, the time mark and the spatial position of the fault.

[0047] In the fault traveling wave current signal sampling step, the entire fault signal is divided into multiple time segments according to the time axis by performing high-frequency sampling on the current waveform on the distribution line. The length of each segment is dynamically adjusted according to the frequency and time changes of the signal to ensure that the subtle changes of the signal are fully captured. Subsequently, the time and frequency components of each segment are extracted, and the time domain signal is converted into a frequency domain signal through Fourier transform. The frequency distribution and phase information of each time segment are calculated. At the same time, the fluctuation characteristics of the current waveform are captured by calculating the amplitude change, and a time-frequency segment set is generated. The generated time-frequency segment set contains the frequency distribution, phase information and amplitude change of the time segment.

[0048] In the transient waveform interpolation and reconstruction step, the transient waveform in each segment is interpolated based on a set of time-frequency segments. First, the waveform differences between adjacent segments are analyzed and the minimum difference path is calculated to ensure that the key features of the waveform are retained during the segment merging process. The waveform is then smoothed using fitting technology, and the spline interpolation method is used to smoothly transition the waveform within the segment, thereby eliminating the jumping phenomenon at the segment boundaries. Finally, the smoothed waveforms of each segment are gradually merged to reconstruct a complete waveform signal. In this process, the time series and amplitude information of the waveform are linearly or nonlinearly fitted to ensure the continuity and smoothness of the waveform. The generated synthetic signal sequence includes the overall trend of the reconstructed waveform, the continuity of the time series, and the smoothed amplitude information.

[0049] In the waveform feature extraction step, the key feature points in the current waveform are screened based on the synthetic signal sequence to extract the time position and amplitude information of each waveform. The time domain and frequency domain data of the signal are calculated one by one, and the time difference and amplitude difference of each feature point with its adjacent points are compared to screen out representative feature points. The feature points reflect the significant change positions in the waveform and are important indicators of fault signals. The valid feature points are determined by cumulative calculation, and the difference values ​​between the feature points are screened and verified to ensure the validity and accuracy of each feature point. Finally, a key signal feature point set is obtained. The feature point set includes the target time point, amplitude and difference between the feature points. The feature data will be used for further fault identification analysis in subsequent steps to ensure the accuracy and stability of feature extraction.

[0050] In the feature point correction and optimization step, the feature point distribution in the distribution line current is analyzed based on the key signal feature point set, and the time difference and amplitude difference of each feature point are calculated to correct the position of the feature point. In the correction process, the time distribution and spatial layout of the feature points are analyzed, the temporal distribution of the feature points is gradually adjusted, and the temporal and spatial relationships between different feature points are optimized. The feature points are globally optimized by introducing the Lagrange multiplier method to ensure the uniform distribution of the feature points on the time axis. At the same time, the variational method is used to adjust the layout of the feature points in space to achieve the optimal arrangement of the feature points and generate an optimized signal feature distribution, which includes the adjusted feature point positions, the corrected time intervals and the optimized spatial layout. The data will be used for accurate analysis and identification of fault signals in subsequent steps.

[0051] In the fault signal identification and classification step, the time domain and frequency domain characteristics of the distribution line fault signal are analyzed based on the optimized signal feature distribution, the time and frequency characteristics of the feature points are compared one by one, and the signals that meet the fault characteristics are classified and labeled by comparing the distribution of the feature points on the time axis and the response in the frequency domain. The fault type corresponding to each feature point is determined by cross-analyzing the time domain data and frequency domain data of the signal and combining it with the comparison of historical fault data, and a fault identification result is generated. The result includes the identified fault type, the time mark and spatial location of the fault. The generated fault identification result will be used in the early warning and maintenance plan of power grid faults to ensure the stable operation of the power system.

