Small current identification method and device

Abstract

The invention discloses a small current identification method and device, and the method comprises the steps: S1, obtaining the generation mechanisms and waveform features of different power distribution network disturbance events, and constructing a noise model of a power distribution network according to the generation mechanisms and the waveform features; s2, carrying out data analysis on the noise model by adopting compressed sensing to obtain monitoring data added with noise; s3, acquiring current distribution information when the power distribution network has a single-phase earth fault according to the monitoring data, and analyzing the single-phase earth fault according to the current distribution information to obtain a current identification result; and S4, inputting the current identification result into a trained KNN traceability model for training to obtain a power distribution network disturbance event traceability result corresponding to the current identification result. According to the method, the power distribution network noise model is constructed, and the compressed sensing technology is combined, so that high-precision low-current identification of the single-phase earth fault of the power distribution network is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network detection, and particularly relates to a small current identification method and device. Background Art

[0002] Since the distribution network is connected to power users, with the improvement of the electrification level on the terminal consumption side and the access of distributed energy to the grid, the load characteristics are more diverse. Multiple factors such as the impact of access equipment and the influence of the external environment will cause disturbances to the grid, reduce the service life of equipment, affect the normal power consumption of users, and may cause huge economic losses to users in severe cases. However, in the distribution network, the existing traditional methods in small current grounding line selection rely on manual feature extraction and are only applicable to specific distribution network structures. Transplanting them to other distribution network structures results in a low line selection accuracy. While using deep learning algorithms requires a large amount of training data. If the trained deep learning line selection model is directly used in other distribution networks, the line selection accuracy will be severely reduced. Currently, the common solution is to regenerate a large amount of data for the training of the new model, which is time-consuming, laborious and cannot meet the actual application requirements. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a small current identification method and device to improve the efficiency and accuracy of small current identification.

[0004] To solve the above technical problem, the present invention provides a small current identification method, including:

[0005] Step S1, obtaining the occurrence mechanism and waveform characteristics of different distribution network disturbance events, and constructing a noise model of the distribution network according to the occurrence mechanism and the waveform characteristics;

[0006] Step S2, using compressive sensing to perform data analysis on the noise model to obtain the monitored data with added noise;

[0007] Step S3, obtaining the current distribution information when a single-phase grounding fault occurs in the distribution network according to the monitored data, and analyzing the single-phase grounding fault according to the current distribution information to obtain a current identification result;

[0008] Step S4, inputting the current identification result into a trained KNN traceability model for training to obtain a traceability result of the distribution network disturbance event corresponding to the current identification result.

[0009] Preferably, the obtaining of the occurrence mechanism and waveform characteristics of different distribution network disturbance events in step S1 specifically includes:

[0010] Step S10, sampling the waveform data in the actual operation process by using the Nyquist theorem;

[0011] Step S11: Compress the data using the corresponding algorithm and transform it to another mapping domain;

[0012] Step S12: Reconstruct the original data through the reconstruction algorithm, and compare the reconstructed data with the original data to obtain the compression performance;

[0013] Step S13: Extract features from the data in the compression domain corresponding to the compression performance to characterize the features of the disturbance event.

[0014] Preferably, in step S2, compressive sensing is used to perform data analysis on the noise model, specifically including:

[0015] Perform compressive sensing on a signal X of length N to obtain a compressed observed value Y, where is the measurement matrix, and each row in the matrix forms a partial information of the signal by multiplying with the expansion coefficients to form a linear observation vector of length M;

[0016] Represent the signal X as the product of a sparse dictionary matrix ψ and a sparse signal S, where ψ is an N×N orthonormal basis matrix, and S is an N×1 sparse signal column vector, and S is K-sparse, K << N;

[0017] Through the sensing matrix A cs Represent the observed value Y as Y = A cs S.

