Power distribution network traveling wave head identification method and system based on kurtosis adaptive threshold

Abstract

The present application relates to power distribution network fault traveling wave ranging technical field, especially in power distribution network traveling wave head recognition method and system based on kurtosis adaptive threshold. Method includes the following steps: collecting the traveling wave signal of power distribution network, the traveling wave signal is pretreated, and zero mean sequence is obtained; wavelet decomposition is carried out to the zero mean sequence, and high frequency detail component is extracted; the high frequency detail component is calculated in real time by using sliding window, and noise root mean square and kurtosis coefficient are obtained, and adaptive detection threshold model is established, and current detection threshold is dynamically output; based on the current detection threshold, the high frequency detail component is checked jointly by using multidimensional waveform constraint condition, and effective fault traveling wave head is identified, and wave head arrival time is recorded; according to the wave head arrival time synchronously collected at line two ends, combined with time synchronization mechanism, power distribution network fault double-end ranging is completed.

Description

Technical Field

[0001] This invention relates to the field of distribution network fault traveling wave location technology, and in particular to a method and system for identifying the wavefront of a distribution network traveling wave based on a kurtosis adaptive threshold. Background Technology

[0002] Fault traveling wave ranging in power distribution networks relies on collecting transient high-frequency voltage and current traveling waves generated by the fault and recording the time it takes for the traveling waves to arrive at the monitoring equipment to calculate the distance between the fault point and the measurement point. Since traveling waves propagate at near the speed of light in the line, the accuracy of capturing the arrival time of the traveling wavefront directly determines the magnitude of the error in the fault ranging result.

[0003] Currently, the most commonly used wavefront identification methods in the industry include the fixed threshold method and the wavelet transform modulus maxima method. The fixed threshold method sets a fixed amplitude threshold in advance; as long as the signal exceeds the threshold, a wavefront is detected. However, noise in power distribution networks varies continuously with load, weather, and season, resulting in significant differences in noise fluctuation amplitude. The fixed threshold cannot adapt to time-varying noise environments, easily leading to false noise identification or missed detection of weak fault traveling waves. While the wavelet transform modulus maxima method can effectively extract high-frequency detail components containing fault information through layered signal processing using db4 wavelets and locate signal abrupt changes based on the modulus maxima of detail coefficients, this method still requires manually setting a fixed judgment threshold and cannot automatically adjust the identification standard according to the real-time noise conditions on site.

[0004] Existing technologies have introduced kurtosis into traveling wave detection to determine whether there is an impulse disturbance in the signal. However, kurtosis is only used as a post-fault statistical auxiliary indicator after the fault occurs. There is no quantitative correlation between kurtosis and the detection threshold, and the detection cannot be dynamically adjusted in real time based on kurtosis.

[0005] Distribution network noise can be categorized into two types: steady-state Gaussian white noise and impulse noise generated by lightning strikes or switching operations. Impulse noise's time-domain waveform closely resembles the wavefront of a real fault traveling wave, making it a primary factor in interference wavefront identification. Existing traveling wave identification schemes generally rely on a single amplitude as the basis for judgment, making it difficult to distinguish between occasional noise spikes and genuine fault traveling waves. This results in low wavefront identification accuracy and ultimately compromises the accuracy of fault location in the distribution network. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for identifying traveling wave fronts in distribution networks based on kurtosis adaptive thresholds, so as to solve the above-mentioned problems.

[0007] To achieve the above objectives, one aspect of the present invention provides a method for identifying the traveling wavefront of a distribution network based on a kurtosis adaptive threshold, comprising the following steps: The traveling wave signal of the power distribution network is collected, and the traveling wave signal is preprocessed to obtain a zero-mean sequence; Wavelet decomposition is performed on the zero-mean sequence to extract high-frequency detail components; A sliding window is used to calculate the high-frequency detail components in real time to obtain the root mean square noise and kurtosis coefficient, and an adaptive detection threshold model is established to dynamically output the current detection threshold. Based on the current detection threshold, multi-dimensional waveform constraints are used to jointly verify the high-frequency detail components, identify the valid fault traveling wave front, and record the arrival time of the wave front. Based on the arrival time of the wavefronts collected synchronously at both ends of the line, and combined with the time synchronization mechanism, the distance measurement of the fault at both ends of the distribution network is completed.

[0008] Furthermore, in the aforementioned method for identifying the wavefront of a distribution network based on a kurtosis adaptive threshold, when acquiring the distribution network traveling wave signal, a 1MHz ADC is used for sampling, and the sampled data is stored in a circular buffer queue; a sliding window with a length of 1024 points is used to calculate the window sampling mean in real time and remove the DC component, outputting a zero-mean sequence; the sliding window refreshes the window data every 128 sampling points.

[0009] Furthermore, in the aforementioned method for identifying traveling wave fronts in distribution networks based on kurtosis adaptive thresholds, the wavelet decomposition uses the db4 wavelet to complete a 5-level decomposition, extracting the high-frequency detail components of the d1, d2, and d3 levels; the wavelet decomposition adopts a point-by-point incremental sliding calculation method, and the wavelet decomposition results are updated synchronously for each new sampling point input.

[0010] Furthermore, in the aforementioned method for identifying traveling wavefronts in distribution networks based on kurtosis adaptive thresholds, the formula for calculating the kurtosis coefficient is as follows: ,in, The second-order central moments of the high-frequency detail components, The fourth-order central moment of the high-frequency detail components; The adaptive detection threshold model satisfies: Where Th is the current detection threshold, and k is a fixed baseline coefficient. denoted as the root mean square of the noise, and f(K) is the kurtosis correction factor.