[0052] In this embodiment, the steps of obtaining the time-frequency segment set are specifically as follows:

[0053] Based on the fault traveling wave current signal, the current waveform is sampled at high frequency, the signal is divided into multiple time segments, and a fixed time window is used to extract the time and frequency components of each segment one by one, calculate the time-frequency characteristics, and generate segmented time-frequency characteristics by analyzing the waveform change trend;

[0054] Based on the segmented time-frequency characteristics, the current fluctuation characteristics in each time segment are analyzed. By gradually matching the frequency and time between segments, the frequency changes within the time period are correlated, the frequency relationship between the segments is established, and the time-frequency correlation data is generated.

[0055] Based on the time-frequency correlation data, the frequency and time information of each time segment are integrated, and the correlation between the segments is merged in an inductive and organized manner. Through the unified integration of the time-frequency information between the segments, a time-frequency segment set is generated.

[0056] In the sub-step of obtaining the segmented time-frequency characteristics based on the fault traveling wave current signal, the entire signal is divided into multiple time segments by high-frequency sampling of the current waveform. The length of each segment is determined by a fixed time window to ensure that the signal captured in different time periods can reflect the slight difference in fluctuations. During the sampling process, the data of each time segment is stored in the form of a matrix, the rows of the matrix represent the time points, and the columns represent the sampled current amplitudes. In each time segment, the Fourier transform is performed on the collected current data to convert the time domain signal into a frequency domain signal and extract the frequency component. The extracted frequency component is represented by a data dot matrix in the frequency domain, where each column corresponds to the amplitude intensity at a specific frequency. In order to calculate the time-frequency characteristics, the frequency distribution and amplitude changes in the time segment are further analyzed. By calculating the gradient and amplitude differences of the frequency components in time, the instantaneous change trend in the current signal is identified, and the changes are marked as feature points on the time-frequency diagram to generate segmented time-frequency characteristics. In this sub-step, the generated segmented time-frequency feature data includes the frequency distribution, phase information and amplitude changes of the time segment.

[0057] In the sub-step of analyzing the current fluctuation characteristics based on the segmented time-frequency characteristics and generating time-frequency correlation data, the current fluctuation characteristics in each segment are identified by deeply analyzing the segmented time-frequency characteristics of each time segment. The specific operation includes calculating and comparing the time series data of the frequency and phase in the time segment, and identifying the frequency drift and phase jump phenomena by comparing the frequency distribution and amplitude changes of adjacent time segments, and marking the phenomena as key indicators of current fluctuation. In order to establish the frequency relationship between the segments, the matching degree of the frequency distribution is calculated segment by segment, and the minimum error matching method is used to quantify the frequency relationship between different segments. The segments with the highest matching degree are associated to generate frequency correlation data between the segments. Specifically, by calculating the frequency matching degree, the continuity of the frequency components on the time axis is identified, and the correlation matrix between the segments is constructed. The rows and columns of the matrix correspond to different time segments, and the matrix element values ​​represent the frequency correlation strength between the segments. The time-frequency correlation data finally generated is stored in the form of a two-dimensional array, which contains the frequency matching relationship between each time segment and the corresponding matching degree score.

[0058] In the sub-step of integrating time segment information based on time-frequency correlation data and generating a set of time-frequency segments, the time-frequency correlation data generated in the previous sub-step is summarized and sorted to achieve unified integration of time-frequency information between time segments. The specific process of integration includes merging the frequency distribution and time information of the time segments, and merging the strongly correlated time segments into longer time-frequency segments according to the correlation strength in the time-frequency correlation matrix to reduce the number of time-frequency segments and improve the concentration of analysis. The integration process uses the weighted average method to smooth the frequency distribution of each time segment to ensure that the continuity of frequency information and the consistency of time sequence are retained during the segment merging process. In the merging process, the frequency and time data of different segments are merged by frequency weighted average and time weighted average. The frequency weighted average is used to merge segments with similar frequency distribution, and the time weighted average is used to maintain the balance of time distribution between segments. After the operation, the final generated set of time-frequency segments includes not only the frequency and time information of each time segment, but also the integrated frequency distribution, phase information and time series data. The set is stored in the form of a multidimensional matrix as an important basic data for subsequent fault feature extraction and identification.

