Distribution network line fault detection method, system and device based on pulse reflection waves and medium

Abstract

The invention discloses a distribution network line fault detection method, system and device based on pulse reflection waves and a medium, and belongs to the technical field of fault detection, and the method comprises the steps: deploying pulse sources, generating pulse signals, recording a reference waveform of each pulse source, receiving reflection wave signals, extracting features, and comparing reflection wave features with the reference waveforms to recognize abnormal signals. Filtering the abnormal signal, calculating an error index and a harmonic index of the filtered signal to preliminarily judge a fault point, calculating a path difference of the fault point, constructing a path matrix, and optimizing the path matrix according to a three-phase signal of the distribution network; and extracting a reflected wave signal corresponding to the position of the fault point in the path matrix, converting the reflected wave signal into a two-dimensional image, and classifying the fault point through a neural network. According to the method, the accurate preliminary judgment of the fault point is realized, the anti-noise capability and the analysis precision of the signal are effectively improved, the accuracy of the path matrix is remarkably enhanced, and the efficient detection, positioning and classification of the complex distribution network fault are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault detection, and in particular to a distribution network line fault detection method, system, device and medium based on pulse reflection wave. BACKGROUND

[0002] With the rapid expansion of modern distribution network scale and the increasing complexity, the stable operation of power lines becomes particularly critical. However, the distribution network line is easily affected by various factors during operation, such as external environmental interference, equipment aging or line defects, leading to the occurrence of faults. These faults not only pose a threat to the reliability of the power system, but also increase the maintenance cost and the risk of power outage. Traditional distribution network line fault detection methods mainly use current and voltage signals, usually adopting time domain analysis, frequency domain analysis or model-based detection algorithms. However, these methods have limitations in complex fault scenarios, such as difficulty in accurately locating fault points, insufficient signal noise suppression capability, and poor ability to distinguish multiple types of faults. In recent years, pulse reflection wave technology has attracted attention due to its unique advantages in signal propagation characteristics. This technology uses the reflection characteristics of pulse signals to effectively identify fault points and their locations. However, existing pulse reflection wave-based fault detection technologies are mostly limited to single feature extraction methods, failing to fully combine multiple signal analysis and optimization techniques, and having low robustness in complex distribution network scenarios. SUMMARY

[0003] In view of the above problems, the present application is proposed.

[0004] Therefore, the technical problem solved by the present application is: how to solve the problem that existing pulse reflection wave-based fault detection technologies are mostly limited to single feature extraction methods, fail to fully combine multiple signal analysis and optimization techniques, and have low robustness in complex distribution network scenarios.

[0005] To solve the above technical problems, the present application provides the following technical solutions: a distribution network line fault detection method based on pulse reflection wave, which includes deploying pulse sources at key nodes of the distribution network to generate pulse signals and recording the reference waveform of each pulse source; receiving reflected wave signals reflected from different positions in the line and extracting features, comparing the reflected wave features with the reference waveform, and calculating the error to judge abnormal signals; performing dynamic Kalman filtering on the abnormal signals, calculating the error index and harmonic index of the filtered signals to preliminarily judge the fault point; calculating the path difference of the fault point and generating a path matrix, optimizing the path matrix according to the three-phase signals of the distribution network, obtaining the time difference and path length of the fault point from the path matrix, and extracting the reflected wave signal corresponding to the fault point position; converting the reflected wave signal into a two-dimensional image, and classifying the fault point through a neural network.

[0006] As a preferred scheme of the pulse reflection wave-based distribution network line fault detection method, the method comprises the following steps: deploying pulse sources, generating pulse signals, and recording reference waveforms of each pulse source; after receiving an activation instruction, sequentially starting the pulse sources, making each pulse source inject a high-frequency narrow pulse signal into the distribution network line, recording waveform data at the output end, extracting instantaneous frequency and amplitude characteristics of the waveform data, and storing the characteristics in a standard waveform characteristic library.

[0007] As a preferred scheme of the pulse reflection wave-based distribution network line fault detection method, the method comprises the following steps: deploying pulse sources, generating pulse signals, and recording reference waveforms of each pulse source; after receiving an activation instruction, sequentially starting the pulse sources, making each pulse source inject a high-frequency narrow pulse signal into the distribution network line, recording waveform data at the output end, extracting instantaneous frequency and amplitude characteristics of the waveform data, and storing the characteristics in a standard waveform characteristic library.

[0008] As a preferred scheme of the pulse reflection wave-based distribution network line fault detection method, the method comprises the following steps: deploying pulse sources, generating pulse signals, and recording reference waveforms of each pulse source; after receiving an activation instruction, sequentially starting the pulse sources, making each pulse source inject a high-frequency narrow pulse signal into the distribution network line, recording waveform data at the output end, extracting instantaneous frequency and amplitude characteristics of the waveform data, and storing the characteristics in a standard waveform characteristic library.

[0009] As a preferred scheme of the pulse reflection wave-based distribution network line fault detection method, the method comprises the following steps: deploying pulse sources, generating pulse signals, and recording reference waveforms of each pulse source; after receiving an activation instruction, sequentially starting the pulse sources, making each pulse source inject a high-frequency narrow pulse signal into the distribution network line, recording waveform data at the output end, extracting instantaneous frequency and amplitude characteristics of the waveform data, and storing the characteristics in a standard waveform characteristic library.

