An adaptive identification and extraction method for transient traveling wave signals in complex environment

Abstract

This invention relates to the field of signal recognition technology, specifically an adaptive recognition and extraction method for transient traveling wave signals in complex environments. The method includes: establishing a mapping relationship between real-time traveling wave signals and standard traveling wave signals using convolutional integrals; extracting the voltage and current fluctuation amplitudes within each time window to determine the signal characteristics of the real-time traveling wave signal; performing scene sorting on the signal characteristics, dividing them sequentially into scene components and interference components, and configuring component matrices corresponding to the signal characteristics; using the component matrices to solve for the similarity of the real-time traveling wave signals, and determining the sum of similarities across multiple time windows based on the change amplitude of the signal characteristics to obtain the baseline signal characteristics; based on the number of time windows corresponding to the baseline signal characteristics, selecting the baseline signal characteristics corresponding to the largest time window percentage as the target signal, and fitting the fault location of the real-time traveling wave signal based on the preliminary positioning range of the target signal. This method achieves both accuracy and efficiency in traveling wave signal recognition.

Description

Technical Field

[0001] This invention relates to the field of signal recognition technology, specifically an adaptive recognition and extraction method for transient traveling wave signals under complex environments. Background Technology

[0002] Transient traveling wave signals are instantaneous electromagnetic fluctuations generated when a power system fault occurs. They propagate rapidly and are the core basis for fault location in high-voltage transmission lines. During the operation of high-voltage transmission lines, transient traveling wave signals are easily affected by the superposition of various interferences, making it difficult to determine the distortion and superposition effects of fault traveling wave signals by waveform comparison alone, thus reducing the accuracy of traveling wave signal location.

[0003] For example, Chinese Patent Publication No. CN115616330A discloses a method and system for identifying multiple lightning strikes on transmission lines based on waveform similarity, belonging to the field of intelligent manufacturing and relay protection technology for new power systems. This invention can be integrated into a protection system, using current traveling wave recording analysis to distinguish between single and multiple lightning strikes by comparing the similarity of current traveling waves within a certain time window.

[0004] For example, Chinese Patent Publication No. CN118937889A discloses a method for locating fault sections in a small current grounding system based on the similarity of transient zero-sequence current fuzzy entropy waveforms. The method includes: constructing a distribution network model of the small current grounding system; collecting transient zero-sequence current signals from various sections of the faulty line in the distribution network model; performing CEEMDAN mode decomposition on the transient zero-sequence current signals to obtain the high-frequency imf1 component; calculating the fuzzy entropy value of the imf1 component and defining a disorder value as a first criterion for preliminary screening of fault sections; obtaining a first criterion result based on the first criterion; if the disorder ratio is less than a preset threshold, analyzing the waveform polarity using a second criterion based on the initial polarity of the fuzzy entropy waveform to obtain a second criterion result; and locating the fault section when the first criterion result and the second criterion result are consistent.

[0005] Existing technologies identify lightning strike faults by simulating their waveforms and comparing their similarity; and they locate faulty sections by analyzing entropy disorder. These methods tend to rely on waveform similarity processing after mode decomposition. In complex environments, these methods often focus on extracting features in a single dimension, making it difficult to effectively distinguish different fault types and resulting in a high misjudgment rate in scene sorting. Summary of the Invention

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: an adaptive identification and extraction method for transient traveling wave signals under complex environments, comprising: S1, acquiring real-time traveling wave signals and standard traveling wave signals under different fault types, determining the mapping relationship between real-time traveling wave signals and standard traveling wave signals by using convolution integration, and recording the wavefront arrival time, voltage and current corresponding to the real-time traveling wave signal by using a sliding time window.

[0007] S2, using the time difference of wavefront arrival as the dividing point, performs feature recognition on the real-time traveling wave signal, extracts the fluctuation amplitude of voltage and current within each time window, and determines the signal characteristics of the real-time traveling wave signal.

[0008] S3, based on the mapping relationship between real-time traveling wave signal and standard traveling wave signal, performs scene sorting on signal features, dividing them into scene components and interference components in sequence, and configuring the component matrix corresponding to the signal features.

[0009] S4. The similarity of the real-time traveling wave signal is solved by using the component matrix, and the sum of similarity under multiple time windows is determined by the change amplitude of the signal characteristics, so as to obtain the reference signal characteristics corresponding to the real-time traveling wave signal.

[0010] S5. Based on the number of time windows corresponding to the characteristics of the reference signal, the reference signal characteristics corresponding to the largest time window ratio are used as the target signal. Based on the preliminary positioning range of the target signal, the fault location of the real-time traveling wave signal is obtained by fitting.