[0011] Furthermore, in the aforementioned method for identifying the traveling wave front of a distribution network based on a kurtosis adaptive threshold, the kurtosis correction factor f(K) satisfies the following segmentation rules: When the kurtosis coefficient K > the first threshold, the kurtosis correction factor f(K) = 1.2; When the second threshold of the kurtosis coefficient ≤ K ≤ the first threshold, the kurtosis correction factor f(K) = 1.0; When the kurtosis coefficient K < the second threshold, the kurtosis correction factor f(K) = 0.8.

[0012] Furthermore, in the aforementioned method for identifying traveling wave fronts in distribution networks based on kurtosis adaptive thresholds, the fixed reference coefficient k, the first threshold, and the second threshold can all be modified online through the device configuration interface, with the first threshold being 3.5 and the second threshold being 2.5.

[0013] Furthermore, in the aforementioned method for identifying the traveling wave front of a distribution network based on kurtosis adaptive threshold, the multi-dimensional waveform constraints include: Condition 1: The absolute value of the signal amplitude at N consecutive sampling points all exceeds the current detection threshold. N≥3; Condition 2: The difference signs of N consecutive sampling points exceeding the threshold are consistent, and the waveform shows a monotonically increasing or monotonically decreasing trend; Condition 3: The mean polarities of the preset number of sampling points before and after the wavefront are opposite; If any high-frequency detail component in any layer simultaneously meets the above three conditions, then a valid faulty traveling wave head is determined to exist.

[0014] Furthermore, in the aforementioned method for identifying traveling wave fronts in a distribution network based on kurtosis adaptive threshold, the time synchronization mechanism uses GPS second pulses to synchronize the clocks of the ranging terminals at both ends of the line. After obtaining the arrival times of the wave fronts at both ends, the waveforms of the traveling wave fronts at both ends are matched through time-domain correlation analysis to obtain the optimal time offset. The distance to the fault point is then calculated in combination with the preset traveling wave propagation speed.

[0015] Furthermore, in the aforementioned method for identifying traveling wave fronts in distribution networks based on kurtosis adaptive thresholds, all algorithms of the method are implemented in a real-time operating system (RTOS) using multi-priority task collaborative scheduling.

[0016] In a second aspect of the invention, a distribution network traveling wavefront identification system based on kurtosis adaptive threshold is also proposed, which performs the method described in the first aspect, the system comprising: The sampling preprocessing module is used to acquire traveling wave signals from the power distribution network and preprocess the traveling wave signals to output a zero-mean sequence; The wavelet decomposition module is used to perform wavelet decomposition on zero-mean sequences to extract high-frequency detail components. The adaptive threshold calculation module is used to calculate the root mean square noise and kurtosis coefficient of high-frequency detail components in real time using a sliding window, and to construct an adaptive detection threshold model based on the root mean square noise and kurtosis coefficient and output the current detection threshold in real time. The multi-constraint wavefront recognition module is used to perform joint verification of high-frequency detail components based on the current detection threshold and multi-dimensional waveform constraints, identify valid fault traveling wavefronts and record the wavefront arrival time; The synchronous ranging module is used to complete the two-end ranging of distribution network faults based on the arrival time of the wavefront combined with the time synchronization mechanism. It also includes a real-time scheduling module, which is used to run the RTOS (Real-Time Operating System) and coordinate the execution of all system algorithms through multi-priority task scheduling.

[0017] Compared with the prior art, the present invention has at least the following technical effects: This invention first preprocesses and decomposes the acquired traveling wave signal from the distribution network to accurately extract high-frequency detail components containing fault characteristics, solving the problems of traditional fixed threshold methods and wavelet modulus maxima methods that rely on manually fixed thresholds and cannot adapt to time-varying noise in the field. Secondly, by calculating the kurtosis coefficient and root mean square noise in real time through a sliding window, an adaptive detection threshold model related to kurtosis is established, enabling dynamic adjustment of the threshold according to Gaussian noise and impulse noise, thus solving the problems of missed fault detection and false noise detection caused by fixed thresholds. Simultaneously, this invention uses multi-dimensional waveform constraints to jointly verify the high-frequency detail components, effectively distinguishing between similarly shaped impulse noise spikes and the actual fault traveling wave fronts, significantly improving the accuracy of wave front identification. Finally, based on the accurately identified wave front arrival time, combined with a time synchronization mechanism at both ends of the line, high-precision dual-end fault location in the distribution network is achieved. Attached Figure Description

[0018] Figure 1 This is a flowchart of a method for identifying the traveling wave front of a distribution network based on a kurtosis adaptive threshold in one embodiment of the present invention; Figure 2 This is a detailed flowchart of a method for identifying the traveling wave front of a distribution network based on a kurtosis adaptive threshold, according to an embodiment of the present invention. Figure 3 This is a system architecture diagram of a distribution network traveling wave front identification system based on kurtosis adaptive threshold in one embodiment of the present invention; Figure 4 This is a block diagram of a distribution network traveling wave front identification system based on kurtosis adaptive threshold in another embodiment of the present invention. Detailed Implementation

[0019] The following will describe in more detail the method and system for identifying traveling wave fronts in distribution networks based on kurtosis adaptive thresholds according to the present invention, with reference to the schematic diagrams. Preferred embodiments of the invention are shown. It should be understood that those skilled in the art can modify the invention described herein while still achieving its advantageous effects. Therefore, the following description should be understood as being of general knowledge to those skilled in the art and is not intended to limit the invention.

[0020] For clarity, not all features of the actual embodiments are described. In the following description, well-known functions and structures are not detailed in detail, as they would obscure the invention with unnecessary detail. It should be understood that in the development of any actual embodiment, numerous implementation details must be made to achieve the developer's specific objectives, such as changes from one embodiment to another according to limitations related to the system or business. Furthermore, it should be understood that such development work may be complex and time-consuming, but is merely routine work for those skilled in the art.

