A fault traveling wave head data acquisition method and device

Abstract

The application discloses a fault traveling wave wave head data acquisition method and device, the method comprises the following steps: acquiring a data time window intercepted from high-frequency sampling data; using a traveling wave inference model to process the data time window, obtaining a probability that the data time window contains a fault traveling wave characteristic; judging the probability based on a preset threshold, obtaining an absolute UTC time when the probability is greater than the preset threshold; obtaining a sampling index of a fault traveling wave wave head starting point according to the data time window; obtaining an absolute arrival time of the fault traveling wave wave head based on the absolute UTC time, the sampling index and a preset high sampling frequency; intercepting the data time window based on the fault traveling wave wave head starting point and a first preset length, obtaining a waveform segment; and obtaining fault traveling wave wave head data based on the waveform segment and the absolute arrival time. The application can solve the problems of low reliability of a fixed threshold method as a starting criterion and great communication pressure caused by uploading of massive original data in the related art.

Description

Technical Field

[0001] This application relates to the field of power distribution network line detection technology, and in particular to a method and device for acquiring fault traveling wave head data. Background Technology

[0002] When a ground fault or short-circuit fault occurs at a point in a power distribution network, the voltage at the fault point changes instantaneously, generating a high-frequency transient current / voltage traveling wave signal. This traveling wave signal propagates at near the speed of light (approximately 300 to 200 meters per second, depending on the line type) to both ends of the line and all branches. The traveling wave is reflected and refracted when it encounters points of impedance discontinuity (such as busbars, branch points, switches, or the fault point itself). When a fault occurs in a power distribution line, a terminal based on traveling wave fault location technology can be used for location; however, this technology still has some limitations.

[0003] The reliability of terminal triggering mechanisms is low. Traditional terminals generally use fixed threshold methods (such as voltage / current exceeding a certain value) or simple frequency-energy methods as activation criteria. The distribution network environment is complex, with frequent interference from switching operations and load switching. Such methods are prone to false tripping under strong noise, while they are prone to failure to trip under weak fault signals such as high-resistance grounding.

[0004] The system lacks local intelligent processing capabilities. Traditional terminals merely act as "pipelines" for data acquisition, lacking the ability to perform preliminary analysis and understanding of waveform content. The inability to effectively refine information at the source of data generation results in low data value density. Furthermore, the indiscriminate uploading of massive amounts of raw data places enormous pressure on the communication network and the main station, severely restricting system scalability and cost-effectiveness.

[0005] In summary, the existing technologies suffer from a series of shortcomings in terms of adaptability, environmental robustness, intelligence, and engineering economics, which collectively limit the effectiveness of traveling wave fault location technology in power distribution networks. Therefore, there is an urgent need for an innovative solution that can fundamentally overcome these bottlenecks. Summary of the Invention

[0006] This application provides a method and apparatus for acquiring fault traveling wave head data, in order to solve the problems of low reliability of the fixed threshold method as a start-up criterion and high communication pressure caused by uploading massive amounts of raw data in related technologies.

[0007] In a first aspect, embodiments of this application provide a method for acquiring fault traveling wave head data, the method comprising:

[0008] A data time window extracted from high-frequency sampled data is obtained, and the data time window is processed using a traveling wave inference model to obtain the probability that the data time window contains fault traveling wave characteristics.

[0009] The probability is judged based on a preset threshold to obtain the absolute UTC time when the probability is determined to be greater than the preset threshold.

[0010] Based on the data time window, obtain the sampling index of the starting point of the fault traveling wave head;

[0011] Based on the absolute UTC time, sampling index, and preset high sampling frequency, the absolute arrival time of the fault traveling wave front is obtained;

[0012] Based on the starting point of the fault traveling wavefront and the first preset length, the data time window is truncated to obtain waveform segments.

[0013] Based on the waveform segment and absolute arrival time, fault traveling wave head data are obtained.

[0014] In conjunction with the first aspect, in one implementation, before acquiring the data time window, the method further includes: sampling at a preset high sampling frequency to acquire high-frequency sampling data and storing it in a circular buffer.

[0015] Obtaining the data time window extracted from the high-frequency sampled data includes: reading the high-frequency sampled data from the circular buffer and extracting a continuous data sequence of a second preset length as the data time window.

[0016] In conjunction with the first aspect, in one implementation, the traveling wave inference model employs a one-dimensional CNN model.

[0017] In conjunction with the first aspect, in one implementation, the probability is determined based on a preset threshold to obtain the absolute UTC time when the probability is determined to be greater than the preset threshold, including:

[0018] Determine the magnitude of the probability relative to a preset threshold;

[0019] If the probability is greater than or equal to a preset threshold, it is determined that a fault traveling wave has been detected, and the absolute UTC time locked by the local precision clock source is read.

[0020] Otherwise, it is determined that no fault traveling wave was detected.

[0021] In conjunction with the first aspect, in one embodiment, the method further includes: when it is determined that a fault traveling wave has been detected, temporarily locking the write operation of the circular buffer on high-frequency sampled data;

[0022] And / or, the method further includes: when it is determined that no fault traveling wave is detected, returning to the data time window from which the high-frequency sampled data was acquired.