[0011] The beneficial effects of this invention are as follows: First, this invention uses time synchronization alignment and convolution integral mapping to filter the optimal matching standard traveling wave signal using the maximum value of the convolution integral, and establishes a mapping relationship with the current real-time traveling wave signal; it uses preliminary similarity matching to match and map the real-time traveling wave signal with the standard traveling wave signal, providing a data foundation for subsequent time window division and basic signal characteristics.

[0012] Second, this invention dynamically adjusts the sliding time window based on the wavefront arrival time difference to ensure that the window is adapted to the transient characteristics of the signal. Then, it performs differentiated processing based on current and voltage, and selects differentiated features such as information entropy and intra-cluster mean to describe the energy characteristics of the traveling wave signal in different scenarios, thereby improving the identification of sorting in each scenario and the matching efficiency of the current real-time traveling wave signal.

[0013] Third, this invention distinguishes the waveforms in each scenario by using candidate scenarios and feature thresholds, and determines the scenario components directly related to the fault in each scenario, as well as the interference components that cause current interference. The structured storage of the component matrix enables subsequent similarity calculations to accurately focus on the scenario components, avoid the cross-scenario influence of interference components, and improve the recognition accuracy of traveling wave signals under multi-time window processing.

[0014] Fourth, this invention quantifies the weights in component matrix calculation by using the ratio of the difference between adjacent time windows to the total difference. This suppresses the impact of abnormal fluctuations in some time windows. Subsequently, the multi-time-window weighted reference signal features avoid feature shifts caused by interference, thus providing reliable data support for fault location. Finally, the wavefront arrival time difference corresponding to the basic signal features is used to select the target signal using its corresponding direct and reflected waves, followed by distance calculation to obtain the initial location range. This initial location range is then combined with the values ​​of multiple time windows to optimize the result, ensuring that the target signal selection process always focuses on the location distribution of the dominant fault, avoiding location deviations caused by transient interference. Attached Figure Description

[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0016] Figure 1 This is a flowchart illustrating an adaptive identification and extraction method for transient traveling wave signals under complex environments.

[0017] Figure 2 This is a flowchart illustrating step S1 of an adaptive identification and extraction method for transient traveling wave signals under complex environments.

[0018] Figure 3 This is a flowchart illustrating step S2 of an adaptive identification and extraction method for transient traveling wave signals under complex environments.

[0019] Figure 4 This is a flowchart illustrating step S3 of an adaptive identification and extraction method for transient traveling wave signals under complex environments.

[0020] Figure 5 This is a flowchart illustrating step S4 of an adaptive identification and extraction method for transient traveling wave signals under complex environments.

[0021] Figure 6 This is a flowchart illustrating step S5 of an adaptive identification and extraction method for transient traveling wave signals under complex environments. Detailed Implementation

[0022] The embodiments of the present invention are described in detail below. The embodiments described below are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Where specific techniques or conditions are not specified in the embodiments, they shall be performed in accordance with the techniques or conditions described in the literature in the art or in accordance with the product manual.

[0023] See Figure 1An adaptive identification and extraction method for transient traveling wave signals under complex environments includes: S1, acquiring real-time traveling wave signals and standard traveling wave signals under different fault types, determining the mapping relationship between real-time traveling wave signals and standard traveling wave signals using convolution integration, and recording the wavefront arrival time, voltage, and current corresponding to the real-time traveling wave signal using a sliding time window.

[0024] S2, using the time difference of wavefront arrival as the dividing point, performs feature recognition on the real-time traveling wave signal, extracts the fluctuation amplitude of voltage and current within each time window, and determines the signal characteristics of the real-time traveling wave signal.

[0025] S3, based on the mapping relationship between real-time traveling wave signal and standard traveling wave signal, performs scene sorting on signal features, dividing them into scene components and interference components in sequence, and configuring the component matrix corresponding to the signal features.

[0026] S4. The similarity of the real-time traveling wave signal is solved by using the component matrix, and the sum of similarity under multiple time windows is determined by the change amplitude of the signal characteristics, so as to obtain the reference signal characteristics corresponding to the real-time traveling wave signal.

[0027] S5. Based on the number of time windows corresponding to the characteristics of the reference signal, the reference signal characteristics corresponding to the largest time window ratio are used as the target signal. Based on the preliminary positioning range of the target signal, the fault location of the real-time traveling wave signal is obtained by fitting.

[0028] In the processing of this invention, essentially two types of continuous traveling wave signals are acquired: a standard traveling wave signal with an accompanying fault type and a real-time input traveling wave signal. The two traveling wave signals are continuously convolved and integrated in the time domain to adapt to the characteristics of the traveling wave signal such as fluctuation, polarity, quantity and energy, thereby completing the extraction and processing of the real-time input traveling wave signal.