[0018] Preferably, step S2 further includes:

[0019] Use the l1-norm reconstruction algorithm to transform the reconstruction problem into a convex optimization problem, and solve the following optimization problem to obtain the sparse signal S:

[0020]

[0021] Use the orthogonal matching pursuit OMP algorithm to perform sparse decomposition on the observed value Y obtained by compressive sensing. Specifically:

[0022] Take each column in the sensing matrix A cs and take the inner product with the residual R t t represents the number of iterations, and the initial value of the residual is r1. Then, in the t-th iteration process, find a set of column vectors with the maximum inner product through the calculation results and preset the column vector as η t , numbered λ, satisfying the following conditions:

[0023] η t = arg max t＝1,2,...,T |<r t , ξ λt >|

[0024] Among them, ξ λt represents the λ-th column of the sensing matrix A cs at the t-th iteration, and r t represents the residual value obtained at the t-th iteration;

[0025] After completing the t-th iteration, the number Γ of the column vector with the best match is obtained t , and the set Γ of the numbers of the best-matching class vectors is updated through multiple iterations t = Γ t- ∪{η t}, and the atom set Among them, the λ-th column ξ CS of the sensor matrix A λt is set to zero;

[0026] The approximation value of the sparse signal S at the λ-th number at the t-th iteration is calculated by the least squares method as follows:

[0027]

[0028] Update the residual:

[0029]

[0030] Iteratively update the residual value successively until the number of iterations is greater than the sparsity K of the sparse signal, and the sparse signal

[0031] is obtained. The original signal is reconstructed through the inverse transform of the sparse basis.

[0032] Preferably, the step S3 specifically includes:

[0033] According to the monitoring data, obtain the current distribution information when a single-phase grounding fault occurs in the distribution network. Among them, the current distribution information includes the fault-phase current, the non-fault-phase current, and the arc suppression coil current;

[0034] Based on the current distribution information, calculate the neutral point-to-ground voltage U0 and the three-phase-to-ground voltages U AK , U BK and U CK vector trajectories varying with the ground fault transition resistance R f ;

[0035] According to the vector trajectories, determine the characteristics of the fault-phase voltage drop and the non-fault-phase voltage rise, and combine the compensation method of the arc suppression coil to diagnose the single-phase grounding fault and generate a current recognition result;

[0036] Among them, the compensation methods of the arc suppression coil include full compensation, under-compensation, and over-compensation, and the harmonic components are used as the fault line selection characteristic quantities in the over-compensation method.

[0037] Preferably, the calculation method of the neutral point voltage U0 to the ground is as follows:

[0038]

[0039] Wherein, E A is the power supply electromotive force of the faulty phase, C ∑ is the sum of the capacitances to the ground of the resonant grounding system, ω is the angular frequency, and L is the equivalent inductance of the arc suppression coil;

[0040] According to the change of the transition resistance R f at the grounding point, the neutral point voltage U0 to the ground and the three-phase voltages U AK , U BK and U CK changing with R f are obtained, and the locus of the neutral point voltage U0 to the ground is a semi-circular arc o-k-a with E A as the diameter.

[0041] Preferably, the step S4 specifically includes:

[0042] Input the current recognition result into the trained KNN traceability model, and calculate the distance between the current recognition result and the known category samples in the sample set. Among them, the sample set contains s categories (G1, G2,..., G s ), and the number of samples of each category is N j (j = 1, 2,..., s);

[0043] Based on the distance metric formula, select the k samples with the closest distance to the current recognition result from the sample set, and count the number of samples of each category in the k samples;

[0044] According to the statistical result, determine the current recognition result as the category with the most occurrences. The category discriminant function H u (x i ) satisfies:

[0045] H u (x i ) = max(N u ), u = (1, 2,..., s)

[0046] Wherein, x i represents the value of the i-th sampling point in the unknown sample X, and N u represents the number of samples of the u-th category.

[0047] Preferably, the distance metric formula includes Manhattan distance, Euclidean distance, and cosine distance, where:

[0048] The calculation method of Manhattan distance Md(X,G) is as follows:

[0049]

[0050] The calculation method of Euclidean distance Ed(X,G) is as follows:

[0051]

[0052] Cosine distance cosθ (X,G) The calculation method is as follows:

[0053]

[0054] where i = 1, 2,..., m, x i and g i are the values at the i-th position of the unknown sample X and G respectively.

[0055] Preferably, the training process of the KNN traceability model includes:

[0056] Extract sample data from historical distribution network disturbance events and construct a sample set (G1, G2,..., G s ) containing s categories;

[0057] Preprocess the sample data, including normalization, denoising, and feature extraction;

[0058] Based on the preprocessed sample data, use the cross-validation method to determine the optimal distance metric formula and parameter k.