[0023] In conjunction with the first aspect, in one implementation, obtaining the sampling index of the fault traveling wavefront start point based on the data time window includes:

[0024] In the temporarily locked ring buffer, find the sampling point with the largest current change rate among all sampling points included in the data time window, take this sampling point as the starting point of the fault traveling wave front, and record the absolute array index n of this sampling point in the ring buffer.

[0025] Using the absolute array index p in the circular buffer corresponding to the latest sampling point when the probability is greater than the preset threshold as the time reference zero point, the integer sampling interval offset of the fault traveling wave wavehead starting point relative to the time reference zero point is calculated, and the integer sampling interval offset is defined as the sampling index Index_peak of the fault traveling wave wavehead starting point, Index_peak=np.

[0026] In conjunction with the first aspect, in one implementation, obtaining the absolute arrival time of the fault traveling wavefront based on the absolute UTC time, sampling index, and preset high sampling frequency includes:

[0027] The subsampling offset between the fault traveling wavefront start point and the sampling points adjacent to the fault traveling wavefront start point among all sampling points included in the data time window is obtained using the amplitude of the fault traveling wavefront start point.

[0028] Based on the absolute UTC time, sampling index, subsampling offset, and preset high sampling frequency, the absolute arrival time of the fault traveling wave front is obtained.

[0029] In conjunction with the first aspect, in one implementation, based on the starting point of the fault traveling wavefront and a first preset length, the data time window is truncated to obtain a waveform segment, including:

[0030] Centered on the starting point of the fault traveling wave front, continuous data sequences of a first preset length are extracted from the data time window forward and backward to obtain waveform segments.

[0031] In conjunction with the first aspect, in one implementation, obtaining fault traveling wave front data based on the waveform segment and absolute arrival time includes:

[0032] Feature extraction is performed on the waveform segment to obtain traveling wave features, which include at least one of the following: the top K wavelet coefficients with the highest energy, the amplitude of the fault traveling wave front, the polarity of the fault traveling wave front, the steepness of the fault traveling wave front, and the main frequency of the fault traveling wave front.

[0033] The traveling wave characteristics and absolute arrival time are encapsulated into fault traveling wave head data.

[0034] Secondly, embodiments of this application provide a fault traveling wave head data acquisition device, the fault traveling wave head data acquisition device comprising:

[0035] The traveling wave inference module is used to obtain a data time window extracted from high-frequency sampled data, and to process the data time window using a traveling wave inference model to obtain the probability that the data time window contains fault traveling wave characteristics.

[0036] A probability determination module is used to determine the probability based on a preset threshold in order to obtain the absolute UTC time when the probability is determined to be greater than the preset threshold.

[0037] The index acquisition module is used to acquire the sampling index of the starting point of the fault traveling wave head according to the data time window;

[0038] The time determination module is used to obtain the absolute arrival time of the fault traveling wave front based on the absolute UTC time, the sampling index, and the preset high sampling frequency.

[0039] The waveform acquisition module is used to extract the data time window based on the starting point of the fault traveling wave head and a first preset length to obtain waveform segments.

[0040] The wavefront data acquisition module is used to obtain fault traveling wavefront data based on the waveform segment and the absolute arrival time.

[0041] The beneficial effects of the technical solution provided in this application include:

[0042] This application utilizes a traveling wave inference model to perform real-time pattern recognition of waveforms. It can reliably distinguish between genuine fault traveling waves and various interference signals in complex electromagnetic noise backgrounds, fundamentally solving the long-standing problems of false operation and refusal to operate in the traditional fixed threshold method.

[0043] This application automatically extracts features that characterize the nature of the fault for uploading and performs intelligent compression at the source of data generation. This reduces the amount of uplink data by an order of magnitude and solves the problem of heavy pressure on the communication network and the main station caused by uploading massive amounts of raw data indiscriminately. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating an embodiment of the fault traveling wave head data acquisition method of this application;

[0046] Figure 2 This is a functional module diagram of an embodiment of the fault traveling wave head data acquisition device of this application. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0049] In a first aspect, embodiments of this application provide a method for acquiring fault traveling wave head data.

[0050] In one embodiment, reference is made to Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the fault traveling wave head data acquisition method of this application. Figure 1 As shown, the fault traveling wave head data acquisition method includes:

[0051] 101: Obtain a data time window extracted from high-frequency sampled data, and process the data time window using a traveling wave inference model to obtain the probability that the data time window contains a fault traveling wave.

[0052] In power distribution networks, data acquisition and conversion typically require instrument transformers to convert the high voltage and current on the primary side of the transformer into a low voltage and current on the secondary side. When a fault occurs in a power distribution network, the fault point generates a sudden change in voltage and current, triggering a high-frequency transient current / voltage traveling wave signal. Since this traveling wave primarily contains high-frequency components (kHz to MHz), traditional power frequency current transformers (CTs) have insufficient bandwidth and cannot accurately transmit this signal. Therefore, this application employs a high-bandwidth (typically ≥5MHz), linear Rogowski coil, or a low-current transformer with a specific frequency response range to acquire the fault current signal containing the high-frequency traveling wave component. Furthermore, this application includes a signal conditioning circuit, including an anti-aliasing filter and a programmable gain amplifier. The filter uses a passive LC or high-performance active filter with a cutoff frequency set at 2-5MHz, aiming to retain the useful high-frequency signal while suppressing higher-frequency radio frequency interference. This facilitates providing a clean signal with appropriate amplitude for subsequent sampling.