[0059] In this embodiment, the steps for obtaining the synthetic signal sequence are specifically as follows:

[0060] Based on the set of time-frequency segments, the transient waveform in each segment is interpolated, the waveform difference path between the segments is selected, and the waveform between each segment is gradually and smoothly connected by calculating the minimum difference path, so as to minimize the difference between the segments and generate the waveform interpolation result;

[0061] Based on the waveform interpolation result, each interpolated waveform is smoothed, and the waveform differences of adjacent segments are gradually merged, and the waveform transition at the segment connection is processed by weighted smoothing. The merged segment waveforms are sorted to generate a smooth waveform sequence;

[0062] Based on the smooth waveform sequence, the reconstructed waveform signal is integrated, and the time series within the fragments are gradually merged, the integrated time series is matched with the waveform, and the reconstructed fragment signals are synthesized into a continuous waveform signal to generate a synthetic signal sequence.

[0063] In the sub-step of interpolation processing based on the set of time-frequency segments, the transient waveform in each segment is analyzed in detail, and the interpolation calculation is performed on each key point of the transient waveform by combining linear interpolation and nonlinear interpolation. First, the waveform data of each segment is decomposed into discrete time points and corresponding amplitude data, which are represented in array form. During the interpolation process, the difference path is determined by calculating the waveform difference between adjacent segments, and the path with the smallest difference is selected for interpolation calculation. The specific operation includes interpolation calculation of the amplitude at each time point, and the piecewise linear interpolation method is used to make the waveform change between each segment smoothly transition to avoid mutations at the connection. Nonlinear interpolation is used to process segments with more drastic changes to ensure that the waveform can be naturally connected in high-frequency changing segments. The generated waveform interpolation results are saved in the form of serialized data, including the interpolation amplitude at each time point and the change trend of the waveform after interpolation.

[0064] In the sub-step of smoothing based on the waveform interpolation result and generating a smoothed waveform sequence, each segment waveform after interpolation is systematically smoothed, the waveform differences between adjacent segments are first accumulated and calculated, and the segments with large differences are weighted, and the waveform mutations at the segment connection are reduced by weighted smoothing. Specifically, the weighted smoothing process processes the waveform differences of each segment in segments, and uses the local weighted regression method to recalculate the boundary waveform of each segment according to the difference weight to ensure the continuity of the waveform at the segment connection. In the weighted processing process, the correlation of the time series is considered. By introducing the time weight, the time difference between the segments is adjusted so that the time series of the waveform remains consistent during the smooth transition process to avoid phase shift. The waveform after weighted smoothing is rearranged, and the smoothed waveforms of adjacent segments are integrated. The generated smoothed waveform sequence is saved in the form of a time series, including time points, smoothed amplitude data and trend information of the overall waveform.

[0065] In the sub-step of reconstructing and generating a synthetic signal sequence based on a smooth waveform sequence, each time segment in the smooth waveform sequence is integrated, and the waveform signals of adjacent segments are gradually merged. The weighted accumulation and nonlinear fitting methods are used to accurately match the integrated time series with the waveform signal. The specific operation includes aligning the time series of each segment with the waveform amplitude data, introducing timing weights in the alignment process, minimizing the errors of time points and amplitude points, ensuring that the time series remains consistent during the merging process, and adjusting the waveforms between segments through nonlinear fitting to make the waveforms continuously connected on the time axis to avoid fluctuations or irregular changes. The integrated waveform signal is calculated through multiple iterations to gradually correct the differences between the segments and finally synthesize a continuous waveform signal. The generated synthetic signal sequence is stored in matrix form, containing the integrated time points, waveform amplitudes, and phase information.

