Lightning arrester leakage current meter calibration method and system based on small sample nonlinear mapping and two-dimensional evaluation

Abstract

The application discloses a lightning arrester leakage current meter calibration method and system based on small sample nonlinear mapping and double-dimension evaluation, and the method comprises the following steps: a three-dimensional simulation model of a lightning arrester is constructed, and two-dimensional scattered sample sets of the simulation model under multiple working conditions are obtained; boundary envelopes of each working condition are constructed; the areas defined by the boundary envelopes are respectively taken as input spaces, and corresponding SOM neural networks are constructed; sample points are randomly extracted from the input spaces, input into the SOM neural networks for training, and standard test signals are output; the standard test signals are input into the meter, actual measurement coordinates output by the meter are obtained, and the actual measurement coordinates are compared with standard calibration point coordinates; double-dimension evaluation is performed on the comparison results, so that a comprehensive diagnostic conclusion is obtained; the application expands the traditional single numerical precision evaluation into local numerical precision evaluation and global topological mapping evaluation, can identify hidden cross-border misdiagnosis risks that cannot be found by the traditional method, and effectively distinguishes abnormal states of different properties.

Description

Technical Field

[0001] This invention relates to the field of surge arrester leakage current meter calibration technology, and in particular to a surge arrester leakage current meter calibration method and system based on small sample nonlinear mapping and dual-dimensional evaluation. Background Technology

[0002] Metal oxide surge arresters are critical protection devices in power systems used to limit overvoltages, and their operating status directly affects the reliability of power grid supply. In actual operation, metal oxide surge arresters may experience abnormal temperature rise or uneven voltage distribution due to factors such as internal moisture, surface contamination, or component aging. In severe cases, this can lead to thermal breakdown or even explosion. To monitor the insulation status of the surge arrester in real time, a leakage current measuring meter is usually connected in series in its grounding circuit. By measuring the total current amplitude and its phase angle change relative to the operating voltage, the core criterion for assessing the operating status of the surge arrester is used.

[0003] Since the accuracy of leakage current measuring instruments directly affects the reliability of surge arrester condition diagnosis, a dedicated calibration device must be used to verify the accuracy of the instruments before commissioning and during periodic maintenance. The functions of existing calibration devices mainly cover current amplitude and phase angle verification as well as operation count verification, among which current amplitude and phase angle verification are key links to ensure measurement accuracy.

[0004] However, existing verification techniques have the following shortcomings in practical applications:

[0005] First, the selection of verification points lacks specificity; traditional verification methods usually select a few isolated combinations of current amplitude and phase angle for output calibration based on factory settings or human experience. This method does not consider the nonlinear coupling distribution characteristics of the total current amplitude and phase angle of the surge arrester under different fault conditions. The output verification signal cannot effectively cover the continuous distribution area of ​​the real operating conditions, resulting in insufficient verification coverage.

[0006] Second, the evaluation of verification results is based on a single dimension. Existing technologies mostly use independent calculation of amplitude error and phase angle error to evaluate meter accuracy, which cannot identify the risk of cross-boundary misdiagnosis caused by measurement fluctuations. For example, when a surge arrester is on the verge of evolving between two fault conditions, even if the meter's measurement error is within the factory allowable range, its slight measurement fluctuations may cause the measurement coordinates to cross the boundaries of different conditions, leading to a substantial misjudgment of the fault type and affecting the accuracy of operation and maintenance decisions.

[0007] Therefore, how to extract representative verification points based on the distribution characteristics of the total current amplitude and phase angle under the actual operating conditions of surge arresters to improve the verification coverage, and at the same time construct a multi-dimensional evaluation mechanism that takes into account both numerical accuracy and false alarm prevention capability, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0008] In view of the shortcomings of the prior art, the present invention provides a method and system for verifying the leakage current meter of surge arrester based on small sample nonlinear mapping and dual-dimensional evaluation, so as to solve the problems of insufficient verification coverage caused by the selection of verification points deviating from the actual working condition distribution, and the inability to identify cross-border misdiagnosis risks due to the single evaluation dimension in the prior art.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] In a first aspect, the present invention provides a method for verifying the leakage current meter of a surge arrester based on small-sample nonlinear mapping and two-dimensional evaluation, comprising the following steps:

[0011] A three-dimensional simulation model of the surge arrester is constructed, and a two-dimensional scatter sample set of the total current amplitude and phase angle of the simulation model under various operating conditions is obtained through parametric scanning calculation.

[0012] After normalizing the two-dimensional scattered sample set, a support vector data description algorithm is introduced to construct the boundary envelope of each working condition.

[0013] The regions defined by each of the aforementioned boundary envelopes are used as input spaces to construct corresponding SOM neural networks, and the number of neuron nodes in the SOM neural network is set according to the manifold span of each of the aforementioned input spaces.

[0014] Randomly sample points from the input space and input them into the corresponding SOM neural network for iterative training to extract the coordinates of representative verification points for each working condition and generate a small-sample adaptive verification protocol.

[0015] The calibration device outputs a standard test signal according to the small sample adaptive calibration protocol.

[0016] The standard test signal is input into the leakage current meter of the surge arrester to obtain the actual measurement coordinates output by the meter. The coordinates of the standard verification point corresponding to the standard test signal are compared with the actual measurement coordinates, and the comparison results are evaluated in two dimensions to obtain a comprehensive diagnostic conclusion.

[0017] In one optional implementation, a three-dimensional simulation model of the surge arrester is constructed. Through parametric scanning calculations, a two-dimensional scatter sample set of the total current amplitude and phase angle of the simulation model under various operating conditions is obtained. Specifically, this includes the following steps:

[0018] Obtain the physical dimensions and structural characteristics of the surge arrester;

[0019] Based on the physical dimensions and structural characteristics of the surge arrester, a three-dimensional simulation model of the surge arrester is constructed using simulation software.

[0020] The relative permittivity and conductivity parameters are input into each component of the simulation model, and the current characteristic distribution is calculated to obtain the total current amplitude and phase angle of the simulation model.

[0021] The key parameter variation ranges under various operating conditions are set, and a two-dimensional scatter sample set of the full current amplitude and phase angle of the simulation model under each operating condition is obtained through parametric scanning calculation.

[0022] In an optional implementation, after normalizing the two-dimensional scattered sample set, a support vector data description algorithm is introduced to construct the boundary envelope of each working condition, specifically including the following steps:

[0023] Normalize the two-dimensional scattered sample set corresponding to each working condition, map the sample data to a preset interval, and construct the training sample set corresponding to each working condition.

