10KV power distribution network grounding fault line selection method based on ICEEMDAN and Stacking integrated learning

Abstract

The invention relates to the technical field of power system fault detection, in particular to a 10KV power distribution network ground fault line selection method based on ICEEMDAN and Stacking ensemble learning, which comprises the following steps: determining the state of a power distribution network according to a line danger coefficient and a fault transient threshold, and determining a line selection mode according to the state of the power distribution network to obtain a pre-fault line; a zero-sequence current signal corresponding to the pre-fault line is collected, and the zero-sequence current signal is decomposed based on the empirical mode decomposition frequency to obtain an IMF; detecting a sample entropy corresponding to each IMF, and determining a fault feature matrix construction mode according to an entropy dislocation threshold; inputting the fault feature matrix into a pre-trained Stacking ensemble learning model, and determining a base classification mode according to a matrix feature coefficient and a matrix attribution coefficient; splicing output results of the base classifiers into input features of the element classifiers so as to select fault lines; the fault line selection accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system fault detection, and in particular to a 10KV distribution network grounding fault line selection method based on ICEEMDAN and Stacking integrated learning. Background Art

[0002] In 10KV distribution networks, the neutral point resonant grounding method is widely used. However, when a single-phase grounding fault occurs, the traditional line selection method has many problems. The existing single-phase grounding fault line selection method is affected by the asymmetry of the power grid and environmental interference, and has problems such as low detection accuracy, high false alarm rate and poor classification accuracy. Therefore, how to improve the accuracy of fault line selection is a technical problem that needs to be solved urgently by technical personnel in this field.

[0003] Chinese Patent Publication No. CN112485595A discloses a method and device for line selection protection for ground faults in distribution networks. The device comprises a high-frequency voltage monitoring sensor, a high-frequency zero-sequence current sensor, and a line selection module. By real-time monitoring of the system busbar's three-phase voltage, zero-sequence voltage, and line zero-sequence current, the device determines ground faults and the ground fault phase using the high-frequency characteristics of the system voltage signal. Furthermore, the device determines the faulty line based on the polarity characteristics of the system zero-sequence current and the high-frequency signals of the faulty phase voltage. This technical solution suffers from the following problems: It relies on high-frequency signal characteristics to determine faults, lacks comprehensive analysis of multi-dimensional features under complex operating conditions, and fails to utilize an integrated learning model for optimization, resulting in low line selection accuracy. Summary of the Invention

[0004] To this end, the present invention provides a 10KV distribution network grounding fault line selection method based on ICEEMDAN and Stacking ensemble learning, which is used to overcome the problems in the prior art that fault judgment relies on high-frequency signal characteristics, lacks comprehensive analysis of multi-dimensional features under complex working conditions, and does not utilize ensemble learning models for optimization, resulting in low line selection accuracy.

[0005] To achieve the above objectives, the present invention provides a 10KV distribution network ground fault line selection method based on ICEEMDAN and Stacking ensemble learning, comprising:

[0006] When the bus side zero sequence voltage is greater than the preset bus side zero sequence voltage, the distribution network state is determined according to the line risk factor and the fault transient threshold. The line selection method is determined according to the distribution network state to obtain the pre-fault line. The line selection method is feature selection based on the line influence coefficient or association selection based on the line interaction degree.

[0007] The zero-sequence current signal corresponding to the pre-fault line is collected, the number of empirical mode decompositions is determined according to the decomposition interaction degree, and the zero-sequence current signal is decomposed based on the number of empirical mode decompositions to obtain IMF;

[0008] Detect the sample entropy corresponding to each IMF, and determine the fault feature matrix construction method based on the entropy value deviation threshold: construct the fault feature matrix based on the effective sample entropy or IMF impact representation value;

[0009] The fault feature matrix is ​​input into the pre-trained Stacking ensemble learning model. The base classification method is determined based on the matrix feature coefficient and the matrix attribution coefficient. The base classifier is selected based on the conversion correlation degree or the simulation matching coefficient.

[0010] According to the base classification defect coefficient and the evaluation instability threshold, the optimization method is determined to adjust the fault feature matrix dimension or the number of cross-validation times;

[0011] The output results of the base classifier are concatenated as the input features of the meta-classifier to select the faulty lines.