[0029] When processing traveling wave signals, the standard traveling wave signals of known fault types are first processed, and multiple sets of scene sorting features are obtained after processing this part of the data. The current input traveling wave signal is then compared with the features under these different scenarios to determine the reference signal features.

[0030] The general method of convolution integration is as follows: ;in, Represents a real-time traveling wave signal. This represents the value of the real-time traveling wave signal at time t. In the current convolution process, the amplitude dimension can be used, that is, to record the similarity of the traveling wave signal at the peak, and to calculate based on the value at the peak position. Indicates a standard traveling wave signal. This indicates the standard traveling wave signal at time [time]. The value, This represents the convolution delay time, which is set by the deviation generated during peak alignment; Represents real-time traveling wave signal Compared with standard traveling wave signals Convolution calculation, The symbol representing the convolution operation.

[0031] After the convolution integral is calculated, the standard traveling wave signal with the largest value among all fault types is regarded as the mapping data corresponding to the current real-time traveling wave signal, thereby completing the peak comparison processing.

[0032] like Figure 2 As shown, the implementation of step S1 includes: S11, performing time synchronization and alignment between the real-time traveling wave signal and the standard traveling wave signal to align the peak positions of the real-time traveling wave signal and the standard traveling wave signal.

[0033] S12 performs convolution integration on the real-time traveling wave signal and the standard traveling wave signal according to the corresponding time length, outputs the standard traveling wave signal with the largest convolution integral value, and constructs a mapping relationship with the current real-time traveling wave signal.

[0034] The currently set standard traveling wave signal not only covers the signals corresponding to various fault types, but also includes the standard traveling wave signal corresponding to interference signals under lightning strike scenarios, thereby distinguishing the traveling wave signal under pure fault scenarios and lightning-interference scenarios.

[0035] For traveling wave signals in these scenarios, the arrival time and number of wavefronts of the traveling wave signal in real time are recorded. For example, in the lightning strike scenario, a single lightning strike usually has 1-2 main wavefronts. Based on the wavefront arrival time difference, wavefronts that are misjudged by reflected waves can be eliminated to avoid abnormal wavefront identification. Then, these data are used as wavefront features of the traveling wave signal. Combined with the current and voltage in the corresponding scenario, the data acquisition of the real-time traveling wave signal is completed.

[0036] Therefore, when recording the arrival time of the wavefront corresponding to the real-time traveling wave signal in step S1, the implementation method also includes: taking the detection point that identifies the wavefront of the real-time traveling wave signal as the main body, retrieving the wavefront arrival time difference of all adjacent detection points; the wavefront arrival time difference can represent the difference in the arrival time of the wave detected at two adjacent positions, thereby indicating the propagation speed of the real-time traveling wave signal.

[0037] The validity is tested based on the mean and standard deviation of the wavefront arrival time difference, and the wavefront arrival time corresponding to the validity test is regarded as the output data.

[0038] During the validity test, the wavefront arrival time difference is filtered in the form of mean ± 3 standard deviations. The filtered data is then divided into time series consisting of wave arrival times, and the voltage and current values ​​in these time series are recorded to illustrate the specific characteristics of the current real-time traveling wave signal.

[0039] It should be noted that when processing real-time traveling wave signals, standardization is performed using methods such as median filtering and wavelet transform to divide the corresponding signal into multiple modal data.

[0040] A sliding time window is a data window that represents a certain time period and is used to record the waveform of a real-time traveling wave signal. Each sliding time window needs to cover the rising edge, peak value, and falling edge of the wave front to describe the waveform characteristics of the real-time traveling wave signal to a relatively complete extent, which facilitates the verification and processing of waveform amplitude, voltage value, and other information.

[0041] In step S2, the energy characteristics of the real-time traveling wave signal are quantified by the fluctuation amplitude of voltage and current. For example, in the case of lightning strike, lightning strikes usually directly interfere with the information entropy of the current. By observing the fluctuation amplitude of the current, it can be understood whether the real-time traveling wave signal is close to the lightning strike fault. In other scenarios, the voltage fluctuation is used to identify the voltage fluctuation during the fault, thereby obtaining the signal characteristics of the real-time traveling wave signal with respect to energy.

[0042] like Figure 3 As shown, the implementation of step S2 includes: S21, adjusting the sliding time window based on the value of the wavefront arrival time difference, and dividing it into multiple time windows.

