Section positioning method for power distribution network fault time sequence data imaging intelligent identification

Abstract

The invention discloses a section positioning method for power distribution network fault time sequence data image intelligent identification, which comprises the following steps: acquiring a zero sequence current short time window sequence ci with the length of Li, i being the serial number of the zero sequence current short time window sequence; judging whether the length Li is stable mapping or not, and if the length Li is not stable mapping, performing minimum zero filling on the length Li to obtain a zero-sequence current short-time window sequence length L'i; if the length Li is stable mapping, determining a factor pair (r, c) with the minimum distance, rearranging the zero-sequence current short-time window sequence ci into a two-dimensional array, and carrying out transposition to obtain F = DT, otherwise, determining a factor pair (r ', c') with the minimum distance, rearranging the zero-filled short-time window sequence ci into a two-dimensional array, and carrying out transposition to obtain F = D 'T; and taking a global minimum value and a global maximum value which are jointly calculated by short time windows of all sections as upper and lower quantized limits, and linearly mapping F to a [0, 1] gray domain to obtain a two-dimensional image Imi.

Description

Technical Field

[0001] This invention relates to a section location method for intelligent identification of time-series fault data in distribution networks, belonging to the field of distribution network fault location technology. Background Technology

[0002] In medium- and low-voltage distribution networks ranging from 6 to 35 kV, both ungrounded neutral points and grounded faults via arc suppression coils coexist, with single-phase grounding being the most common fault type. In arc suppression coil grounding systems, coil compensation significantly weakens the zero-sequence current amplitude in the fault circuit and may cause directional instability, resulting in a reduction of the electrical quantity differences between the faulty and healthy sections. Parallel operation of sections and frequent load and switch operations further disperse the distinguishable information within transient and high-resistance grounding faults, making traditional section location methods unreliable in such scenarios.

[0003] Existing methods for fault location often extract single or limited features from transient quantities to construct criteria, such as attenuated DC components, transient current amplitude and polarity, correlation, and transient energy or power direction, or employ simple voting for multi-feature fusion. These methods lack robustness under noise, operating condition disturbances, and parameter uncertainties; cross-segment dimensions and scales are often inconsistent, leading to distortion in lateral comparisons; they are highly dependent on topology and prior parameters, resulting in high migration and maintenance costs. These limitations are further amplified by the "weak differences and weak directionality" caused by coil compensation, making accurate fault segment selection even more difficult. Summary of the Invention

[0004] The technical problem this invention aims to solve is to overcome the shortcomings of existing technologies and provide a segment location method for intelligent image-based identification of time-series fault data in distribution networks. In arc-suppression coil grounding systems, this method achieves cross-segment dimensional comparability of short-time windows of zero-sequence current in segments without relying on labeled data or sophisticated prior models. It also enables stable mapping from one-dimensional signals to two-dimensional images and unsupervised clustering based on direction-frequency information for adaptive selection of faulty segments. This invention requires no sophisticated prior models, performs unified cross-segment calibration and structured representation on synchronous short-time window data, and completes differential clustering under label-free conditions—a lightweight method that improves the accuracy and engineering robustness of segment location under complex operating conditions.

[0005] Prior to this invention, a method for segment location based on the image-based intelligent identification of time-series fault data in a distribution network is provided, comprising:

[0006] At the moment of a fault in the distribution network, several short-time window sequences of zero-sequence current of equal length from different sections of the distribution network are collected. i Zero-sequence current short-time window sequence c i The length is L i , where i is the sequence number of the short-time window sequence of the zero-sequence current;

[0007] Determine whether the zero-sequence current short-time window sequence satisfies the stable mapping. If the zero-sequence current short-time window sequence does not satisfy the stable mapping, then perform minimum zero-padding on the zero-sequence current short-time window sequence to obtain a new zero-sequence current short-time window sequence c'. i The new zero-sequence current short-time window sequence c' i The new length is L' i ;

[0008] If the zero-sequence current short-time window sequence c i If a stable mapping is satisfied, then the relationship with... The factor pair with the smallest distance (r,c) will be the short-time window sequence c of the zero-sequence current. i Rearranged into a two-dimensional array And by transposing, we get F=D T Otherwise, confirm with The factor pair with the smallest distance (r', c') will be the new zero-sequence current short-time window sequence after zero-filling. Rearranged into a two-dimensional array And transpose to obtain the transpose matrix F=D' T ;

[0009] Using the global minimum and global maximum values ​​calculated jointly by short time windows of all segments as the upper and lower limits of quantization, the transpose matrix F is linearly mapped to the [0,1] grayscale domain to obtain the two-dimensional image Im. i .