[0010] The preferred scheme realizes the elimination of abnormal paths and the correction of path differences, effectively improves the stability and accuracy of the path matrix, reduces the path misjudgment probability caused by unbalanced signals or external disturbances, and significantly improves the spatial resolution of fault point positioning, by mean and deviation analysis on the path difference and combination of Karrenbauer transformation of three-phase signals to extract sequence components, normalization and proportional coefficient adjustment.

[0011] As a preferred scheme of the pulse reflection wave-based distribution network line fault detection method, the reflection wave signal is converted into a two-dimensional image, including extracting the time difference and path length corresponding to each fault point from the path matrix, obtaining the corresponding reflection wave signal, normalizing the reflection wave signal, converting the reflection wave signal into a two-dimensional image by using the Gramian Angular Field method, and constructing a training data set by taking each image as a sample; based on the training data set, a generative adversarial network is introduced to generate enhanced samples, and combined with the original samples to construct a training data set containing diversified fault features.

[0012] The preferred scheme converts the reflection wave signal into a two-dimensional image and introduces a generative adversarial network to expand sample data, constructs an image sample set containing various fault features, makes the training data have stronger diversity and coverage, enhances the generalization ability and adaptability of the classification model to different fault forms, and improves the stability and accuracy of the fault detection system in complex power grid environments.

[0013] As a preferred scheme of the pulse reflection wave-based distribution network line fault detection method, the fault point is classified by a neural network, including selecting image samples from the training data set as input, extracting image feature vectors by using a convolutional neural network, calculating the Euclidean distance between samples, optimizing and training the model by combining a contrast loss function, and mapping the input samples to corresponding fault type labels.

[0014] The preferred scheme extracts high-dimensional features of the waveform image by using a convolutional neural network and introduces a contrast loss function to optimize the model, so that the model can effectively capture the slight differences between different fault types, improve the accuracy of fault recognition, and still achieve high-accuracy classification and recognition in actual distribution networks with various noise interference and boundary ambiguity.

[0015] The application provides a pulse reflection wave-based distribution network line fault detection system.

[0016] To solve the above technical problems, the application provides the following technical scheme: a distribution network line fault detection system based on pulse reflected waves, comprising a pulse source generation module, a reflected wave receiving module, a fault judgment module, a fault waveform extraction module and a fault classification module; the pulse source generation module is used for deploying pulse source generation pulse signals at key nodes of a distribution network and recording reference waveforms of each pulse source; the reflected wave receiving module is used for receiving reflected wave signals reflected from different positions in the line and extracting features, comparing the reflected wave features with the reference waveforms, and calculating errors to judge abnormal signals; the fault judgment module is used for dynamically Kalman filtering the abnormal signals, calculating error indexes and harmonic indexes of the filtered signals, and preliminarily judging fault points; the fault waveform extraction module is used for calculating path differences of the fault points and generating a path matrix, optimizing the path matrix according to three-phase signals of the distribution network, obtaining time differences and path lengths of the fault points from the path matrix, and extracting reflected wave signals of corresponding fault point positions; and the fault classification module is used for converting the reflected wave signals into two-dimensional images and classifying the fault points through a neural network.

[0017] The application provides a computer device, comprising a memory and a processor, and the memory stores a computer program, characterized in that the processor implements the steps of the distribution network line fault detection method based on pulse reflected waves when executing the computer program.

[0018] The application provides a computer readable storage medium, which stores a computer program, characterized in that the computer program is executed by a processor to implement the steps of the distribution network line fault detection method based on pulse reflected waves.

[0019] The application has the following beneficial effects: the application filters abnormal signals, separates fundamental frequency components and non-fundamental frequency components through dynamic Kalman filtering, combines error indexes and harmonic indexes to realize accurate preliminary judgment of fault points, effectively improves the noise resistance and analysis accuracy of signals, calculates path differences of fault points, constructs a path matrix, introduces decomposition and proportional optimization of three-phase signals of the distribution network, significantly enhances the robustness and accuracy of the path matrix, solves the limitations of traditional methods in a multi-path interference scene, and fully utilizes the propagation characteristics of pulse reflected waves, combines modern signal analysis and optimization techniques, and realizes efficient detection, positioning and classification of complex distribution network faults. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0021] Figure 1 A general flowchart of a pulse reflection wave-based distribution network line fault detection method is provided for an embodiment of the present application.

[0022] Figure 2 A scheme module diagram of a pulse reflection wave-based distribution network line fault detection system is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to make the above objectives, features and advantages of the present application more apparent, more comprehensible, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0024] Embodiment 1, with reference to Figure 1 For an embodiment of the present application, the embodiment provides a pulse reflection wave-based distribution network line fault detection method, comprising:

[0025] S1, deploying pulse sources at key nodes of the distribution network to generate pulse signals and recording reference waveforms of each pulse source.

[0026] S2, receiving reflected wave signals reflected from different positions in the line and extracting features, comparing the reflected wave features with the reference waveforms, and calculating errors to judge abnormal signals.

[0027] S3, performing dynamic Kalman filtering on the abnormal signals to calculate error indexes and harmonic indexes of the filtered signals to preliminarily judge fault points.