[0021] The invention is described more specifically by way of example in the following paragraphs with reference to the accompanying drawings. The advantages and features of the invention will become clearer from the following description. It should be noted that the drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the invention.

[0022] Based on the teachings of this specification, those skilled in the art can form new technical solutions through cross-combination of different implementation methods without creating technical contradictions. Such variations should all be considered to fall within the protection scope of this invention.

[0023] Example 1 like Figure 1 and Figure 2 This embodiment provides a method for identifying the wavefront of a traveling wave in a distribution network based on a kurtosis adaptive threshold. This method addresses the problems of traditional fixed-threshold traveling wave detection methods, which are susceptible to interference from Gaussian white noise and impulse noise, leading to low wavefront identification accuracy, frequent false alarms and missed alarms, and poor fault location accuracy. The method includes the following steps: S1: Acquire traveling wave signals from the power distribution network, preprocess the traveling wave signals to obtain a zero-mean sequence; S2: Perform wavelet decomposition on the zero-mean sequence to extract high-frequency detail components; S3: A sliding window is used to calculate the high-frequency detail components in real time to obtain the root mean square noise and kurtosis coefficient, and an adaptive detection threshold model is established to dynamically output the current detection threshold. S4: Based on the current detection threshold, use multi-dimensional waveform constraints to jointly verify the high-frequency detail components, identify the valid fault traveling wave front, and record the arrival time of the wave front; S5: Based on the arrival time of the wavefronts synchronously collected at both ends of the line, and combined with the time synchronization mechanism, complete the dual-end ranging of the distribution network fault.

[0024] For step S1, when acquiring the traveling wave signal of the distribution network, a 1MHz ADC is used for sampling, and the sampled data is stored in a circular buffer queue; a sliding window with a length of 1024 points is used to calculate the window sampling mean in real time and remove the DC component, and output a zero-mean sequence; the sliding window refreshes the window data once every 128 sampling points.

[0025] Specifically, the traveling wave signals acquired in this embodiment include current traveling waves and voltage traveling waves. The voltage and current traveling waves are obtained through a capacitive voltage divider (CVT) and a Rogowski coil current transformer, respectively. After high-frequency noise interference is filtered out by a front-end analog filter circuit, the signals are sent to a high-speed ADC sampling chip. The ADC chip is configured with a fixed sampling rate of 1MHz for signal digitization, with a single sampling interval of 1μs. The raw traveling wave data obtained from sampling is written to the embedded system's circular buffer queue in real time to avoid data loss and frame misalignment during high-speed sampling, ensuring the integrity and continuity of the sampled data.

[0026] Furthermore, to eliminate detection errors caused by grid baseline drift and DC bias, this embodiment employs a sliding window of 1024 points to preprocess the buffered sampled data. This window corresponds to a time length of 1.024ms, which can cover multiple power frequency cycles, ensuring the stability of statistical data. The system calculates the signal mean of all sampling points within the sliding window in real time, subtracts the mean from each original sampled value within the window to remove the DC component of the signal, and finally outputs a zero-mean sampling sequence without DC offset.

[0027] It should be noted that the sliding window in this embodiment adopts an incremental refresh method, refreshing the window data once every 128 sampling points (corresponding to 128μs). This can balance the real-time performance of the algorithm with the stability of data statistics, and avoid the waste of computing power and processing delay caused by refreshing the entire window frame.

[0028] For step S2, the wavelet decomposition uses the db4 wavelet to complete a 5-level decomposition, extracting the high-frequency detail components of the d1, d2, and d3 levels; the wavelet decomposition adopts a point-by-point incremental sliding calculation method, and the wavelet decomposition result is updated synchronously for each new sampling point input.

[0029] Specifically, this embodiment selects the db4 wavelet, which has excellent regularity and compact support characteristics suitable for power traveling wave analysis, to perform a 5-level hierarchical decomposition on the zero-mean sequence output in step S1. During the wavelet decomposition process, parallel filtering is performed using db4 wavelet-specific high-pass and low-pass filter coefficients, followed by a 2x downsampling. Each level of decomposition outputs low-frequency approximate components (cA1-cA5) and high-frequency detail components (d1-d5), and the approximate components of the previous level are iteratively decomposed layer by layer.

[0030] Since the low-frequency approximation components cA1~cA5 contain a 50Hz power frequency useless signal, this embodiment directly discards cA1~cA5 and ultra-low frequency d4, d5, and only extracts the three layers of high-frequency detail components d1, d2 and d3 as the data source for subsequent wavefront detection. The corresponding frequency bands are 250kHz~500kHz, 125kHz~250kHz and 62.5kHz~125kHz, respectively, which completely covers the effective spectrum range of the transient fault traveling wave in the distribution network.

[0031] Furthermore, in order to solve the problems of large delay and poor real-time performance of traditional whole-frame wavelet decomposition, this embodiment adopts a point-by-point incremental sliding calculation method. That is, for each input of a new zero-mean sampling point after step S1 and processing, the wavelet decomposition result will be updated in real time, and the latest d1, d2, d3 high-frequency detail components will be output synchronously without waiting for the whole frame of 1024 points of data to be collected, which greatly reduces the output delay of wavelet decomposition.

[0032] For step S3, the three high-frequency detail components d1, d2 and d3 extracted in step S2 are each configured with an independent sliding window of 1024 points, and the new sliding period of the window is also 128 sampling points (128μs).