[0010] First, determine whether the zero-sequence current short-time window sequence satisfies a stable mapping. If the zero-sequence current short-time window sequence does not satisfy a stable mapping, then perform minimum zero-padding on the zero-sequence current short-time window sequence to obtain a new zero-sequence current short-time window sequence c'. i The new zero-sequence current short-time window sequence c' i The new length is L' i ,include:

[0011] If the length L i It can be decomposed into factor pairs (r,c) and has a square property index. Not greater than the threshold ρ max If ∈[2,3], then the zero-sequence current short-time window sequence is determined to satisfy the "stable mapping"; otherwise, the zero-sequence current short-time window sequence c is determined to satisfy the "stable mapping". i It does not satisfy the "stable mapping" condition, where max(r,c) is the maximum factor pair and min(r,c) is the minimum factor pair, and r and c represent the row factor and column factor, respectively.

[0012] If the short-time window sequence of zero-sequence current does not satisfy the "stable mapping", then for length L i Perform minimum zero padding.

[0013] Prioritize the short-time window sequence c of the zero-sequence current. iPerform minimal zero padding, including:

[0014] Determine the minimum fill value ΔL i This makes the short-time window sequence length L' of the zero-sequence current so that i Satisfy L' i =L i +ΔL i Furthermore, the factor pairs (r', c') of the decomposition satisfy ρ' ≤ ρ max ,in The quadraticity index is calculated after minimum zero-padding, where max(r',c') is the maximum factor pair and min(r',c') is the minimum factor pair; ρ max Let r' be the threshold, r' be the row factor obtained after minimum zero-padding, c' be the column factor obtained after minimum zero-padding, and L' be the column factor. i This is the length of the short-time window sequence of the zero-sequence current obtained after minimum zero-filling.

[0015] First, determine whether the zero-sequence current short-time window sequence satisfies a stable mapping. If the zero-sequence current short-time window sequence does not satisfy a stable mapping, then perform minimum zero-padding on the zero-sequence current short-time window sequence to obtain a new zero-sequence current short-time window sequence c'. i The new zero-sequence current short-time window sequence c' i The new length is L' i ,include:

[0016] If the length L i A prime number or length L i =1, then determine the length L i The "stable mapping" is not satisfied; if the length L i It is a composite number and the square property index ρ≤ρ max In determining the length L i It satisfies the "stable mapping".

[0017] Prioritizes the following: Minimum zero-padding satisfies:

[0018] ΔL i =inf{Δ∈R + ∪{0}∣(L i +Δ)∈S d , ρ(L i +Δ)≤ρ max},

[0019] In the formula, ΔL i Let R be the minimum zero-padding length, and let inf{} be the minimum value among all Δ values ​​in the infimum, where Δ represents the length of the zero-padding. + R is the set of positive real numbers. + ∪{0} is the set of non-negative real numbers, S d For the set of composite numbers, L iρ is the length. max The threshold value is used.

[0020] Prior to this, the factor pair (r,c) satisfies:

[0021] ,

[0022] In the formula, a is the first positive real number, b is the second positive real number, L is the length of the short-time window sequence of zero-sequence current, and argmin is a function of the minimum value.

[0023] Prioritize normalization of the two-dimensional image Im i Perform contrast-limited adaptive histogram equalization enhancement with a clipping limit factor ClipLimit ∈ [0.01, 0.1].

[0024] Firstly, obtain the two-dimensional image Im. i Then, based on Gabor filtering, the two-dimensional image Im... i Perform unsupervised clustering;

[0025] Among them, the two-dimensional image Im is based on Gabor filtering. i Unsupervised clustering includes:

[0026] Construct a Gabor filter bank g(λ,θ), where λ∈[2,16] pixels, θ is 4 to 8 directions with equal angular intervals of 0° to 180°, ψ=0 or ψ∈[-π,π], γ∈[0.3,1.0], σ=κλ, κ∈[0.5,1.0];

[0027] For two-dimensional image Im i Apply g(λ,θ) one by one to obtain the response R(λ,θ) and calculate the two-dimensional image Im. i Gabor features g are obtained by taking the mean amplitude and standard deviation of the amplitude. i ;

[0028] Gabor feature g i Perform Z-score standardization;

[0029] Unsupervised clustering is used to analyze Gabor features g based on the Z-score standard. i Classify and identify the faulty sections.