[0059] The present invention also provides a small current identification device, including:

[0060] A noise model construction module, configured to obtain the occurrence mechanism and waveform characteristics of different distribution network disturbance events, and construct a noise model of the distribution network according to the occurrence mechanism and the waveform characteristics;

[0061] A monitoring data acquisition module, configured to perform data analysis on the noise model by using compressive sensing to obtain the monitored data after adding noise;

[0062] A current identification module, configured to obtain the current distribution information when a single-phase grounding fault occurs in the distribution network according to the monitored data, and analyze the single-phase grounding fault according to the current distribution information to obtain a current identification result;

[0063] A traceability model training module, configured to input the current identification result into the trained KNN traceability model for training to obtain the traceability result of the distribution network disturbance event corresponding to the current identification result.

[0064] Implementing the present invention has the following beneficial effects: Based on the occurrence mechanisms and waveform characteristics of different disturbance events, the present invention constructs a noise model and uses compressive sensing technology to extract sparsely represented data, effectively reducing noise interference and improving the accuracy of monitoring data. By analyzing the current distribution information during a single-phase grounding fault and combining the compensation method of the arc suppression coil and the neutral point voltage change characteristics, accurate fault diagnosis is achieved. In addition, the present invention inputs the current recognition result into the trained KNN traceability model, uses various distance measurement methods to trace the disturbance events, and further improves the accuracy and robustness of fault recognition. By generating diverse waveform data and mapping it to different sparse domains, the present invention optimizes the performance of the sparsely represented data, enhances the adaptability and generalization ability of model training, and provides reliable technical support for distribution network fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0066] Figure 1 is a schematic flowchart of a small current recognition method according to Embodiment 1 of the present invention.

[0067] Figure 2 is a schematic flowchart of the specific process for obtaining the occurrence mechanisms and waveform characteristics of different distribution network disturbance events in the embodiments of the present invention.

[0068] Figure 3 is the current distribution diagram of the resonant grounding system in the embodiments of the present invention.

[0069] Figure 4 is a schematic diagram of the neutral point voltage offset trajectory of the resonant grounding system in the embodiments of the present invention.

[0070] Figure 5 is a schematic structural diagram of a small current recognition device according to Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0071] The following descriptions of the embodiments are with reference to the drawings to illustrate specific embodiments in which the present invention can be implemented.

[0072] Please refer to Figure 1 as shown, Embodiment 1 of the present invention provides a small current recognition method, including:

[0073] Step S1: Obtain the occurrence mechanisms and waveform characteristics of different distribution network disturbance events, and construct a noise model of the distribution network according to the occurrence mechanisms and the waveform characteristics;

[0074] Step S2: Use compressive sensing to perform data analysis on the noise model to obtain the monitored data with added noise;

[0075] Step S3: Obtain the current distribution information when a single-phase grounding fault occurs in the distribution network according to the monitored data, and analyze the single-phase grounding fault according to the current distribution information to obtain a current identification result;

[0076] Step S4: Input the current identification result into the trained KNN traceability model to obtain the traceability result of the distribution network disturbance event corresponding to the current identification result.

[0077] As can be seen from the above steps, in the embodiment of the present invention, through the analysis of the characteristics and occurrence mechanisms of different distribution network disturbance events, different distribution network disturbance event modules are built in the software to generate disturbance data. In order to simulate the diverse characteristics of disturbance data in the actual process, different disturbance data are obtained by controlling the duration, position, and module parameters of the disturbance data for subsequent analysis; since the obtained data is not sparse data and does not meet the conditions of compressive sensing, a suitable sparse basis needs to be selected to map the original disturbance data to the sparse domain to obtain a sparse representation signal; important parameters of the compressive sensing process are selected, including the measurement matrix and the compression ratio, and the reconstructed sparse representation signal is obtained through an optimization algorithm. Based on the obtained sparse representation signal, the original signal after reconstruction can be further obtained; performance evaluation indicators are proposed to evaluate the performance of the proposed compressive sensing of distribution network disturbance data. If the performance indicators of the reconstructed data meet the requirements, the compressive sensing sampling data is obtained as the analysis data for the subsequent distribution network disturbance traceability model. Otherwise, the compression ratio is continuously adjusted until the performance indicators of the reconstructed data meet the requirements.