[0024] For each working condition, the following steps are performed iteratively on the training sample set to obtain the boundary envelope of each working condition:

[0025] Based on the Gaussian radial basis kernel function, a support vector data description model is constructed according to the training sample set;

[0026] Solve the support vector data description model to obtain the minimum hypersphere that encloses the training sample set, and determine the center of the minimum hypersphere;

[0027] Multiple grid points are generated in the original two-dimensional plane. Each grid point is mapped to a high-dimensional feature space through the Gaussian radial basis kernel function. The Euclidean distance from each grid point to the center of the minimum hypersphere in the high-dimensional feature space is calculated.

[0028] Filter grid points where the difference between the Euclidean distance and the minimum hypersphere radius is less than a preset tolerance;

[0029] Based on the grid points, continuous contour lines are fitted to form a continuous contour line, and the continuous contour line is used as the nonlinear continuous closed boundary envelope of the corresponding working condition.

[0030] In an optional implementation, the regions defined by each of the boundary envelopes are used as input spaces to construct corresponding SOM neural networks, and the number of neurons in the SOM neural network is set according to the manifold span of each input space, specifically including the following steps:

[0031] The regions defined by the boundary envelopes of each working condition are used as independent input spaces to construct corresponding SOM neural networks for each working condition.

[0032] The maximum range of amplitude and the maximum range of phase angle of the two-dimensional scattered sample set are extracted as the normalization benchmark;

[0033] Calculate the manifold span of the input space corresponding to each working condition;

[0034] The manifold span includes the amplitude span and phase angle span of sample points within the input space;

[0035] Based on the normalization benchmark, the amplitude span and the phase angle span are normalized respectively, and the comprehensive total span index of the corresponding working condition is calculated based on the normalized amplitude span and phase angle span.

[0036] The number of neuron nodes for the corresponding working condition is set according to the comprehensive total span index.

[0037] In an optional implementation, sample points are randomly selected from the input space and input into the corresponding SOM neural network for iterative training to extract the coordinates of representative verification points for each working condition, generating a small-sample adaptive verification protocol, specifically including:

[0038] For each working condition, the weight vector of the neuron node in each SOM neural network is randomly initialized.

[0039] Each of the aforementioned SOM neural networks is iteratively trained to extract the coordinates of representative verification points for each working condition;

[0040] The coordinates of all representative verification points under each working condition are summarized to generate a small-sample adaptive verification protocol.

[0041] The iterative training of each of the SOM neural networks includes the following steps:

[0042] A1. Randomly sample points from the input space and input them into the corresponding SOM neural network, and calculate the Euclidean distance between the sample points and each weight vector, and determine the neuron with the smallest distance as the best matching unit.

[0043] A2. Based on the topological neighborhood function, update the weight vectors of the best matching unit and its topological neighboring neurons, so that the updated weight vectors are shifted toward the sample point.

[0044] A3. Repeat steps A1 and A2 until the preset convergence condition is met, then end the iterative training and use the final weight vector of each neuron node in each SOM neural network as the representative verification point coordinates for each working condition.

[0045] In one optional implementation, a standard test signal is input to a surge arrester leakage current meter to obtain the actual measurement coordinates output by the meter. The coordinates of the standard verification point corresponding to the standard test signal are compared with the actual measurement coordinates, and the comparison results are verified in two dimensions to obtain a comprehensive diagnostic conclusion. Specifically, this includes the following steps:

[0046] Input the standard test signal into the surge arrester leakage current meter to obtain the actual measurement coordinates output by the meter.

[0047] The coordinates of the standard verification point are compared with the actual measured coordinates, and the comparison results are verified in two dimensions to obtain a global local precision flag that characterizes the numerical error and a global cross-boundary flag that characterizes the consistency of the working condition.

[0048] The dual-dimensional verification includes local numerical accuracy evaluation and global topology mapping evaluation;

[0049] Risk diagnosis is performed based on the global local precision flag and the global cross-boundary flag, and a comprehensive diagnostic conclusion is output.

[0050] In one optional implementation, the local numerical accuracy evaluation specifically includes the following steps:

[0051] B1. Call the covariance matrix of the two-dimensional scattered sample set corresponding to each working condition to calculate the Mahalanobis distance between the normalized actual measured coordinates and the coordinates of the standard verification point.

[0052] B2. Compare the Mahalanobis distance with a preset accuracy threshold to determine whether the local accuracy meets the standard;

[0053] If the preset accuracy threshold is not exceeded, the local accuracy is determined to be up to standard;

[0054] If the preset accuracy threshold is exceeded, the local accuracy is determined to be substandard.

[0055] B3. Traverse all standard verification points and execute steps B1 and B2 respectively to obtain the local precision flag bit corresponding to each standard verification point.

[0056] B4. Globally summarize the local precision flags of all the standard verification points to obtain the global local precision flags;

[0057] If the local accuracy of all verification points meets the standard, then the global local accuracy is deemed to meet the standard.

[0058] If the local accuracy of any verification point is substandard, then the global local accuracy is determined to be substandard.

[0059] In one optional implementation, the global topology mapping evaluation specifically includes the following steps:

[0060] C1. Input the coordinates of the standard verification point into the support vector data description model of all working conditions respectively. The decision function of each working condition determines whether the standard verification point belongs to the corresponding working condition. All working condition labels that are determined to belong are collected into the expected topology belonging set.

[0061] C2. Input the normalized actual measured coordinates into the support vector data description model for all working conditions to obtain the actual topology set;

[0062] C3. Compare the actual topology attribution set with the expected topology attribution set;

[0063] If the actual topology attribution set is consistent with the expected topology attribution set, then it is determined that there is no risk of cross-boundary misdiagnosis.

[0064] If the actual topological affiliation set is inconsistent with the expected topological affiliation set, it is determined that there is a risk of cross-boundary misdiagnosis;

[0065] C4. Traverse all standard checkpoints and execute steps C1 to C3 respectively to obtain the cross-boundary flag bit corresponding to each standard checkpoint.

[0066] C5. Globally summarize the cross-boundary flags of all the standard check points to obtain the global cross-boundary flags;

[0067] If all standard checkpoints meet the cross-boundary flag, then the global cross-boundary flag is determined to have no risk of cross-boundary misdiagnosis.

[0068] If any standard checkpoint fails to meet the cross-boundary flag, the global cross-boundary flag is deemed to indicate a risk of cross-boundary misdiagnosis.

[0069] In an optional implementation, risk diagnosis is performed based on the global local precision flag and the global cross-boundary flag, and a comprehensive diagnostic conclusion is output, specifically including the following steps:

[0070] When the global local accuracy flag is compliant and the global cross-boundary flag is free of cross-boundary misdiagnosis risk, the surge arrester leakage current meter is determined to be in a qualified state.