[0030] Prioritize unsupervised clustering for Gabor features g based on the Z-score standard. i Classification includes:

[0031] Calculate the average profile coefficient for all candidate cluster sets, and select the cluster number K* corresponding to the one with the largest average profile coefficient;

[0032] If the relative increase in the average silhouette coefficient between the number of clusters K* and the number of adjacent smaller clusters does not exceed the threshold τ, then the number of adjacent smaller clusters is selected as the final number of clusters, and K-means clustering is performed with the final number of clusters to obtain the category labels;

[0033] If there is one or more potential fault segments in a fault segment cluster, the potential fault segment closest to the end of the fault segment cluster is selected as the fault segment.

[0034] Prior to this, a segment location system for intelligent identification of distribution network fault time-series data images, based on the segment location method for intelligent identification of distribution network fault time-series data images as described above, includes:

[0035] The imaging mapping module is used to acquire short-time window sequences of zero-sequence current. i The length is L i , i is the sequence number of the short-time window sequence of the zero-sequence current; determine the length L i Is it a stable mapping if the length L i If it is not a stable mapping, then for length L i Perform minimum zero-padding to obtain the length L' of the short-time window sequence of the zero-sequence current. i If the length L i For a stable mapping, then determine the relationship with... The factor pair with the smallest distance (r,c) will be the short-time window sequence c of the zero-sequence current. i Rearranged into a two-dimensional array And by transposing, we get F=D T Otherwise, confirm with The factor pair (r', c') with the smallest distance will be zero-padded. Rearranged into a two-dimensional array And transpose to get F=D' T Using the global minimum and global maximum values ​​calculated jointly by short time windows of all segments as the upper and lower limits of quantization, F is linearly mapped to the [0,1] grayscale range to obtain the two-dimensional image Im. i ;

[0036] The feature and clustering module is used to obtain the two-dimensional image Im. i Then, based on Gabor filtering, the two-dimensional image Im... i Unsupervised clustering is performed; among which, Gabor filtering is used to perform clustering on the two-dimensional image Im. i Unsupervised clustering is performed, including: constructing a Gabor filter bank g {λ,θ} Where λ∈[2,16] pixels, θ is 4 to 8 angularly spaced directions from 0° to 180°, ψ=0 or ψ∈[-π,π], γ∈[0.3,1.0], σ=κλ, κ∈[0.5,1.0]; for the two-dimensional image Im iApply g one by one λ,θ Receive response R λ,θ And calculate the two-dimensional image Im i Gabor features g are obtained by taking the mean amplitude and standard deviation of the amplitude. i ; for Gabor feature g i Z-score standardization is performed; unsupervised clustering is used to standardize the Gabor features g according to the Z-score standard. i Classify and identify the faulty sections.

[0037] Preferably, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the methods described herein.

[0038] Preferably, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described herein.

[0039] The beneficial effects achieved by this invention are as follows:

[0040] 1. By adopting an unsupervised learning mechanism, fault detection can be achieved without relying on fault labels and complex prior models (such as Mayr or Cassie models), which reduces the method's dependence on field data accumulation and model parameter tuning, and simplifies the complexity of engineering deployment.

[0041] 2. By combining the image texture features after signal conversion with Gabor filters for feature extraction, the weak transient current features exhibited by transient faults and high-resistance grounding faults within a short time window can be effectively captured. This overcomes the problem of weak fault features and difficulty in line selection caused by the compensation effect in the traditional steady-state component method in the grounding system with arc suppression coil.

[0042] 3. By constructing transient energy indices and combining principal component analysis (PCA) and cluster analysis for fault segment identification, it is possible to effectively distinguish faulty lines from healthy lines under conditions of short transient process duration and non-stationary signal, thereby improving the identification sensitivity and reliability in high-resistance grounding fault scenarios of resonant systems.

[0043] 4. This method is designed as a pre-screening module that can be run both online and offline. Its core algorithm relies on widely collected zero-sequence current signals, eliminating the need for additional dedicated coupling equipment. Furthermore, it significantly reduces the workload of manual intervention and tuning through automated feature extraction and judgment.