[0043] When adjusting the sliding time window, it can be based on 1.2 times the arrival time difference of the wavefront, so that the rising edge, peak, and falling edge of the wavefront are covered in each time window. If any time window cannot cover the corresponding wavefront feature, it is adjusted by ±10% until the time window can completely cover the real-time traveling wave signal in a certain time period. The maximum number of adjustments is no more than 3 times to prevent over-adjustment, which may lead to inaccurate extraction of wavefront features.

[0044] S22, for the current value within each time window, extract the information entropy of the current value under the corresponding fluctuation amplitude, and use the information entropy of the current value as the feature value of the current.

[0045] Information entropy calculation typically involves recording the probability of current amplitude falling within a certain range, thereby statistically calculating the information entropy of the current to distinguish between lightning strike faults and short-circuit faults. When a line experiences a lightning strike or short-circuit fault, the difference in high-frequency components in the current leads to different peak waveform contents, thus information entropy can be used for current analysis. When lightning strikes the line causing insulator breakdown, the lightning current enters the transmission line, containing more mid-to-high frequency components. The current signal exhibits single / double main wavefronts with rapid attenuation, and its probability distribution is concentrated at the poles; therefore, the information entropy is relatively low during lightning strike faults. Conversely, during short-circuit faults, the current signal exhibits main wavefronts with multiple reflected wavefronts. The fault current not only contains steady-state short-circuit components but also generates multiple wavefronts due to reflections at both ends and branch lines, resulting in a wide range of current fluctuations and thus a higher information entropy. By determining the information entropy value of the current, the distribution of current value fluctuations across various ranges can be identified, thereby explaining the current values ​​under different conditions.

[0046] When calculating the information entropy of the current value, the sampling point corresponding to the current value is first determined. The fluctuation amplitude difference between adjacent points is calculated point by point to form a sequence of current amplitude differences. Then, the difference is normalized according to the range of amplitude differences and divided into multiple intervals according to the normalized interval range, such as 32 intervals, to adapt to 16-bit sampling precision. The probability value for calculating the information entropy is then set according to the ratio of the number of sampling points in each interval to the total number of sampling points to illustrate the clustering of current fluctuation amplitude. Finally, the information entropy of the current value is obtained.

[0047] S23. For the voltage value within each time window, perform clustering calculation based on the voltage value fluctuation amplitude. Based on the clusters of voltage values ​​under the corresponding fluctuation amplitude, select the cluster mean, cluster standard deviation and cluster proportion corresponding to the cluster as the characteristic values ​​corresponding to the voltage.

[0048] The voltage value is calculated in the same way as the current value. However, the voltage value is clustered based on the difference in fluctuation amplitude. The mean, standard deviation, and the ratio of the number of sampling points in each cluster to the number of sampling points in the current time window (which can be simply described as the cluster ratio) are normalized and used as the feature value of the current voltage value to explain the dominant situation of different voltage values ​​in the current time window.

[0049] When clustering voltage fluctuations, algorithms such as K-means and DBSCAN can be used to obtain the corresponding clusters.

[0050] S24 couples the characteristic values ​​of voltage and current, and considers the sum of the coupled characteristic values ​​under multiple time windows as the output signal characteristic.

[0051] When performing feature coupling, the current value and the voltage values ​​are first concatenated to form a feature vector. Then, weights are assigned to each feature value to complete the summarization of signal features.

[0052] Therefore, the processing method of step S24 also includes: extracting the intra-cluster mean, intra-cluster standard deviation and cluster percentage of the voltage value in each time window, and obtaining the information entropy of the current value in each time window.

[0053] The information entropy, intra-cluster mean, intra-cluster standard deviation, and cluster proportion are concatenated into a feature vector. The dispersion of voltage values ​​and the information entropy of current values ​​in the current time window are used to assign weights to each feature value. The weighted data are then summed to obtain the signal features after summing for each time window.

[0054] The weight of the current value can be set directly based on its information entropy value to represent the complex fluctuation of the current value.

[0055] When assigning weights to voltage values, the dispersion represents the ratio of the standard deviation of the voltage value fluctuation amplitude within the current cluster to the standard deviation of the voltage value fluctuation amplitude within the current time window. This quantifies the dispersion of each cluster, and the weight of the corresponding voltage value is based on this ratio. The weights corresponding to the cluster mean, cluster standard deviation, and cluster percentage are all set using the weights corresponding to their dispersion. The weights of these three values ​​are normalized with the weights of the current values ​​to complete the weight configuration within the current time window.

[0056] In addition to setting the signal features through weighted summation of feature values, the data after statistics for each time window is also recorded for subsequent component judgment and similarity calculation. After weighted summation, the signal feature values ​​are normalized to serve as the basis for subsequent scene sorting.