[0053] Before acquiring the data time window, sampling is performed at a preset high sampling frequency to obtain high-frequency sampled data, which is then stored in a circular buffer.

[0054] According to the Nyquist sampling theorem, to reconstruct a signal with a maximum frequency of f_max losslessly, the sampling frequency must be greater than 2f_max. The main energy of a traveling wavefront is concentrated below 1MHz, but crucial information determining the wavefront arrival time may reside in higher frequency components. A 10MHz sampling rate is sufficient to ensure accurate characterization of the wavefront rising edge (nanosecond to microsecond level). Therefore, this application uses an ADC (analog-to-digital converter) with a sampling rate of at least 10 MSPS (million samples per second) and a resolution of ≥14 bits.

[0055] For the ring buffer, it is necessary to cover the time window of the initial traveling wave arrival of the fault and the return of the reflected wave from the opposite bus (for a typical 10km line, the round-trip time is about 70 microseconds, but multi-branch reflection and system margin need to be considered). Therefore, a large-capacity SRAM or DDR memory is equipped to cyclically buffer at least 10-20 milliseconds of raw sampled data.

[0056] As an example, sampling is performed at a preset high sampling frequency fs, and the collected high-frequency sampled data (such as 14-bit signed integers) is written to a pre-allocated circular buffer in real time and continuously. The size of fs can be set according to actual needs; for example, if fs = 10MHz, then sampling is performed at 10 samples per second. 6 Sampling is performed at a rate of one sampling point at a time. The data collected at each sampling point includes the current value. The data collected at all sampling points constitute high-frequency sampling data. It can be seen that high-frequency sampling data can be understood as a continuous data sequence composed of data such as current values ​​corresponding to the sampling point sequence.

[0057] Logically, the circular buffer is considered as a sliding time window, which always retains the latest L sampling points or the latest high-frequency sampling data of a certain duration. For example, L = fs × 0.02 seconds, which corresponds to 20 milliseconds of high-frequency sampling data. That is, the circular buffer always retains the latest 20 milliseconds of high-frequency sampling data, while the rest of the high-frequency sampling data is overwritten.

[0058] The data is managed using two pointers (write pointer and read pointer). The write pointer is updated in real time by the ADC interrupt service routine; the read pointer lags behind the write pointer, and the ripple inference model reads data from historical high-frequency sampling data through the write pointer.

[0059] This design ensures that data is never stopped being overwritten, while providing recent historical data of a fixed time period for subsequent analysis, ensuring that the initial traveling wave of any failure that occurs at any time can be completely recorded in the cache.

[0060] Obtaining the data time window extracted from the high-frequency sampled data includes: reading the high-frequency sampled data from the circular buffer and extracting a continuous data sequence of a second preset length as the data time window. For example, based on the read pointer, a continuous data sequence of length W is extracted backward (in the historical direction). The second preset length W can be set according to actual needs, for example, the second preset length W = 1024 sampling points. This continuous data sequence is the data time window for this inference.

[0061] The data time window is preprocessed by standardization, such as subtracting the mean, dividing by the variance, or normalizing the amplitude, to eliminate the influence of dimensions and improve the stability of subsequent model inference. The data time window after standardization is a W×1 dimensional array.

[0062] The traveling wave inference model in step 101 can use existing models for real-time waveform scanning and inference. For example, as an example, the traveling wave inference model uses a one-dimensional CNN model (i.e., a 1D-CNN model) for real-time waveform scanning and inference. The W×1 dimensional array after data standardization and preprocessing is input into a lightweight 1D-CNN model deployed on the chip. It receives the data time window after data standardization and preprocessing, and automatically extracts the deep spatiotemporal features of the fault traveling wave through multi-layer convolution and pooling operations. The 1D-CNN model performs a forward propagation calculation on the chip and finally outputs a two-dimensional probability value [P_normal, P_fault] through forward inference. Here, P_normal represents the probability that the signal in the current data time window "belongs to normal operation or background noise" calculated by the model; P_fault represents the probability that the data time window contains fault traveling wave features.

[0063] By utilizing a lightweight artificial intelligence model to perform real-time pattern recognition on waveforms, it is possible to reliably distinguish between genuine fault traveling waves and various interference signals in complex electromagnetic noise backgrounds, fundamentally solving the long-standing problems of false activation and refusal to activate in the traditional fixed threshold method.

[0064] 102: The probability is judged based on a preset threshold to obtain the absolute UTC time when the probability is determined to be greater than the preset threshold.

[0065] In step 102, P_fault is compared with a preset high-confidence threshold θ_trigger. The value of this preset threshold θ_trigger can be set according to actual needs, such as 0.95. The setting of this high-confidence preset threshold aims to reduce the false alarm rate and ensure that the trigger is highly reliable.

[0066] Specifically, the probability P_fault is compared with the preset threshold θ_trigger.