[0066] In this embodiment, the steps for obtaining the key signal feature point set are specifically as follows:

[0067] Based on the synthetic signal sequence, the key feature points in the reconstructed waveform are screened, the time position and amplitude information of the feature points are extracted one by one, the time position of each feature point is compared with the adjacent points, and the difference between adjacent feature points is calculated to generate a preliminary feature point set;

[0068] Based on the preliminary feature point set, the time difference and amplitude difference between the feature point and the adjacent points are calculated one by one, and the effective feature points are determined by cumulative calculation. The effective feature points are matched with the adjacent points, and the difference values ​​between the feature points are screened to generate a valid feature point set;

[0069] Based on the effective feature point set, the effectiveness and accuracy of the feature points are judged by analyzing the time difference and amplitude change between the feature points, and the effective feature points are screened and classified to generate a key signal feature point set.

[0070] In the sub-step of screening key feature points based on the synthetic signal sequence, the time position and amplitude information of each feature point in the waveform are extracted one by one by deeply analyzing the reconstructed waveform signal. The information is stored in the form of a data array, with the time position as the index and the amplitude as the corresponding amplitude data. During the extraction process, the time position of the feature point is compared with the position of the adjacent feature point one by one, and the difference between the adjacent feature points is calculated, including the difference in time position and the amplitude of the change in amplitude, so as to identify the significant change points in the waveform. The screening of feature points is done by setting a threshold, and the difference points exceeding the threshold are taken as elements in the preliminary feature point set. The preliminary feature point set contains all possible key change points. The generated preliminary feature point set is saved in the form of a multidimensional array, which contains the time position, amplitude information of each feature point and its difference parameters with the adjacent points. Further screening and verification will be carried out based on this set in the future.

[0071] In the sub-step of calculating the effective feature points based on the preliminary feature point set, the time difference and amplitude difference between each feature point and the adjacent points are calculated one by one, the difference is quantified by cumulative calculation, and the effective feature points are further screened out. In the specific operation, each feature point in the preliminary feature point set is arranged in chronological order, the time difference between adjacent points is calculated, and the time difference is accumulated to identify points with large differences on the time axis. At the same time, for the calculation of the amplitude difference, by comparing the amplitude change of the feature point with the adjacent points, a threshold of the cumulative change is set, and the feature points with large amplitude changes are screened out as effective feature points. To ensure the accuracy of the screening, the effective feature points are matched again with their adjacent points, and the outliers or noise points are excluded by calculating the difference value between the feature points. Finally, the generated effective feature point set contains the feature points that have been screened and verified. The data format is a multidimensional array, which records the time position, amplitude information and matching degree of each effective feature point with its adjacent points.

[0072] In the sub-step of generating a key signal feature point set based on the valid feature point set, the data in the valid feature point set is further analyzed to determine the validity and accuracy of the feature points. First, the time position and amplitude information of each feature point are secondary analyzed, and the time difference and amplitude change between the feature points are calculated to further verify whether the feature point can accurately reflect the key changes of the waveform. In order to improve the accuracy of the judgment, the time difference and amplitude change of each feature point are normalized to eliminate the scale difference between different feature points. Next, the valid feature points are classified according to the time and amplitude relationship between the feature points, and the points with similar time difference and amplitude change characteristics are classified into one category to form a more concentrated feature point group. In the classification process, the overall trend of the waveform is taken into account. By calculating the similarity between the feature point groups, the most representative feature points are further screened out, and finally the key signal feature point set is generated. The generated key signal feature point set is saved in the form of a structured data table, which contains the time position, amplitude information and classification label of each key feature point.

[0073] In this embodiment, the steps for obtaining the optimized signal feature distribution are as follows:

[0074] Based on the key signal feature point set, the distribution of feature points in the distribution line current is analyzed. By calculating the time difference of each feature point, the time distribution data of the feature points are classified and sorted, and the time difference information of adjacent feature points is integrated to generate a feature time series.