[0028] S4, calculating path differences of the fault points and generating a path matrix, optimizing the path matrix according to three-phase signals of the distribution network, obtaining time differences and path lengths of the fault points from the path matrix, and extracting reflected wave signals corresponding to the fault point positions.

[0029] S5, converting the reflected wave signals into two-dimensional images, and classifying the fault points through a neural network.

[0030] It should be noted that the power distribution network line structure is complex, the operation environment is disturbed, the traditional detection method based on voltage and current waveform has significant deficiencies in fault point identification accuracy and response speed, especially in the background of high-frequency weak signal interference, the conventional method is difficult to accurately identify the fault position and type. In the application, a distributed detection network is formed by deploying a time-synchronized multi-pulse source, high-precision signal injection and echo reception are realized; combined with high-sensitivity reflected wave extraction, dynamic filtering and double-index judgment mechanism, abnormal signals and fault points can be accurately identified; the spatial positioning accuracy is improved through path matrix construction and three-phase signal optimization; and then the fault type identification is completed by means of signal image conversion and neural network classification. The whole method consists of five steps of signal acquisition, abnormal identification, fault judgment, path analysis and intelligent classification, which can adapt to the actual demand of multiple interference and diversified fault types in distribution network, and improve the response accuracy and intelligent level of fault detection system.

[0031] Embodiment 2 is an embodiment of the application, which provides a power distribution network line fault detection method based on pulse reflected wave based on the previous embodiment, comprising:

[0032] In the embodiment of the application, in step S1, pulse sources are deployed at key nodes of the power distribution network to generate pulse signals and record the reference waveform of each pulse source.

[0033] In the embodiment of the application, in step S1, pulse sources are deployed at key nodes of the power distribution network to generate pulse signals and record the reference waveform of each pulse source.

[0034] Specifically, deploying pulse sources to generate pulse signals and recording the reference waveform of each pulse source is to deploy multiple pulse sources at key nodes (line branch points, load concentration points) of the power distribution network line, number each pulse source S1, S2,..., S n , form a distributed detection network, and use GPS time service technology to time synchronize all pulse sources, ensure the time sequence consistency between pulse sources, and control the error within 10 nanoseconds.

[0035] When the activation instruction is received, the pulse sources S1, S2,..., S n are started in sequence to ensure the accuracy of time sequence control, each pulse source generates a high-frequency narrow pulse signal and injects the signal into the power distribution network line, records the transmission time and signal strength, uses a high-precision oscilloscope to record waveform data at the output end of each pulse source, and extracts the instantaneous frequency and amplitude characteristics of the recorded waveform data by Hilbert-Huang transformation.

[0036] The waveform data includes the amplitude and frequency of the signal.

[0037] A plurality of pulse sources are deployed at key nodes of the distribution network line, and each pulse source is numbered to form a distributed detection network. Through an accurate arrangement strategy, it is ensured that each line is covered by at least one pulse source, the coverage of fault detection is improved, the position of the fault point can be accurately located, the signal comparison data between different pulse sources have time consistency, the measurement accuracy of the reflection wave arrival time is improved, high-precision input is provided for subsequent path difference calculation, the high-frequency narrow pulse signal has good spectral distribution characteristics, can effectively propagate and reflect defects or abnormalities in the line, the instantaneous frequency and amplitude extracted by HHT can reflect the real characteristics of the pulse signal, and provide a more comprehensive description for identifying abnormal signals, and eliminate the limitations of linear transformation methods when processing non-stationary signals, and adapt to the complex operating environment of the distribution network line.

[0038] In an alternative embodiment, deploying pulse sources at key nodes of the distribution network to generate pulse signals and recording the reference waveform of each pulse source can also be achieved by selecting the end of the feeder and the start of the branch in the typical operating section to arrange the pulse sources, and using a unified clock synchronization module to uniformly schedule the transmission time, so as to realize centralized monitoring and standard waveform acquisition of the line state in a specific area.

[0039] In another alternative embodiment, deploying pulse sources at key nodes of the distribution network to generate pulse signals and recording the reference waveform of each pulse source can also be achieved by integrating the pulse sources with edge intelligent devices and arranging them at switch stations or segmenter positions, and combining local high-precision clocks to realize time synchronization and on-site waveform recording, thereby improving the flexibility of pulse signal injection and the on-site adaptability of waveform acquisition.

[0040] The present application can ensure the comparability and consistency of the pulse signal at different nodes through the high timing accuracy synchronous pulse source arrangement and reference waveform acquisition mechanism, provide accurate reference for subsequent reflection wave error analysis and abnormality identification, and improve the signal coverage and time domain identification accuracy in complex areas of the distribution network structure.

[0041] In the embodiment of the present application, the reflected wave signals reflected from different positions in the line are received and features are extracted in step S2, the reflected wave features are compared with the reference waveform, and the error is calculated to judge the abnormal signal.

[0042] The reflected wave is received by a high-sensitivity receiver, the reflected wave signal is normalized after denoising by a low-pass filter, and the error between the reflected wave features and the reference waveform is calculated and compared with a set threshold to judge whether it is an abnormal signal, if it is abnormal, the original features and normalized features are stored in an abnormal data set.