[0044] 5. This invention is particularly suitable for the location of distribution network sections with neutral points grounded by arc suppression coils. It exhibits good adaptability to intermittent faults and high-resistance faults (transition resistance can reach thousands of ohms or more), which helps to improve the efficiency of fault section identification under complex working conditions and shorten the fault investigation time. It provides a low-cost, fast-response, and efficient solution for the stable operation of distribution networks and has significant engineering promotion value. Attached Figure Description

[0045] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a schematic diagram of Embodiment 1 of this application;

[0047] Figure 2 This is a schematic diagram of the simulated topology of a distribution network with the neutral point grounded through an arc suppression coil in Embodiment 1 of this application;

[0048] Figure 3 This is a waveform diagram of the zero-sequence current in 13 sections in Embodiment 1 of this application;

[0049] Figure 4 This is a schematic diagram of the two-dimensional images of the zero-sequence current in segments 1, 2, and 8 in Embodiment 1 of this application;

[0050] Figure 5 This is a waveform diagram of the zero-sequence current in 13 sections in Embodiment 2 of this application;

[0051] Figure 6 This is a schematic diagram of the two-dimensional zero-sequence current images of segments 1, 2, 10 and 13 in Embodiment 2 of this application; Detailed Implementation

[0052] Example 1

[0053] See Figure 1 This application discloses a segment location method for intelligent identification of fault time-series data in distribution networks, including: S1, acquiring i synchronously sampled zero-sequence current short-time window sequences c i The length is L i ; i is the sequence number of the short-time window sequence of the zero-sequence current, and more specifically, c i This represents the first zero-sequence current short-time window sequence.

[0054] S2, regarding length L i Perform stability factor decomposition to determine: if the length L i It can be decomposed into factor pairs (r,c) and has a square property index. Not greater than the threshold ρ max If ∈[2,3], then the decision length L i If it is a "stable mapping", otherwise determine the length L. i This is an "unstable mapping"; max(r,c) is the maximum factor pair, and min(r,c) is the minimum factor pair, where r and c represent the row factor and column factor, respectively.

[0055] S3. If it is an "unstable mapping", then for length L i Perform minimum zero padding: based on a predetermined minimum padding value ΔL i This makes the short-time window sequence length L' of the zero-sequence current so that i =L i +ΔL i Decompose into (r', c') and ρ' ≤ ρ max Where ρ' is the squareness index calculated after minimum zero-padding, ρ max Let r' be the threshold, r' be the row factor obtained after minimum zero-padding, c' be the column factor obtained after minimum zero-padding, and L' be the column factor. i This is the length of the short-time window sequence of the zero-sequence current obtained after minimum zero-filling.

[0056] S4. If it is a stable mapping, then determine the relationship with... The factor pair with the smallest distance (r,c) will be the short-time window sequence c of the zero-sequence current. i Rearranged into a two-dimensional array And by transposing, we get F=D T Otherwise, confirm with The factor pair (r', c') with the smallest distance will be zero-padded. Rearranged into a two-dimensional array And transpose to get F=D' T Where D' is the two-dimensional array obtained after minimum zero-padding, and R... r×c Let r be the set of all real-valued r×c matrices. F is the set of all real-valued r'×c' matrices, and F is the transpose of a two-dimensional array D or a two-dimensional array D'.

[0057] S5. Normalization using cross-segment global extremum calibration: Using the global minimum and global maximum values ​​calculated jointly by short time windows of all segments as the upper and lower limits of quantization, F is linearly mapped to the [0,1] grayscale domain to obtain the two-dimensional image Im. i .

[0058] The quantization upper and lower limits are the global minimum and global maximum values ​​calculated jointly by all short-time windows in all segments, including: directly calculating all zero-sequence current short-time window sequences c that have been collected. iThe global minimum and global maximum values ​​are calculated. These two values ​​together constitute the amplitude range of the zero-sequence current signal in all segments within this time window. This global extreme value is used as the normalization benchmark to ensure the comparability of characteristics between zero-sequence current data from different segments.

[0059] In this embodiment of the application, step S2 includes: when the length L i A prime number or length L i When the length L is 1, it is directly determined as an "unstable mapping"; i It is a composite number and the square property index ρ≤ρ max It is determined to be a "stable mapping".