[0078] It should be noted that by obtaining the occurrence mechanisms and waveform characteristics of different distribution network disturbance events, constructing a noise model of the distribution network according to the occurrence mechanisms and the waveform characteristics, using compressive sensing to perform data analysis on the noise model to obtain the monitored data with added noise, obtaining the current distribution information when a single-phase grounding fault occurs in the distribution network according to the monitored data, analyzing the single-phase grounding fault according to the current distribution information to obtain a current identification result, and inputting the current identification result into the trained KNN traceability model for training to obtain the traceability result of the distribution network disturbance event corresponding to the current identification result, the accuracy of small current identification of distribution network noise is improved. By obtaining diverse waveform data and mapping the original waveform data to different sparse domains, more optimal sparse representation data is determined, and the robustness of model training is improved.

[0079] Please refer to again Figure 2 As shown, the specific steps of obtaining the occurrence mechanism and waveform characteristics of different distribution network disturbance events in step S1 include:

[0080] Step S10, sampling the waveform data during the actual operation using the Nyquist theorem;

[0081] Step S11, compressing the data using the corresponding algorithm and transforming it to another mapping domain;

[0082] Step S12, reconstructing the original data through the reconstruction algorithm, comparing the reconstructed data with the original data to obtain the compression performance;

[0083] Step S13, extracting features based on the compressed domain data corresponding to the compression performance to characterize the features of the disturbance event.

[0084] In step S2 of the embodiment of the present invention, compressed sensing is used to perform data analysis on the noise model to obtain the monitored data after adding noise, which specifically includes:

[0085] If the signal X meets the sparsity requirement, compressed sensing is performed on the signal X with length N to obtain the compressed observed value:

[0086]

[0087] where Y is the observation vector, is the measurement matrix, and each row in the matrix can be visualized as a sensor; each row in the matrix multiplies with the expansion coefficient to obtain partial information of the signal, forming a linear observation vector with length M;

[0088] If there is an N-dimensional real-valued signal X ∈ R N×1 represented by the linear combination of a certain N×1-dimensional orthonormal basis vector and taking as column vectors, using the N×N sparse dictionary matrix ψ = [ψ1, ψ2,... ψ N to represent, satisfying ψψ T = ψ T ψ = 1, then the signal X can be represented as:

[0089] or X = ψS(2)

[0090] where S is the N×1 column vector formed by the expanded sparse signal S = [s1, s2,..., s N T The preset sparse signal S is K-term sparse, that is, the number K of non-zero coefficients of S << N;

[0091] ​The transform domain that converts a non-sparse signal into a sparse signal becomes a sparse basis. The data of the sparse basis satisfies the premise of the compressive sensing process, and the corresponding expression is:

[0092]

[0093] Among them, A cs is the sensing matrix.

[0094] It should be noted that the embodiment of the present invention uses the l1-norm reconstruction algorithm to transform the reconstruction problem into a convex optimization problem to simplify the solution of the linear programming problem. Solving the simple l1-norm optimization problem produces the same solution:

[0095]

[0096] The orthogonal matching pursuit (OMP) algorithm is used to transform the signal reconstruction process into a sparse decomposition process of the compressed data obtained through compressive sensing, and orthogonalization processing is performed on the selected atoms during the sparse decomposition process. The execution process of the OMP algorithm includes:

[0097] Take the inner product of each column in the sensing matrix A cs with the residual R t . Let t represent the number of iterations, and the initial value of the residual is r1. Then, in the t-th iteration process, find a set of column vectors with the maximum inner product through the calculation results and preset the column vector as η t , with the number λ. The condition satisfied is:

[0098] η t = arg max t＝1,2,...,T | <r t , ξ λt > | (5)

[0099] Among them, ξ λt represents the λ-th column in the sensing matrix A cs in the t-th iteration, and r t represents the residual value obtained in the t-th iteration;

[0100] After completing the t-th iteration, obtain the number Γ t of the column vector with the best match. Update the set Γ t of the numbers of the best-matching class vectors through multiple iterations t = Γ t - ∪ {η t}, and obtain the atom set Among them, the λ-th column ξ CS of the sensor matrix A λt is set to zero;

[0101] Obtain the approximation value of the sparse signal S at the λ-th number in the t-th iteration through the least squares method according to formula (6):

[0102]

[0103] Updated residual:

[0104]

[0105] Iteratively update the residual value. If the number of iterations satisfies being greater than the sparsity of the sparse signal, proceed to the next step; otherwise, perform the first step where formula (5) is located.