[0075] Based on the feature time series, the positions of the feature points are corrected in sequence, the time distribution and spatial layout of the feature points are gradually adjusted, the corrected feature points are sorted, the adjusted feature point positions are matched with the time distribution, and the corrected feature distribution is generated;

[0076] Based on the correction feature distribution, the timing distribution of the feature points is optimized. By analyzing the spatial layout of the feature points, the timing information of the feature points is gradually adjusted. The optimized feature points and the timing information are unified and integrated to generate an optimized signal feature distribution.

[0077] In the sub-step of analyzing the distribution of characteristic points in the distribution line current based on the key signal characteristic point set, the characteristic point set is systematically analyzed, and the time information of each characteristic point is first compared with the adjacent characteristic points to calculate the time difference between each characteristic point. The time information of the characteristic point is stored in the form of a timestamp, and the time difference between each characteristic point is represented by a time difference matrix, and the elements of the matrix represent the time interval between adjacent characteristic points. By processing the time difference matrix, the characteristic points with similar time differences are classified and sorted. The classification process uses a hierarchical clustering method to group the characteristic points according to the time difference by setting the time difference threshold, and further refine the characteristic points in each group. The classified and sorted characteristic point time data is stored in a serialized form, and combined with the time difference information, a characteristic time series is generated. The characteristic time series is used for the basic data of the subsequent correction of the characteristic point position, and serves as the key input for optimizing the distribution of characteristic points in subsequent steps.

[0078] In the sub-step of correcting the position of feature points based on the feature time series, the time series of feature points is first corrected by gradually adjusting the time distribution and spatial layout of feature points. During the correction process, the position of feature points is optimized using the least squares method to ensure the uniformity of the distribution of feature points on the time axis. The position of each feature point is represented in the form of spatial coordinates and timestamps. During the correction process, the offset of each feature point on the time axis is calculated, and the offset is corrected by weighted processing to ensure that the position of the corrected feature point in the time series is reasonable. On this basis, the spatial layout of feature points is adjusted. The specific operation includes calculating the spatial distance between feature points, optimizing the spatial layout by multidimensional scaling analysis, and adjusting the relative position between feature points, so that the feature points are more uniform in spatial distribution. The corrected feature point position and time distribution are combined through a matching algorithm, and the adjusted feature points are sorted in chronological order to generate a corrected feature distribution. The corrected feature distribution contains the corrected time information and spatial position of each feature point, which is used as the basis for subsequent optimization of the time series distribution of feature points.

[0079] In the sub-step of optimizing the time series distribution of feature points based on the corrected feature distribution, the corrected feature points are further optimized. First, the time series information of each feature point is analyzed, and the time series information of the feature points is aligned using the dynamic time warping method to eliminate the time deviation between the feature points. The time series information includes the time position and time series of the feature points. By analyzing the spatial layout of the feature points, the time difference between the feature points is dynamically adjusted to make the distribution of the feature points in the time series more reasonable. In the adjustment process, the spatial position and time difference of the feature points are comprehensively considered, and the time series of the feature points is optimized by the weighted average method. The optimized time series is unified and integrated with the spatial layout of the feature points to generate the final optimized signal feature distribution. The optimized signal feature distribution contains the optimized time series and spatial position of the feature points, which is the key data basis for fault identification and ensures the accuracy and consistency of the signal features.

[0080] In this embodiment, the steps for obtaining the fault identification result are specifically as follows:

[0081] Based on the optimization of signal feature distribution, long short-term memory neural network is used to analyze time series data, gradually extract the time feature information of each time segment, analyze the dynamic change trend of each time feature, and generate time domain fault features;

[0082] Based on the time domain fault characteristics, the frequency domain characteristics of the fault signal are analyzed, and the frequency information of the characteristic points is compared section by section. By matching the frequency domain changes of the characteristic points, the frequency domain data that meets the fault characteristics are classified and sorted, and the frequency domain characteristics are uniformly summarized to generate frequency domain fault characteristics;

[0083] Based on the frequency domain fault characteristics, the signals that meet the fault characteristics are uniformly classified and labeled, and each labeled fault signal is sorted and classified to generate a fault identification result.