[0060] In this embodiment of the application, the minimum zero-padding in step S3 satisfies:

[0061] ΔL i =inf{Δ∈R + ∪{0}∣(L i +Δ)∈S d , ρ(L i +Δ)≤ρ max},

[0062] In the formula, ΔL i To find the minimum zero-padding length that satisfies the condition, inf{} is the infimum used to find the minimum value among all Δ values ​​that satisfy the condition, where Δ represents the length of the zero-padding, R. + R is the set of positive real numbers. + ∪{0} is the set of non-negative real numbers, S d It is the set of composite numbers;

[0063] In this embodiment of the application, the selection of the factor pair (r,c) in step S4 satisfies:

[0064] ,

[0065] In the formula, a is the first positive real number, b is the second positive real number, L is the length of the short-time window sequence of zero-sequence current, and argmin is the independent variable that minimizes the value.

[0066] In this embodiment of the application, the normalized two-dimensional image Im i Perform contrast-limited adaptive histogram equalization (CLAHE) enhancement with clipping limit factor ClipLimit∈[0.01,0.1].

[0067] In this embodiment of the application, once all i short-time window sequences of zero-sequence current collected have been processed (this process has been executed i times in a loop), it is determined that all lines have been processed.

[0068] Gabor filtering for two-dimensional image Im iUnsupervised clustering includes:

[0069] S6. Constructing a Gabor filter bank g {λ,θ} , where λ∈[2,16] pixels, θ is 4 to 8 angular intervals of 0°~180°, ψ=0 or ψ∈[-π,π], γ∈[0.3,1.0], σ=κλ, κ∈[0.5,1.0];

[0070] S7, for the two-dimensional image Im i Apply g one by one λ,θ Receive response R λ,θ And calculate the two-dimensional image Im i Gabor features g are obtained by taking the mean amplitude and standard deviation of the amplitude. i ;

[0071] S8, Gabor feature g i Perform Z-score standardization, and optionally perform principal component analysis (PCA) dimensionality reduction;

[0072] S9. Without introducing any label information, use unsupervised clustering to normalize (or reduce) the Gabor features g by Z-score. i The classification process is as follows: For all candidate cluster sets, calculate the average profile coefficient, and select the cluster number K* corresponding to the one with the largest average profile coefficient. When the relative increase in the average profile coefficient between cluster number K* and its smaller neighboring clusters does not exceed a threshold τ, select the smaller neighboring clusters and perform K-means clustering with the final cluster number to obtain the category label. Within a fault segment cluster, if there is more than one potential fault segment, select the segment closest to the end of the fault segment cluster as the fault segment.

[0073] In this embodiment, step S7 further fuses texture features obtained from the gray-level co-occurrence matrix (GLCM). These texture features include contrast, correlation, energy, and homogeneity. The texture features are then combined with Gabor features. i The cascaded features form a fusion feature, and the clustering of the fusion feature is still unsupervised clustering.

[0074] In this embodiment of the application, the unsupervised clustering in step S9 is K-means clustering, which uses 10 to 50 random initializations and an upper limit of 100 to 300 iterations to improve stability.

[0075] In the embodiments of this application, no labeled data, annotations or supervision signals are used in the entire clustering process, and supervised classifiers or discriminators, including logistic regression, support vector machines, random forests, gradient boosting trees, and neural network classifiers, are not trained or invoked.

[0076] The core innovation of this invention lies in its pioneering stable mapping mechanism of "one-dimensional zero-sequence current short-time window signal - two-dimensional image" to address the limitations of traditional positioning methods under complex operating conditions and the "weak difference and weak directionality" of arc suppression coil grounding systems. This mechanism achieves the adaptation and conversion of signals of different lengths through stability factor decomposition and minimum zero-filling, and achieves dimensional comparison by combining cross-segment global extreme value calibration. At the same time, it innovatively uses Gabor filter banks to extract multi-directional frequency features, combined with K-means unsupervised clustering based on average profile coefficient to adaptively determine the number of clusters. This allows for accurate differentiation between faulty and healthy sections without the need for labeled data, fine prior models, or supervision signals. It constructs a structured representation system of "signal-image-feature-clustering", which not only overcomes the bottlenecks of poor robustness and high dependence of traditional methods, but also achieves rapid and lightweight positioning, providing an innovative technical solution for intelligent operation and maintenance of distribution networks.

[0077] To verify the effectiveness of the proposed image-based distribution network fault location method based on Gabor filters, a system was built in PSCAD / EMTDC as follows: Figure 2 The neutral point shown is grounded through an arc suppression coil in the distribution network. All lines use YJV22-6 / 10kV-3*70mm2 cable lines, with a unit zero-sequence capacitance to ground parameter of C0=124.28×10-9F / km. Each outgoing line is connected to a constant impedance load model of 1MW+0.1MVar per phase.