[0057] In one embodiment of the present invention, when performing scene sorting for signal features in step S3, the essence is to select scene components as the target set and interference components as the supplementary set, and to divide the input signal features into data combinations under multiple scenes through multi-threshold division processing, so as to complete the identification and selection of the current real-time traveling wave signal.

[0058] like Figure 4 As shown, the implementation of step S3 includes: S31, based on the mapping relationship between the real-time traveling wave signal and the standard traveling wave signal, selecting candidate scenarios of the current real-time traveling wave signal, classifying the candidate scenarios into fault scenarios and interference scenarios, and outputting a scenario candidate set.

[0059] At this point, based on the fault type at the time of mapping, several candidate scenarios most relevant to this fault type will be selected. For example, if the current fault type is a short circuit fault, then the candidate scenarios will be selected as the most relevant fault scenarios to the current location, such as three-phase short circuit at the beginning of a 220kV line and single-phase ground fault at the middle of a 110kV line. Then, interference scenarios highly relevant to this fault will be extracted, such as interference from closing operation of a 110kV line and partial discharge interference from insulators of a 220kV line, to describe the scenario corresponding to the current real-time traveling wave signal.

[0060] As for the selection logic of candidate scenarios, based on the data processed by convolution integral in step S1, first, the mapped fault type is locked, then all sub-scenarios under the fault type are found, the convolution integral values ​​of the real-time traveling wave signal and these sub-scenarios are selected, and the convolution integral values ​​are sorted in descending order to select the top few scenarios; for example, the first two scenarios are selected according to the order of fault scenarios and interference scenarios, or the two top scenarios are directly selected as candidate scenarios.

[0061] S32, construct a scene-feature threshold library by configuring the threshold space for each data in the scene candidate set according to the scene configuration; the feature threshold library contains the convolution integral value of the corresponding scene and the real-time traveling wave signal, the weighted sum of the signal features in each time window, and the feature values ​​corresponding to voltage and current in the signal features; these values ​​will all be set with a standard value according to their scene, that is, the average value and standard deviation calculated by the corresponding data in the corresponding scene are used, and the confidence interval is set in the form of average value ± 3 times standard deviation to complete the configuration of its threshold space.

[0062] S33, based on the scene-feature threshold library, performs matching analysis on each candidate scene and outputs the successfully matched data as a scene component set.

[0063] When performing matching analysis on candidate scenes in step S33, the implementation method includes: for signal features in any time window, firstly, matching is performed based on the feature value of the signal feature, which is the weighted sum of the signal features in each time window. If the feature value of the signal feature belongs to the threshold range, then proceed to the next step of screening; otherwise, the candidate scene is directly eliminated.

[0064] Secondly, the characteristic values ​​corresponding to voltage and current in the signal features are used for matching. If the characteristic values ​​corresponding to voltage and current belong to the threshold range, the current candidate scene is determined to be successfully matched, and the corresponding signal features are output as scene components.

[0065] If multiple candidate scenarios match, the values ​​obtained by the convolution integral of each candidate scenario are sorted in descending order, and the top two candidate scenarios are retained as output. In this case, two candidate scenarios are selected to set the primary candidate scenario and the secondary candidate scenario, so as to complete the analysis of the real-time traveling wave signal in the form of two scenarios, and prevent the problem of missing data judgment and inaccurate matching when relying on only a single candidate scenario.

[0066] S34: Filter the interference components of the data that failed to match, and output the filtered interference component set.

[0067] The method for filtering interference components is the same as that for scene components. The interference components are biased towards identifying fixed interference and random interference, and their feature data is output to clarify the type and degree of interference in different scenarios.

[0068] If the data belongs to the threshold range corresponding to the interference component, the corresponding data is regarded as a fixed interference component, and the remaining interference components are determined as random interference components to complete the processing of the interference component set. Random interference components are mostly random noise of traveling wave signals. After filtering, the relevant data are marked and the fixed interference components generated by the scene are recorded.

[0069] S35: Using the data corresponding to the scene component set and the interference component set as the vertical axis of the matrix, and the dimension corresponding to the time window as the horizontal axis of the matrix, configure the component matrix corresponding to the signal features. The depth of the matrix will be set based on the current amount of data to complete the configuration of the corresponding component matrix.

[0070] In one embodiment of the present invention, during step S4, the similarity between each independent component in the component matrix and the standard traveling wave signal in the corresponding component is calculated to match the specific similarity in the corresponding scenario, thereby obtaining the reference signal features of the current real-time traveling wave signal in the corresponding scenario.