[0043] Specifically, the activation instruction sequentially activates the pulse sources S1, S2,..., S n , ensuring the accuracy of timing control, each pulse source generates a high-frequency narrow pulse signal and injects the signal into the power distribution network line, records the transmission time and signal strength, uses a high-precision oscilloscope to record waveform data at the output end of each pulse source, and performs Hilbert-Huang transformation on the recorded waveform data to extract instantaneous frequency and amplitude characteristics.

[0044] The waveform data includes the amplitude and frequency of the signal.

[0045] The extracted reference waveform characteristic values (instantaneous frequency and amplitude) are stored in a database to form a standard waveform characteristic library.

[0046] Further, the reflected wave signal is received and the characteristics are extracted, and the reflected wave characteristics and the reference wave are compared to identify abnormal signals. After each pulse source transmits a pulse signal, the signal reflected back from different positions of the line is received through a high-sensitivity receiver, the arrival time of the received signal is recorded, the amplitude, direction and time difference of the signal are extracted from the collected reflected wave signal, the collected reflected wave signal is smoothed through a Bessel low-pass filter, and the amplitude and time difference after noise reduction are normalized, and the time difference, direction and amplitude are integrated into a feature data set,

[0047] The error between the current reflected wave and the reference wave is calculated, the error threshold O is set through real-time measurement of the signal-to-noise ratio (SNR) of the system, if the error value is greater than or equal to the threshold O, the corresponding signal is marked as abnormal, otherwise it is not marked, the original characteristics and normalized characteristics of the signals marked as abnormal are extracted, and an abnormal data set is generated; the abnormal data set includes an abnormal signal sequence, corresponding time difference, amplitude and marked abnormal point index.

[0048] The high-sensitivity receiver can overcome the influence of signal attenuation and environmental interference, capture weak signals with high precision, and ensure that signals reflected from different positions on the line can be recorded completely. In a complex distribution network environment, the device can accurately capture the details of the reflected waveform, providing a reliable data foundation for subsequent feature extraction. The extracted features, such as amplitude, direction, and time difference, can fully represent the physical characteristics of the reflected wave and provide multi-dimensional support for distinguishing different fault types. The direction feature combined with the time difference feature point can achieve preliminary judgment of fault location and type. The Bessel low-pass filter filters out high-frequency noise, effectively improves the signal-to-noise ratio of the signal, and at the same time preserves the phase characteristics of the signal, so that the filtered reflected waveform can accurately reflect the true characteristics of the line fault point. Compared with ordinary filters, the Bessel filter can better protect the time characteristics, making the subsequent time difference analysis more accurate. Normalization processing can adjust the amplitude and time difference features to a unified numerical range, eliminating the uneven weight problem caused by the value difference of the feature values, and ensuring the stability and accuracy of the subsequent error calculation and classification algorithm. Normalization can also improve the sensitivity of abnormal signal detection, so that small feature deviations can be captured. The reference waveform serves as an ideal waveform reference, and by comparison, abnormal signal points that deviate from the normal range can be accurately identified. The setting of the error threshold O can flexibly adapt to the detection needs of different lines, realizing real-time response to various fault scenarios.

[0049] In the embodiments of the present application, the abnormal signal is dynamically Kalman filtered in step S3 to calculate the error index and harmonic index of the filtered signal to preliminarily judge the fault point.

[0050] The state of the abnormal signal is estimated by the dynamic filtering method, and the difference between the fundamental component and the non-fundamental component is extracted based on the filtering result to construct the error index, and the harmonic index is calculated by combining the ratio of the high-order harmonic component to the fundamental component. The fault point is judged according to the comparison result of the error index and the harmonic index with the corresponding threshold value.

[0051] Specifically, the abnormal signal is filtered to calculate the error index and harmonic index of the filtered signal to preliminarily judge the fault point. The abnormal signal sequence is standardized, and the dynamic state transition model and observation model of the signal are defined:

[0052] x k+1 =ax k +bw k

[0053] z k =Hx k +v k

[0054] wherein, x kis the state vector, describing the fundamental frequency characteristics of the signal, a is the state transition matrix, representing the law of state change over time, b is the input weight matrix, w k is the signal dynamic noise, z k is the observation value, H is the observation matrix, v k is the observation noise.

[0055] Set the initial state to zero, initialize the state vector, state covariance matrix P, signal dynamic noise covariance matrix Q and observation noise covariance matrix R.

[0056] Use Gaussian-Student's t distribution to describe the dynamic noise:

[0057] p(v k )=π k N(v k ;0,R)+(1-π k )Student's t(v k ;0,R,ν)

[0058] Where v k is the observation noise, representing the random error in the observed signal, π k is the mixing coefficient of Gaussian noise, representing the weight of Gaussian distribution in noise modeling, N(v k ; 0, R) is the probability density function of Gaussian distribution, used to describe random noise in general case, Student's t(v k ; 0, R, ν) is the Student's t distribution part, used to capture outliers and long-tailed noise, ν is the degree of freedom of t distribution, controlling the long-tail characteristics of the tail of the distribution, usually ranging between 3 to 5, R is the observation noise covariance matrix, describing the noise intensity and its correlation.

[0059]

[0060] Where d is the dimension of the observation noise vector, T is the transpose operation.

[0061]

[0062] Where Γ() is the gamma function, T is the transpose operation.

