Cross-dimensional transient fingerprint elastic alignment and collaborative discrimination method based on event pseudo-synchronization

Abstract

This invention discloses a cross-dimensional transient fingerprint elastic warping and collaborative discrimination method based on event pseudo-synchronization, belonging to the field of distribution network automation and relay protection technology. Addressing the problems of misjudgment in multi-terminal waveform comparison due to the lack of high-precision time synchronization in distribution terminals, and the limited communication bandwidth, this method first establishes event pseudo-synchronization time anchors using a nonlinear mutation operator with an adaptive sliding window, extracting a dimension-reduced cross-dimensional transient fingerprint containing macroscopic steady-state, microscopic frequency domain energy distribution, and morphological trajectory envelopes that avoid DC components. In the same-source fault discrimination stage, an elastic dynamic warping distance function with a time offset penalty factor bound to the physical error boundary of the power grid is constructed, and collaborative verification is performed using frequency domain cosine space mapping. This invention completely decouples the dependence on absolute time synchronization, adaptively eliminating asynchronous sampling phase shift errors while reducing the amount of communication interaction data, significantly improving the robustness of same-source fault identification in complex distribution networks.

Description

Technical Field

[0001] This invention discloses a cross-dimensional transient fingerprint elastic regularization and collaborative discrimination method based on event pseudo-synchronization, belonging to the field of distribution network automation and relay protection technology. Background Technology

[0002] In modern power distribution networks, accurately identifying whether adjacent distribution terminals detect the same fault (i.e., fault origin identification) is a prerequisite for achieving precise fault location and isolation. Currently, differential protection or origin identification technologies based on multi-terminal waveform comparison rely heavily on strict microsecond-level time synchronization (such as GPS, BeiDou, or the IEEE 1588PTP protocol).

[0003] The existing technology has the following drawbacks when applied to conventional power distribution terminals:

[0004] 1. High synchronization cost and difficult maintenance: Most medium and low voltage power distribution terminals are limited by cost and communication conditions and cannot be configured with high-precision hardware synchronization clocks.

[0005] 2. Traditional asynchronous comparison is prone to misjudgment: Without absolute time synchronization, the time point at which both terminals detect a sudden fault change will inevitably have a nonlinear delay error at the millisecond level. Traditional waveform comparison methods based on the difference of instantaneous values ​​or Pearson correlation coefficients will experience a sharp drop in correlation when encountering slight time offsets, causing "same-source faults" to be misjudged as "different-source faults".

[0006] 3. Communication bandwidth bottleneck: Transmitting millisecond-level high-frequency waveform recording files involves extremely large amounts of data, which can easily cause congestion in 5G / 4G-based wireless power distribution communication networks. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a cross-dimensional transient fingerprint elastic regularization and collaborative discrimination method based on event pseudo-synchronization.

[0008] To address the aforementioned technical problems, this invention provides a method for cross-dimensional transient fingerprint elastic regularization and collaborative discrimination based on event pseudo-synchronization, comprising the following steps:

[0009] S1: Real-time acquisition of electrical quantity sequences of power distribution terminals, based on nonlinear mutation detection algorithm, establishing the mutation moment as the event pseudo-synchronization time anchor point t0, and using this as a benchmark to extract transient data windows;

[0010] S2: Perform multi-scale dimensionality reduction on the waveform within the transient data window to extract and generate a cross-dimensional transient fingerprint vector F. The cross-dimensional transient fingerprint vector F includes: a macroscopic steady-state feature vector F. macro Frequency domain energy distribution vector F freq and the morphological trajectory envelope sequence F shape ;

[0011] S3: Exchange the cross-dimensional transient fingerprint vectors extracted from adjacent power distribution terminals through the power distribution communication network;

[0012] S4: Perform macroscopic coarse screening, comparing the macroscopic steady-state eigenvectors F at both ends. macro And the local rough timestamp, if the consistency condition is met, proceed to step S5; otherwise, it is determined to be a non-same-source fault.

[0013] S5: Perform transient collaborative discrimination and calculate the frequency domain energy distribution vectors F at both ends. freq Spatial cosine similarity; simultaneously, an elastic warping algorithm with a time offset penalty factor is used to calculate the envelope sequence F of the morphological trajectories at both ends. shape The matching distance is used to calculate the comprehensive confidence score by weighting and fusing the two indicators. When the comprehensive confidence score is greater than the preset threshold, it is determined that the two ends have detected the same source fault.