[0012] Furthermore, if the distribution network status is that the line risk coefficient is greater than or equal to the preset line risk coefficient or the fault transient threshold is greater than or equal to the preset fault transient threshold, the line selection method is to perform associated line selection based on the line interaction degree.

[0013] Furthermore, if the distribution network status is that the line risk coefficient is less than the preset line risk coefficient and the fault transient threshold is less than the preset fault transient threshold, the line selection method is to perform feature line selection according to the line influence coefficient.

[0014] Furthermore, the fault transient threshold is determined based on the line matching threshold and the change impact threshold;

[0015] The fault transient threshold is positively correlated with the line matching threshold and the change impact threshold.

[0016] Furthermore, the number of empirical mode decompositions is determined according to the decomposition interaction degree, including:

[0017] If the decomposition interaction degree is greater than or equal to the preset decomposition interaction degree, the number of empirical mode decompositions is directly determined based on the historical difference coefficient and the iterative fluctuation value;

[0018] If the decomposition interaction degree is less than the preset decomposition interaction degree, the number of empirical mode decompositions is determined successively according to the amplitude threshold and the attenuation coefficient.

[0019] Furthermore, the sample entropy is calculated as follows: SampEn(m, t) = lnB m (t)-lnB m+1 (t).

[0020] Furthermore, the fault feature matrix construction method is determined according to the entropy value deviation threshold, including:

[0021] If the entropy value deviation threshold is greater than or equal to the preset entropy value deviation threshold, the fault feature matrix is ​​constructed in the manner of constructing the fault feature matrix according to the effective sample entropy;

[0022] If the entropy value deviation threshold is less than the preset entropy value deviation threshold, the fault feature matrix is ​​constructed in a manner of constructing the fault feature matrix according to the IMF impact characterization value.

[0023] Furthermore, a basic classification method is determined according to the matrix characteristic coefficient and the matrix attribution coefficient, including:

[0024] If the matrix characteristic coefficient is greater than or equal to the preset matrix characteristic coefficient or the matrix attribution coefficient is less than the preset matrix attribution coefficient, the base classification method is to select a base classifier according to the conversion correlation degree;

[0025] If the matrix characteristic coefficient is less than the preset matrix characteristic coefficient and the matrix attribution coefficient is greater than or equal to the preset matrix attribution coefficient, the base classification method is to select a base classifier according to the simulated matching coefficient.

[0026] Furthermore, if the base classification defect coefficient is greater than or equal to the preset base classification defect coefficient and the evaluation instability threshold is greater than or equal to the preset evaluation instability threshold, the optimization method is to increase the number of cross-validation times;

[0027] Furthermore, if the base classification defect coefficient is less than the preset base classification defect coefficient and the assessment instability threshold is greater than or equal to the preset assessment instability threshold, the optimization method is to increase the dimension of the fault feature matrix;

[0028] The increase in the dimension of the fault feature matrix is ​​positively correlated with the assessment instability threshold.

[0029] Compared with the prior art, the beneficial effect of the present invention lies in that, in the technical solution of the present invention, the distribution network status is determined according to the line risk coefficient and the fault transient threshold, and the line risk coefficient and the fault transient threshold are used to effectively reflect the fault situation of the distribution network, and then different line selection methods are adaptively selected according to the distribution network status, so that the selection of the line selection method is more in line with the actual application scenario, has higher accuracy and robustness, can effectively improve the safety and reliability of the operation of the 10Kv distribution network, can reduce the time and cost of fault investigation, and improve the efficiency of fault line selection.

[0030] Furthermore, in the present invention, the number of empirical mode decompositions is determined according to the decomposition interaction degree, and the decomposition interaction degree is used to effectively reflect the similarity between the zero-sequence current signal and the zero-sequence current signal decomposed in the historical records. Then, different numbers of empirical mode decompositions are adaptively selected according to the decomposition interaction degree, so that the determined number of empirical mode decompositions can accurately extract fault characteristics and improve line selection sensitivity.