[0106] Obtain the sparse signal from the solution result where K represents the sparsity of the sparse signal;

[0107] Reconstruct the original signal through the inverse transform of the sparse basis. By changing different perturbation parameters, diverse waveform data can be obtained. Map the original waveform data to different sparse domains to determine the sparse representation data with better performance.

[0108] Please also refer to Figure 3 and Figure 4 As shown, step S3 for obtaining the current distribution information when a single-phase grounding fault occurs in the distribution network based on the monitoring data specifically includes:

[0109] Presuppose that a single-phase grounding fault occurs in phase A of the re-feeder L4 of the resonant grounding system, and the magnitudes and directions of the ground capacitances of all feeders are the same. The current flowing out from the power source for each phase is:

[0110]

[0111] where, R f is the transition resistance at the grounding point, U0 is the voltage of the neutral point to the ground, and C ∑ is the sum of the ground capacitances of the resonant grounding system;

[0112] Suppose the equivalent inductance of the arc suppression coil is L, then the current i L of the arc suppression coil is:

[0113]

[0114] The secondary side of the low-current grounding transformer is star-connected. When the arc suppression coil is connected to the resonant grounding system, according to Kirchhoff's node law:

[0115] i A + i B + i C + i L = 0 (10)

[0116] Substitute formulas (8) and (9) into formula (10) to obtain:

[0117]

[0118]

[0119] The neutral point-to-earth voltage is:

[0120]

[0121] In this embodiment, the line model corresponding to the resonant grounding system is preset as a four-outlet line L1, L2, L3, and L4, and the three-phase power source electromotive forces are E A , E B , and E C , the capacitance to earth of each feeder in the resonant grounding system is C 01 , C 02 , C 03 , and C 04 , and the capacitance to earth of the substation equipment is C 0S ;

[0122] According to the change of the transition resistance R f of the grounding point, the neutral point-to-earth voltage U0 and the three-phase voltages to earth U AK , U BK , and U CK can be obtained. As the vector locus of R f changes, the locus of the neutral point-to-earth voltage U0 is a semi-circular arc o-k-a with E A as the diameter;

[0123] Under normal operating conditions, R f = ∞. At this time, point k is located at point o, the neutral point-to-earth voltage U0 = 0 without deviation, and the three-phase voltages to earth are equal to the phase voltages during normal operation:

[0124]

[0125] When R f ≈ 0, at this time point k is located at point a, the neutral point-to-earth voltage U0 rises from 0 to the phase voltage during normal operation, U0 = -E A , the voltage to earth of the non-fault phase rises to the line voltage of the resonant grounding system, the voltage of the fault phase drops to 0, U A = 0;

[0126] When 0 < R f < ∞, at this time the movement locus of point k moves on the semi-circular arc o-k-a with E A as the diameter;

[0127] When the transition resistance R f between the fault phase A and the grounding point is not zero, the voltage of phase A drops and the voltage of phase B rises;

[0128] When the k point is below the extension line of b - o, the voltage of phase C decreases, and the voltage of phase A is higher than that of phase C;

[0129] When the k point is above the extension line of b - o, the voltage of phase C increases, and the voltage of phase A is lower than that of phase C;

[0130] When an inductive current i is added to the grounding point after using an arc suppression coil for grounding L , the total current flowing back from the grounding point at this time is:

[0131]

[0132] Among them, is the capacitive current of the resonant grounding system to the ground, and i L is the current of the arc suppression coil.