[0071] When the global local accuracy flag is substandard and the global cross-boundary flag is not at risk of cross-boundary misdiagnosis, the surge arrester leakage current meter is determined to be in a state of accuracy deterioration.

[0072] When the global local accuracy flag is compliant and the global cross-boundary flag indicates a risk of cross-boundary misdiagnosis, the surge arrester leakage current meter is determined to be in a critical misdiagnosis state.

[0073] When the global local accuracy flag is not up to standard and the global cross-boundary flag indicates a risk of cross-boundary misdiagnosis, the surge arrester leakage current meter is determined to be in a failed state.

[0074] Secondly, the present invention provides a surge arrester leakage current meter calibration system based on small-sample nonlinear mapping and dual-dimensional evaluation, comprising:

[0075] The sample generation module is used to construct a three-dimensional simulation model of the surge arrester. Through parametric scanning calculation, it obtains a two-dimensional scatter sample set of the full current amplitude and phase angle of the simulation model under various operating conditions.

[0076] The envelope construction module is used to normalize the two-dimensional scattered sample set and then introduce a support vector data description algorithm to construct the boundary envelope of each working condition.

[0077] The network construction module is used to take the regions defined by each of the boundary envelopes as input spaces to construct the corresponding SOM neural network, and to set the number of neurons in the SOM neural network according to the manifold span of each of the input spaces.

[0078] The verification point extraction module is used to randomly extract sample points from the input space, input them into the corresponding SOM neural network for iterative training, in order to extract the coordinates of representative verification points for each working condition and generate a small sample adaptive verification protocol.

[0079] The signal output module is used to output a standard test signal by the calibration device according to the small sample adaptive calibration protocol.

[0080] The dual-dimensional diagnostic module is used to input the standard test signal into the surge arrester leakage current meter, obtain the actual measurement coordinates output by the meter, compare the coordinates of the standard verification point corresponding to the standard test signal with the actual measurement coordinates, and perform a dual-dimensional evaluation on the comparison results to obtain a comprehensive diagnostic conclusion.

[0081] The beneficial effects of the embodiments provided by the present invention include:

[0082] This invention uses a support vector data description algorithm to perform nonlinear spatial mapping on the total current amplitude and phase angle scatter points under various operating conditions, constructing a continuous closed boundary that accurately envelops each typical operating condition. This allows the selection of verification signals to be based on the distribution characteristics of real operating conditions, rather than relying on experience or isolated set values, effectively improving the test coverage of the verification device for complex fault states.

[0083] This invention adaptively sets the number of neuron nodes based on the manifold span of the closed area under each working condition, and differentiates the number of verification points for working conditions with different degrees of divergence. While ensuring accurate tracking of the entire process of severe degradation evolution, it avoids outputting redundant test signals in high-density overlapping areas, thereby maximizing information acquisition under limited verification resources.

[0084] This invention extends the traditional single numerical accuracy assessment into a two-dimensional system that combines local numerical accuracy assessment and global topological mapping assessment. It can identify hidden cross-boundary misdiagnosis risks that traditional methods cannot detect, and effectively distinguish between two different types of abnormal states: accuracy degradation and critical misdiagnosis. This provides a reliable basis for accurate calibration and operation and maintenance decisions of meters. Attached Figure Description

[0085] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other embodiments can be obtained based on these drawings.

[0086] Figure 1 A flowchart illustrating the surge arrester leakage current meter verification method in the embodiments of this specification is shown;

[0087] Figure 2 A two-dimensional scatter plot of the total current amplitude and phase angle distribution for each operating condition in the embodiments of this specification is shown.

[0088] Figure 3 The following diagram illustrates the nonlinear continuous closed boundary envelope diagram of the surge arrester under multiple operating conditions in the embodiments of this specification; wherein, Figure 3 (a) is the case of small disturbance. Figure 3 (b) is a severely degraded chemical condition;

[0089] Figure 4 This document illustrates the distribution diagram of specific verification points extracted from a small sample of surge arresters in embodiments of this specification; wherein, Figure 4 (a) is the case of small disturbance. Figure 4 (b) is a severely degraded chemical condition;

[0090] Figure 5 A block diagram of the surge arrester leakage current meter calibration system according to an embodiment of this specification is shown. Detailed Implementation

[0091] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention may be practiced without requiring some of these specific details. The following description of embodiments is merely intended to provide a better understanding of the invention by illustrating examples of the invention.

[0092] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appear in a specific order. However, it should be clearly understood that these operations may not be performed in the order they appear herein, or they may be performed in parallel. The operation numbers, such as S1, S2, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be performed sequentially or in parallel.

[0093] The following explains the technical terms used in the embodiments:

[0094] SOM Neural Network: Self-Organizing Map Neural Network is a competitive artificial neural network based on unsupervised learning. By simulating the mechanism of ordered arrangement and competitive response of neurons in the cerebral cortex, the SOM Neural Network utilizes competitive learning and topological neighborhood functions during training to enable neighboring neurons to respond to similar input patterns, thereby mapping high-dimensional input data onto a low-dimensional discrete grid while maintaining the original topological structure of the data.

[0095] Support Vector Data Description (SVDD) is a single-class classification algorithm based on statistical learning theory. SVDD introduces a kernel function to nonlinearly map sample data in the original space to a high-dimensional feature space, and then finds a hypersphere with the smallest volume in this high-dimensional feature space that encloses as much of the sample data of the target class as possible. When this hypersphere is projected back into the original space, it forms a nonlinear continuous closed boundary that closely follows the data distribution pattern, which is used to determine whether a new sample point belongs to the target class.

[0096] Example 1

[0097] like Figure 1 As shown, this embodiment provides a method for verifying the leakage current meter of a surge arrester based on small-sample nonlinear mapping and dual-dimensional evaluation, including the following steps:

[0098] S1. Construct a three-dimensional simulation model of the surge arrester, and obtain a two-dimensional scatter sample set of the total current amplitude and phase angle of the simulation model under various operating conditions through parametric scanning calculation;

[0099] In this embodiment, since it is difficult to exhaustively collect the actual operating data of the surge arrester under various critical faults in the actual field, this embodiment uses simulation methods to construct a three-dimensional simulation model of the surge arrester in order to obtain a complete sample library of operating conditions.

[0100] For example, step S1 specifically includes the following steps:

[0101] Obtain the physical dimensions and structural characteristics of the surge arrester;

[0102] Based on the physical dimensions and structural characteristics of the surge arrester, a three-dimensional simulation model of the surge arrester is constructed using simulation software.

[0103] The three-dimensional simulation model includes key components of the surge arrester such as the core valve plate, insulating sleeve, flange, and internal gas layer.