[0071] like Figure 5 As shown, step S4 is implemented as follows: S41, if the current component matrix corresponds to a single scene, for the scene component in the component matrix, calculate its similarity with the standard traveling wave signal to obtain the similarity value of the component matrix under the corresponding time window. The similarity calculated here can be based on the Pearson correlation coefficient, using the timestamp corresponding to its time window to complete the similarity calculation. This quantizes the similarity between the real-time traveling wave signal after interference removal and the standard traveling wave signal, thus completing the verification of the convolution integral.

[0072] S42, when the input component matrix corresponds to multiple scenes, uses the interference components to perform time-varying statistical analysis on the scene components, and calculates the similarity value based on the analyzed scene components.

[0073] When the component matrix corresponds to multiple scenarios, the scenario components and interference components are prone to coexist and overlap, which leads to the distortion of fault identification and affects the accuracy of subsequent similarity calculation. Even if the scenario components and interference components are distinguished by template comparison in the above steps, there are still cases where interference overlap exists in some scenario components, which need to be processed for time-varying differences under multiple scenarios.

[0074] Because the time-varying physical characteristics of interference components (such as lightning interference) and scene components (such as short-circuit faults) are fundamentally different, auxiliary statistical analysis can be performed based on their time-varying characteristics to achieve the correction processing of scene components.

[0075] When performing time-varying statistical analysis in step S42, the implementation method also includes: viewing the mean, variance, peak value, and difference between adjacent time windows of the interference component in each time window; these four values ​​represent the stability of the interference component and its sensitivity to the influence of other time windows.

[0076] Using the peak-to-average value and the ratio of the difference between adjacent time windows to the mean as matching indicators, we can examine the interference type of the current interference component. Specifically, the peak value divided by the mean indicates the degree of abrupt change in the interference component within a single time window, while the ratio of the difference between adjacent time windows to the mean indicates the trend of change in the interference over continuous time windows. By utilizing trend-related and abrupt change-related data, we can further describe and filter the current interference component. For example, interference types can be categorized as stationary interference, impulse interference, trend-based interference, and weak interference. Each type of interference component has a preset weight value, which can be used to correct the scene component. For instance, stationary interference has the highest weight, so a fixed weight of 0.8 can be used. Impulse interference mainly manifests in peak value changes, and its weight can be based on the normalized value of the peak-to-average value. Trend-based interference emphasizes the difference between adjacent windows, so the ratio of the difference between adjacent time windows to the mean can be used, and the weight can be set after normalization. Weak interference generally has the lowest weight, and a fixed value of 0.1-0.2 can be used to describe the specific value under weak interference.

[0077] For example, under certain conditions, the peak mean of stationary interference is ≤1.5 and the difference / mean of adjacent windows is ≤0.2; the peak mean of impulse interference is >3 and the difference / mean of adjacent windows is >0.5; the peak mean of trending interference is 1.5-3 and the difference / mean of adjacent windows is 0.2-0.5; and the peak mean of weak interference is ≤1.5 and the difference / mean of adjacent windows is ≤0.1. Based on these values, the type of interference corresponding to the current interference component can be distinguished, and thus the weight used can be determined.

[0078] Weights are assigned to corresponding interference components based on their types. The scene component is then subtracted from the weighted interference component value to obtain the analyzed scene component. During correction, the scene component is typically corrected by subtracting the product of the average value of the corresponding interference component and its weight. If multiple interference components exist within a continuous time window, a weighted average of all interference components is calculated. Using the weights of each interference type as the basis, the scene component is subtracted from this weighted average. This avoids over-correction of the scene component, which could distort the similarity calculation and prevent direct matching to the baseline signal in specific situations.

[0079] S43, calculate the difference between the signal features corresponding to the scene components in adjacent time windows, and simultaneously calculate the total difference across all time windows. The ratio of the difference between adjacent time windows to the total difference is used as the weight. The difference between signal features in adjacent time windows is based on a weighted sum of voltage and current feature values. This sum represents the changes in voltage and current within the corresponding time window, thus illustrating the changes in adjacent time windows, quantifying the similarity between real-time features and standard features, and avoiding feature shifts caused by real-time signal noise.

[0080] S44 performs weighted summation on scene components within multiple time windows, and uses the weighted summation value as the output reference signal feature.

[0081] In one embodiment of the present invention, such as Figure 6 As shown, the implementation of step S5 includes: S51, judging the reference signal characteristics, determining the reference signal characteristics at both ends respectively, and splitting the reflected wave and direct wave corresponding to each reference signal characteristic according to the wavefront arrival time difference corresponding to the reference signal characteristics; this part uses the reference characteristics and wavefront arrival time difference of the monitoring points at both ends of the line to split the direct wave and the reflected wave; through the waveform corresponding to the reference signal characteristics, the wave that directly propagates to the monitoring points at both ends is identified and marked as the direct wave, and the wave that propagates to the monitoring point after being reflected by the line endpoint or branch is marked as the reflected wave, and then according to the time difference between the two, the reflected wave and direct wave data under multiple time windows are obtained.