[0133] It should be noted that according to the different degrees of compensation for the capacitive current, the arc suppression coil has three compensation methods: full compensation, under - compensation, and over - compensation:

[0134] Full compensation: The grounding point current i D ≈0. When fully compensated, ωL = 1 / 3ωC ∑ , the inductor and the three - phase capacitances to the ground are in series resonance, and the series resonance raises the neutral point voltage to the ground;

[0135] Under - compensation: When the operating mode of the resonant grounding system changes, the capacitive current decreases, and at this time it will change to

[0136] Over - compensation: The grounding point current after compensation is inductive. The degree to which the arc suppression coil current is greater than the capacitive current of the resonant grounding system to the ground is represented by the over - compensation degree P. Selecting the over - compensation degree P = 5% - 10%, then the relationship is:

[0137]

[0138] Among them, when a single - phase grounding fault occurs in the resonant grounding system, in the over - compensation mode, it is impossible to use the fundamental component characteristics of the zero - sequence current for fault line selection, and only the harmonic components can be used as the fault line selection feature quantity, thereby improving the small - current recognition efficiency.

[0139] Optionally, the current recognition result is input into the trained KNN traceability model for training to obtain the traceability result of the distribution network disturbance event corresponding to the current recognition result, including:

[0140] In a sample set containing s categories (G1, G2,..., G s ), each category of sample has N jFor each \(j = 1,2,\cdots,s\), the class discrimination function in the KNN algorithm is \(H\), and formula (17) represents the total number of samples where \((G_1,G_2,\cdots,G s ) belong to different classes:

[0141] H u (G u ) = N u , \(u=(1,2,\cdots,s)\) (17)

[0142] When an unknown-class sample \(X\in R 1×m is input into the feature space, by calculating the distance between the sample and the known samples \(g\) in the sample set, and counting the number of the most occurrences of sample classes among the \(k\) nearest distances obtained, the unknown-class sample is judged as the type with the most occurrences:

[0143] H u (x i ) = max(N u ), \(u=(1,2,\cdots,s)\) (18)

[0144] where \(x i represents the value of the \(i\)-th sampling point of the unknown sample \(X\).

[0145] In this embodiment, by selecting an appropriate distance metric formula to calculate the distance between different samples and based on the given \(k\) value to implement the class decision of the unknown sample, Manhattan distance, Euclidean distance, and cosine distance are adopted, and the corresponding calculation formulas are respectively:

[0146]

[0147]

[0148]

[0149] where \(i = 1,2,\cdots,m\), \(x i and \(g i are respectively the values of the \(i\)-th position of the samples \(X\) and \(G\). The smaller the Manhattan distance \(Md(X,G)\) and the Euclidean distance \(Ed(X,G)\) are, the higher the similarity between the two samples is. The closer the cosine distance \(\cos\theta (X,G) is to 1, the more the directions of the two vectors are the same, that is, the higher the sample similarity is.

[0150] Please refer to Figure 5 shown. Corresponding to the small current identification method described in Embodiment 1 of the present invention, Embodiment 2 of the present invention also provides a small current identification device, including:

[0151] A noise model construction module, configured to obtain the occurrence mechanisms and waveform characteristics of different distribution network disturbance events, and construct a noise model of the distribution network according to the occurrence mechanisms and the waveform characteristics;

[0152] A monitoring data acquisition module, configured to perform data analysis on the noise model by using compressive sensing to obtain the monitored data with added noise;

[0153] A current identification module, configured to obtain the current distribution information when a single-phase grounding fault occurs in the distribution network according to the monitored data, and analyze the single-phase grounding fault according to the current distribution information to obtain a current identification result;

[0154] A traceability model training module, configured to input the current identification result into a trained KNN traceability model to obtain a traceability result of the distribution network disturbance event corresponding to the current identification result.

[0155] Corresponding to the small current identification method according to the first embodiment of the present invention, the third embodiment of the present invention further provides a small current identification device, including:

[0156] One or more processors;

[0157] A memory;

[0158] One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the one or more processors, and the one or more applications are configured to execute the small current identification method.

[0159] Corresponding to the small current identification method according to the first embodiment of the foregoing present invention, the fourth embodiment of the present invention further provides a computer program product, including computer instructions, and the computer instructions instruct a computer device to execute the operations corresponding to the small current identification method according to the first embodiment of the foregoing present invention.

[0160] Preferably, the processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may also be any conventional processor. The processor is the control center of the device and connects various parts of the device by using various interfaces and lines.