[0104] The relative permittivity and conductivity parameters are input into each component of the simulation model, and the current characteristic distribution is calculated to obtain the total current amplitude and phase angle of the simulation model.

[0105] The key parameter variation ranges under various operating conditions are set, and a two-dimensional scatter sample set of the full current amplitude and phase angle of the simulation model under each operating condition is obtained through parametric scanning calculation.

[0106] In this embodiment, three typical fault conditions—insulating sleeve aging, internal moisture absorption, and wet contamination—as well as normal operation were simulated. To obtain the data characteristics under each condition, specific variation ranges were set for the variables under each condition, and parametric scanning calculations were performed. The specific simulation parameter settings for different fault conditions are shown in Table 1.

[0107] Table 1 Simulation settings for different fault conditions

[0108]

[0109] By iterating through the parameters shown in Table 1, the total current amplitude and phase angle data of the surge arrester at each state point are extracted. The total current data obtained under different fault conditions are then superimposed with interphase interference and projected onto a two-dimensional coordinate system with the total current amplitude as the horizontal axis and the total current phase angle as the vertical axis. The resulting current distribution characteristics are as follows: Figure 2 As shown.

[0110] Figure 2 It demonstrates the nonlinear aggregation and continuous distribution characteristics of surge arrester current amplitude and phase angle during normal operation and various fault evolution processes.

[0111] S2. After normalizing the two-dimensional scattered sample set, a support vector data description algorithm is introduced to construct the boundary envelope of each working condition.

[0112] In this embodiment, because the total current amplitude and phase angle scatter points of the surge arrester under different operating and fault conditions exhibit a complex nonlinear distribution on the two-dimensional plane, in order to accurately extract the actual characteristic range of each operating condition, this embodiment introduces a support vector data description algorithm to enclose the above scatter point data with a nonlinear continuous boundary.

[0113] For example, step S2 specifically includes the following steps:

[0114] Normalize the two-dimensional scattered sample set corresponding to each working condition, map the sample data to a preset interval, and construct the training sample set corresponding to each working condition.

[0115] In order to eliminate the interference of the difference in the dimensions of amplitude and phase angle on the calculation of spatial Euclidean distance, the two-dimensional sample data is uniformly normalized and mapped to the interval [0,1] to construct a training sample set;

[0116] For each working condition, the following steps are performed iteratively on the training sample set to obtain the boundary envelope of each working condition:

[0117] A1. Based on the Gaussian radial basis kernel function, construct a support vector data description model according to the training sample set;

[0118] In this embodiment, considering the arc-shaped distribution of surge arrester fault data on a two-dimensional plane, a Gaussian radial basis function (RBF) kernel is used to map the data to a higher-dimensional space. The expression for the Gaussian RBF kernel is as follows:

[0119]

[0120] In the formula, and For the i-th and j-th two-dimensional coordinate vectors in the training sample set; For sample points and Inner product in a high-dimensional feature space; For sample points and The square of the Euclidean distance on the original two-dimensional plane; The Gaussian kernel width parameter is used to control the smoothness of the mapping.

[0121] A2. Solve the support vector data description model to obtain the smallest hypersphere that encloses the training sample set, and determine the center of the smallest hypersphere;

[0122] In a high-dimensional feature space, by introducing a penalty factor and relaxation variables into the optimization objective function to tolerate a predetermined number of outliers, the minimum hypersphere containing the vast majority of samples is solved.

[0123] The objective function is:

[0124]

[0125] In the formula, To optimize the objective function value; Let C be the radius of the hypersphere in the high-dimensional feature space; C is the penalty factor. is a slack variable; N is the total number of sample points in the training sample set;

[0126] A3. Generate multiple grid points in the original two-dimensional plane, map each grid point to a high-dimensional feature space through the Gaussian radial basis kernel function, and calculate the Euclidean distance of each grid point to the center of the minimum hypersphere in the high-dimensional feature space.

[0127] A4. Filter grid points where the difference between the Euclidean distance and the minimum hypersphere radius is less than a preset tolerance;

[0128] A5. Based on the grid points, fit to form continuous contour lines, and use the continuous contour lines as the nonlinear continuous closed boundary envelope of the corresponding working condition.

[0129] Following steps A1 to A5 above, the training sample set for each operating condition is traversed to obtain the boundary envelope of each operating condition, resulting in an independent envelope map of the nonlinear characteristic region of the surge arrester under multiple operating conditions, as shown below. Figure 3 As shown.

[0130] The boundary envelopes of each operating condition in this embodiment accurately reproduce the nonlinear multidimensional topological space of the surge arrester under different operating and fault states, laying a precise regional foundation for subsequent small sample feature point extraction.

[0131] S3. The regions defined by each of the boundary envelopes are used as input spaces to construct the corresponding SOM neural network, and the number of neurons in the SOM neural network is set according to the manifold span of each of the input spaces.

[0132] In this embodiment, since the closed area defined by the boundary envelope of each working condition constructed in the above steps contains an infinite number of state points, it is difficult for traditional verification methods to exhaustively count them. In order to achieve efficient and comprehensive small sample testing, this embodiment introduces the SOM neural network to extract specific verification points.

[0133] For example, step S3 specifically includes the following steps:

[0134] The regions defined by the boundary envelopes of each working condition are used as independent input spaces to construct corresponding SOM neural networks for each working condition.

[0135] The maximum range of amplitude and the maximum range of phase angle of the two-dimensional scattered sample set are extracted as the normalization benchmark;

[0136] Calculate the manifold span of the input space corresponding to each working condition;

[0137] The manifold span includes the amplitude span and phase angle span of sample points within the input space;

[0138] Specifically, for the k-th type of working condition, the formula for calculating the manifold span of its input space is as follows:

[0139]

[0140] In the formula, The amplitude span of the sample points within the input space; This represents the maximum amplitude of the sample points within the input space. This represents the minimum amplitude of the sample points within the input space. The phase angle span of the sample points within the input space; This represents the maximum phase angle of the sample points within the input space. This represents the minimum phase angle of the sample points within the input space.

[0141] Based on the normalization benchmark, the amplitude span and the phase angle span are normalized respectively, and the comprehensive total span index of the corresponding working condition is calculated based on the normalized amplitude span and phase angle span.

[0142] To optimize the allocation of verification resources, this embodiment adopts an asymmetric point allocation strategy that is positively correlated with the manifold span of the region. The manifold span characteristics are quantified by calculating the comprehensive total span index of each working condition.

[0143] The overall span index is as follows:

[0144]

[0145] In the formula, The maximum range of magnitude for the global sample set; The maximum range of phase angles for the global sample set;

[0146] The number of neuron nodes for the corresponding working condition is set according to the comprehensive total span index;

[0147] Substituting the scatter sample sets of each working condition in this embodiment into the above formula, the calculated comprehensive span index of each working condition is shown in Table 2.