[0033] To reduce the computational overhead of embedded terminals, this embodiment employs an incremental update algorithm to perform iterative calculations of the second and fourth central moments of the window: when a new sampling point slides into the window, its contribution to the second and fourth central moments is accumulated in real time; when an old sampling point slides out of the window, its corresponding moment contribution is subtracted synchronously. This eliminates the need for repeated calculations every 128μs by traversing all 1024 sampling points of the window, significantly reducing floating-point operations. Each sliding window layer synchronously outputs two types of parameters: one is the root mean square noise σ. n One is used to quantify the overall amplitude fluctuation level of the current line background noise; the other is the second-order central moment m2 and the fourth-order central moment m4 of the window sequence.

[0034] Kurtosis is a fourth-order normalized central moment that characterizes the distribution pattern of a random variable. It can intuitively reflect the sharpness of the peaks and the thickness of the tails of the waveform distribution. The kurtosis of a Gaussian distribution signal is fixed at 3, which is well-suited to the noise differentiation requirements of the power distribution network.

[0035] There are two types of background noise in the power distribution network operation site. One type is Gaussian white noise formed by the thermal noise of transformers and acquisition chips, with a uniform power spectrum and a kurtosis coefficient stable at around 3. The other type is pulse noise generated by switch opening and closing, thyristor commutation, and lightning strike induction. Its time domain waveform is highly similar to the fault traveling wave front, and its duration is short. Its kurtosis coefficient will be much greater than 3, which is also the main interference source causing wave front misidentification.

[0036] In this embodiment, the formula for calculating the kurtosis coefficient K is: ,in, The second-order central moments of the high-frequency detail components, This represents the fourth-order central moment of the high-frequency detail components. Specifically, the second-order central moment... That is, the variance of the window signal and the fourth central moment. The fourth power average represents the deviation of the signal from the mean. By iteratively calculating the kurtosis coefficient K of each high-frequency detail component in real time, the severity of impulse interference in the current background noise can be determined in real time. The larger the kurtosis coefficient value, the more severe the impulse noise, and the higher the detection threshold should be to avoid false identification; the closer the kurtosis coefficient is to 3, the more serious the noise is, indicating that the noise is mainly stationary Gaussian noise, and the threshold can be appropriately lowered to improve the detection capability of weak ground fault traveling waves.

[0037] It should be noted that the kurtosis coefficient is updated every 128μs, but the single-point detection delay of the front-end wavefront is determined only by the 1MHz sampling interval and is fixed at 1μs. The real-time wavefront identification performance is not affected by the kurtosis window refresh interval.

[0038] Furthermore, an adaptive detection threshold model is constructed using the root mean square of noise and the kurtosis coefficient, satisfying: Where Th is the current detection threshold. The root mean square of the noise is given, and k is a fixed reference coefficient. This is the kurtosis correction factor. Based on extensive field measurements in distribution networks, the factory default optimal empirical value for the fixed baseline coefficient k is 5.

[0039] The kurtosis correction factor f(K) in this embodiment satisfies the following segmentation rules: When the kurtosis coefficient K > the first threshold, the kurtosis correction factor f(K) = 1.2; When the second threshold of the kurtosis coefficient ≤ K ≤ the first threshold, the kurtosis correction factor f(K) = 1.0; When the kurtosis coefficient K < the second threshold, the kurtosis correction factor f(K) = 0.8.

[0040] In a preferred embodiment, the first threshold is 3.5 and the second threshold is 2.5.

[0041] When K > 3.5, it indicates severe impulse noise interference. Setting f(K) = 1.2 raises the overall detection threshold, suppressing false wavefronts triggered by numerous short-term impulse spikes. When 2.5 ≤ K ≤ 3.5, the line noise is close to standard Gaussian stationary noise, commonly seen in suburban light-load distribution lines. Setting f(K) = 1.0 keeps the baseline detection threshold unchanged, balancing identification sensitivity and anti-interference capability. When K < 2.5, it represents an extremely low noise scenario with almost no impulse interference. Setting f(K) = 0.8 lowers the detection threshold, amplifying the detection probability of weak fault transient signals and preventing the missed identification of weak traveling wave signals such as high-resistance grounding.

[0042] The first threshold, second threshold, and fixed reference coefficient k mentioned above are all optimal empirical parameters obtained from multi-scenario experiments. However, they are not fixed and cannot be modified. They can all be modified online through the device configuration interface without stopping or restarting the device. For example, in heavy industrial high-noise corridor scenarios, the fixed reference coefficient k can be increased to 6~8, while in suburban low-noise branch lines, the fixed reference coefficient k can be decreased to 3~4. In densely industrial areas, the first threshold and second threshold can be fine-tuned to 3.8 and 2.8, respectively, to match the local long-term noise statistical distribution.

[0043] For step S4, to address the issue that a single frame threshold cannot distinguish between noise spikes and true traveling waves, this embodiment sets triple waveform constraints based on an adaptive dynamic threshold. The three layers of high-frequency detail components (d1, d2, and d3) extracted after wavelet decomposition are independently verified layer by layer. Each layer of high-frequency detail components undergoes the same set of triple waveform constraints, and the three layers are judged independently. If any one of the three layers (d1, d2, and d3) simultaneously satisfies the triple constraints, a valid fault traveling wave front can be detected. If none of the three layers of high-frequency detail components simultaneously satisfy the multiple constraints, the current sampling point is skipped, and the cycle continues to verify the next sampling point.

[0044] Specifically, the multi-dimensional waveform constraint conditions include: Condition 1: The absolute value of the signal amplitude at N consecutive sampling points (N≥3) all exceeds the current detection threshold. It can avoid triggering single-point instantaneous noise spikes caused by lightning strikes and switching interference. Such interference only exceeds the standard at a single sampling point and cannot meet the requirements for continuous multi-point operation.