[0014] Furthermore, the electrical quantity sequence of the distribution terminal is acquired in real time. Based on a nonlinear mutation detection algorithm, the mutation moment is established as the pseudo-synchronization time anchor point t0 of the event, and a transient data window is extracted based on this, including:

[0015] Real-time acquisition of electrical quantity sequences is performed, and the discrete Teager energy operator is used to perform nonlinear optimization on the sampled sequence x(n) to amplify high-frequency abrupt changes. The calculation formula is as follows:

[0016]

[0017] Where: x(n) represents the instantaneous amplitude of the electrical quantity collected at the current nth discrete sampling point; x(n-1) and x(n+1) represent the instantaneous amplitudes of the previous and next sampling points, respectively; E(n) represents the instantaneous change energy value of the current nth sampling point calculated by the TEO operator;

[0018] Maintain a sliding data window of fixed length L in front of the current sampling point, and calculate the average background energy reference value within this window:

[0019]

[0020] Where: L represents the length of the sliding data window; k is the temporal traversal index for the summation operation; E base (n): represents the average absolute value of the background energy reference within a historical window of length L before the current time n, used to characterize the current steady-state noise level of the power grid;

[0021] Real-time generation of dynamic trigger thresholds that adapt to and float with background noise:

[0022]

[0023] Among them: E th (n) represents the adaptive trigger threshold corresponding to the current nth sampling point; K m The set sensitivity factor is used to filter out normal load switching energy fluctuations; E min To prevent zero drift, a minimum threshold constant is required;

[0024] When consecutive N c The instantaneous energy E(n) at each sampling point is greater than the corresponding dynamic trigger threshold E. th When (n), the first point exceeding the limit is formally established as the pseudo-synchronization time anchor point t0 of the event; N c To confirm the counter value for image stabilization;

[0025] The transient data window is reconstructed using t0 as the origin of the time coordinate system.

[0026] Furthermore, in step S2, the macroscopic steady-state feature vector F macro The extraction methods include:

[0027] Based on t0, select Select the steady-state data window before the fault. This is the data window after the fault, where T cycle It is the power frequency cycle;

[0028] The fundamental amplitude values ​​within the two data windows are extracted using the full-wave Fourier algorithm. The steady-state current RMS value mutation ΔI and voltage RMS value mutation ΔU are calculated, and combined with the fault phase determined by the sequence component, the macroscopic steady-state feature vector F is formed. macro .

[0029] Furthermore, in step S2, the frequency domain energy distribution vector F freq The extraction methods include:

[0030] The transient high-frequency sequence after t0 is decomposed into multiple layers using wavelet packet transform.

[0031] The lowest frequency band containing the fundamental frequency and low harmonics, as well as the highest frequency band affected by the attenuation of the line distributed capacitance, are discarded, and only K sub-frequency bands within the middle characteristic frequency band range are extracted.

[0032] Calculate the wavelet packet reconstruction energy of the K sub-bands, divide by the total energy of these K bands for normalization, and generate the frequency domain energy distribution vector. Its element calculation formula is:

[0033]

[0034] Among them: E ke represents the absolute value of the reconstructed energy of the k-th feature sub-band extracted after wavelet packet decomposition. k This represents the energy percentage of the k-th frequency band.

[0035] Furthermore, in step S2, the morphological trajectory envelope sequence F shape The extraction methods include:

[0036] The Hilbert transform is used to construct an analytical signal from the transient current waveform after t0 to obtain the upper envelope. The upper envelope is then downsampled proportionally with a fixed step size to generate a discrete feature point sequence F of length M. shape ={S1,S2,...,S M}; M is the total number of points in the dimensionality-reduced sequence, S m This represents the amplitude of the m-th envelope feature point after downsampling.

[0037] Furthermore, in step S4, a macroscopic coarse screening is performed, comparing the macroscopic steady-state feature vectors F at both ends. macro And a local coarse timestamp. If the consistency condition is met, proceed to step S5; otherwise, it is determined to be a non-homogeneous fault, including:

[0038] Compare the macroscopic steady-state eigenvectors F at both ends macro Whether they are completely consistent, and compare whether the difference between the rough timestamps issued locally by the two devices is within the tolerance range;

[0039] If any condition is not met, the result is a veto and the fault is directly determined to be non-homogeneous; if the coarse screening is passed, proceed to step S5.