[0148] Table 2 Characteristic span and comprehensive total span index for each typical working condition

[0149]

[0150] Based on the quantitative data distribution characteristics in Table 2 above, this embodiment sets the first span threshold to 0.2 and the second span threshold to 0.7, and sets the number of neuron nodes of the SOM neural network corresponding to each working condition according to the following rules:

[0151] When the overall span index is ≤0.2, it is judged as a small span level, and the number of neuron nodes is set to 3;

[0152] The operating conditions belonging to this level include normal operating conditions and wet and dirty operating conditions, and their comprehensive total span index is between 0.11 and 0.16. The total current amplitude and phase angle of this type of operating condition are highly dense and overlap each other in two-dimensional space, and the span of the characteristic distribution area is small. If more neuron nodes are set, the verification signal will be redundant. Therefore, three reference nodes are set to improve the verification efficiency while ensuring verification coverage.

[0153] When 0.2 < Total Span Index ≤ 0.7, it is judged as a medium span level, and the number of neuron nodes is set to 5;

[0154] The operating conditions belonging to this level include the upper section damp condition and the upper section aging condition of the insulation, and their comprehensive total span index is between 0.45 and 0.50. The characteristic distribution of this type of operating condition shows a moderate degree of nonlinear divergence, with a moderate manifold span. A moderate number of neuron nodes are set to ensure effective coverage of arc-shaped or long strip boundaries.

[0155] When the overall span index is greater than 0.7, it is judged as the extreme span level, and the number of neuron nodes is set to 7.

[0156] The operating conditions belonging to this level include the lower section damp condition and the lower insulation section aging condition, with a comprehensive total span index between 0.91 and 0.98. The characteristics of this type of operating condition are distributed in a two-dimensional space with a significant long span spread, representing the most serious deterioration evolution process. A large number of neuron nodes are set to cover the evolution range of the entire fault life cycle, so as to achieve accurate tracking and comprehensive coverage of fault characteristics.

[0157] S4. Randomly select sample points from the input space and input them into the corresponding SOM neural network for iterative training to extract the coordinates of representative verification points for each working condition and generate a small sample adaptive verification protocol.

[0158] In this embodiment, after determining the number of neuron nodes for each working condition, the SOM neural network corresponding to each working condition is iteratively trained so that each neuron node converges within the manifold region of the real data distribution, thereby extracting the coordinates of the verification point that are both physically meaningful and representative.

[0159] For example, step S4 specifically includes the following steps:

[0160] S401. For each working condition, randomly initialize the weight vector of the neuron node in each SOM neural network.

[0161] S402. Iteratively train each of the SOM neural networks to extract the coordinates of representative verification points for each working condition;

[0162] Specifically, step S402 includes the following steps:

[0163] S4021. Randomly sample points from the input space and input them into the corresponding SOM neural network, and calculate the Euclidean distance between the sample points and each weight vector, and determine the neuron with the smallest distance as the best matching unit.

[0164] S4022. Based on the topological neighborhood function, update the weight vectors of the best matching unit and its topological neighboring neurons, so that the updated weight vectors are shifted toward the sample point.

[0165] The update rule for the weight vector is as follows:

[0166]

[0167] In the formula, t is the current iteration number; The learning rate decays over time; is a time-shrinking topological neighborhood function used to determine the magnitude of the influence on the neuron at a distance j from the best-matching unit in this iteration.

[0168] S4023. Repeat steps S4021 and S4022 until the preset convergence condition is met, then end the iterative training and use the final weight vector of each neuron node in each SOM neural network as the representative verification point coordinates for each working condition.

[0169] S403. Summarize the coordinates of all representative verification points for each working condition and generate a small-sample adaptive verification protocol.

[0170] In some embodiments, the preset convergence conditions include: the number of iterations reaches a preset maximum number of iterations and / or the weight change of each neuron node is lower than a preset convergence threshold.

[0171] In this embodiment, the extracted coordinates of representative verification points for each operating condition are converted into standard electrical signal parameters to form the adaptive output protocol of the surge arrester verification device. The full current amplitude and phase angle coordinates of the representative verification points for each operating condition are shown in Table 3.

[0172] Table 3. Total Current Amplitude and Phase Angle Coordinates of SOM Characteristic Verification Points

[0173]

[0174] The coordinates of each verification point in the aforementioned small-sample adaptive verification protocol not only cover the geometric center features of the corresponding areas for each working condition, but also include the limit boundary features of the data probability density. This enables accurate tracking and comprehensive coverage of the entire lifecycle of a fault, from minor to serious, with a limited number of verification points. Figure 4 As shown;

[0175] Figure 4 The diagram shows the boundary envelope of each working condition and the distribution of representative verification points extracted by the SOM neural network in this embodiment. Figure 4 Mid-star marker , which are the coordinates of representative verification points extracted after the SOM neural network corresponding to each working condition has converged iteratively. Each verification point is located within the area defined by the boundary envelope of the corresponding working condition.

[0176] S5. The calibration device outputs a standard test signal according to the small sample adaptive calibration protocol.

[0177] In this embodiment, the calibration device outputs standard test signals with corresponding amplitudes and phase angles sequentially according to the coordinates of each calibration point listed in the small sample adaptive calibration protocol generated in step S4, so that the surge arrester leakage current meter can measure them.

[0178] S6. Input the standard test signal into the surge arrester leakage current meter, obtain the actual measurement coordinates output by the meter, compare the standard verification point coordinates corresponding to the standard test signal with the actual measurement coordinates, and evaluate the comparison results in two dimensions to obtain a comprehensive diagnostic conclusion.

[0179] For example, step S6 specifically includes the following steps:

[0180] S601. Input the standard test signal into the surge arrester leakage current meter to obtain the actual measurement coordinates output by the meter.

[0181] This involves obtaining the actual two-dimensional coordinate vector measured by the meter and then mapping it to a test vector using normalization parameters.

[0182] S602. Compare the coordinates of the standard verification point with the actual measured coordinates, and perform dual-dimensional verification on the comparison results to obtain a global local precision flag that characterizes the numerical error and a global cross-boundary flag that characterizes the consistency of the working condition.

[0183] The two-dimensional verification includes local numerical accuracy evaluation and global topology mapping evaluation.

[0184] In some embodiments, local numerical accuracy evaluation specifically includes the following steps:

[0185] B1. Call the covariance matrix of the two-dimensional scattered sample set corresponding to each working condition to calculate the Mahalanobis distance between the normalized actual measured coordinates and the coordinates of the standard verification point.