[0067] If P_fault < θ_trigger, it is determined that no fault traveling wave was detected, indicating noise or normal disturbance. The current data time window is discarded, and a new data time window extracted from high-frequency sampling data is acquired before proceeding to the next inference. It is understandable that after completing the current inference, the next inference can proceed directly, or it can proceed only after triggering an inference condition. The inference condition can be preset, for example, re-acquiring a data time window extracted from high-frequency sampling data after a fixed time interval, which can be set according to actual conditions. Another example is acquiring a preset number of sampling points and then re-acquiring a data time window extracted from high-frequency sampling data before proceeding to the next inference. This preset number can be set according to actual conditions, such as 256 points.

[0068] If P_fault ≥ θ_trigger, it is determined that a fault traveling wave has been detected, and the absolute UTC time locked by the local precision clock source is read. Specifically, a high-priority interrupt is generated at this time to temporarily lock the write operation of high-frequency sampled data in the circular buffer to prevent critical data from being overwritten. It can be understood that the above temporary lock time is extremely short, which can be set to the microsecond level. At the same time, the absolute UTC time when the probability of the local precision clock source being locked is greater than the preset threshold is atomically read from the high-precision synchronization clock module, and this absolute UTC time is recorded as the trigger time stamp T_trigger.

[0069] It adopts an integrated BeiDou / GPS dual-mode timing module, which receives synchronization messages sent by the master clock source through the communication network, and realizes microsecond-level synchronization of the local clock at the hardware level.

[0070] 103: Based on the data time window, obtain the sampling index of the starting point of the fault traveling wave head.

[0071] Since the fault traveling wave has been detected in the data time window, the starting point of the fault traveling wave head can be located in step 103 using the data time window.

[0072] Specifically, within the temporarily locked annular buffer, the sampling point with the largest current change rate among all sampling points included in the data time window is identified. This sampling point is taken as the starting point of the fault traveling wavefront, and its absolute array index n in the annular buffer is recorded. Then, using the absolute array index p of the latest sampling point in the annular buffer corresponding to the absolute UTC time when the probability is greater than a preset threshold as the time reference zero, the integer sampling interval offset of the fault traveling wavefront starting point relative to this time reference zero is calculated. This integer sampling interval offset is defined as the sampling index Index_peak of the fault traveling wavefront starting point, i.e., Index_peak = np. This sampling index Index_peak is typically an integer less than or equal to zero, and its physical meaning is the number of complete sampling periods that the wavefront leads the trigger time. It is both an integer representing position and can directly and unambiguously participate in time calculation, representing the time offset.

[0073] A simple and commonly used peak lookup algorithm can be used to accurately obtain the starting point of the fault traveling wave head, and then the sampling index Index_peak can be determined.

[0074] 104: Based on the absolute UTC time, sampling index, and preset high sampling frequency, obtain the absolute arrival time of the fault traveling wave front.

[0075] Using step 104, the high-precision time of the fault traveling wave front arriving at the distribution network line detection point can be calculated.

[0076] Specifically, the subsampling offset between the starting point of the fault traveling wave and the sampling points adjacent to the starting point of the fault traveling wave is first obtained using the amplitude of all sampling points included in the data time window from the starting point of the fault traveling wave. The number of sampling points adjacent to the starting point of the fault traveling wave can be selected according to the actual situation, for example, one sampling point before and one sampling point after the starting point of the fault traveling wave.

[0077] Assuming the signal near the peak of the wavefront can be approximated as a quadratic function, we take the starting point of the fault traveling wavefront and one sampling point to its left and right, namely (np-1, y[np-1]), (np, y[np]), and (n-p+1, y[n-p+1]), where y[np-1], y[np], and y[n-p+1] are the amplitudes (i.e., current values) of the three sampling points, and y[np] is the maximum value among the three. By fitting a parabola y=ax²+bx+c through these three sampling points, where a, b, and c are constants obtained from the fitting, the subsampling offset can be directly calculated using the three point values. :

[0078] =(y[np-1]-y[n-p+1]) / [2×(y[np-1]-2y[np]+y[n-p+1])].

[0079] This method requires minimal computation and achieves an accuracy of 0.01-0.05 sampling point intervals (corresponding to 1-5 nanosecond accuracy at 10MHz sampling) when the signal-to-noise ratio is high. It is evident that by calculating the sub-sampling precision time offset, the timestamp accuracy is broken down to the nanosecond level.

[0080] Then, based on the absolute UTC time, sampling index, subsampling offset, and preset high sampling frequency, the absolute arrival time T_absolute of the fault traveling wave front is obtained.

[0081] T_absolute=T_trigger+(Index_peak+Δ)×(1 / fs)

[0082] At the instant when P_fault≥θ_trigger is determined, the high-precision absolute UTC time is atomically recorded. Combined with the subsampling interval interpolation algorithm mentioned above, the measurement accuracy of the wavefront arrival time can be improved to the nanosecond level.

[0083] 105: Based on the starting point of the fault traveling wave head and the first preset length, the data time window is truncated to obtain a waveform segment.

[0084] In step 105, waveform segment extraction is performed to obtain the key waveform of the fault traveling wave. Specifically, taking the starting point of the fault traveling wave as the center, continuous data sequences of a first preset length are extracted from the data time window forward and backward to obtain waveform segments.

[0085] It is understandable that the first preset length of the forward-trunculated continuous data sequence and the backward-trunculated continuous data sequence can be the same or different, depending on the actual needs.