[0040] Furthermore, in step S5, the frequency domain energy distribution vector F at both ends is calculated. freq Spatial cosine similarity includes:

[0041] Calculate the cosine similarity Sim in multidimensional space of the frequency domain energy distribution vectors extracted from adjacent distribution terminals A and B. freq :

[0042] ;

[0043] Where e A,k, e B,k These are the values ​​of the k-th elements in the frequency domain energy distribution vectors extracted from terminal A and terminal B, respectively; Sim freq The frequency domain cosine similarity has a value range of [0,1]. The closer the value is to 1, the more similar the high-frequency energy distributions of the faults at both ends are. This represents the frequency domain energy distribution vector of terminal A. This represents the frequency domain energy distribution vector of terminal B.

[0044] Furthermore, in step S5, an elastic warping algorithm with a time offset penalty factor is used to calculate the envelope sequence F of the morphological trajectories at both ends. shape The matching distance includes:

[0045] Construct a dynamic programming grid matrix of size M×M, and apply the morphological sequences F at both ends. shapeA ={S A,1 ,S A,2 ,...,S A,M} and F shapeB ={S B,1 ,S B,2 ,...,S B,M Perform elastic stretching and compression matching;

[0046] The single-step node cost function of the elastic warping algorithm is reconstructed as follows:

[0047]

[0048] Where i and j are the feature point sequence indices of terminal A and terminal B, respectively, physically representing the temporal position of the feature point after downsampling; S A,i S represents the amplitude of the i-th feature point in the morphological sequence of terminal A; B,j This represents the magnitude of the j-th feature point in the terminal B morphology sequence;

[0049] |S A,i -S B,j | represents the absolute difference cost between the two points in the vertical magnitude dimension; (ij) 2 This represents the squared offset of these two points on the horizontal time axis; λ is the time offset penalty coefficient.

[0050] After calculating the single-step cost c(i,j) of each node in the matrix, the cumulative normalized distance matrix D(i,j) is calculated using a dynamic programming algorithm, and its state transition equation is as follows:

[0051]

[0052] The global minimum normalization distance between two feature sequences of length M is found to be D. warp =D(M,M);

[0053] By using a negative exponential normalized mapping function, the global normalized distance D is... warp Converted into morphological similarity score Sim shape :

[0054]

[0055] Where κ is the smoothing decay constant, used to map the distance value to the (0,1] interval; Sim shape The final output is the morphological similarity score; if the two sequences are completely identical and have no time shift, D... warp =0, Sim shape The maximum value of 1 has been reached.

[0056] Furthermore, the value of the time offset penalty coefficient λ is strictly constrained by the physical error boundary of the power grid, satisfying the inequality:

[0057]

[0058] Among them, W max This represents the maximum allowable index deviation value converted from the maximum asynchronous misalignment time of the hardware to the downsampled sequence; C amp_max γ is the per-unit value of the maximum possible amplitude difference of the same transient waveform under normal transformer transmission error; γ is the blocking constant greater than 1.

[0059] When the difference between the horizontal and vertical coordinates of the matching path |ij| exceeds the maximum physical deviation W max At that time, the cost generated by the quadratic penalty term will be directly greater than the normal amplitude difference limit γC. amp_max This forces the blocking of the matching path, thereby blocking the excessive stretching matching that violates the transient physical evolution law of the power grid.

[0060] Furthermore, in step S5, a weighted fusion calculation is performed on the two indicators to determine the overall confidence level, including:

[0061] The formula for calculating the overall confidence level using weighted fusion is:

[0062] ;

[0063] Score is the overall confidence score; α and β are empirical weighting coefficients adjusted according to the field channel and transformer type, and α+β=1 is guaranteed.

[0064] Furthermore, K m The value is 3.0~5.0;

[0065] N c Set to 3;

[0066] K=4;

[0067] M is 10~20.

[0068] In a second aspect, the present invention provides a power distribution terminal fault source identification system employing the method described in the first aspect, comprising:

[0069] The data acquisition module is used to acquire electrical quantity sequences in real time.

[0070] The pseudo-synchronization anchor point generation module is used to establish the pseudo-synchronization time anchor point t0 of an event based on a nonlinear mutation detection algorithm.

[0071] The fingerprint extraction module is used to extract a cross-dimensional transient fingerprint vector F, which includes a macroscopic steady-state feature vector F. macro Frequency domain energy distribution vector F freq and morphological trajectory envelope sequence F shape ;

[0072] The communication module is used to exchange the cross-dimensional transient fingerprint vector with adjacent power distribution terminals;

[0073] The coarse screening module is used to compare the macroscopic steady-state eigenvector F. macro and a local rough timestamp;

[0074] The collaborative discrimination module is used to calculate the frequency domain cosine similarity and the elastic regularized matching distance with a time offset penalty factor, and then weighted and fused to output the comprehensive confidence score.