[0031] Furthermore, in the present invention, the entropy value deviation threshold is used to effectively reflect the discrete degree of the sample entropy corresponding to each IMF, and then different fault feature matrix construction methods are adaptively selected according to the entropy value deviation threshold. The fault feature matrix construction method can ensure that the matrix matches the complexity of the fault characteristics, improve the signal-to-noise ratio of the feature matrix, enhance the detection capability of weak fault signals, and thus improve the line selection efficiency.

[0032] Furthermore, the present invention effectively reflects the actual characteristics of the fault feature matrix through the matrix characteristic coefficients and the matrix attribution coefficients, and then adaptively selects different base classification methods according to the matrix characteristic coefficients and the matrix attribution coefficients, which can significantly improve the model training efficiency while ensuring the classification accuracy.

[0033] Furthermore, the present invention effectively reflects the defect situation of the base classification through the base classification defect coefficient and the evaluation instability threshold, and then adaptively selects different optimization methods according to the base classification defect coefficient and the evaluation instability threshold, so that the selected optimization method can supplement the key features and improve the ability to distinguish defects. At the same time, it can improve the accuracy and stability of the model, thereby improving the line selection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 Schematic diagram of the 10KV distribution network ground fault line selection method based on ICEEMDAN and Stacking integrated learning of the present invention;

[0035] Figure 2 This is a flow chart of the present invention for determining a line selection method according to the state of the distribution network;

[0036] Figure 3 This is a flow chart of the method for determining the fault feature matrix construction method according to the entropy value deviation threshold of the present invention;

[0037] Figure 4 The present invention is a flow chart of determining a base classification method according to matrix characteristic coefficients and matrix dependency coefficients. DETAILED DESCRIPTION

[0038] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0039] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0040] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside", and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0041] Furthermore, it should be noted that, in the description of the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0042] See also Figures 1 to 4 As shown, the present invention provides a 10KV distribution network ground fault line selection method based on ICEEMDAN and Stacking integrated learning, including:

[0043] When the bus side zero sequence voltage is greater than the preset bus side zero sequence voltage, the distribution network state is determined according to the line risk factor and the fault transient threshold. The line selection method is determined according to the distribution network state to obtain the pre-fault line. The line selection method is feature selection based on the line influence coefficient or association selection based on the line interaction degree.

[0044] The zero-sequence current signal corresponding to the pre-fault line is collected, the number of empirical mode decompositions is determined according to the decomposition interaction degree, and the zero-sequence current signal is decomposed based on the number of empirical mode decompositions to obtain IMF;

[0045] Detect the sample entropy corresponding to each IMF, and determine the fault feature matrix construction method based on the entropy value deviation threshold: construct the fault feature matrix based on the effective sample entropy or IMF impact representation value;

[0046] The fault feature matrix is ​​input into the pre-trained Stacking ensemble learning model. The base classification method is determined based on the matrix feature coefficient and the matrix attribution coefficient. The base classifier is selected based on the conversion correlation degree or the simulation matching coefficient.

[0047] According to the base classification defect coefficient and the evaluation instability threshold, the optimization method is determined to adjust the fault feature matrix dimension or the number of cross-validation times;

[0048] The output results of the base classifier are concatenated as the input features of the meta-classifier to select the faulty lines.

[0049] The application scenario of the present invention is line selection when a single-phase grounding fault occurs in a distribution network. In the present invention, several historical records are correspondingly provided, and any historical record records the line risk factor, fault transient threshold, line influence coefficient, decomposition interaction, amplitude threshold and attenuation coefficient, etc. in the historical process of line selection when a single-phase grounding fault occurs in the distribution network at least once, and each historical record corresponds to a qualified mark, which records whether the line selection process when a single-phase grounding fault occurs in the distribution network meets user requirements. The qualified mark can be recorded manually. It can be understood that the user can determine whether the line selection process when a single-phase grounding fault occurs in the distribution network meets the requirements based on self-set indicators. The self-set indicators can be but are not limited to line selection time, which will not be elaborated here. Among them, the line selection time is the time consumed from the start of line selection to the selection of the faulty line when a single-phase grounding fault occurs in the distribution network;

[0050] The zero-sequence voltage on the bus side is measured by a zero-sequence voltage transformer. When collecting the zero-sequence current signal corresponding to the pre-fault line, it is collected by a zero-sequence current transformer. This is a common technical means in this field and will not be described in detail.