[0045] Condition 2: The differential signs of N consecutive sampling points exceeding the threshold are consistent, and the waveform shows a monotonically rising or monotonically falling trend. A real fault traveling wave is generated by abrupt changes in line parameters, and its wavefront waveform is smooth and monotonous. In contrast, impulse noise is accompanied by high-frequency oscillations, and the waveform differential signs alternate between positive and negative, failing to meet this condition. Therefore, noise and the real wavefront can be accurately distinguished.

[0046] Condition 3: The mean polarities of the sampling points before and after the wavefront are opposite. In this embodiment, the mean polarities of the signals at 3-5 sampling points before and after the wavefront are opposite. A real fault is a transient change in line voltage, which will produce a significant jump in voltage or current polarity. Interference such as baseline drift and slow load fluctuations do not exhibit polarity reversal characteristics, thus completely eliminating false detections caused by baseline offset.

[0047] After completing the triple constraint joint verification of the three layers of high-frequency detail components, the valid wavefront determination and information storage process will begin. When any one of the high-frequency detail components in d1, d2, or d3 simultaneously satisfies all three conditions, the system immediately determines that a valid fault traveling wavefront has been captured at the current sampling time. At this point, the program immediately reads the high-precision local clock calibrated by the GPS time synchronization module, extracts the precise arrival time of the wavefront, and synchronously records the wavelet layer number and relative amplitude of the wavefront. The complete wavefront information, including layer information, timestamp, and amplitude characteristics, is stored in the system and then used as the input data for the two-end ranging in step S5 for subsequent use.

[0048] For step S5, the time synchronization mechanism uses GPS pulse-per-second (PPS) to synchronize the clocks of the ranging terminals at both ends of the line. After obtaining the arrival times of the wavefronts at both ends, the traveling wavefront waveforms at both ends are matched by time-domain correlation analysis to obtain the optimal time offset. The distance to the fault point is then calculated in combination with the preset traveling wave propagation speed.

[0049] Specifically, the dual-end ranging in step S5 is completed collaboratively by two independent ranging terminals at both ends of the line. Each ranging device will independently and completely execute the entire process from step S1 to step S4. The two devices independently complete the fault wavefront detection. After each device identifies a valid fault traveling wavefront, it records the arrival timestamp of the wavefront with a unified GPS time base, which is denoted as the local timestamp t1 and the remote timestamp t2.

[0050] In this embodiment, the dual-end ranging scenario uses GPS pulse-per-second (PPS) to achieve high-precision time synchronization between the two ranging devices. Both ranging terminals at either end of the line are connected to external GPS receivers, continuously receiving the PPS signal output by GPS to align and calibrate their local real-time clocks. Every second, the GPS PPS triggers a high-priority GPS synchronization task within the RTOS to refresh the local clock reference. The GPS synchronization accuracy can control the clock synchronization accuracy at both ends within 0.5 μs. This time error corresponds to a traveling wave propagation distance of approximately 150 m, effectively controlling the basic system error in the ranging process. The wavefront timestamps t1 and t2 recorded by all terminals are based on the same GPS standard time base, ensuring the validity of direct comparison and calculation of the timestamps at both ends.

[0051] After the traveling waves generated by the line fault are captured and wavefronts are identified by the terminals at both ends, the local and remote ranging devices exchange locally stored high-frequency wavelet waveform segments via a communication link. The waveform segment extracted by the local end is denoted as S1, and the waveform segment transmitted from the remote end is denoted as S2. After data exchange, the RTOS schedules low-priority correlation analysis and ranging tasks to perform time-domain correlation matching operations to eliminate timestamp matching deviations caused by waveform distortion and interference. The processor then executes the relevant function formula... Where i is the sampling point number, all offsets τ are traversed for calculation, the peak value of the correlation function is solved, and the offset corresponding to the peak value is the optimal matching offset of the wavefronts at both ends. Based on relevant analysis, the automatic and accurate matching of fault waveforms at both ends is achieved to distinguish between real fault traveling waves and false waveforms caused by line noise.

[0052] After solving for the optimal matching offset, first calculate the combined wavefront time difference Δt at both ends. The calculation formula is as follows: This time difference integrates the GPS synchronization timestamp difference and waveform matching correction offset, thus fully reflecting the true time difference of the fault traveling wave arriving at both ends of the terminal. Then, the fault distance is calculated by combining this with the pre-calibrated traveling wave propagation speed v within the equipment. In this embodiment, the traveling wave propagation speed is taken as 0.96 to 0.98 times the speed of light in a vacuum. The calculation formula for the distance from the fault point to the local ranging device is as follows: Finally, the fault distance is calculated.

[0053] Furthermore, all algorithms in this embodiment are implemented in a real-time operating system (RTOS) using multi-priority task collaborative scheduling. The algorithm is applicable to embedded processor platforms such as ARM Cortex-A9, abandoning the traditional batch frame processing mode and following a priority-driven real-time scheduling principle for task architecture partitioning. Critical paths use interrupt service routines to ensure hard real-time performance, while non-critical paths are implemented using periodic tasks. The specific task architecture is as follows... Figure 3 and Table 1 below: Table 1

[0054] Specifically, the traveling wave signal acquisition and preprocessing process involved in step S1 is implemented collaboratively by the system's highest-priority sampling interrupt service routine and the second-highest-priority deDC cycle task. The ADC sampling interrupt is triggered every 1μs, and the highest-priority sampling interrupt service routine acquires the raw traveling wave signal from the distribution network in real time, writing the sampled data into a circular buffer to achieve continuous high-speed signal acquisition. The deDC cycle task operates with a fixed cycle of 128μs, periodically reading 1024 points of sliding window sampling data from the circular buffer, calculating the window mean to remove the DC drift component of the signal, and outputting a standardized zero-mean sequence, ultimately completing the preprocessing operation of the traveling wave signal.