[0075] The beneficial effects achieved by this invention are as follows:

[0076] 1. This invention proposes a fault origin identification method for distribution networks without high-precision time synchronization. This method establishes pseudo-synchronization anchor points for events through an adaptive nonlinear operator, completely decoupling from the dependence on absolute hardware clocks.

[0077] 2. By combining anti-DC filtering and frequency band normalization strategies, cross-dimensional reduction transient fingerprints are extracted within one power frequency cycle, reducing the amount of multi-terminal communication data by two orders of magnitude, effectively breaking through the bandwidth bottleneck and overcoming signal attenuation and distortion.

[0078] 3. To address the millisecond-level timing deviation of asynchronous sampling, this scheme constructs a time offset penalty mechanism that integrates the physical boundary constraints of the power grid to reconstruct the regularization algorithm. This mechanism can adaptively compensate for reasonable phase shift errors and strictly prevent over-matching that violates transient laws, significantly improving the accuracy and robustness of distributed fault collaborative judgment under complex operating conditions. Attached Figure Description

[0079] Figure 1 This is the overall execution flowchart of the cross-dimensional transient fingerprint elastic regularization and collaborative discrimination method based on event pseudo-synchronization. Detailed Implementation

[0080] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0081] Example 1:

[0082] This embodiment provides a cross-dimensional transient fingerprint elastic regularization and collaborative discrimination method based on event pseudo-synchronization, such as... Figure 1 As shown, it includes the following steps:

[0083] Step 1: Adaptive extraction of event pseudo-synchronization time anchors based on background energy tracking

[0084] To address the issues of the lack of a high-precision hardware clock and drastic changes in background noise in power distribution terminals, this step establishes a relative time reference locally using nonlinear operators and a sliding window dynamic threshold.

[0085] 1.1 Instantaneous Energy Calculation: The distribution terminal collects electrical quantity sequences (such as phase current or phase voltage) in real time, and uses the Discrete Teager Energy Operator (TEO) for nonlinear optimization to amplify high-frequency abrupt changes. The calculation formula is as follows:

[0086]

[0087] Where: x(n) represents the instantaneous amplitude of the electrical quantity collected at the current nth discrete sampling point; x(n-1) and x(n+1) represent the instantaneous amplitudes of the previous and next sampling points, respectively; E(n) represents the instantaneous change energy value of the current nth sampling point calculated by the TEO operator.

[0088] 1.2 Background Sliding Window Maintenance: To overcome false triggering caused by load fluctuations, a fixed-length sliding data window is always maintained before the current sampling point to calculate the background energy baseline value in real time.

[0089]

[0090] Where: L represents the length of the sliding data window (i.e., the number of sampling points it contains), preferably the total number of sampling points corresponding to one power frequency cycle; k is the time-series traversal index for the summation operation; E base (n): represents the average absolute value of the background energy reference within a historical window of length L before the current time n, used to characterize the current steady-state noise level of the power grid.

[0091] 1.3 Dynamic threshold generation and stabilization confirmation:

[0092] Real-time generation of dynamic trigger thresholds that adapt to and float with background noise:

[0093]

[0094] Among them: E th (n) represents the adaptive trigger threshold corresponding to the current nth sampling point; K m The set sensitivity factor (preferably 3.0~5.0) is used to filter out normal load switching energy fluctuations; Emin To prevent zero drift, a minimum threshold constant is used to prevent E from drifting when the power grid is in an absolute dead zone (no background noise). base Approaching zero leads to weak quantization noise that causes false triggering.

[0095] When consecutive N c (The anti-shake confirmation count value, preferably set to 3) The instantaneous energy E(n) of each sampling point is greater than the corresponding E th When (n), a genuine and valid physical mutation is detected, and the system local count value corresponding to the first limit violation point is officially marked as the event pseudo-synchronization time anchor point t0. The transient data window is reconstructed using t0 as the origin of the time coordinate system (time zero).

[0096] Step 2: Multi-scale dimensionality reduction extraction of cross-dimensional transient waveform fingerprints

[0097] Using t0 as the absolute zero baseline, fingerprint matrix vectors containing three dimensions are extracted in parallel. :

[0098] 2.1 Extracting macroscopic steady-state features Fmacro:

[0099] Based on t0, select Select the steady-state data window before the fault. The data windows are defined as follows: Tcycle represents the power frequency cycle. The fundamental amplitude values ​​within the two data windows are extracted using the full-wave Fourier transform algorithm. The steady-state current RMS value mutation ΔI and voltage RMS value mutation ΔU are calculated, and combined with the fault phase determined by the sequence component, the macroscopic steady-state feature vector F is formed. macro .