[0063] Gaussian distribution handles random noise within normal range, Student's t distribution captures abnormal noise (such as long-tailed noise), effectively avoiding the sensitivity of traditional Gaussian distribution model to outliers. Enhance the robustness of the filter in non-stationary environment, improve the quality of the filtered signal, and provide more reliable data support for subsequent steps (such as fault point determination).

[0064] Calculate the next time state vector according to the state transition model

[0065]

[0066] where u k-1 is the external input vector, is the updated state vector at the last time k-1, and is the filtered result at the last time.

[0067] Update the uncertainty of the predicted state according to the signal dynamic noise covariance matrix:

[0068] P k|k-1 = aP k-1|k-1 b T + Q

[0069] where P k|k-1 is the predicted covariance matrix, P k-1|k-1 is the updated covariance matrix at the last time k-1, is the estimation error covariance of the filtered state, and T is the transpose of the matrix.

[0070] Calculate the state change amplitude:

[0071]

[0072] where is the current state prediction value, is the last state filtering value.

[0073] Adjust the weight of the noise model according to f k

[0074]

[0075] where f k is the state change amplitude, represents the difference between the filter's predicted state and the actual state, λ is the adjustment rate, controls the sensitivity of weight adjustment, and the typical value is between 0.1 and 1.0, and δ is the trigger threshold.

[0076] δ = μ f + 3σ f

[0077] where μ f is the mean of the historical state change amplitude, and σ f is the standard deviation of the historical state change amplitude.

[0078] If f k > δ, the current state change is significant and the trigger condition is met, the weight of the Gaussian distribution is increased to adapt to the influence of long-tail noise, and the observation update process is performed to correct the state prediction. ​

[0079] If f k ≤ δ, the current state change is not significant and the triggering condition is not met, the weight of Student's t distribution is increased, the filtering effect is stabilized, the observation update is skipped, and the prediction of the next moment is directly performed.

[0080] By f k threshold control, unnecessary weight adjustment is reduced, calculation resource use is optimized, noise model dynamic adaptability to abnormal signal scenarios is improved, and filter processing capacity for complex signals is enhanced.

[0081] When the triggering condition is met, the Kalman gain K k is calculated:

[0082] K k = P k∣k-1 H T (HP k∣k-1 H T +R k ) -1

[0083] Where P k∣k-1 is the covariance matrix, H is the observation matrix, and R k is the measurement noise covariance.

[0084] The state vector x is corrected using the current observation signal and the covariance matrix P k∣k is updated:

[0085]

[0086] P k∣k = (U-K k H)P k∣k-1

[0087] Where z k is the observation signal and U is the identity matrix.

[0088] The fundamental component I is calculated based on the updated value of the state vector:

[0089]

[0090] The non-fundamental component L is calculated based on the fundamental component and the current observation signal:

[0091] L = z k -I

[0092] The error A between the non-fundamental component and the non-fundamental component is calculated: A = L-I.

[0093] Traverse each abnormal signal point, step by step to calculate the error A, according to a large number of historical data, statistics normal operation signal of the fundamental frequency and non-base frequency component deviation range, set error threshold A1, if the error A is greater than the threshold A, then directly determine that the abnormal point is a fault point, otherwise further calculate the harmonic index B to confirm the fault condition of the abnormal point.

[0094] Further calculate the harmonic index B refers to the root mean square value D of the high harmonic component 2 And the fundamental frequency component signal amplitude I1 to calculate the harmonic index B:

[0095]

[0096] Where, n is the order of the harmonic.

[0097] Traverse each abnormal signal point, step by step to calculate the harmonic index B, according to the harmonic characteristics introduced by common faults (such as ground, short circuit, etc.), statistical analysis of the amplitude of high harmonic component and the ratio of fundamental frequency component distribution range set index threshold B1, if the harmonic index is greater than the threshold B1, then the abnormal point is determined as a fault point.

[0098] By calculating A first, only when needed to calculate B, greatly reduce the complexity of calculation, A is sensitive to amplitude deviation and non-base frequency abnormal, such as short circuit, open circuit fault, B focuses on high harmonic distortion, can detect small impedance ground and complex harmonic caused fault, combined with two kinds of index can cover a variety of fault scene, effectively avoid the false negative or false positive caused by a single detection index, prefer to use A for rapid determination, to ensure that the fault point of high amplitude deviation can be marked in time, when A can't judge, use B to provide supplementary analysis, through the joint determination of double index, maximize the detection accuracy, reduce false positive, improve the reliability of fault point positioning.