[0055] The zero-mean sequence wavelet decomposition and high-frequency component extraction process involved in step S2 is independently implemented by the system's second-highest priority wavelet decomposition task. The wavelet decomposition task employs a point-by-point incremental sliding calculation method, synchronously updating the decomposition results of the zero-mean sequence output from S1 point by point. Based on the db4 wavelet, it completes five levels of wavelet decomposition operations, extracting and outputting the three levels of high-frequency detail components (d1, d2, and d3) in real time. This avoids the algorithmic delay caused by batch calculations across entire frames, ultimately completing the extraction of the effective high-frequency feature components of the traveling wave.

[0056] The high-frequency detail component calculation and adaptive threshold modeling process involved in step S3 is achieved collaboratively by the system's priority kurtosis calculation task and the adaptive threshold calculation task. The kurtosis calculation task operates with a fixed cycle of 128 μs, using an incremental update algorithm within a 1024-point sliding window to calculate the noise root mean square, second-order central moment, and fourth-order central moment of the high-frequency detail components in real time, thus obtaining the real-time kurtosis coefficients. The adaptive threshold calculation task runs synchronously with the kurtosis calculation task, matching the corresponding kurtosis correction factor based on the real-time kurtosis coefficients, and dynamically iteratively calculating the current detection threshold using the noise root mean square and a fixed baseline coefficient, ultimately completing the real-time output of the adaptive detection threshold model.

[0057] The multi-dimensional waveform constraint verification and valid wavefront identification process involved in step S4 is independently implemented by the highest priority wavefront criterion interrupt service program in the system. Specifically, the wavefront criterion interrupt is executed continuously in a sampling-point manner, retrieving the real-time detection threshold output in step S3. It simultaneously performs joint verification of three waveform constraints—continuous threshold exceedance, consistent differential sign, and opposite wavefront polarity—on the three layers of high-frequency detail components. This process filters out truly valid faulty traveling wavefronts, accurately captures and records the precise arrival time of the wavefront, and ultimately completes the identification and time-marking operation of the faulty traveling wavefront.

[0058] The time synchronization and dual-end fault location process involved in step S5 is achieved collaboratively by the system's high-priority GPS synchronization task and the low-priority correlation analysis ranging task. The GPS synchronization task is triggered by a second-by-second GPS pulse interruption, completing the local clock calibration of the ranging terminals at both ends of the line, establishing a high-precision unified time synchronization mechanism, and accurately providing time synchronization for the wavefront moment. The correlation analysis ranging task starts after the fault wavefront event is triggered, reading the arrival time and waveform data of the synchronization wavefronts at both ends of the line, matching the optimal time offset through time-domain correlation analysis, and iteratively solving the fault point distance by combining the traveling wave propagation speed, ultimately completing the dual-end fault location of the distribution network.

[0059] To ensure the stability and data consistency of multi-task parallel operation, data transfer between tasks and interrupts is accomplished through shared memory and a circular buffer, resolving the rate matching issue between high-speed sampling and algorithm processing. For the high-frequency real-time data path of kurtosis threshold update and wavefront criterion reading, a double buffering mechanism is specifically adopted for data isolation, which can avoid multi-task read / write conflicts and reduce recognition errors caused by data anomalies.

[0060] Furthermore, to verify the detection performance and engineering applicability of the method of the present invention, comparative tests were conducted under laboratory conditions in this embodiment. The test adopted a 1MHz sampling rate and a 20dB signal-to-noise ratio environment, targeting a simulated 10kV distribution line, injecting a typical double-exponential fault traveling wave signal with a rise time of 1μs and a decay time constant of 50μs, and superimposing Gaussian white noise and impulse noise to form a mixed background noise. A total of 200 sets of samples were tested, and the performance indicators of the adaptive threshold method of the present invention and the traditional fixed threshold method were compared, as shown in Table 2 below.

[0061] Table 2

[0062] The test results show that the core performance indicators of the method of this invention are significantly better than those of the traditional fixed threshold method. The traditional fixed threshold method has a wavefront recognition accuracy of only 68%, with false recognition and missed recognition rates as high as 30% and 32% respectively, an average fault location error of approximately 500m, and a software processing delay of less than 1ms. In contrast, the method of this invention achieves a wavefront recognition accuracy of 94%, with false recognition and missed recognition rates both controlled below 6%, and an average fault location error reduced to approximately 200m. Furthermore, relying on a point-by-point interruption processing mechanism, the signal processing delay is compressed to less than 1μs, resulting in a qualitative improvement in real-time detection performance.

[0063] In summary, this invention identifies the statistical characteristics of on-site noise by calculating the kurtosis coefficient in real time and constructs an adaptive threshold model, effectively adapting to complex noise environments. This significantly improves the accuracy of fault traveling wave front identification, increasing it by 26 percentage points compared to traditional methods. Subsequently, by jointly judging three constraints, noise spikes and oscillation interference are precisely filtered out, thoroughly overcoming the shortcomings of traditional fixed threshold methods that are susceptible to noise interference and significantly reducing the probability of false and missed identifications. Simultaneously, the method combines GPS time synchronization at both ends of the line with a time-domain correlation waveform matching algorithm, effectively optimizing fault ranging accuracy and keeping its error within the required range to meet engineering positioning accuracy requirements.

[0064] In addition, this invention possesses excellent real-time performance and environmental adaptability. The algorithm implements multi-task deterministic scheduling based on an RTOS real-time operating system. The wavefront criterion adopts a sampling-point interrupt execution method, with a processing latency of no more than 1μs. The kurtosis threshold calculation uses an incremental update algorithm to reduce computational power consumption, ensuring the system's hard real-time operating characteristics, and the overall ranging response remains stable at the millisecond level. Simultaneously, the detection threshold can dynamically and adaptively adjust according to the on-site noise conditions, eliminating the need for manual presets and on-site parameter debugging, thus adapting to different operating conditions and noise levels in distribution network scenarios.