[0078] Example 1

[0079] In this simulation case, the sampling frequency fs = 25kHz and the fault time is set to 0.04s. To verify the accuracy of the image-based distribution network fault location method implemented based on Gabor filters, a single-phase ground fault with a transition resistance of 1500Ω is set at the midpoint of section l8. The zero-sequence current waveforms flowing through the beginning of each section are shown in the figure below. Figure 3 As shown.

[0080] The sampling time window is uniformly set to 0.035–0.050 s; a short time window of equal length c is extracted from each segment. i Length L i =375 points. Factorization yields factor pairs (r,c) of (25*15), classifying it as a "stable mapping". Further cross-segment normalization is performed after mapping, resulting in a two-dimensional image Im of the pseudo-color map for 13 segments. i Two-dimensional images of some typical sections are shown below. Figure 4 As shown.

[0081] Construct a Gabor filter bank with θ∈{0°,30°,60°,90°,120°,150°}, and obtain the Gabor feature g by statistically analyzing the mean and standard deviation of the amplitude for each response. i ; for Gabor feature gi Z-score standardization was performed; after PCA dimensionality reduction, the first 2-3 principal components were retained for visualization. By comparing the average silhouette coefficients of various clustering schemes, the final number of clusters, K, was determined to be 3. The clustering results are shown in Table 1.

[0082] Table 1. Segment Fault Clustering Classification Table

[0083]

[0084] As shown in Table 1, the clustering results can eliminate interference segment 1 (the upstream segment of segment 8) and accurately identify the fault segment as segment 8. It has the ability to distinguish single-phase grounding faults in distribution networks grounded through arc suppression coils, thus verifying the effectiveness and reliability of this method.

[0085] Example 2

[0086] In this simulation case, the sampling frequency fs = 25kHz and the fault time is set to 0.04s. To verify the accuracy of the image-based distribution network fault location method implemented based on Gabor filters, in section l... 13 For a single-phase ground fault with a transition resistance of 2000Ω at the midpoint, the zero-sequence current waveforms flowing through the beginning of each section are shown in the figure below. Figure 5 As shown.

[0087] The sampling time window is uniformly set to 0.035–0.050 s; a short time window of equal length c is extracted from each segment. i Length L i =375 points. Factorization yields factor pairs (r,c) of (25*15), classifying it as a "stable mapping". Further cross-segment normalization is performed after mapping, resulting in a two-dimensional image Im of 13 segments. i Two-dimensional images of some typical sections are shown below. Figure 6 As shown.

[0088] Construct a Gabor filter bank with θ∈{0°,30°,60°,90°,120°,150°}, and obtain the Gabor feature g by statistically analyzing the mean and standard deviation of the amplitude for each response. i ; for Gabor feature g i Z-score standardization was performed; after PCA dimensionality reduction, the first 2-3 principal components were retained for visualization. By comparing the average silhouette coefficients of various clustering schemes, the final number of clusters K was determined to be 2. The clustering results are shown in Table 2.

[0089] Table 2. Segment Fault Clustering Classification Table

[0090]

[0091] Since faulty segment cluster 1 has two members (segment 10 and segment 13), according to the principle that "if there is more than one potential faulty segment in the faulty segment cluster, the segment closer to the end of the segment is selected as the faulty segment", segment 13 is determined to be the faulty segment.

[0092] As shown in Table 2, the clustering results can eliminate interference items segment 2 and segment 10 (the upstream segment of segment 13), and accurately identify the fault segment as segment 13. It has the ability to distinguish single-phase grounding faults in distribution networks grounded through arc suppression coils, thus verifying the effectiveness and reliability of this method.

[0093] In the embodiments of this application, the unsupervised clustering output includes the category label, principal component value and contour coefficient of each segment, and is used to form a pre-screening list of faulty segments or a candidate set for subsequent segment location.

[0094] In this embodiment of the application, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the methods described above.

[0095] In this application embodiment, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0096] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0097] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention described herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not invented herein. The specification and embodiments are to be considered exemplary only.

[0098] The above specific embodiments further illustrate the purpose, technical solution and beneficial effects of this application. It should be understood that the above are only specific embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of this application should be included within the scope of protection of this application.