[0100] 2.2 Extraction of Microscopic Frequency Domain Energy Map F freq :

[0101] The transient high-frequency sequence after t0 is subjected to wavelet packet multi-level decomposition. After removing low frequencies affected by power flow and extremely high frequencies attenuated by line attenuation, only K sub-bands within the middle frequency range are retained (e.g., the 500Hz~2500Hz range is divided into 4 sub-bands, then K=4). The wavelet packet reconstruction energy of these K sub-bands is calculated and normalized by dividing by the total energy of these K bands to generate a frequency domain energy distribution vector. Its element calculation formula is:

[0102]

[0103] Among them: E k e represents the absolute value of the reconstructed energy of the k-th feature sub-band extracted after wavelet packet decomposition. k The energy percentage of the k-th frequency band ensures that the sum of all elements in the vector is always 1.

[0104] 2.3 Extracting the morphological trajectory envelope sequence F shape :

[0105] The Hilbert transform is used to construct an analytical signal from the transient current waveform after t0 to obtain the upper envelope. The upper envelope is then downsampled proportionally with a fixed step size, compressing the high-frequency sampling points into a discrete feature point sequence F of length M. shape ={S1,S2,...,S M}. M is the total number of points in the dimensionality-reduced sequence (preferably 10~20), S m This represents the amplitude of the m-th envelope feature point after downsampling.

[0106] Step 3: Fingerprint interaction under low-bandwidth network:

[0107] Two adjacent power distribution terminals (defined as terminal A and terminal B) will send the cross-dimensional transient fingerprint data packets FA and FB extracted in step 2, which have a total data volume of only tens of bytes, to each other through the power distribution communication network (5G / 4G / wireless self-organizing network, etc.) or send them to the next-level edge computing master station.

[0108] Step 4: Macroscopic coarse screening

[0109] After receiving the fingerprint data from the other party, the system compares whether the fault characteristics of the two terminals are completely consistent, and compares whether the difference in the coarse timestamps sent locally by the two devices is within the tolerance range (e.g., ±2 seconds). If either condition is not met, the system rejects the data and directly determines it to be a non-homogeneous fault; if the coarse screening is passed, the system proceeds to in-depth transient collaborative discrimination.

[0110] Step 5: Transient Co-discrimination with Integrating Elastic Regularization Constraints

[0111] This step consists of two parallel mathematical operation branches, which calculate the similarity in the frequency domain and morphology respectively, and finally make a comprehensive decision:

[0112] 5.1: Microscopic Frequency Domain Spatial Cosine Mapping

[0113] Calculate the cosine similarity Sim in multidimensional space of the frequency domain energy distribution vectors extracted from adjacent distribution terminals A and B. freq Because this indicator compares the relative proportions of energy in each frequency band, it is not affected by the offset of the pseudo-synchronization anchor point.

[0114]

[0115] Where e A,k ,e B,k These are the values ​​of the k-th elements in the frequency domain energy distribution vectors extracted from terminal A and terminal B, respectively; Sim freqThis represents the frequency domain cosine similarity, with a value range of [0,1]. The closer the value is to 1, the more similar the high-frequency energy distributions of the faults at both ends are.

[0116] 5.2: Elastic morphology regularity with physical time constraints

[0117] Construct a dynamic programming grid matrix of size M×M, and apply the morphological sequences F at both ends. shapeA ={S A,1 ,S A,2 ,...,S A,M} and F shapeB ={S B,1 ,S B,2 ,...,S B,M Elastic stretching and compression matching are performed. The single-step node cost function of this invention is reconstructed as follows:

[0118]

[0119] Where i and j are the feature point sequence indices of terminal A (1≤i≤M) and terminal B (1≤j≤M), respectively. Physically, they represent the temporal positions of the feature points after downsampling. A,i S represents the amplitude of the i-th feature point in the morphological sequence of terminal A; B,j |S represents the amplitude of the j-th feature point in the terminal B morphological sequence; A,i -S B,j | represents the absolute difference cost between the two points in the vertical magnitude dimension; (ij) 2 This represents the squared offset of these two points on the horizontal time axis (index axis). If the difference between i and j is too large, it indicates that the algorithm attempts to forcibly match the waveform over an extremely long period of time; λ is the time offset penalty coefficient (weight parameter).