[0086] For example, taking the starting point of the fault traveling wavefront as the center, 64 sampling points are taken forward and 191 sampling points are taken backward, for a total of 256 points. This segment completely contains the rising edge, peak value, and early oscillations of the initial traveling wave of the fault, which is the most core raw information.

[0087] 106: Based on the waveform segment and absolute arrival time, obtain the fault traveling wave head data.

[0088] In step 106, feature extraction can be performed on the waveform segment to obtain traveling wave features, which include at least one of time-frequency domain features and physical features.

[0089] The time-frequency domain features include the top K wavelet coefficients with the highest energy, which are obtained by performing discrete wavelet transform on the waveform segment.

[0090] The physical characteristics include at least one of the following: fault traveling wavefront amplitude, fault traveling wavefront polarity, fault traveling wavefront steepness, and fault traveling wavefront dominant frequency.

[0091] Finally, the traveling wave characteristics and absolute arrival time are encapsulated into fault traveling wave head data.

[0092] The traveling wave features are combined into a one-dimensional feature vector, Feature_Vector, whose length is typically only 1% to 5% of the original cached data.

[0093] Furthermore, the encapsulated fault traveling wave header data, as a data packet, is a structured data object or message whose fields at least include:

[0094] a. Header: Device unique ID, protocol version.

[0095] b. Time core: absolute time scale T_absolute, sampling index Index_peak, preset high sampling frequency fs.

[0096] c. Event content: trigger phase, feature vector (Feature_Vector) and its type description.

[0097] d. Quality control: Model probability P_fault, signal quality indicators.

[0098] Finally, communication reporting is performed. The industrial communication interface used supports Ethernet, Industrial Ethernet, 5G, or HPLC, and has precise timestamp marking and data uplink capabilities. Data packets are sent to the transmission queue and uploaded to the designated edge gateway via a communication module (such as a 5G module). After reporting is completed, the ring buffer is unlocked, and the normal continuous acquisition and monitoring cycle resumes, waiting for the next trigger. It should be noted that, depending on the design requirements, the lock can also be released after obtaining the waveform segment, and the write operation can continue to write high-frequency sampling data to the ring buffer without waiting for reporting to be completed.

[0099] In step 106, this application automatically extracts and uploads only feature vectors that can characterize the nature of the fault (such as time-frequency domain features, physical features, etc.), and achieves intelligent compression at the source of data generation. This can reduce the amount of uplink data by 1-2 orders of magnitude, clearing the communication bandwidth obstacle for large-scale deployment of the system.

[0100] The fault traveling wave head data acquisition method provided in this application has at least the following advantages:

[0101] (1) A fundamental breakthrough in positioning accuracy

[0102] Existing single / double-ended methods are limited by wavefront identification ambiguity and synchronization costs, resulting in limited accuracy and instability in complex multi-domain dominance networks. This application overcomes the theoretical limitations of traditional methods by introducing a traveling wave inference model, such as a one-dimensional CNN model, to fuse and optimize the entire network topology and data. This enables intelligent analysis of traveling wave propagation paths, thereby achieving more stable and accurate fault location without the need for costly double-ended deployments across the entire network.

[0103] (2) The system reliability is systematically enhanced.

[0104] Traditional methods rely on single-point threshold judgment, resulting in weak anti-interference capabilities and low sensitivity to high-impedance faults. This application introduces a traveling wave inference model, such as a one-dimensional CNN model, for signal processing and decision-making. This method not only significantly improves noise interference resistance but also provides higher detection sensitivity to weak fault signals, systematically enhancing overall reliability.

[0105] (3) Significantly improved economic efficiency and scalability

[0106] Existing centralized processing models face immense pressure from massive data transmission and computation, resulting in high costs for large-scale deployment. This application addresses this by implementing intelligent data compression on the terminal side and offloading real-time computing load at the edge, significantly reducing the infrastructure requirements for communication networks and cloud centers. This leads to lower overall system costs, greater controllability, and easier large-scale deployment and expansion.

[0107] (4) Possesses adaptive and continuous evolutionary capabilities

[0108] Existing devices have fixed algorithms that cannot adapt to network changes. The traveling wave inference model introduced in this application, such as a one-dimensional CNN model, supports online updates and iterative learning. When the distribution network topology is modified or the operation mode changes, it can automatically learn and adapt through new data, achieving continuous performance evolution that is "more accurate and smarter with use," thus solving the problems of rigidity and high maintenance difficulty of traditional methods.

[0109] Understandably, the aforementioned traveling wave inference model can be further trained. For example, firstly, a training dataset is constructed by using electromagnetic transient simulation software to generate a massive number of waveform samples of traveling waves from power distribution network faults and normal noise, and accurately labeling their categories. Secondly, for the initial structure of the one-dimensional CNN model, the training dataset is used to train the model with binary cross-entropy as the loss function, and iterative optimization enables the model to learn to distinguish fault features from background noise. Next, the trained model is pruned and quantized to remove redundant parameters and convert the computational precision to fixed-point numbers to obtain a lightweight model that meets the resource constraints of the terminal. Finally, the lightweight model is converted into a format executable by embedded chips, integrated into the terminal firmware, and its triggering accuracy and real-time performance are verified through historical data and simulation tests, completing the entire preparation process from data to a usable model.