[0186] The Mahalanobis distance is expressed as follows:

[0187]

[0188] In the formula, The Mahalanobis distance; These are the actual measured coordinates; These are the coordinates of the standard verification point.

[0189] B2. The Mahalanobis distance Compared with the preset accuracy threshold Compare the results to determine whether the local accuracy meets the standard;

[0190] Among them, the preset precision threshold Corresponding to the factory accuracy class of the leakage current meter of the surge arrester;

[0191] If the preset precision threshold is not exceeded, then the local precision flag corresponding to the standard verification point is determined. This indicates that the local accuracy meets the standard;

[0192] If the accuracy exceeds the preset accuracy threshold, then the local accuracy flag corresponding to the standard verification point is determined. This indicates that the local accuracy is not up to standard;

[0193] B3. Traverse all standard calibration points, and execute steps B1 and B2 respectively to obtain the local precision flag bit corresponding to each standard calibration point.

[0194] B4. Globally summarize the local precision flags of all the standard verification points to obtain the global local precision flags;

[0195] If the local accuracy of all verification points meets the standard, then the global local accuracy is deemed to meet the standard.

[0196] If the local accuracy of any verification point is substandard, then the global local accuracy is determined to be substandard.

[0197] In some embodiments, global topology mapping evaluation specifically includes the following steps:

[0198] C1. Input the coordinates of the standard verification points into the support vector data description models for all working conditions. The decision function for each working condition determines whether the standard verification point belongs to the corresponding working condition. All working condition labels that are determined to belong are then aggregated into the desired topology affiliation set. ;

[0199] In some embodiments, the first Coordinates of one standard feature verification point Iterate through all inputs In the support vector data description model for similar working conditions; when the first Decision function for similar working conditions Then the label Add to the expected topological affiliation set middle.

[0200] C2. Input the normalized actual measured coordinates into the support vector data description model for all operating conditions to obtain the actual topology set. ;

[0201] C3. Assign the actual topology to the set. With the desired topological affiliation set Perform a Boolean comparison to obtain the cross-boundary flag corresponding to the standard check point;

[0202] If the actual topology belongs to the set With the desired topological affiliation set Consistency indicates that even if there is a measurement offset, the actual measured coordinates have not changed their attribute of being within the original working condition area. Therefore, the cross-boundary marker corresponding to the standard verification point is determined. That is, there is no risk of cross-disciplinary misdiagnosis;

[0203] If the actual topology belongs to the set With the desired topological affiliation set Inconsistency, meaning the presence of unexpected labels or the loss of key labels in the actual topology attribution set, indicates that the actual measured coordinates have crossed the topology boundary of the operating condition. In this case, the cross-boundary marker corresponding to the standard verification point is determined. This indicates a risk of cross-disciplinary misdiagnosis.

[0204] C4. Traverse all standard checkpoints and execute steps C1 to C3 respectively to obtain the cross-boundary flag bit corresponding to each standard checkpoint.

[0205] C5. Globally summarize the cross-boundary flags of all the standard check points to obtain the global cross-boundary flags;

[0206] If all standard checkpoints meet the cross-boundary flag, then the global cross-boundary flag is determined to have no risk of cross-boundary misdiagnosis.

[0207] If any standard checkpoint fails to meet the cross-boundary flag, the global cross-boundary flag is deemed to indicate a risk of cross-boundary misdiagnosis.

[0208] S603. Perform risk diagnosis based on the global local precision flag and the global cross-boundary flag, and output a comprehensive diagnostic conclusion.

[0209] Specifically, step S603 includes the following steps:

[0210] When the global local accuracy flag is compliant and the global cross-boundary flag is free of cross-boundary misdiagnosis risk, the surge arrester leakage current meter is determined to be in a qualified state.

[0211] Among them, under qualified conditions, the numerical measurement accuracy of the surge arrester leakage current meter meets the standard, and there is no risk of false alarm when dealing with real complex working conditions, so the verification is passed.

[0212] When the global local accuracy flag is substandard and the global cross-boundary flag is not at risk of cross-boundary misdiagnosis, the surge arrester leakage current meter is determined to be in a state of accuracy deterioration.

[0213] In the case of deteriorated accuracy, the numerical measurement accuracy of the surge arrester leakage current meter has exceeded the allowable error range. However, the current measurement deviation has not yet caused it to fall out of the physical boundary of the original operating condition and has not triggered a false alarm. The internal measuring components of the meter have deteriorated and need to be fine-tuned and calibrated.

[0214] When the global local accuracy flag is compliant and the global cross-boundary flag indicates a risk of cross-boundary misdiagnosis, the surge arrester leakage current meter is determined to be in a critical misdiagnosis state.

[0215] In the critical misdiagnosis state, although the numerical accuracy of the surge arrester leakage current meter is within the factory error range, its normal measurement fluctuations are sufficient to trigger the crossing of the topological boundary for physical states that are on the edge of critical evolution. The state identification function has hidden dangers and the meter calculation method threshold must be adjusted or the meter should be downgraded.

[0216] When the global local accuracy flag is not up to standard and the global cross-boundary flag indicates a risk of cross-boundary misdiagnosis, the surge arrester leakage current meter is determined to be in a failed state.

[0217] In particular, when the surge arrester is in a failure state, the accuracy of the leakage current meter reading is seriously out of tolerance, which has led to a substantial risk of misjudging the fault type. Therefore, the equipment must be returned or replaced as unqualified.

[0218] It should be noted that, due to the objective physical overlap between the total current amplitude and phase angle characteristics of the surge arrester under different operating and fault conditions, there may be some overlapping areas between the boundary envelopes of each operating condition constructed in step S2. Therefore, when a certain standard verification point falls within such an overlapping area, the point will be determined to belong to multiple operating conditions by the support vector data description models of multiple operating conditions at the same time. Thus, its expected topology belonging set contains multiple operating condition labels.

[0219] In this case, if the meter under test measures accurately, its actual measurement result should also be located within the overlapping area. That is, the actual topology attribution set contains multiple operating condition labels that are exactly the same as the expected topology attribution set. If the two are consistent, it is determined that there is no risk of cross-boundary misdiagnosis. If the meter under test has a measurement deviation, its actual measurement result may deviate from the overlapping area, resulting in the actual topology attribution set being inconsistent with the expected topology attribution set, and there is a risk of cross-boundary misdiagnosis.

[0220] Example 2

[0221] like Figure 5 As shown, this embodiment provides a surge arrester leakage current meter calibration system 100 based on small-sample nonlinear mapping and dual-dimensional evaluation, including:

[0222] The sample generation module 101 is used to construct a three-dimensional simulation model of the surge arrester and obtain a two-dimensional scatter sample set of the full current amplitude and phase angle of the simulation model under various operating conditions through parametric scanning calculation.