[0099] The standardization normalizes the amplitude and time characteristics of the abnormal signal to a unified range, reducing the calculation deviation caused by the difference in signal scale. Through the standardization processing, the characteristics of the abnormal signal are more easily captured by the dynamic model, improving the reliability of signal filtering and fault point judgment. The state transition model predicts the next time state of the abnormal signal, and the observation model corrects the prediction error through the measurement value. The combination of the two dynamically adjusts the prediction result through Kalman filtering, making the filtering process more adaptable to complex signal changes. In a dynamic noise environment, it reduces signal distortion, improves the dynamic tracking ability of abnormal signals, and improves the accuracy of fault point positioning, especially the adaptability to long tail noise. By describing the dynamic noise with Gaussian distribution and combining the long tail noise weight adjustment mechanism, the filter can flexibly respond to signal mutations and complex noise environments. If the current state changes significantly, the weight of the Gaussian distribution is increased to improve the response ability to abnormal signal mutations. If the change is not significant, the weight of the t-distribution is increased to enhance the stability of the filtering. The error index is used to quantify the influence of non-base frequency components on the signal, directly reflecting the dynamic deviation of the abnormal signal. The harmonic index is used to analyze the relative intensity of high-order harmonics in the signal, supplementing the shortcomings of the error index. Through the combination of the two, the precise judgment of complex fault points is realized. In the case of greater noise influence, the joint judgment of the error index and the harmonic index improves the accuracy of fault point identification. Through Kalman gain calculation and state vector correction, the system can balance between real-time and accuracy. When the trigger condition is met, the filter further optimizes state estimation by updating the covariance matrix in real time. According to the calculation results of the error index and the harmonic index, combined with the preset threshold, abnormal signal points are gradually screened, and the specific location and characteristics of the fault point are finally confirmed to realize high-precision judgment of complex fault points, improving the applicability of the system to diversified fault scenarios and providing high-quality input for subsequent path matrix optimization and fault classification.

[0100] In the embodiments of the present application, the path difference of the fault point is calculated and the path matrix is generated in step S4. The path matrix is optimized according to the three-phase signal of the distribution network. The time difference and path length of the fault point are obtained from the path matrix, and the reflected wave signal corresponding to the fault point position is extracted.

[0101] The time difference is calculated according to the arrival time of the fault point signal and the arrival time of the reference wave, and the path difference is calculated in combination with the wave propagation speed. The path difference of the corresponding fault point is recorded for each pulse source, a path difference list is constructed, and a path matrix is combined. The path difference average of each path is calculated and the deviation value is evaluated. The deviation threshold K is set. If the path deviation exceeds the threshold, the path is removed. The three-phase signal of the distribution network is collected. The Karrenbauer transformation is used to extract the zero sequence, positive sequence and negative sequence components. Each component is normalized and the proportion coefficient is calculated. The proportion coefficient is used to adjust the path difference, and the optimized path matrix is generated.

[0102] Specifically, the path difference of the fault point is calculated and a path matrix is constructed, the time difference of each fault point is calculated according to the arrival time of the fault point signal and the arrival time of the reference wave, and the fault time difference is recorded in the corresponding pulse source, the path difference Δl is calculated according to the wave propagation speed and the time difference i :

[0103] Δl i =g·Δt i

[0104] Wherein, g is the wave speed, Δt i is the time difference of the fault point.

[0105] For each pulse source, the time difference of the corresponding fault point is read in turn, and the calculation result of each fault point is stored as a path difference list.

[0106] According to the path difference data of each pulse source, a path matrix is combined.

[0107] The path difference average of the fault point is calculated, the deviation of each path is calculated according to the average and the path difference, the threshold value K is set by analyzing the distribution of the path difference in the historical data, if the deviation is greater than the threshold value K, the path is eliminated; the Karrenbauer transformation is used to decompose the three-phase signal into zero sequence, positive sequence and negative sequence components.

[0108] Each component is normalized, and the proportion coefficient G is calculated according to the normalized component:

[0109]

[0110] Wherein, V0, V1 and V2 are zero sequence, positive sequence and negative sequence components extracted by Karrenbauer transformation. The proportion coefficient is used to adjust the path difference Δl' i :

[0111] Δl′ i =Δl i ·G

[0112] The optimized path matrix is obtained.

[0113] The time difference is calculated by the arrival time of the fault point signal and the arrival time of the reference wave, and the path difference is further calculated, an intuitive physical quantity is provided, the relative position of the fault point is reflected, basic data is provided for the construction and optimization of the path matrix, the complex distribution network topology is adapted, subsequent positioning is facilitated, the path difference data recorded by multiple pulse sources is integrated into a path matrix, and optimization is carried out according to the matrix data, a systematic description of the fault point path difference is provided, the error of single-point path difference calculation is reduced by combining multiple source data, the matrix form is convenient for algorithm processing and analysis, the mean and deviation of the calculated path difference are calculated, the paths with deviation exceeding the set threshold are removed, the credibility of the matrix is enhanced, the error paths caused by noise or external interference are reduced, the decomposed components are convenient for analyzing the signal characteristics in the unbalanced state, the normalization processing improves the comparability between signal components, reduces the calculation error, the threshold K is set by calculating the deviation value of the path difference, the paths exceeding the range are removed to eliminate the influence of abnormal data on fault positioning, and the optimized path matrix can more accurately reflect the actual situation. The optimized path difference is used for further fault point positioning and analysis, and the fault positioning precision is improved.

[0114] In the embodiment of the application, the reflected wave signal is converted into a two-dimensional image in step S5, and the fault point is classified through a neural network.

[0115] The time difference and path length corresponding to each fault point are extracted from the path matrix, the corresponding reflected wave signal is obtained, the reflected wave signal is normalized, the Gramian Angular Field (GAF) method is used to convert the reflected wave signal into a two-dimensional image, and each image is used as a sample to construct a training data set; based on the training data set, a generative adversarial network is introduced to generate enhanced samples, and a training data set containing diversified fault features is constructed by combining the original samples.

[0116] Image samples are selected from the training data set as input, a convolutional neural network is used to extract image feature vectors, the Euclidean distance between samples is calculated, the model is optimized and trained in combination with a contrast loss function, and the input samples are mapped to corresponding fault type labels.