[0110] The one-dimensional convolutional neural network model in this application is specifically designed and optimized for the task of "real-time identification of traveling waves at the edge". The lightweight 1D-CNN model adopts a modular design with sequential stacking, mainly including an input layer, a feature extraction backbone network, a global feature pooling layer, and a classification output layer.

[0111] Specific functions and design considerations of each module:

[0112] Input layer: Receives a fixed-length (e.g., 1024 sampling points) one-dimensional current signal time window.

[0113] Feature extraction backbone network: Consists of 2-3 lightweight convolutional modules connected sequentially. Each lightweight convolutional module contains:

[0114] ① One-dimensional convolutional layers: Use small-sized convolutional kernels (e.g., length 3 or 5) to focus on extracting local temporal features (e.g., steep wavefronts). To control the number of parameters, the number of convolutional channels increases layer by layer but remains low.

[0115] ②Activation layer: Employ computationally efficient ReLU or its variants (such as Leaky ReLU) functions.

[0116] ③ Pooling layer: One-dimensional max pooling (pooling window length is usually 2) is used for downsampling, expanding the receptive field and enhancing feature translation invariance.

[0117] This repetitive "convolution-activation-pooling" structure is a classic feature extraction pattern in CNNs, but by strictly controlling the number of layers, channels, and kernel size, a customized lightweight design for this task has been achieved.

[0118] Global Feature Pooling Layer: After the last convolutional module, a global average pooling layer is used instead of a traditional fully connected layer. This layer compresses all the time synchronization information of each feature channel into a scalar, greatly reducing the number of model parameters and effectively preventing overfitting.

[0119] The classification output layer consists of a very small fully connected layer (e.g., 2 neurons), followed by a Softmax activation function, which maps the features into a two-dimensional probability vector [P_normal, P_fault].

[0120] Lightweighting includes pruning and quantization. Pruning removes redundant weights from the model, specifically by setting connections with absolute values ​​below a set threshold to zero based on weight size, followed by fine-tuning to restore accuracy. Quantization converts the model from 32-bit floating-point precision to 8-bit fixed-point precision. This is achieved by statistically analyzing the numerical range of each layer using calibration data, determining the linear mapping parameters, and converting weights and activation values ​​to INT8 format. After pruning and quantization, the model size is reduced to less than a quarter of its original size and can be efficiently executed on embedded processors that support integer arithmetic.

[0121] Redundant parameters are those weights (i.e., connections between neurons) that contribute very little or almost nothing to the final output decision of the model. Numerically, the absolute values ​​of these weights are usually close to zero.

[0122] Fixed-point quantization refers to the convention that the position of the decimal point is fixed. For example, INT8 quantization means using an integer in the range [-128, 127] to represent the numerical range originally covered by FP32. This is essentially a lossy compression, but through a meticulous calibration process, the loss of accuracy can be minimized.

[0123] Secondly, embodiments of this application also provide a fault traveling wave head data acquisition device.

[0124] In one embodiment, reference is made to Figure 2 , Figure 2 This is a functional module diagram of an embodiment of the fault traveling wave head data acquisition device of this application. Figure 2 As shown, the fault traveling wave head data acquisition device includes:

[0125] The traveling wave inference module is used to obtain a data time window extracted from high-frequency sampled data, and to process the data time window using a traveling wave inference model to obtain the probability that the data time window contains fault traveling wave characteristics.

[0126] The probability determination module is used to determine the probability based on a preset threshold in order to obtain the absolute UTC time when the probability is determined to be greater than the preset threshold.

[0127] The index acquisition module is used to acquire the sampling index of the starting point of the fault traveling wave head according to the data time window.

[0128] The time determination module is used to obtain the absolute arrival time of the fault traveling wave front based on the absolute UTC time, the sampling index, and the preset high sampling frequency.

[0129] The waveform acquisition module is used to extract waveform segments by truncating the data time window based on the starting point of the fault traveling wave head and a first preset length.

[0130] The wavefront data acquisition module is used to obtain fault traveling wavefront data based on the waveform segment and the absolute arrival time.

[0131] Furthermore, the traveling wave inference module is also used to read the high-frequency sampled data of the annular buffer and extract a continuous data sequence of a second preset length as a data time window.

[0132] Furthermore, the fault traveling wave head data acquisition device also includes a sampling data acquisition module. The sampling data acquisition module is used to sample at a preset high sampling frequency before the data acquisition time window to obtain high-frequency sampling data and store it in a ring buffer.

[0133] Furthermore, the sampling data acquisition module includes a current sensor, a signal conditioning circuit, an ADC analog-to-digital converter, and memory.

[0134] Current sensors employ high-bandwidth (typically ≥5MHz) and linear Rogowski coils or small current transformers with specific frequency response ranges to acquire fault current signals containing high-frequency traveling wave components.

[0135] The signal conditioning circuit includes an anti-aliasing filter and a programmable gain amplifier. The filter uses a passive LC or high-performance active filter with a cutoff frequency set between 2-5MHz. This is designed to retain useful high-frequency signals while suppressing higher-frequency RF interference, thus providing a clean signal with appropriate amplitude for subsequent ADC sampling.