[0084] In the sub-step of analyzing time series data based on optimizing signal feature distribution, the optimized signal feature distribution is analyzed by using a long short-term memory neural network. First, the data of the signal feature distribution is input in the format of a time series, and the time series data includes the timestamp and corresponding amplitude information of each feature point. The long short-term memory neural network will gradually process the data of each time series segment and extract the time feature information. Specifically, the network will capture the short-term fluctuations and long-term trends of each time feature through the dynamic update of the hidden layer state, and gradually extract the change information of the time feature. Each level in the neural network will gradually output the corresponding time feature information according to different time segments of the input data. Through further analysis of the time feature information, the network can identify the dynamic change trend in the signal and finally generate the time domain fault feature. The time domain fault feature is stored as a structured time series data set, which contains the time feature value of each time period and its corresponding dynamic change trend.

[0085] In the sub-step of analyzing the frequency domain characteristics of the fault signal based on the time domain fault characteristics, the frequency information of each feature point is extracted segment by segment by performing frequency domain analysis on the time domain fault characteristic data, and then compared with the optimized frequency domain data. First, the time series data of the time domain fault characteristics are converted into frequency domain data, and the frequency domain data includes frequency components and their corresponding amplitude information. The feature points in each time series are processed by fast Fourier transform to convert the time characteristics into frequency characteristics. During the comparison process, the frequency domain feature points matching the known fault mode are identified by comparing the time domain feature data with the frequency characteristics in the historical fault data one by one. For each matched frequency domain feature point, the frequency domain data that meets the fault characteristics are marked by classification and sorting, and the frequency domain data of different categories are uniformly summarized. The frequency domain fault characteristics finally generated are stored in the form of a two-dimensional frequency matrix, which contains frequency values ​​and amplitude strengths, and is classified according to the fault type, providing basic data for further identification of the fault signal.

[0086] In the sub-step of generating fault identification results based on frequency domain fault features, the frequency domain fault features are uniformly classified and labeled to identify signals that meet the fault features and organize and classify them. First, each feature point in the frequency domain fault features is labeled. The basis for the labeling is the degree of matching between the frequency domain features and the known fault model. The labeling content includes the fault type, the time location of occurrence, and the corresponding frequency information. Then, all the labeled feature points are sorted, and the fault features of the same type are classified and summarized. During the aggregation process, the weighted average method is used to calculate the frequency and amplitude of the same type of fault features to ensure the centralization and consistency of the data. After the classification and sorting is completed, the generated fault identification results are indexed by the fault type, and the specific time location, frequency information, and amplitude data of each fault are recorded, and stored as a fault identification report file.

[0087] In this embodiment, the long short-term memory neural network is based on the formula:

[0088]

[0089] Calculate the time domain fault characteristics, where h t is the hidden state at the current moment, W h is the weight matrix of hidden state and input, h t-1 is the hidden state of the previous moment, x t is the input data at the current moment, b h is the bias term, α t is the dynamically adjusted time delay parameter, δ t is the time series feature weight, γ is the weight coefficient related to the input, λ is the hidden state compensation coefficient, ρ t is the adjustment factor, βt is the environmental noise correction term, ξ is the noise amplification factor, and σ is the activation function.

[0090] Calculation process: First, x t is the input data at the current moment, representing the time domain data input of the distribution line fault signal, and is combined with the hidden state h at the previous moment. t-1 For splicing, through the matrix W h Perform a linear transformation on the input data and hidden state. The dynamically adjusted time delay parameter α is introduced t This parameter is calculated based on the time deviation of historical data and is calculated by comparing it with the time series feature weight δ t A nonlinear combination of To adjust the alignment accuracy and importance of time domain data;

[0091] Secondly, the bias term b h Perform weighted calculation, combine the weight coefficient γ associated with the fluctuation of the input data and the hidden state compensation coefficient λ, perform weighted average processing, and then divide by the adjustment factor ρ t , this adjustment factor is dynamically generated by analyzing the rate of change of the current input data to ensure that the impact of the bias term is coordinated with the change of the hidden state. The formula is

[0092] Finally, add the environmental noise correction term β t , and is adjusted by the noise amplification factor ξ to offset the interference of the external environment on the time domain data, and finally the hidden state h at the current moment is generated through the activation function σ t .