[0117] Specifically, the reflected wave signal corresponding to the fault point position in the path matrix is extracted, the positions of all fault points are obtained from the path matrix, each position corresponds to the time difference and path length of the reflected wave, the reflected wave signal is normalized after being extracted, the Gramian Angular Field method is used to convert the reflected wave signal into a two-dimensional image, the image of each fault point is used as a sample, an initial training data set is constructed, a generative adversarial network is used to expand the sample set, a training data set containing diversified fault features is constructed by combining the generated samples and real samples.

[0118] The fault point is marked by the path matrix, the signal is accurately extracted and normalized, the time series signal is converted into an image format convenient for machine learning analysis by combining the GAF image process, the seamless connection from signal space to feature space is completed, and the generative adversarial network (GAN) technology not only makes up for the problem of insufficient number of real fault samples, but also expands the diversity through pseudo samples, so that the training data set has good representativeness and coverage. Combined with real samples and generated samples, the diversified training set greatly improves the generalization ability and applicability of the classification model. Through the combination of multiple technologies, the innovative application in the fault detection of distribution network lines is realized: signal image and GAN sample expansion. This not only improves the efficiency and accuracy of fault detection, but also provides a solution for the adaptability of smart grid complex scenarios.

[0119] The classification result of the waveform image is obtained by classifying the fault point. The sample is selected from the training data set as input, the convolutional neural network is used to extract the feature vector of the input image, the Euclidean distance between the samples is calculated, the contrast loss function is used to optimize the model, and the input sample is mapped to the fault type label (such as open circuit, short circuit, large impedance grounding, and small impedance grounding).

[0120] Through automatic feature extraction and high-dimensional feature mapping, the complexity of traditional manual feature engineering is greatly reduced, and the classification efficiency is improved. By introducing the contrast loss function and sample expansion technology, the classification model can still maintain high accuracy in a complex noise environment, and the fault point classification result can be directly applied to the fault maintenance and optimization decision of the distribution network line, providing strong support for improving the reliability of the power system.

[0121] In an optional implementation, converting the reflected wave signal into a two-dimensional image and classifying the fault point by using a neural network can also use the Markov Transition Field method to image encode the reflected wave time series, and construct a neural network model based on the residual network structure to extract features and classify images, in order to enhance the sensitivity of the model to signal detail changes.

[0122] In another optional implementation, converting the reflected wave signal into a two-dimensional image and classifying the fault point by using a neural network can also combine GAF images and time-frequency graphs as multi-modal inputs, use a multi-branch neural network to extract features of different types of images and perform fusion classification, and improve the recognition ability of the model to different types of fault features.

[0123] The application can effectively map the feature difference difficult to directly distinguish in the time sequence signal into a spatial image feature, so that the model can automatically extract the discriminative feature of the fault mode, and improve the fault type recognition accuracy and system intelligent level in the complex distribution network environment.

[0124] Embodiment 3, refer to Figure 2 As an embodiment of the application, the embodiment provides a distribution network line fault detection system based on pulse reflection wave, comprising a pulse source generation module, a reflection wave receiving module, a fault judgment module, a fault waveform extraction module and a fault classification module.

[0125] The pulse source generation module is used for deploying pulse sources at key nodes of the distribution network to generate pulse signals and record the reference waveform of each pulse source.

[0126] The reflection wave receiving module is used for receiving reflection wave signals reflected from different positions in the line and extracting features, comparing the reflection wave features with the reference waveform, and calculating the error to judge abnormal signals.

[0127] The fault judgment module is used for dynamically Kalman filtering the abnormal signals, calculating the error index and harmonic index of the filtered signals, and preliminarily judging the fault point.

[0128] The fault waveform extraction module is used for calculating the path difference of the fault point and generating a path matrix, optimizing the path matrix according to the three-phase signal of the distribution network, obtaining the time difference and path length of the fault point from the path matrix, and extracting the reflection wave signal of the corresponding fault point position.

[0129] The fault classification module is used for converting the reflection wave signal into a two-dimensional image, and classifying the fault point through a neural network.

[0130] The embodiment also provides an electronic device suitable for the case of the distribution network line fault detection method based on pulse reflection wave, comprising a memory and a processor; the memory is used for storing computer executable instructions, and the processor is used for executing the computer executable instructions to realize the distribution network line fault detection method based on pulse reflection wave as proposed in the above embodiment.

[0131] The embodiment also provides a storage medium having a computer program stored thereon, and the program is executed by a processor to realize the distribution network line fault detection method based on pulse reflection wave as proposed in the above embodiment.

[0132] The storage medium proposed in the embodiment and the distribution network line fault detection method based on pulse reflection wave proposed in the above embodiment belong to the same inventive concept, and the technical details not described in detail in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0133] From the above description of the embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software and necessary universal hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a floppy disk, a ROM, a RAM, a FLASH, a hard disk, or an optical disc, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of various embodiments of the present application.