[0136] The ADC analog-to-digital converter uses a high-speed ADC with a sampling rate of not less than 10 MSPS (million samples per second) and a resolution of ≥14 bits.

[0137] The memory uses high-capacity SRAM or DDR memory, which has a circular cache.

[0138] Furthermore, in one embodiment, the probability determination module determines the probability based on a preset threshold to obtain the absolute UTC time when the probability is determined to be greater than the preset threshold, including: determining the magnitude of the probability and the preset threshold; if the probability is greater than or equal to the preset threshold, determining that a fault traveling wave has been detected, temporarily locking the write operation of the circular buffer on the high-frequency sampled data, and reading the absolute UTC time locked by the local precision clock source; otherwise, determining that no fault traveling wave has been detected, and returning to the data time window obtained from the high-frequency sampled data.

[0139] Further, in one embodiment, the index acquisition module acquires the sampling index of the fault traveling wavefront start point according to the data time window, including: in the temporarily locked annular buffer, finding the sampling point with the largest current change rate among all sampling points included in the data time window, and taking this sampling point as the fault traveling wavefront start point, and recording the absolute array index n of the sampling point in the annular buffer; taking the absolute array index p of the latest sampling point in the annular buffer corresponding to the absolute UTC time when the probability is greater than a preset threshold as the time reference zero point, calculating the integer sampling interval offset of the fault traveling wavefront start point relative to the time reference zero point, and defining the integer sampling interval offset as the sampling index Index_peak of the fault traveling wavefront start point, where Index_peak=np.

[0140] Further, in one embodiment, the time determination module obtains the absolute arrival time of the fault traveling wave front based on the absolute UTC time, the sampling index, and the preset high sampling frequency, including: obtaining the subsampling offset between the fault traveling wave front start point and the sampling points adjacent to the fault traveling wave front start point from the amplitude of the sampling points near the fault traveling wave front start point among all sampling points included in the data time window; and obtaining the absolute arrival time of the fault traveling wave front based on the absolute UTC time, the sampling index, the subsampling offset, and the preset high sampling frequency.

[0141] Furthermore, in one embodiment, the waveform acquisition module truncates the data time window based on the starting point of the fault traveling wave head and a first preset length to obtain waveform segments, including: taking the starting point of the fault traveling wave head as the center, truncating the data time window forward and backward by a continuous data sequence of the first preset length to obtain waveform segments.

[0142] Furthermore, in one embodiment, the wavefront data acquisition module obtains fault traveling wave wavefront data based on the waveform segment and the absolute arrival time, including: performing feature extraction on the waveform segment to obtain traveling wave features, wherein the traveling wave features include at least one of the top K wavelet coefficients with the highest energy, fault traveling wave wavefront amplitude, fault traveling wave wavefront polarity, fault traveling wave wavefront steepness, and fault traveling wave wavefront dominant frequency; and encapsulating the traveling wave features and the absolute arrival time into fault traveling wave wavefront data.

[0143] The functions of each module in the above-mentioned fault traveling wave head data acquisition device correspond to the steps in the above-mentioned fault traveling wave head data acquisition method embodiment, and their functions and implementation processes will not be described in detail here.

[0144] This application differs from traditional traveling wave acquisition units that merely function as "full sampling, full uploading" data pipelines. Through the deep integration of three core technologies, it creates a fault traveling wave head data acquisition method and device with local intelligent cognitive capabilities. First, it innovatively introduces a traveling wave inference model, achieving real-time pattern recognition of the original waveform. "Shape recognition" replaces the traditional "amplitude threshold comparison," fundamentally solving the problems of false triggering under complex noise interference and failure to trigger under high-impedance faults. Second, it designs an intelligent triggering decision mechanism based on a high-confidence threshold. It uses the probability that the data time window output by the traveling wave inference model contains fault traveling wave characteristics for rigorous decision-making, ensuring that only highly certain fault events initiate subsequent processes, thus constructing a highly reliable triggering defense. Third, it proposes a collaborative method for sub-sampling precision time offset calculation and event-driven feature extraction, breaking through the timestamp accuracy to the nanosecond level on the terminal side and simultaneously generating highly compressed standardized feature vectors, achieving a substantial improvement in data value density. The above three points are interconnected and together realize the intelligent closed loop of "perception-decision-refinement" at the data source, providing the upper-level system with accurate, reliable and efficient structured event reports, realizing the paradigm shift from "passive collection" to "active perception", and forming a complete, rigorous and highly competitive technical solution package.