[0093] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0094] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0095] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for identifying tree-line discharge faults based on fault traveling wave current characteristics, characterized in that: The following steps are involved: Based on the fault traveling wave current signal, the current waveform on the distribution line is sampled at high frequency, and the signal is divided into multiple time segments. By calculating the change trend of the time and frequency components of each segment, the current fluctuation characteristics of each period are analyzed, and the time-frequency changes are gradually matched to generate a set of time-frequency segments. Based on the set of time-frequency segments, interpolation processing is performed on the transient waveform in each segment, a path with the minimum waveform difference between the segments is selected, a smooth transition of the segments is completed by fitting the waveform, and the complete waveform signal is gradually merged and reconstructed, and a synthetic signal sequence is obtained by reconstructing the segment signals; Based on the synthetic signal sequence, the waveform features in the current waveform are screened, the time position and amplitude of the characteristic points in the waveform are extracted, the time difference and amplitude difference between each characteristic point and the adjacent points are calculated one by one, the validity and accuracy of the characteristic points in the waveform are screened, and a key signal characteristic point set is obtained; Based on the key signal feature point set, the distribution of feature points in the distribution line current is analyzed, and the positions of the feature points are corrected in sequence by calculating the time difference and amplitude difference of each feature point, the time series distribution of the feature points is optimized, the spatial distribution of the feature points is adjusted, and the optimized signal feature distribution is generated; Based on the optimized signal feature distribution, the time domain and frequency domain features of the distribution line fault signal are analyzed, and by comparing the time-frequency characteristics of the feature points one by one, the signals that meet the fault features are classified and labeled to generate a fault identification result.

2. The tree-line discharge fault identification method based on fault traveling wave current characteristics according to claim 1 is characterized in that: The time-frequency segment set includes the frequency distribution, phase information and amplitude change of the time segment, the synthetic signal sequence includes the overall trend of the reconstructed waveform, the continuity of the time series and the smoothed amplitude information, the key signal feature point set includes the target time point, amplitude and the difference between the feature points, the optimized signal feature distribution includes the adjusted feature point position, the corrected time interval and the optimized spatial layout, and the fault identification result includes the identified fault type, the time mark and the spatial position of the fault.

3. The tree-line discharge fault identification method based on fault traveling wave current characteristics according to claim 1 is characterized in that: The steps for obtaining the time-frequency segment set are specifically as follows: Based on the fault traveling wave current signal, the current waveform is sampled at high frequency, the signal is divided into multiple time segments, and a fixed time window is used to extract the time and frequency components of each segment one by one, calculate the time-frequency characteristics, and generate segmented time-frequency characteristics by analyzing the waveform change trend; Based on the segmented time-frequency characteristics, the current fluctuation characteristics in each time segment are analyzed, and the frequency relationship between the segments is established by gradually matching the frequency and time between the segments, and the frequency changes within the time period are associated, so as to generate time-frequency association data; Based on the time-frequency association data, the frequency and time information of each time segment are integrated, and the association relationship between the segments is merged in an inductive and organized manner. Through the unified integration of the time-frequency information between the segments, a time-frequency segment set is generated.