[0120] After calculating the single-step cost c(i,j) of each node in the matrix, the cumulative normalized distance matrix D(i,j) is calculated using a dynamic programming algorithm, and its state transition equation is as follows:

[0121]

[0122] Finally, the global minimum normalization distance between two feature sequences of length M is found to be D. warp =D(M,M).

[0123] Subsequently, the global normalized distance D is mapped using a negative exponential normalization mapping function. warp Converted into morphological similarity score Sim shape :

[0124]

[0125] Where κ is the smoothing attenuation constant (which can be tuned according to the downsampling amplitude normalization reference), used to map the distance value to the (0,1] interval; Sim shape This is the morphological similarity score for the final output. If the two sequences are completely identical and have no time shift, D... warp =0, Sim shape It reached its maximum value of 1.

[0126] In this invention, the value of λ is strictly constrained by the physical error boundary of the power grid and must satisfy the inequality:

[0127]

[0128] Among them, W max Convert the maximum asynchronous misalignment time of the hardware (e.g., 2ms) to the maximum allowable index deviation value (e.g., 4 points) in the downsampling sequence. C amp_max γ is the per-unit value representing the maximum possible amplitude difference of the same-origin transient waveform under normal transformer transmission error. γ is a blocking constant greater than 1 (e.g., γ=2.5).

[0129] When the difference between the horizontal and vertical coordinates of the matching path |ij| exceeds the maximum physical deviation W max At that time, the cost generated by the quadratic penalty term will be directly greater than the normal amplitude difference limit γC. amp_max This forces the blocking of the matching path. Excessive stretching matching that deviates from the transient physical evolution of the power grid is strictly prohibited. The global cumulative minimum regularization distance is obtained through dynamic programming and then mapped back to the morphological similarity score Sim. shape .

[0130] 5.3: Weighted Decision

[0131] Constructing a weighted fusion decision model:

[0132]

[0133] Score is the overall confidence score (between 0 and 1.0); α and β are empirical weighting coefficients adjusted according to the field channel and transformer type, and α+β=1 is guaranteed.

[0134] When the score is greater than the set same-source discrimination threshold (e.g., 0.85), it is finally determined that the fault detected by both terminals is of the same source within the same line segment; otherwise, it is determined to be a non-same-source disturbance.

[0135] Example 2:

[0136] This embodiment provides a power distribution terminal fault source identification system using the method described in Embodiment 1, including:

[0137] The data acquisition module is used to acquire electrical quantity sequences in real time.

[0138] The pseudo-synchronization anchor point generation module is used to establish the pseudo-synchronization time anchor point t0 of an event based on a nonlinear mutation detection algorithm.

[0139] The fingerprint extraction module is used to extract a cross-dimensional transient fingerprint vector F, which includes a macroscopic steady-state feature vector F. macro Frequency domain energy distribution vector F freq and morphological trajectory envelope sequence F shape ;

[0140] The communication module is used to exchange the cross-dimensional transient fingerprint vector with adjacent power distribution terminals;

[0141] The coarse screening module is used to compare the macroscopic steady-state eigenvector F. macro and a local rough timestamp;

[0142] The collaborative discrimination module is used to calculate the frequency domain cosine similarity and the elastic regularized matching distance with a time offset penalty factor, and then weighted and fused to output the comprehensive confidence score.

[0143] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0144] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0145] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0146] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0147] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for elastic regularization and collaborative discrimination of cross-dimensional transient fingerprints based on event pseudo-synchronization, characterized in that, Includes the following steps: S1: Real-time acquisition of electrical quantity sequences of power distribution terminals, based on nonlinear mutation detection algorithm, establishing the mutation moment as the event pseudo-synchronization time anchor point t0, and using this as a benchmark to extract transient data windows; S2: Perform multi-scale dimensionality reduction on the waveform within the transient data window to extract and generate a cross-dimensional transient fingerprint vector F. The cross-dimensional transient fingerprint vector F includes: a macroscopic steady-state feature vector F. macro Frequency domain energy distribution vector F freq and the morphological trajectory envelope sequence F shape ; S3: Exchange the cross-dimensional transient fingerprint vector F extracted from adjacent power distribution terminals through the power distribution communication network; S4: Perform macroscopic coarse screening, comparing the macroscopic steady-state eigenvectors F at both ends. macro And the local rough timestamp, if the consistency condition is met, proceed to step S5; otherwise, it is determined to be a non-same-source fault. S5: Perform transient collaborative discrimination and calculate the frequency domain energy distribution vectors F at both ends. freq Spatial cosine similarity; simultaneously, an elastic warping algorithm with a time offset penalty factor is used to calculate the envelope sequence F of the morphological trajectories at both ends. shape The matching distance is used to calculate the comprehensive confidence score by weighting and fusing the two indicators. When the comprehensive confidence score is greater than the preset threshold, it is determined that the two ends have detected the same source fault.