[0223] The envelope construction module 102 is used to normalize the two-dimensional scattered sample set and then introduce a support vector data description algorithm to construct the boundary envelope of each working condition.

[0224] The network construction module 103 is used to take the regions defined by each of the boundary envelopes as input spaces to construct the corresponding SOM neural network, and to set the number of neurons in the SOM neural network according to the manifold span of each of the input spaces.

[0225] The verification point extraction module 104 is used to randomly extract sample points from the input space and input them into the corresponding SOM neural network for iterative training in order to extract the coordinates of representative verification points for each working condition and generate a small sample adaptive verification protocol.

[0226] The signal output module 105 is used to output a standard test signal by the calibration device according to the small sample adaptive calibration protocol.

[0227] The dual-dimensional diagnostic module 106 is used to input the standard test signal into the surge arrester leakage current meter, obtain the actual measurement coordinates output by the meter, compare the standard verification point coordinates corresponding to the standard test signal with the actual measurement coordinates, and perform a dual-dimensional evaluation on the comparison results to obtain a comprehensive diagnostic conclusion.

[0228] Example 3

[0229] This embodiment provides an electronic device, including at least one control processor and a memory for communicatively connecting to the at least one control processor;

[0230] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0231] The non-transient software program and instructions required to implement the surge arrester leakage current meter verification method based on small-sample nonlinear mapping and two-dimensional evaluation in the above embodiments are stored in memory. When executed by the processor, the surge arrester leakage current meter verification method based on small-sample nonlinear mapping and two-dimensional evaluation in the above embodiments is executed. For example, the above-described method is executed. Figure 1 Method steps S1 to S6;

[0232] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0233] Example 4

[0234] This embodiment provides a computer-readable storage medium storing computer-executable instructions for causing a computer to execute a surge arrester leakage current meter verification method based on small-sample nonlinear mapping and dual-dimensional evaluation as described in Embodiment 1.

[0235] It should be noted that the computer-readable storage medium in this embodiment may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof.

[0236] More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0237] In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this embodiment, the computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable program. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof.

[0238] The computer-readable signal medium may also be any computer-readable storage medium other than a computer-readable storage medium, which can send, propagate or transmit a program for use by or in connection with an instruction execution system, apparatus or device.

[0239] The computer program contained on the computer-readable storage medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof. The aforementioned computer-readable storage medium can be used to write a computer program for performing this embodiment in one or more programming languages ​​or combinations thereof. These programming languages ​​include object-oriented programming languages—such as Python, Java, and C++—and conventional procedural programming languages—such as C or similar programming languages. The program can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0240] The above description is merely a preferred embodiment of the present invention. It will be apparent to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects. The scope of the present invention is defined by the appended claims rather than the foregoing description, and thus all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention.

Claims

1. A method for verifying leakage current meters of surge arresters based on small-sample nonlinear mapping and dual-dimensional evaluation, characterized in that, Includes the following steps: A three-dimensional simulation model of the surge arrester is constructed, and a two-dimensional scatter sample set of the total current amplitude and phase angle of the simulation model under various operating conditions is obtained through parametric scanning calculation. After normalizing the two-dimensional scattered sample set, a support vector data description algorithm is introduced to construct the boundary envelope of each working condition. The regions defined by each of the aforementioned boundary envelopes are used as input spaces to construct corresponding SOM neural networks, and the number of neuron nodes in the SOM neural network is set according to the manifold span of each of the aforementioned input spaces. Randomly sample points from the input space and input them into the corresponding SOM neural network for iterative training to extract the coordinates of representative verification points for each working condition and generate a small-sample adaptive verification protocol. The calibration device outputs a standard test signal according to the small sample adaptive calibration protocol. The standard test signal is input into the leakage current meter of the surge arrester to obtain the actual measurement coordinates output by the meter. The coordinates of the standard verification point corresponding to the standard test signal are compared with the actual measurement coordinates, and the comparison results are evaluated in two dimensions to obtain a comprehensive diagnostic conclusion.

2. The method according to claim 1, characterized in that, A three-dimensional simulation model of the surge arrester is constructed. Through parametric scanning calculations, a two-dimensional scatter sample set of the total current amplitude and phase angle of the simulation model under various operating conditions is obtained. This process includes the following steps: Obtain the physical dimensions and structural characteristics of the surge arrester; Based on the physical dimensions and structural characteristics of the surge arrester, a three-dimensional simulation model of the surge arrester is constructed using simulation software. The relative permittivity and conductivity parameters are input into each component of the simulation model, and the current characteristic distribution is calculated to obtain the total current amplitude and phase angle of the simulation model. The key parameter variation ranges under various operating conditions are set, and a two-dimensional scatter sample set of the full current amplitude and phase angle of the simulation model under each operating condition is obtained through parametric scanning calculation.

3. The method according to claim 1, characterized in that, After normalizing the two-dimensional scattered sample set, a support vector data description algorithm is introduced to construct the boundary envelope of each working condition, specifically including the following steps: Normalize the two-dimensional scattered sample set corresponding to each working condition, map the sample data to a preset interval, and construct the training sample set corresponding to each working condition. For each working condition, the following steps are performed iteratively on the training sample set to obtain the boundary envelope of each working condition: Based on the Gaussian radial basis kernel function, a support vector data description model is constructed according to the training sample set; Solve the support vector data description model to obtain the minimum hypersphere that encloses the training sample set, and determine the center of the minimum hypersphere; Multiple grid points are generated in the original two-dimensional plane. Each grid point is mapped to a high-dimensional feature space through the Gaussian radial basis kernel function. The Euclidean distance from each grid point to the center of the minimum hypersphere in the high-dimensional feature space is calculated. Filter grid points where the difference between the Euclidean distance and the minimum hypersphere radius is less than a preset tolerance; Based on the grid points, continuous contour lines are fitted to form a continuous contour line, and the continuous contour line is used as the nonlinear continuous closed boundary envelope of the corresponding working condition.

4. The method according to claim 1, characterized in that, The regions defined by each of the aforementioned boundary envelopes are used as input spaces to construct corresponding SOM neural networks. The number of neurons in the SOM neural network is determined based on the manifold span of each input space, specifically including the following steps: The regions defined by the boundary envelopes of each working condition are used as independent input spaces to construct corresponding SOM neural networks for each working condition. The maximum range of amplitude and the maximum range of phase angle of the two-dimensional scattered sample set are extracted as the normalization benchmark; Calculate the manifold span of the input space corresponding to each working condition; The manifold span includes the amplitude span and phase angle span of sample points within the input space; Based on the normalization benchmark, the amplitude span and the phase angle span are normalized respectively, and the comprehensive total span index of the corresponding working condition is calculated based on the normalized amplitude span and phase angle span. The number of neuron nodes for the corresponding working condition is set according to the comprehensive total span index.