[0051] The value of the preset bus side zero-sequence voltage can be determined by the user according to the actual application scenario. The greater the user's demand for improving the fault processing accuracy, the smaller the value of the preset bus side zero-sequence voltage is. A preset bus side zero-sequence voltage value is provided. The preset bus side zero-sequence voltage is 0.15U, where U is the rated voltage of the distribution network. In the present invention, U=10kV;

[0052] The outputs of the base classifiers are concatenated as input features for the meta-classifier to select faulty lines. For a single base model, after the first training run, the output on the validation set is denoted as a1, and the output on the test set is denoted as b1. After the second training run, the output on the validation set is denoted as a2, and the output on the test set is denoted as b2. Following the above steps, after z-fold cross-validation of Model 1, the results a1, a2, ..., az and b1, b2, ..., bz are obtained. a1, a2, ..., az are the outputs on the validation set after each training run. These are concatenated to obtain A1, which is the prediction result on the original training set after training. b1, b2, ..., bz are the outputs on the test set after each training run. These are summed and averaged to obtain B1, which is the prediction result on the entire original test set after training. After the model is trained, A1 and B1 are obtained. Similarly, z-fold cross-validation is performed on the other base models. Assuming there are s base models in total, after the above operation, we will obtain A1, A2, ..., As and B1, B2, ..., Bs. We then combine A1, A2, ..., As as the training set and use B1, B2, ..., Bs as the test set to train and test our meta-classifier model. To prevent overfitting, we choose the Logistic Regression model as the meta-classifier, directly modeling the classification probability to select the faulty line.

[0053] Specifically, if the distribution network status is that the line risk coefficient is greater than or equal to the preset line risk coefficient or the fault transient threshold is greater than or equal to the preset fault transient threshold, the line selection method is to perform associated line selection based on the line interaction degree.

[0054] The distribution network state includes a first distribution network state and a second distribution network state. The first distribution network state is that the line risk coefficient is greater than or equal to the preset line risk coefficient or the fault transient threshold is greater than or equal to the preset fault transient threshold. The second distribution network state is that the line risk coefficient is less than the preset line risk coefficient and the fault transient threshold is less than the preset fault transient threshold.

[0055] The present invention includes several nodes, including bus nodes, load nodes, and branch nodes. The bus node is the central point in the distribution network for connecting multiple lines and electrical equipment. The load node is the point where the power user accesses the distribution network. The branch node is the point where the line branches off and is usually used to distribute power to different areas or users. This is easy for those skilled in the art to understand and will not be described in detail. The reference line is the line connecting two nodes and a single reference line only includes two nodes at both ends of the reference line.

[0056] The line risk coefficient is the average value of the sub-risk coefficients corresponding to each reference line. For a single reference line, this reference line is recorded as the target reference line. The calculation formula for the sub-risk coefficient w corresponding to the target reference line is: α i is the weight of the i-th influencing factor, and the weight corresponding to each influencing factor is 0.2, d i is the hazard reference value corresponding to the i-th influencing factor. The influencing factors include line length, historical faults, and load level. For n=3, i is 1, 2, and n is hazard reference value corresponding to line length = length of target reference line / average length of each reference line. The hazard reference value corresponding to historical faults = number of faults of target reference line in the past year / (average number of faults of each reference line in the past year + 1). The hazard reference value corresponding to load level = current current of target reference line / maximum current allowed when designing the reference line.

[0057] The values ​​of the preset line risk coefficient and the preset fault transient threshold can be determined by the user according to the actual application scenario. The smaller the values ​​of the preset line risk coefficient and the preset fault transient threshold, the greater the user's demand for feature line selection based on the line influence coefficient. A value of the preset line risk coefficient and the preset fault transient threshold is provided, and the historical records of the user's feature line selection based on the line influence coefficient are detected. The average value of the line risk coefficient corresponding to the historical records that can meet the user's needs is recorded as the preset line risk coefficient, and the average value of the fault transient threshold corresponding to the historical records that can meet the user's needs is recorded as the preset fault transient threshold;