Claims

1. A method for segment location based on the image-based intelligent identification of time-series fault data in a distribution network, characterized in that, include: At the moment of a fault in the distribution network, several short-time window sequences of zero-sequence current of equal length from different sections of the distribution network are collected. i Zero-sequence current short-time window sequence c i The length is L i , where i is the sequence number of the short-time window sequence of the zero-sequence current; Determine whether the zero-sequence current short-time window sequence satisfies the stable mapping. If the zero-sequence current short-time window sequence does not satisfy the stable mapping, then perform minimum zero-padding on the zero-sequence current short-time window sequence to obtain a new zero-sequence current short-time window sequence c'. i The new zero-sequence current short-time window sequence c' i The new length is L' i ; If the zero-sequence current short-time window sequence c i If a stable mapping is satisfied, then the relationship with... The factor pair with the smallest distance (r,c) will be the short-time window sequence c of the zero-sequence current. i Rearranged into a two-dimensional array And by transposing, we get F=D T Otherwise, confirm with The factor pair with the smallest distance (r', c') will be the new zero-sequence current short-time window sequence after zero-filling. Rearranged into a two-dimensional array And transpose to obtain the transpose matrix F=D' T ; Using the global minimum and global maximum values ​​calculated jointly by short time windows of all segments as the upper and lower limits of quantization, the transpose matrix F is linearly mapped to the [0,1] grayscale domain to obtain the two-dimensional image Im. i .

2. The method for segment location based on the image-based intelligent identification of time-series fault data in a distribution network according to claim 1, characterized in that, Determine whether the zero-sequence current short-time window sequence satisfies the stable mapping. If the zero-sequence current short-time window sequence does not satisfy the stable mapping, then perform minimum zero-padding on the zero-sequence current short-time window sequence to obtain a new zero-sequence current short-time window sequence c'. i The new zero-sequence current short-time window sequence c' i The new length is L' i ,include: If the length L i It can be decomposed into factor pairs (r,c) and has a square property index. Not greater than the threshold ρ max If ∈[2,3], then the zero-sequence current short-time window sequence is determined to satisfy the "stable mapping"; otherwise, the zero-sequence current short-time window sequence c is determined to satisfy the "stable mapping". i The "stable mapping" is not satisfied, where max(r,c) is the maximum factor pair and min(r,c) is the minimum factor pair, and r and c represent the row factor and column factor, respectively. If the short-time window sequence of zero-sequence current does not satisfy the "stable mapping", then for length L i Perform minimum zero padding.

3. The method for segment location based on the image-based intelligent identification of time-series fault data in a distribution network according to claim 2, characterized in that, For the short-time window sequence c of zero-sequence current i Perform minimal zero padding, including: Determine the minimum fill value ΔL i This makes the short-time window sequence length L' of the zero-sequence current so that i Satisfy L' i =L i +ΔL i Furthermore, the factor pairs (r', c') of the decomposition satisfy ρ' ≤ ρ max ,in The quadraticity index is calculated after minimum zero-padding, where max(r',c') is the maximum factor pair and min(r',c') is the minimum factor pair; ρ max Let r' be the threshold, r' be the row factor obtained after minimum zero-padding, c' be the column factor obtained after minimum zero-padding, and L' be the column factor. i This is the length of the short-time window sequence of the zero-sequence current obtained after minimum zero-filling.

4. The method for segment location based on the image-based intelligent identification of time-series fault data in a distribution network according to claim 2, characterized in that, Determine whether the zero-sequence current short-time window sequence satisfies the stable mapping. If the zero-sequence current short-time window sequence does not satisfy the stable mapping, then perform minimum zero-padding on the zero-sequence current short-time window sequence to obtain a new zero-sequence current short-time window sequence c'. i The new zero-sequence current short-time window sequence c' i The new length is L' i ,include: If the length L i A prime number or length L i =1, then determine the length L i The "stable mapping" is not satisfied; if the length L i It is a composite number and the square property index ρ≤ρ max In determining the length L i It satisfies the "stable mapping".

5. The method for segment location based on the image-based intelligent identification of time-series fault data in a distribution network according to claim 1, characterized in that, Minimum zero-padding satisfies: ΔL i =inf{Δ∈R + ∪{0}∣(L i +Δ)∈S d , ρ(L i +Δ)≤ρ max }, In the formula, ΔL i Let R be the minimum zero-padding length, and let inf{} be the minimum value among all Δ values ​​in the infimum, where Δ represents the length of the zero-padding. + R is the set of positive real numbers. + ∪{0} is the set of non-negative real numbers, S d For the set of composite numbers, L i ρ is the length. max The threshold value is used.