[0145] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0146] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0147] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0148] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0149] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0150] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0151] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for acquiring fault traveling wave head data, characterized in that, The fault traveling wave head data acquisition method includes: A data time window extracted from high-frequency sampled data is obtained, and the data time window is processed using a traveling wave inference model to obtain the probability that the data time window contains fault traveling wave characteristics. The probability is judged based on a preset threshold to obtain the absolute UTC time when the probability is greater than the preset threshold, including: judging the magnitude of the probability and the preset threshold; if the probability is greater than or equal to the preset threshold, it is determined that a fault traveling wave has been detected, the write operation of the ring buffer to the high-frequency sampling data is temporarily locked, and the absolute UTC time locked by the local precision clock source when the probability is greater than the preset threshold is read. According to the data time window, the sampling index of the fault traveling wave wavefront start point is obtained, including: in the temporarily locked ring buffer, finding the sampling point with the largest current change rate among all sampling points included in the data time window, taking this sampling point as the fault traveling wave wavefront start point, and recording the absolute array index n of this sampling point in the ring buffer; taking the absolute array index p of the latest sampling point in the ring buffer corresponding to the absolute UTC time when the probability is greater than the preset threshold as the time reference zero point, calculating the integer sampling interval offset of the fault traveling wave wavefront start point relative to the time reference zero point, and defining the integer sampling interval offset as the sampling index Index_peak of the fault traveling wave wavefront start point, Index_peak=np; Based on the absolute UTC time, sampling index, and preset high sampling frequency, the absolute arrival time of the fault traveling wave front is obtained, including: obtaining the subsampling offset between the fault traveling wave front start point and the sampling points adjacent to the fault traveling wave front start point among all sampling points included in the data time window, using the amplitude of the sampling points adjacent to the fault traveling wave front start point; and obtaining the absolute arrival time of the fault traveling wave front based on the absolute UTC time, sampling index, subsampling offset, and preset high sampling frequency. Based on the starting point of the fault traveling wavefront and the first preset length, the data time window is truncated to obtain waveform segments. Based on the waveform segment and absolute arrival time, fault traveling wave head data are obtained.

2. The fault traveling wave head data acquisition method as described in claim 1, characterized in that: Before acquiring the data time window, the method further includes: sampling at a preset high sampling frequency to obtain high-frequency sampling data and storing it in a circular buffer. Obtaining the data time window extracted from the high-frequency sampled data includes: reading the high-frequency sampled data from the circular buffer and extracting a continuous data sequence of a second preset length as the data time window.

3. The fault traveling wave head data acquisition method as described in claim 1, characterized in that, The traveling wave inference model adopts a one-dimensional CNN model.

4. The fault traveling wave head data acquisition method as described in claim 1, characterized in that, If the probability is less than a preset threshold, it is determined that no fault traveling wave was detected.

5. The fault traveling wave head data acquisition method as described in claim 4, characterized in that: The method further includes: when it is determined that no fault traveling wave is detected, returning to the data time window from which the high-frequency sampling data is obtained.

6. The fault traveling wave head data acquisition method as described in claim 1, characterized in that, Based on the starting point of the fault traveling wavefront and the first preset length, the data time window is truncated to obtain waveform segments, including: Centered on the starting point of the fault traveling wave front, continuous data sequences of a first preset length are extracted from the data time window forward and backward to obtain waveform segments.

7. The fault traveling wave head data acquisition method as described in claim 1, characterized in that, Based on the waveform segment and absolute arrival time, fault traveling wave front data are obtained, including: Feature extraction is performed on the waveform segment to obtain traveling wave features, which include at least one of the following: the top K wavelet coefficients with the highest energy, the amplitude of the fault traveling wave front, the polarity of the fault traveling wave front, the steepness of the fault traveling wave front, and the main frequency of the fault traveling wave front. The traveling wave characteristics and absolute arrival time are encapsulated into fault traveling wave head data.

8. A fault traveling wave head data acquisition device, characterized in that, The fault traveling wave head data acquisition device includes: The traveling wave inference module is used to obtain a data time window extracted from high-frequency sampled data, and to process the data time window using a traveling wave inference model to obtain the probability that the data time window contains fault traveling wave characteristics. The probability determination module is used to determine the probability based on a preset threshold in order to obtain the absolute UTC time when the probability is determined to be greater than the preset threshold. The module includes: determining the magnitude of the probability and the preset threshold; if the probability is greater than or equal to the preset threshold, determining that a fault traveling wave has been detected, temporarily locking the write operation of the ring buffer to the high-frequency sampling data, and reading the absolute UTC time locked by the local precision clock source when the probability is determined to be greater than the preset threshold. The index acquisition module is used to acquire the sampling index of the fault traveling wave wavefront start point according to the data time window, including: in a temporarily locked ring buffer, finding the sampling point with the largest current change rate among all sampling points included in the data time window, taking this sampling point as the fault traveling wave wavefront start point, and recording the absolute array index n of this sampling point in the ring buffer; taking the absolute array index p of the latest sampling point in the ring buffer corresponding to the absolute UTC time when the probability is greater than a preset threshold as the time reference zero point, calculating the integer sampling interval offset of the fault traveling wave wavefront start point relative to the time reference zero point, and defining the integer sampling interval offset as the sampling index Index_peak of the fault traveling wave wavefront start point, Index_peak=np; The time determination module is used to obtain the absolute arrival time of the fault traveling wave front based on the absolute UTC time, sampling index, and preset high sampling frequency. This includes: obtaining the subsampling offset between the fault traveling wave front start point and the sampling points adjacent to the fault traveling wave front start point from the amplitude of the sampling points among all sampling points included in the data time window; and obtaining the absolute arrival time of the fault traveling wave front based on the absolute UTC time, sampling index, subsampling offset, and preset high sampling frequency. The waveform acquisition module is used to extract the data time window based on the starting point of the fault traveling wave head and a first preset length to obtain waveform segments. The wavefront data acquisition module is used to obtain fault traveling wavefront data based on the waveform segment and the absolute arrival time.

Images