4. The tree-line discharge fault identification method based on fault traveling wave current characteristics according to claim 1 is characterized in that: The steps for obtaining the synthetic signal sequence are specifically as follows: Based on the set of time-frequency segments, interpolation processing is performed on the transient waveform in each segment, a waveform difference path between the segments is selected, and the waveform between each segment is gradually and smoothly connected by calculating the minimum difference path, so as to minimize the difference between the segments and generate a waveform interpolation result; Based on the waveform interpolation result, each segment waveform after interpolation is smoothed, by gradually merging the waveform differences of adjacent segments, using a weighted smoothing method to process the waveform transition at the segment connection, and arranging the merged segment waveforms to generate a smoothed waveform sequence; Based on the smoothed waveform sequence, the reconstructed waveform signals are integrated, and by gradually merging the time series within the segments, the integrated time series are matched with the waveform, and the reconstructed segment signals are synthesized into a continuous waveform signal to generate a synthetic signal sequence.

5. The tree-line discharge fault identification method based on fault traveling wave current characteristics according to claim 1 is characterized in that: The steps for obtaining the key signal feature point set are specifically as follows: Based on the synthetic signal sequence, key feature points in the reconstructed waveform are screened, the time position and amplitude information of the feature points are extracted one by one, the time position of each feature point is compared with the adjacent points, and the difference between adjacent feature points is calculated to generate a preliminary feature point set; Based on the preliminary feature point set, the time difference and amplitude difference between the feature point and the adjacent points are calculated one by one, the valid feature point is determined by cumulative calculation, the valid feature point is matched with the adjacent points, and the difference values ​​between the feature points are screened to generate a valid feature point set; Based on the effective feature point set, the effectiveness and accuracy of the feature points are judged by analyzing the time difference and amplitude change between the feature points, and the effective feature points are screened and classified to generate a key signal feature point set.

6. The method for identifying tree-line discharge faults based on fault traveling wave current characteristics according to claim 1 is characterized in that: The steps for obtaining the optimized signal feature distribution are specifically as follows: Based on the key signal feature point set, the distribution of feature points in the distribution line current is analyzed, the time distribution data of the feature points are classified and sorted by calculating the time difference of each feature point, and the time difference information of adjacent feature points is integrated to generate a feature time series; Based on the feature time series, the positions of the feature points are corrected in sequence, the time distribution and spatial layout of the feature points are gradually adjusted, the corrected feature points are sorted, and the adjusted feature point positions are matched with the time distribution to generate a corrected feature distribution; Based on the corrected feature distribution, the time series distribution of the feature points is optimized, the spatial layout of the feature points is analyzed, the time series information of the feature points is gradually adjusted, the optimized feature points and the time series information are unified and integrated to generate an optimized signal feature distribution.

7. The tree-line discharge fault identification method based on fault traveling wave current characteristics according to claim 1 is characterized in that: The steps for obtaining the fault identification result are as follows: Based on the optimized signal feature distribution, a long short-term memory neural network is used to perform time series data analysis, gradually extract the time feature information of each time series segment, analyze the dynamic change trend of each time feature, and generate time domain fault features; Based on the time domain fault characteristics, the frequency domain characteristics of the fault signal are analyzed, the frequency information of the characteristic points are compared section by section, the frequency domain data that meets the fault characteristics are classified and sorted by matching the frequency domain changes of the characteristic points, and the frequency domain characteristics are uniformly summarized to generate the frequency domain fault characteristics; Based on the frequency domain fault characteristics, the signals that meet the fault characteristics are uniformly classified and labeled, each labeled fault signal is sorted and classified to generate a fault identification result.

8. The method for identifying tree-line discharge faults based on fault traveling wave current characteristics according to claim 7 is characterized in that: The long short-term memory neural network is according to the formula: Calculate the time domain fault characteristics, where h t is the hidden state at the current moment, W h is the weight matrix of hidden state and input, h t-1 is the hidden state of the previous moment, x t is the input data at the current moment, b h is the bias term, α t is the dynamically adjusted time delay parameter, δ t is the time series feature weight, γ is the weight coefficient related to the input, λ is the hidden state compensation coefficient, ρ t is the adjustment factor, β t is the environmental noise correction term, ξ is the noise amplification factor, and σ is the activation function.

Images