[0082] S52, after dividing the reference signal features, select the reference signal feature with the largest time window proportion as the target signal.

[0083] The baseline signal feature representing the largest time window percentage indicates the dominant fault type in the time dimension. This is achieved by extracting the number and duration of time windows for each scene component, as well as the total number of time windows and the total duration, to determine the dominant fault type under multiple time window statistics. For example, when the ratio of the number of time windows to the total number of time windows, the ratio of duration, and the ratio of total duration are all at their maximum values, the corresponding baseline signal feature is used as the target signal to determine the dominant problem under multi-time window analysis.

[0084] S53, perform positioning processing based on the segment location of the target signal to obtain the preliminary positioning range of the target signal.

[0085] The initial positioning range will be calculated based on the dual-end traveling wave positioning method, using the time difference of wavefront arrival at the two monitoring points. That is, the positioning point coordinates = traveling wave propagation speed × wavefront arrival time difference / 2. This completes the initial positioning range setting. At the same time, when considering errors such as propagation speed error and time difference measurement error, the initial positioning range can also be set in the form of positioning point coordinates ±5%, where 5% represents the proportion of positioning point coordinates. This determines the initial positioning range measured under a single time window.

[0086] S54, based on the initial positioning range of the target signal under multiple time windows, corrects and fits the initial positioning range to obtain the output fault location.

[0087] When performing preliminary positioning range correction fitting, the positioning point coordinates of each time window are first used to filter out some positioning point coordinates that exceed the global mean ± 3 times the standard deviation. Then, the positioning point coordinates are weighted and averaged according to the ratio of the feature value of the reference signal feature under each time window to the feature value of the reference signal feature of all time windows, so as to obtain the center value of the preliminary positioning range after fitting, and then the fitted fault location can be output.

[0088] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention, which are still covered within the protection scope of the present invention.

Claims

1. An adaptive identification and extraction method for transient traveling wave signals under complex environments, characterized in that, include: S1. Acquire real-time traveling wave signals and standard traveling wave signals under different fault types. Use convolution integration to determine the mapping relationship between real-time traveling wave signals and standard traveling wave signals. Use a sliding time window to record the wavefront arrival time, voltage, and current corresponding to the real-time traveling wave signal. S2, taking the time difference of wavefront arrival as the dividing point, performs feature recognition on the real-time traveling wave signal, extracts the fluctuation amplitude of voltage and current within each time window, and determines the signal characteristics of the real-time traveling wave signal; S3, based on the mapping relationship between real-time traveling wave signal and standard traveling wave signal, performs scene sorting on signal features, dividing them into scene components and interference components in sequence, and configuring the component matrix corresponding to the signal features; S4. The similarity of the real-time traveling wave signal is solved by using the component matrix, and the sum of similarity under multiple time windows is determined by the change amplitude of the signal characteristics, so as to obtain the reference signal characteristics corresponding to the real-time traveling wave signal. S5. Based on the number of time windows corresponding to the characteristics of the reference signal, the characteristics of the reference signal corresponding to the largest time window ratio are taken as the target signal. Based on the preliminary positioning range of the target signal, the fault location of the real-time traveling wave signal is obtained by fitting. Step S2 can be implemented in the following ways: S21, Based on the value of the wavefront arrival time difference, the sliding time window is adjusted and divided into multiple time windows; S22, For the current value in each time window, extract the information entropy of the current value under the corresponding fluctuation amplitude, and use the information entropy of the current value as the feature value of the current. S23, For the voltage value within each time window, perform clustering calculation based on the voltage value fluctuation amplitude, and select the cluster mean, cluster standard deviation and cluster proportion corresponding to the cluster as the characteristic value corresponding to the voltage based on the cluster of voltage value under the corresponding fluctuation amplitude. S24, the characteristic values ​​of voltage and current are coupled, and the sum of the coupled characteristic values ​​under multiple time windows is regarded as the output signal characteristic; Step S4 can be implemented in the following ways: S41, If ​​the current component matrix corresponds to a single scene, calculate the similarity between the scene component in the component matrix and the standard traveling wave signal to obtain the similarity value of the component matrix in the corresponding time window. S42, when the input component matrix corresponds to multiple scenes, use the interference components to perform time-varying statistical analysis on the scene components, and calculate the similarity value based on the analyzed scene components. S43, calculate the difference of signal features corresponding to scene components in adjacent time windows, and calculate the total difference of all time windows at the same time. Use the ratio of the difference between adjacent time windows to the total difference as the weight. S44 performs weighted summation on scene components within multiple time windows, and uses the weighted summation value as the output reference signal feature.