2. The cross-dimensional transient fingerprint elastic regularization and collaborative discrimination method based on event pseudo-synchronization according to claim 1, characterized in that, The electrical quantity sequence of the power distribution terminal is acquired in real time. Based on a nonlinear mutation detection algorithm, the mutation moment is established as the pseudo-synchronization time anchor point t0 of the event, and a transient data window is extracted based on this as a reference, including: Real-time acquisition of electrical quantity sequences is performed, and the discrete Teager energy operator is used to perform nonlinear optimization on the sampled sequence x(n) to amplify high-frequency abrupt changes. The calculation formula is as follows: ； Where: x(n) represents the instantaneous amplitude of the electrical quantity collected at the current nth discrete sampling point; x(n-1), x(n+1): represent the instantaneous amplitude of the electrical quantity at the previous sampling point and the next sampling point, respectively; E(n) represents the instantaneous change energy value of the current nth sampling point calculated by the TEO operator; Maintain a sliding data window of fixed length L in front of the current sampling point, and calculate the average background energy reference value within this window: ； Where: L represents the length of the sliding data window; k is the temporal traversal index for the summation operation; E base (n): represents the average absolute value of the background energy reference within a historical window of length L before the current time n, used to characterize the current steady-state noise level of the power grid; Real-time generation of dynamic trigger thresholds that adapt to and float with background noise: ； Among them: E th (n) represents the adaptive trigger threshold corresponding to the current nth sampling point; K m The set sensitivity factor is used to filter out normal load switching energy fluctuations; E min To prevent zero drift, a minimum threshold constant is required; When consecutive N c The instantaneous energy E(n) at each sampling point is greater than the corresponding dynamic trigger threshold E. th When (n), the first point exceeding the limit is formally established as the pseudo-synchronization time anchor point t0 of the event; N c To confirm the counter value for image stabilization; The transient data window is reconstructed using t0 as the origin of the time coordinate system.

3. The cross-dimensional transient fingerprint elastic regularization and collaborative discrimination method based on event pseudo-synchronization according to claim 2, characterized in that, In step S2, the macroscopic steady-state eigenvector F macro The extraction methods include: Based on t0, select Select the steady-state data window before the fault. This is the data window after the fault, where T cycle It is the power frequency cycle; The fundamental amplitude values ​​within the two data windows are extracted using the full-wave Fourier algorithm. The steady-state current RMS value mutation ΔI and voltage RMS value mutation ΔU are calculated, and combined with the fault phase determined by the sequence component, the macroscopic steady-state feature vector F is formed. macro .

4. The cross-dimensional transient fingerprint elastic regularization and collaborative discrimination method based on event pseudo-synchronization according to claim 2, characterized in that, In step S2, the frequency domain energy distribution vector F freq The extraction methods include: The transient high-frequency sequence after t0 is decomposed into multiple layers using wavelet packet transform. The lowest frequency band containing the fundamental frequency and low harmonics, as well as the highest frequency band affected by the attenuation of the line distributed capacitance, are discarded, and only K sub-frequency bands within the middle characteristic frequency band range are extracted. Calculate the wavelet packet reconstruction energy of the K sub-bands, divide by the total energy of these K bands for normalization, and generate the frequency domain energy distribution vector. Its element calculation formula is: ； Among them: E k e represents the absolute value of the reconstructed energy of the k-th feature sub-band extracted after wavelet packet decomposition. k Let K be the energy percentage of the k-th frequency band, where K is the total number of sub-bands.

5. The cross-dimensional transient fingerprint elastic regularization and collaborative discrimination method based on event pseudo-synchronization according to claim 4, characterized in that, In step S2, the morphological trajectory envelope sequence F shape The extraction methods include: The Hilbert transform is used to construct an analytical signal from the transient current waveform after t0 to obtain the upper envelope. The upper envelope is then downsampled proportionally with a fixed step size to generate a discrete feature point sequence F of length M. shape ={S1,S2,...,S M }; M is the total number of points in the dimensionality-reduced sequence, S m This represents the amplitude of the m-th envelope feature point after downsampling.