5. The method according to claim 1, characterized in that, Randomly sample points are drawn from the input space and input into the corresponding SOM neural network for iterative training to extract the coordinates of representative verification points for each working condition, generating a small-sample adaptive verification protocol, specifically including: For each working condition, the weight vector of the neuron node in each SOM neural network is randomly initialized. Each of the aforementioned SOM neural networks is iteratively trained to extract the coordinates of representative verification points for each working condition; The coordinates of all representative verification points under each working condition are summarized to generate a small-sample adaptive verification protocol. The iterative training of each of the SOM neural networks includes the following steps: A1. Randomly sample points from the input space and input them into the corresponding SOM neural network, and calculate the Euclidean distance between the sample points and each weight vector, and determine the neuron with the smallest distance as the best matching unit. A2. Based on the topological neighborhood function, update the weight vectors of the best matching unit and its topological neighboring neurons, so that the updated weight vectors are shifted toward the sample point. A3. Repeat steps A1 and A2 until the preset convergence condition is met, then end the iterative training and use the final weight vector of each neuron node in each SOM neural network as the representative verification point coordinates for each working condition.

6. The method according to claim 1, characterized in that, The standard test signal is input into the surge arrester leakage current meter to obtain the actual measurement coordinates output by the meter. The coordinates of the standard verification point corresponding to the standard test signal are compared with the actual measurement coordinates, and the comparison results are verified in two dimensions to obtain a comprehensive diagnostic conclusion. The specific steps include: Input the standard test signal into the surge arrester leakage current meter to obtain the actual measurement coordinates output by the meter. The coordinates of the standard verification point are compared with the actual measured coordinates, and the comparison results are verified in two dimensions to obtain a global local precision flag that characterizes the numerical error and a global cross-boundary flag that characterizes the consistency of the working condition. The dual-dimensional verification includes local numerical accuracy evaluation and global topology mapping evaluation; Risk diagnosis is performed based on the global local precision flag and the global cross-boundary flag, and a comprehensive diagnostic conclusion is output.

7. The method according to claim 6, characterized in that, The local numerical accuracy assessment specifically includes the following steps: B1. Call the covariance matrix of the two-dimensional scattered sample set corresponding to each working condition to calculate the Mahalanobis distance between the normalized actual measured coordinates and the coordinates of the standard verification point. B2. Compare the Mahalanobis distance with a preset accuracy threshold to determine whether the local accuracy meets the standard; If the preset accuracy threshold is not exceeded, the local accuracy is determined to be up to standard; If the preset accuracy threshold is exceeded, the local accuracy is determined to be substandard. B3. Traverse all standard verification points and execute steps B1 and B2 respectively to obtain the local precision flag bit corresponding to each standard verification point. B4. Globally summarize the local precision flags of all the standard verification points to obtain the global local precision flags; If the local accuracy of all verification points meets the standard, then the global local accuracy is deemed to meet the standard. If the local accuracy of any verification point is substandard, then the global local accuracy is determined to be substandard.

8. The method according to claim 6, characterized in that, The global topology mapping evaluation specifically includes the following steps: C1. Input the coordinates of the standard verification point into the support vector data description model of all working conditions respectively. The decision function of each working condition determines whether the standard verification point belongs to the corresponding working condition. All working condition labels that are determined to belong are collected into the expected topology belonging set. C2. Input the normalized actual measured coordinates into the support vector data description model for all working conditions to obtain the actual topology set; C3. Compare the actual topology attribution set with the expected topology attribution set; If the actual topology attribution set is consistent with the expected topology attribution set, then it is determined that there is no risk of cross-boundary misdiagnosis. If the actual topological affiliation set is inconsistent with the expected topological affiliation set, it is determined that there is a risk of cross-boundary misdiagnosis; C4. Traverse all standard checkpoints and execute steps C1 to C3 respectively to obtain the cross-boundary flag bit corresponding to each standard checkpoint. C5. Globally summarize the cross-boundary flags of all the standard check points to obtain the global cross-boundary flags; If all standard checkpoints meet the cross-boundary flag, then the global cross-boundary flag is determined to have no risk of cross-boundary misdiagnosis. If any standard checkpoint fails to meet the cross-boundary flag, the global cross-boundary flag is deemed to indicate a risk of cross-boundary misdiagnosis.

9. The method according to claim 6, characterized in that, Risk diagnosis is performed based on the global local precision flag and the global cross-boundary flag, and a comprehensive diagnostic conclusion is output. The specific steps include: When the global local accuracy flag is compliant and the global cross-boundary flag is free of cross-boundary misdiagnosis risk, the surge arrester leakage current meter is determined to be in a qualified state. When the global local accuracy flag is substandard and the global cross-boundary flag is not at risk of cross-boundary misdiagnosis, the surge arrester leakage current meter is determined to be in a state of accuracy deterioration. When the global local accuracy flag is compliant and the global cross-boundary flag indicates a risk of cross-boundary misdiagnosis, the surge arrester leakage current meter is determined to be in a critical misdiagnosis state. When the global local accuracy flag is not up to standard and the global cross-boundary flag indicates a risk of cross-boundary misdiagnosis, the surge arrester leakage current meter is determined to be in a failed state.

10. A surge arrester leakage current meter calibration system based on small-sample nonlinear mapping and dual-dimensional evaluation, characterized in that, include: The sample generation module is used to construct a three-dimensional simulation model of the surge arrester. Through parametric scanning calculation, it obtains a two-dimensional scatter sample set of the full current amplitude and phase angle of the simulation model under various operating conditions. The envelope construction module is used to normalize the two-dimensional scattered sample set and then introduce a support vector data description algorithm to construct the boundary envelope of each working condition. The network construction module is used to take the regions defined by each of the boundary envelopes as input spaces to construct the corresponding SOM neural network, and to set the number of neurons in the SOM neural network according to the manifold span of each of the input spaces. The verification point extraction module is used to randomly extract sample points from the input space, input them into the corresponding SOM neural network for iterative training, in order to extract the coordinates of representative verification points for each working condition and generate a small sample adaptive verification protocol. The signal output module is used to output a standard test signal by the calibration device according to the small sample adaptive calibration protocol. The dual-dimensional diagnostic module is used to input the standard test signal into the surge arrester leakage current meter, obtain the actual measurement coordinates output by the meter, compare the coordinates of the standard verification point corresponding to the standard test signal with the actual measurement coordinates, and perform a dual-dimensional evaluation on the comparison results to obtain a comprehensive diagnostic conclusion.

Images