[0065] Example 2 like Figure 2 This embodiment proposes a distribution network traveling wave front identification system based on kurtosis adaptive threshold, which executes the method described in Embodiment 1. The system includes a sampling preprocessing module, a wavelet decomposition module, an adaptive threshold calculation module, a multi-constraint wave front identification module, a synchronous ranging module, and a real-time scheduling module.

[0066] The sampling preprocessing module is used to acquire and preprocess traveling wave signals from the distribution network, outputting a zero-mean sequence. Specifically, this module uses a 1MHz ADC high-speed sampling chip to acquire fault voltage and current traveling wave signals from the distribution network in real time, storing the raw sampled data into a circular buffer queue in real time to effectively avoid data loss during high-speed sampling. Simultaneously, the module uses a 1024-point sliding window for real-time data preprocessing. By calculating the mean of the sampled data within the window and removing the DC component of the signal, a standard zero-mean sequence is generated. The sliding window refreshes the data every 128 sampling points, ensuring both data preprocessing accuracy and real-time system efficiency, providing a stable, baseline-off input signal for subsequent modules.

[0067] The wavelet decomposition module is used to perform wavelet decomposition on zero-mean sequences to extract high-frequency detail components. Specifically, this module uses the db4 wavelet to perform a 5-level wavelet decomposition on the zero-mean sequence, discarding low-frequency approximations and high-frequency invalid detail components, and accurately extracting the high-frequency detail components (d1, d2, and d3) containing transient fault features as the core data for wavefront identification. Simultaneously, the module adopts a point-by-point incremental sliding calculation mode, updating the wavelet decomposition results synchronously with each new zero-mean sampling point received. This eliminates the need for batch processing of entire frames of data, significantly reducing signal processing latency and enabling real-time, continuous output of high-frequency detail components, ensuring the system's ability to capture short-term, weak fault traveling waves.

[0068] The adaptive threshold calculation module uses a sliding window to calculate the root mean square (RMS) noise and kurtosis coefficient of high-frequency detail components in real time. Based on the RMS and kurtosis coefficients, it constructs an adaptive detection threshold model and outputs the current detection threshold in real time. Specifically, this module calculates the RMS noise, second-order central moment, and fourth-order central moment of the high-frequency detail components in real time using the formula... The module calculates the real-time kurtosis coefficient to accurately characterize the noise distribution characteristics of the current power distribution network. Based on this, the module... An adaptive detection threshold model is constructed, and a corresponding kurtosis correction factor f(K) is matched segmentally according to the numerical range of the kurtosis coefficient. The detection threshold is adaptively adjusted under three operating conditions: high impulse noise, stationary Gaussian noise, and low noise. The details have been described in Implementation Example 1 and will not be repeated here. In addition, the fixed reference coefficient and the upper and lower thresholds of the kurtosis coefficient in the model can be modified online through the device configuration interface, which can flexibly adapt to the power distribution network operating environment of different regions and different operating conditions, and dynamically output the optimal real-time detection threshold.

[0069] The multi-constraint wavefront identification module is used to jointly verify high-frequency detail components based on the current detection threshold using multi-dimensional waveform constraints, identify valid fault traveling wavefronts, and record the wavefront arrival time. Specifically, this module simultaneously executes three constraints: amplitude constraint (at least three consecutive sampling points with amplitudes exceeding the threshold), monotonicity constraint (consistent difference sign of consecutive sampling points exceeding the threshold), and abrupt change constraint (opposite polarity of the mean values ​​of sampling points before and after the wavefront). As long as any layer of high-frequency detail components simultaneously satisfies all three constraints, a valid fault traveling wavefront can be detected, and the wavefront arrival time is immediately and accurately recorded based on GPS synchronization time. By filtering out interference factors such as single-point glitch noise, oscillation interference, and baseline drift from multiple dimensions including amplitude, waveform morphology, and abrupt change characteristics, the accuracy and stability of wavefront identification are greatly improved.

[0070] The synchronous ranging module is used to perform dual-end fault location in distribution networks based on wavefront arrival times combined with a time synchronization mechanism. Specifically, this module uses GPS second pulses to synchronize the clocks of the ranging terminals at both ends of the line, achieving microsecond-level clock alignment between the devices at both ends and ensuring the consistency of wavefront arrival time data at both ends. After obtaining the synchronized wavefront arrival times at both ends of the line, the module performs matching calculations on the traveling wave waveforms at both ends using a time-domain correlation analysis algorithm to solve for the optimal time offset. Combined with the standard propagation speed of traveling waves in the distribution network, the module accurately calculates the fault location using a dual-end ranging formula, effectively avoiding the reflected wave interference problem of single-end ranging and significantly improving fault location accuracy.

[0071] The real-time scheduling module is used to run the entire system's algorithms through multi-priority task collaborative scheduling. Specifically, this module performs priority-based hierarchical scheduling of various tasks such as sampling and acquisition, wavelet decomposition, threshold calculation, wavefront identification, and fault location based on the real-time requirements of each business link, as detailed in Table 1 of Implementation Example 1, which will not be repeated here. Subsequently, through hierarchical task scheduling, double-buffered data isolation, and interrupt-coordinated operation, data read / write conflicts and system scheduling jitter are avoided, ensuring the deterministic and hard real-time operation of the entire wave recognition and ranging algorithm, and guaranteeing stable and reliable system operation under high-speed sampling and continuous computation conditions.