6. The cross-dimensional transient fingerprint elastic regularization and collaborative discrimination method based on event pseudo-synchronization according to claim 5, characterized in that, In step S4, a macroscopic coarse screening is performed, comparing the macroscopic steady-state feature vectors F at both ends. macro And a local coarse timestamp. If the consistency condition is met, proceed to step S5; otherwise, it is determined to be a non-homogeneous fault, including: Compare the macroscopic steady-state eigenvectors F at both ends macro Whether they are completely consistent, and compare whether the difference between the rough timestamps issued locally by the two devices is within the tolerance range; If any condition is not met, the result is a veto and the fault is directly determined to be non-homogeneous; if the coarse screening is passed, proceed to step S5.

7. The cross-dimensional transient fingerprint elastic regularization and collaborative discrimination method based on event pseudo-synchronization according to claim 6, characterized in that, In step S5, the frequency domain energy distribution vector F at both ends is calculated. freq Spatial cosine similarity includes: Calculate the cosine similarity Sim in multidimensional space of the frequency domain energy distribution vectors extracted from adjacent distribution terminals A and B. freq : ； Where e A,k, e B,k These are the values ​​of the k-th elements in the frequency domain energy distribution vectors extracted from terminal A and terminal B, respectively; Sim freq The frequency domain cosine similarity has a value range of [0,1]. The closer the value is to 1, the more similar the high-frequency energy distributions of the faults at both ends are. This represents the frequency domain energy distribution vector of terminal A. This represents the frequency domain energy distribution vector of terminal B.

8. The cross-dimensional transient fingerprint elastic regularization and collaborative discrimination method based on event pseudo-synchronization according to claim 7, characterized in that, In step S5, the elastic warping algorithm with a time offset penalty factor is used to calculate the envelope sequence F of the morphological trajectories at both ends. shape The matching distance includes: Construct a dynamic programming grid matrix of size M×M, and apply the morphological sequences F at both ends. shapeA ={S A,1 ,S A,2 ,...,S A,M } and F shapeB ={S B,1 ,S B,2 ,...,S B,M Perform elastic stretching and compression matching; The single-step node cost function of the elastic warping algorithm is reconstructed as follows: ； Where i and j are the feature point sequence indices of terminal A and terminal B, respectively, physically representing the temporal position of the feature point after downsampling; S A,i S represents the amplitude of the i-th feature point in the morphological sequence of terminal A; B,j This represents the magnitude of the j-th feature point in the terminal B morphology sequence; |S A,i -S B,j | represents the absolute difference cost between the two points in the vertical magnitude dimension; (ij) 2 This represents the squared offset of these two points on the horizontal time axis; λ is the time offset penalty coefficient. After calculating the single-step cost c(i,j) of each node in the matrix, the cumulative normalized distance matrix D(i,j) is calculated using a dynamic programming algorithm, and its state transition equation is as follows: ； The global minimum normalization distance between two feature sequences of length M is found to be D. warp =D(M,M); By using a negative exponential normalized mapping function, the global normalized distance D is... warp Converted into morphological similarity score Sim shape : ； Where κ is the smoothing decay constant, used to map the distance value to the (0,1] interval; Sim shape The final output is the morphological similarity score; if the two sequences are completely identical and have no time shift, D... warp =0, Sim shape It reached its maximum value of 1.

9. The cross-dimensional transient fingerprint elastic regularization and collaborative discrimination method based on event pseudo-synchronization according to claim 8, characterized in that, The value of the time offset penalty coefficient λ is strictly constrained by the physical error boundary of the power grid, satisfying the inequality: ； Among them, W max This represents the maximum allowable index deviation value converted from the maximum asynchronous misalignment time of the hardware to the downsampled sequence; C amp_max γ is the per-unit value of the maximum possible amplitude difference of the same transient waveform under normal transformer transmission error; γ is the blocking constant greater than 1. When the difference between the horizontal and vertical coordinates of the matching path |ij| exceeds the maximum physical deviation W max At that time, the cost generated by the quadratic penalty term will be directly greater than the normal amplitude difference limit γC. amp_max This forces the blocking of the matching path, thereby blocking the excessive stretching matching that violates the transient physical evolution law of the power grid.

10. The cross-dimensional transient fingerprint elastic regularization and collaborative discrimination method based on event pseudo-synchronization according to claim 9, characterized in that, In step S5, the two indicators are weighted and fused to calculate the overall confidence level, including: The formula for calculating the overall confidence level using weighted fusion is: ； Wherein, Score is the overall confidence score; α and β are empirical weighting coefficients adjusted according to the field channel and transformer type, and α+β=1 is guaranteed.

Images