[0134] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A method for detecting faults in distribution network lines based on pulse reflection waves, characterized in that: include, Pulse sources are deployed at key nodes of the distribution network to generate pulse signals and the reference waveform of each pulse source is recorded; The system receives reflected wave signals from different locations on the line and extracts their features. It then compares the reflected wave features with a reference waveform, calculates the error, and identifies abnormal signals. The abnormal signal is subjected to dynamic Kalman filtering, and the error exponent and harmonic exponent of the filtered signal are calculated to make a preliminary judgment on the fault point. Calculate the path difference of the fault point and generate a path matrix. Optimize the path matrix based on the three-phase signal of the distribution network. Obtain the time difference and path length of the fault point from the path matrix and extract the reflected wave signal at the corresponding fault point location. The reflected wave signal is converted into a two-dimensional image, and the fault points are classified using a neural network.

2. The method for detecting distribution network line faults based on pulse reflection waves as described in claim 1, characterized in that: The deployment of pulse sources to generate pulse signals and record the reference waveform of each pulse source includes deploying multiple pulse sources at key nodes of the distribution network line, numbering each pulse source, and achieving time synchronization between pulse sources. Upon receiving the activation command, the pulse sources are activated sequentially, each injecting a high-frequency narrow pulse signal into the distribution network line. Waveform data is recorded at the output end, and the instantaneous frequency and amplitude characteristics of the waveform data are extracted and stored in the standard waveform feature library.

3. The method for detecting distribution network line faults based on pulse reflection waves as described in claim 2, characterized in that: The process of receiving reflected wave signals from different locations in the line and extracting features includes: The reflected wave is received by a high-sensitivity receiver. The reflected wave signal is denoised by a low-pass filter and then normalized. The error between the reflected wave characteristics and the reference waveform is calculated and compared with a set threshold to determine whether it is an abnormal signal. If it is abnormal, the original characteristics and normalized characteristics are stored in the abnormal dataset.

4. The method for detecting faults in distribution network lines based on pulse reflection waves as described in claim 3, characterized in that: The calculation of the error index and harmonic index of the filtered signal is used to initially determine the fault point, including: The abnormal signal is estimated by dynamic filtering method, and the difference between the fundamental frequency component and the non-fundamental frequency component is extracted based on the filtering result to construct the error index. At the same time, the harmonic index is calculated by combining the ratio of the higher harmonic component to the fundamental frequency component. The fault point is determined by comparing the error index and the harmonic index with the corresponding threshold.

5. The method for detecting faults in distribution network lines based on pulse reflection waves as described in claim 4, characterized in that: The calculation of the path difference at the fault point and the generation of the path matrix, followed by optimization of the path matrix based on the three-phase signals of the distribution network, includes: The time difference is calculated based on the arrival time of the fault point signal and the arrival time of the reference wave, and the path difference is calculated in combination with the wave propagation speed. For each pulse source, record the path difference corresponding to the fault point, construct a path difference list, and combine them into a path matrix; Calculate the mean path difference for each path and evaluate the deviation value. Set a deviation threshold K. If the path deviation exceeds the threshold, the path is removed. The three-phase signals of the distribution network are collected, and the zero-sequence, positive-sequence, and negative-sequence components are extracted using the Karrenbauer transform. Each component is normalized, and the proportional coefficient is calculated. The path difference is adjusted using the proportional coefficient to generate an optimized path matrix.

6. The method for detecting faults in distribution network lines based on pulse reflection waves as described in claim 5, characterized in that: The process of converting the reflected wave signal into a two-dimensional image includes, Extract the time difference and path length corresponding to each fault point from the path matrix, obtain the corresponding reflected wave signal, normalize the reflected wave signal, use the Gramian Angular Field method to convert the reflected wave signal into a two-dimensional image, and use each image as a sample to construct a training dataset. Generative adversarial networks are introduced based on the training dataset to generate enhanced samples, which are then combined with the original samples to construct a training dataset containing diverse fault features.

7. The method for detecting faults in distribution network lines based on pulse reflection waves as described in claim 6, characterized in that: The classification of fault points using a neural network includes, Image samples are selected from the training dataset as input, and a convolutional neural network is used to extract image feature vectors. The Euclidean distance between samples is calculated, and the model is optimized and trained by combining the contrastive loss function. The input samples are then mapped to the corresponding fault type labels.

8. A distribution network line fault detection system based on pulse reflection waves, employing the distribution network line fault detection method based on pulse reflection waves as described in any one of claims 1 to 7, characterized in that, include: The system includes a pulse source generation module, a reflected wave receiving module, a fault judgment module, a fault waveform extraction module, and a fault classification module. The pulse source generation module is used to deploy pulse sources at key nodes of the distribution network to generate pulse signals and record the reference waveform of each pulse source. The reflected wave receiving module is used to receive reflected wave signals reflected from different positions in the line and extract features, compare the reflected wave features with the reference waveform, calculate the error and judge abnormal signals. The fault diagnosis module is used to perform dynamic Kalman filtering on abnormal signals, calculate the error index and harmonic index of the filtered signal to preliminarily determine the fault point. The fault waveform extraction module is used to calculate the path difference of the fault point and generate a path matrix. It optimizes the path matrix based on the three-phase signal of the distribution network, obtains the time difference and path length of the fault point from the path matrix, and extracts the reflected wave signal at the corresponding fault point location. The fault classification module is used to convert the reflected wave signal into a two-dimensional image and classify the fault points through a neural network.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for detecting faults in distribution network lines based on pulse reflection waves, as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the distribution network line fault detection method based on pulse reflection wave as described in any one of claims 1 to 7.

Images