[0072] The above are merely preferred embodiments of the present invention and do not constitute any limitation on the present invention. Any equivalent substitutions or modifications made by those skilled in the art to the technical solutions and content disclosed in the present invention without departing from the scope of the present invention shall be deemed to have remained within the protection scope of the present invention.

Claims

1. A method for identifying the wavefront of a traveling wave in a distribution network based on kurtosis adaptive threshold, characterized in that, Includes the following steps: The traveling wave signal of the power distribution network is collected, and the traveling wave signal is preprocessed to obtain a zero-mean sequence; Wavelet decomposition is performed on the zero-mean sequence to extract high-frequency detail components; A sliding window is used to calculate the high-frequency detail components in real time to obtain the root mean square noise and kurtosis coefficient, and an adaptive detection threshold model is established to dynamically output the current detection threshold. Based on the current detection threshold, multi-dimensional waveform constraints are used to jointly verify the high-frequency detail components, identify the valid fault traveling wave front, and record the arrival time of the wave front. Based on the arrival time of the wavefronts collected synchronously at both ends of the line, and combined with the time synchronization mechanism, the distance measurement of the fault at both ends of the distribution network is completed.

2. The method for identifying the traveling wave front of a distribution network based on a kurtosis adaptive threshold according to claim 1, characterized in that, When acquiring the traveling wave signal of the distribution network, a 1MHz ADC is used for sampling, and the sampled data is stored in a circular buffer queue. A sliding window with a length of 1024 points is used to calculate the window sampling mean in real time and remove the DC component, outputting a zero-mean sequence. The sliding window refreshes the window data every 128 sampling points.

3. The method for identifying the traveling wave front of a distribution network based on a kurtosis adaptive threshold according to claim 1, characterized in that, The wavelet decomposition uses the db4 wavelet to complete a 5-level decomposition, extracting high-frequency detail components from the d1, d2, and d3 levels. The wavelet decomposition adopts a point-by-point incremental sliding calculation method, and the wavelet decomposition results are updated synchronously for each new sampling point input.

4. The method for identifying the traveling wave front of a distribution network based on a kurtosis adaptive threshold according to claim 1, characterized in that, The formula for calculating the kurtosis coefficient is as follows: ,in, The second-order central moments of the high-frequency detail components, The fourth-order central moment of the high-frequency detail components; The adaptive detection threshold model satisfies: Where Th is the current detection threshold, and k is a fixed baseline coefficient. denoted as the root mean square of the noise, and f(K) is the kurtosis correction factor.

5. The method for identifying the traveling wave front of a distribution network based on a kurtosis adaptive threshold according to claim 4, characterized in that, The kurtosis correction factor f(K) satisfies the following segmentation rules: When the kurtosis coefficient K > the first threshold, the kurtosis correction factor f(K) = 1.2; When the second threshold of the kurtosis coefficient ≤ K ≤ the first threshold, the kurtosis correction factor f(K) = 1.0; When the kurtosis coefficient K < the second threshold, the kurtosis correction factor f(K) = 0.

8.

6. The method for identifying the traveling wave front of a distribution network based on a kurtosis adaptive threshold according to claim 5, characterized in that, The fixed reference coefficient k, the first threshold and the second threshold can all be modified online through the device configuration interface. The first threshold is 3.5 and the second threshold is 2.

5.

7. The method for identifying the traveling wave front of a distribution network based on a kurtosis adaptive threshold according to claim 1, characterized in that, The multi-dimensional waveform constraints include: Condition 1: The absolute value of the signal amplitude at N consecutive sampling points all exceeds the current detection threshold. N≥3; Condition 2: The difference signs of N consecutive sampling points exceeding the threshold are consistent, and the waveform shows a monotonically increasing or monotonically decreasing trend; Condition 3: The mean polarities of the preset number of sampling points before and after the wavefront are opposite; If any high-frequency detail component in any layer simultaneously meets the above three conditions, then a valid faulty traveling wave head is determined to exist.

8. The method for identifying the traveling wave front of a distribution network based on kurtosis adaptive threshold according to claim 1, characterized in that, The time synchronization mechanism uses GPS second pulses to synchronize the clocks of the ranging terminals at both ends of the line. After obtaining the arrival times of the wavefronts at both ends, the traveling wavefront waveforms at both ends are matched through time domain correlation analysis to obtain the optimal time offset. The distance to the fault point is then calculated in combination with the preset traveling wave propagation speed.

9. The method for identifying the traveling wave front of a distribution network based on a kurtosis adaptive threshold according to claim 1, characterized in that, All algorithms described herein are implemented in a real-time operating system (RTOS) using multi-priority task collaborative scheduling.

10. A distribution network traveling wave front identification system based on kurtosis adaptive threshold, characterized in that, The system for performing the method according to any one of claims 1-9 comprises: The sampling preprocessing module is used to acquire traveling wave signals from the power distribution network and preprocess the traveling wave signals to output a zero-mean sequence; The wavelet decomposition module is used to perform wavelet decomposition on zero-mean sequences to extract high-frequency detail components. The adaptive threshold calculation module is used to calculate the root mean square noise and kurtosis coefficient of high-frequency detail components in real time using a sliding window, and to construct an adaptive detection threshold model based on the root mean square noise and kurtosis coefficient and output the current detection threshold in real time. The multi-constraint wavefront recognition module is used to perform joint verification of high-frequency detail components based on the current detection threshold and multi-dimensional waveform constraints, identify valid fault traveling wavefronts and record the wavefront arrival time; The synchronous ranging module is used to complete the two-end ranging of distribution network faults based on the arrival time of the wavefront combined with the time synchronization mechanism. It also includes a real-time scheduling module, which is used to run the RTOS (Real-Time Operating System) and coordinate the execution of all system algorithms through multi-priority task scheduling.

Images