[0161] The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function, etc., and the data storage area can store relevant data, etc. In addition, the memory can be a high-speed random access memory, or a non-volatile memory, such as a plug-in hard disk, a SmartMedia Card (SMC), a Secure Digital (SD) card, a Flash Card, etc., or the memory can also be other volatile solid-state storage devices.

[0162] It should be noted that the above device may include but is not limited to a processor and a memory, which can be understood by those skilled in the art.

[0163] Regarding the working principle and process of this embodiment, please refer to the description of Embodiment 1 of the present invention above, and details will not be repeated here.

[0164] Compared with the prior art, the beneficial effects brought by the embodiments of the present invention are as follows: Based on the occurrence mechanisms and waveform characteristics of different disturbance events, the present invention constructs a noise model and uses compressive sensing technology to extract sparse representation data, effectively reducing noise interference and improving the accuracy of monitoring data. By analyzing the current distribution information during a single-phase grounding fault and combining the compensation method of the arc suppression coil and the change characteristics of the neutral point voltage, accurate fault diagnosis is achieved. In addition, the present invention inputs the current recognition result into a trained KNN traceability model and uses a variety of distance measurement methods to trace the disturbance event, further improving the accuracy and robustness of fault recognition. By generating diverse waveform data and mapping it to different sparse domains, the present invention optimizes the performance of the sparse representation data, enhances the adaptability and generalization ability of model training, and provides reliable technical support for distribution network fault diagnosis.

[0165] The above-disclosed are only the preferred embodiments of the present invention. Of course, the scope of the rights of the present invention cannot be limited by this. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.

Claims

1. A small current identification method, characterized in that, Including: Step S1: Obtain the occurrence mechanisms and waveform characteristics of different distribution network disturbance events, and construct a noise model of the distribution network according to the occurrence mechanisms and the waveform characteristics; Step S2: Use compressive sensing to perform data analysis on the noise model to obtain monitored data with added noise; Step S3: Obtain the current distribution information when a single-phase grounding fault occurs in the distribution network according to the monitored data, and analyze the single-phase grounding fault according to the current distribution information to obtain a current identification result; Step S4: Input the current identification result into a trained KNN traceability model for training to obtain a traceability result of the distribution network disturbance event corresponding to the current identification result.

2. The method according to claim 1, wherein The specific process of step S1 for obtaining the occurrence mechanisms and waveform characteristics of different distribution network disturbance events includes: Step S10: Sample waveform data during actual operation using the Nyquist theorem; Step S11: Use a corresponding algorithm to compress the data and transform it to another mapping domain; Step S12: Reconstruct the original data through a reconstruction algorithm, and compare the reconstructed data with the original data to obtain the compression performance; Step S13: Extract features based on the data in the compression domain corresponding to the compression performance to characterize the features of the disturbance event.

3. The method according to claim 1, wherein In step S2, using compressive sensing to perform data analysis on the noise model specifically includes: Compressively sense the signal X with length N to obtain the compressed observed value Y, where, is the measurement matrix. Each row in the matrix multiplies with the expansion coefficients to obtain partial information of the signal, forming a linear observation vector with length M; Represent the signal X as the product of a sparse dictionary matrix ψ and a sparse signal S, where ψ is an N×N orthonormal basis matrix, S is an N×1 sparse signal column vector, and S is K-term sparse, with K << N; Through the sensing matrix A cs Express the observation value Y as Y = A cs S.

4. The method according to claim 3, characterized in that, Step S2 further includes: Use the l1-norm reconstruction algorithm to transform the reconstruction problem into a convex optimization problem, and solve the following optimization problem to obtain the sparse signal S: Use the orthogonal matching pursuit OMP algorithm to perform sparse decomposition on the observed value Y obtained by compressive sensing. Specifically: Multiply each column in the sensing matrix A cs with the residual R t to calculate the inner product. Let t denote the number of iterations, and the initial value of the residual be r1. Then, during the t-th iteration, find a set of column vectors with the maximum inner product from the calculation results and pre-set the column vectors as η t , numbered as λ, satisfying the following conditions: η t = argmax t＝1,2,...,T | <r t , ξ λt >| Among them, ξ λt represents the λ-th column of the sensing matrix A cs at the t-th iteration, and r t represents the residual value obtained at the t-th iteration; After the t-th iteration, the number Γ with the most matching column vectors is obtained t , and the set Γ of the numbers of the most matching class vectors is updated through multiple iterations t = Γ t- ∪{η t}, and the atomic set is obtained Among them, the λ-th column ξ CS of the sensor matrix A λt is set to zero; Calculate the approximation value of the sparse signal S at the t-th iteration with the number λ by the least squares method according to the following formula: Update the residual: Iteratively update the residual value until the number of iterations is greater than the sparsity K of the sparse signal to obtain the sparse signal Reconstruct the original signal through the inverse transformation of the sparse basis.