6. The method for segment location based on the image-based intelligent identification of time-series fault data in a distribution network according to claim 1, characterized in that, The factor pair (r, c) satisfies: ， In the formula, a is the first positive real number, b is the second positive real number, L is the length of the short-time window sequence of zero-sequence current, and argmin is a function of the minimum value.

7. The method for segment location based on the image-based intelligent identification of time-series fault data in a distribution network according to claim 1, characterized in that, After normalization, the two-dimensional image Im i Perform contrast-limited adaptive histogram equalization enhancement with a clipping limit factor ClipLimit ∈ [0.01, 0.1].

8. The method for segment location based on the image-based intelligent identification of time-series fault data in a distribution network according to claim 1, characterized in that, Obtain the two-dimensional image Im i Then, based on Gabor filtering, the two-dimensional image Im... i Perform unsupervised clustering; Among them, the two-dimensional image Im is based on Gabor filtering. i Unsupervised clustering includes: Construct a Gabor filter bank g(λ,θ), where λ∈[2,16] pixels, θ is 4 to 8 directions with equal angular intervals of 0° to 180°, ψ=0 or ψ∈[-π,π], γ∈[0.3,1.0], σ=κλ, κ∈[0.5,1.0]; For two-dimensional image Im i Apply g(λ,θ) one by one to obtain the response R(λ,θ) and calculate the two-dimensional image Im. i Gabor features g are obtained by taking the mean amplitude and standard deviation of the amplitude. i ; Gabor feature g i Perform Z-score standardization; Unsupervised clustering is used to analyze Gabor features g based on the Z-score standard. i Classify and identify the faulty sections.

9. The method for segment location based on the image-based intelligent identification of time-series fault data in a distribution network according to claim 8, characterized in that, Unsupervised clustering is used to analyze Gabor features g based on the Z-score standard. i Classification includes: Calculate the average profile coefficient for all candidate cluster sets, and select the cluster number K* corresponding to the one with the largest average profile coefficient; If the relative increase in the average silhouette coefficient between the number of clusters K* and the number of adjacent smaller clusters does not exceed the threshold τ, then the number of adjacent smaller clusters is selected as the final number of clusters, and K-means clustering is performed with the final number of clusters to obtain the category labels; If there is one or more potential fault segments in a fault segment cluster, the potential fault segment closest to the end of the fault segment cluster is selected as the fault segment.

10. A section location system for intelligent identification of time-series fault data in a distribution network, characterized in that, A method for segment location based on the image-based intelligent identification of time-series fault data in a distribution network according to any one of claims 1-9 includes: The imaging mapping module is used to acquire short-time window sequences of zero-sequence current. i The length is L i , i is the sequence number of the short-time window sequence of the zero-sequence current; determine the length L i Is it a stable mapping if the length L i If it is not a stable mapping, then for length L i Perform minimum zero-padding to obtain the length L' of the short-time window sequence of the zero-sequence current. i If the length L i For a stable mapping, then determine the relationship with... The factor pair with the smallest distance (r,c) will be the short-time window sequence c of the zero-sequence current. i Rearranged into a two-dimensional array And by transposing, we get F=D T Otherwise, confirm with The factor pair (r', c') with the smallest distance will be zero-padded. Rearranged into a two-dimensional array And transpose to get F=D' T Using the global minimum and global maximum values ​​calculated jointly by short time windows of all segments as the upper and lower limits of quantization, F is linearly mapped to the [0,1] grayscale range to obtain the two-dimensional image Im. i ; The feature and clustering module is used to obtain the two-dimensional image Im. i Then, based on Gabor filtering, the two-dimensional image Im... i Unsupervised clustering is performed; among which, Gabor filtering is used to perform clustering on the two-dimensional image Im. i Unsupervised clustering is performed, including: constructing a Gabor filter bank g {λ,θ} Where λ∈[2,16] pixels, θ is 4 to 8 angularly spaced directions from 0° to 180°, ψ=0 or ψ∈[-π,π], γ∈[0.3,1.0], σ=κλ, κ∈[0.5,1.0]; for the two-dimensional image Im i Apply g one by one λ,θ Receive response R λ,θ And calculate the two-dimensional image Im i Gabor features g are obtained by taking the mean amplitude and standard deviation of the amplitude. i ; for Gabor feature g i Z-score standardization is performed; unsupervised clustering is used to standardize the Gabor features g according to the Z-score standard. i Classify and identify the faulty sections.

Images