2. The adaptive identification and extraction method for transient traveling wave signals under complex environments according to claim 1, characterized in that, The implementation methods for step S1 include: S11, Time synchronization and alignment of the real-time traveling wave signal and the standard traveling wave signal, so that the peak positions of the real-time traveling wave signal and the standard traveling wave signal are aligned; S12 performs convolution integration on the real-time traveling wave signal and the standard traveling wave signal according to their corresponding time lengths, outputs the standard traveling wave signal with the largest convolution integral value, and constructs a mapping relationship with the current real-time traveling wave signal.

3. The adaptive identification and extraction method for transient traveling wave signals under complex environments according to claim 1, characterized in that, When recording the arrival time of the wavefront corresponding to the real-time traveling wave signal in step S1, the implementation method also includes: The detection point is used to identify the wavefront of a real-time traveling wave signal, and the arrival time difference of the wavefront of all adjacent detection points is retrieved. The validity is tested based on the mean and standard deviation of the wavefront arrival time difference, and the wavefront arrival time corresponding to the validity test is regarded as the output data.

4. The adaptive identification and extraction method for transient traveling wave signals under complex environments according to claim 1, characterized in that, The processing method in step S24 also includes: Extract the intra-cluster mean, intra-cluster standard deviation, and cluster percentage of the voltage value within each time window, and obtain the information entropy of the current value within each time window; The information entropy, intra-cluster mean, intra-cluster standard deviation, and cluster percentage are concatenated into a feature vector. The dispersion of voltage values ​​and the information entropy of current values ​​in the current time window are used to assign weights to each feature value. The weighted data are then summed to obtain the signal features after summing for each time window.

5. The adaptive identification and extraction method for transient traveling wave signals under complex environments according to claim 1, characterized in that, Step S3 can be implemented in the following ways: S31, Based on the mapping relationship between real-time traveling wave signal and standard traveling wave signal, select candidate scenarios of the current real-time traveling wave signal, classify the candidate scenarios into fault scenarios and interference scenarios, and output the scenario candidate set; S32, construct a scene-feature threshold library by configuring the threshold space of each data in the scene candidate set; S33, based on the scene-feature threshold library, performs matching analysis on each candidate scene and outputs the successfully matched data as a scene component set; S34, filter the interference components of the data that failed to match, and output the filtered interference component set; S35 uses the data corresponding to the scene component set and the interference component set as the vertical axis of the matrix and the dimension corresponding to the time window as the horizontal axis of the matrix to configure the component matrix corresponding to the signal features.

6. The adaptive identification and extraction method for transient traveling wave signals under complex environments according to claim 5, characterized in that, When performing matching analysis on candidate scenes in step S33, the implementation methods include: For signal features within any time window, the first step is to match based on the feature values ​​of the signal features. If the feature values ​​of the signal features fall within the threshold range, the next step of filtering is performed; otherwise, the candidate scenario is directly eliminated. Secondly, the feature values ​​corresponding to voltage and current in the signal features are used for matching. If the feature values ​​corresponding to voltage and current belong to the threshold range, the current candidate scene is determined to be successfully matched, and the corresponding signal features are output as scene components. If multiple candidate scenes match, sort the values ​​obtained by the convolution integral of each candidate scene in descending order and keep the top two candidate scenes as output.

7. The adaptive identification and extraction method for transient traveling wave signals under complex environments according to claim 1, characterized in that, When performing time-varying statistical analysis in step S42, the implementation method also includes: Examine the mean, variance, peak value, and difference between adjacent time windows for each interference component; The peak mean and the ratio of the difference between adjacent time windows to the mean are used as matching indicators to check the interference type of the current interference component. The weights of the corresponding interference components are set according to the type of interference. The scene components are then subtracted from the weighted interference component values ​​to obtain the analyzed scene components.

8. The adaptive identification and extraction method for transient traveling wave signals under complex environments according to claim 1, characterized in that, Step S5 can be implemented in the following ways: S51, judge the characteristics of the reference signal, determine the characteristics of the reference signal at both ends respectively, and separate the reflected wave and direct wave corresponding to each reference signal characteristic based on the wavefront arrival time difference corresponding to the reference signal characteristics. S52, after dividing the reference signal features, select the reference signal feature with the largest time window proportion as the target signal; S53, perform positioning processing based on the segment location of the target signal to obtain the preliminary positioning range of the target signal; S54, based on the initial positioning range of the target signal under multiple time windows, corrects and fits the initial positioning range to obtain the output fault location.

Images