5. The method according to claim 1, wherein Step S3 specifically includes: According to the monitored data, obtain the current distribution information when a single-phase grounding fault occurs in the distribution network, where the current distribution information includes the fault-phase current, non-fault-phase current, and arc suppression coil current; Based on the current distribution information, calculate the neutral point voltage to ground U0 and the three-phase voltages to ground U AK , U BK and U CK along with the vector locus of the change of the transition resistance R f at the grounding point; According to the vector trajectory, determine the characteristics of the reduction of the fault-phase voltage and the increase of the non-fault-phase voltage, and combine the compensation method of the arc suppression coil to diagnose the single-phase grounding fault and generate a current identification result; Among them, the compensation methods of the arc suppression coil include full compensation, under-compensation, and over-compensation, and under the over-compensation method, harmonic components are used as the fault line selection feature quantity.

6. The method according to claim 5, wherein The calculation method of the neutral point-to-ground voltage U0 is as follows: Among them, E A is the electromotive force of the faulty-phase power supply, C ∑ is the sum of the capacitances to the ground in the resonant grounding system, v is the angular frequency, and L is the equivalent inductance of the arc suppression coil; According to the transition resistance R of the grounding point f the neutral point voltage to ground U0 and the three-phase voltages to ground U AK , U BK and U CK are obtained. As the vector locus of R f changes, the locus of the neutral point voltage to ground U0 is a semi-circular arc o-k-a with E A as the diameter.

7. The method according to claim 1, characterized in that, Step S4 specifically includes: Input the current recognition result into the trained KNN traceability model, and calculate the distance between the current recognition result and the known-class samples in the sample set, where the sample set contains s categories (G1, G2,..., G s ), and the number of samples for each category is N j (j = 1, 2,..., s); Based on the distance metric formula, select the k samples closest to the current identification result from the sample set, and count the quantities of each category in the k samples; Based on the statistical results, determine the current recognition result as the category with the most occurrences, and the category discrimination function H u (x i ) satisfies: H u (x i ) = max(N u ), where u = (1, 2,..., s) where x i represents the value of the ith sampling point in the unknown sample X, and N u represents the number of samples in the uth category.

8. The method according to claim 7, characterized in that, The distance metric formula includes Manhattan distance, Euclidean distance, and cosine distance, where: The calculation method of the Manhattan distance Md(X,G) is: The calculation method of the Euclidean distance Ed(X,G) is: Cosine distance cosθ (X,G) The calculation method is as follows: where \(i = 1, 2, \ldots, m\), \(x\) i and \(g\) i are the values at the \(i\)-th position of the unknown samples \(X\) and \(G\), respectively.

9. The method according to claim 7, characterized in that, The training process of the KNN traceability model includes: Extract sample data from historical distribution network disturbance events and construct a sample set containing s categories (G1, G2,..., G s ); Preprocess the sample data, including normalization, denoising, and feature extraction; Based on the preprocessed sample data, use the cross-validation method to determine the optimal distance metric formula and parameter k.

10. A small current identification device, characterized in that, It includes: A noise model construction module for obtaining the occurrence mechanisms and waveform characteristics of different distribution network disturbance events, and constructing a noise model of the distribution network according to the occurrence mechanisms and the waveform characteristics; A monitoring data acquisition module for using compressive sensing to perform data analysis on the noise model to obtain the monitored data with added noise; A current identification module for obtaining the current distribution information when a single-phase grounding fault occurs in the distribution network according to the monitored data, and analyzing the single-phase grounding fault according to the current distribution information to obtain a current identification result; A traceability model training module for inputting the current identification result into the trained KNN traceability model for training to obtain the traceability result of the distribution network disturbance event corresponding to the current identification result.

Images