[0058] Performing associated line selection based on line interaction, including: recording a reference line whose line influence coefficient is greater than a preset line influence coefficient as an analysis reference line, recording other reference lines other than the analysis reference line as reference lines to be analyzed, and using each analysis reference line and its associated lines as pre-fault lines, wherein the associated lines corresponding to a single analysis reference line are each reference line to be analyzed whose line interaction degree with the analysis reference line is greater than the preset line interaction degree;

[0059] The line influence coefficient corresponding to a single reference line = the sub-hazard coefficient corresponding to the reference line × the fault transient threshold corresponding to the reference line;

[0060] For any two reference lines, the calculation formula of line interaction degree r is:

[0061]

[0062] m is the number of time points in the current data corresponding to a single reference line, x k and y kare the current values ​​corresponding to the kth time point in the current data corresponding to the two reference lines, is x k The average value of the current corresponding to each time point in the current data corresponding to the corresponding reference line, y k The average value of the current value corresponding to each time point in the current data corresponding to the corresponding reference line, k = 1, 2, 3, ..., m; the current data corresponding to a single reference line is the current value monitored in real time on the reference line within 1 hour before the moment when the bus side zero-sequence voltage is greater than the preset bus side zero-sequence voltage, and the 1 hour before the moment when the bus side zero-sequence voltage is greater than the preset bus side zero-sequence voltage is recorded as the reference time period, with the starting time of the reference time period as the starting point, an interval point is set every 5 seconds, and the starting point and each interval point are recorded as time points;

[0063] The values ​​of the preset line influence coefficient and the preset line interactivity can be determined by the user according to the actual application scenario. The greater the user's demand for improving the line selection accuracy, the smaller the values ​​of the preset line influence coefficient and the preset line interactivity. The values ​​of the preset line influence coefficient and the preset line interactivity are provided, and the historical records of users selecting feature lines based on the line influence coefficient are detected. The average value of the line influence coefficient corresponding to each pre-fault line in the historical records that can meet the user's needs is recorded as the preset line influence coefficient, and the preset line interactivity is 0.7.

[0064] Specifically, if the distribution network status is that the line risk coefficient is less than the preset line risk coefficient and the fault transient threshold is less than the preset fault transient threshold, the line selection method is to perform feature line selection according to the line influence coefficient.

[0065] When performing feature line selection based on the line influence coefficient, a reference line whose line influence coefficient is greater than a preset line influence coefficient is used as a pre-fault line.

[0066] Specifically, the fault transient threshold is determined based on the line matching threshold and the change impact threshold;

[0067] The fault transient threshold is positively correlated with the line matching threshold and the change impact threshold.

[0068] Among them, fault transient threshold = line matching threshold × change impact threshold;

[0069] The change impact threshold is the average value of the transient reference values ​​corresponding to each reference line. The transient reference value corresponding to a single reference line = the maximum transient current value of the reference line when the zero-sequence voltage on the bus side is greater than the preset zero-sequence voltage on the bus side × the duration of the transient current of the reference line when the zero-sequence voltage on the bus side is greater than the preset zero-sequence voltage on the bus side. The transient current is measured by a transient recorder. This is easy for those skilled in the art to understand and will not be described in detail.

[0070] The line matching threshold is the average value of the matching reference values ​​corresponding to each reference line. The matching reference value corresponding to a single reference line is the average value of the line interaction degrees corresponding to the reference line and other reference lines.

[0071] Specifically, the number of empirical mode decompositions is determined according to the decomposition interaction degree, including:

[0072] If the decomposition interaction degree is greater than or equal to the preset decomposition interaction degree, the number of empirical mode decompositions is directly determined based on the historical difference coefficient and the iterative fluctuation value;

[0073] If the decomposition interaction degree is less than the preset decomposition interaction degree, the number of empirical mode decompositions is determined successively according to the amplitude threshold and the attenuation coefficient.

[0074] The decomposition interaction degree is determined by recording a single reference line as the target reference line and recording a reference line corresponding to a historical record that meets user requirements and has a confirmed number of empirical mode decompositions as a comparison reference line. The decomposition interaction degree is the maximum value of the sub-interaction coefficients corresponding to the target reference line and each comparison reference line. The sub-interaction coefficient corresponding to the target reference line and a single comparison reference line = 1 / (the absolute value of the difference between the line influence coefficients corresponding to the target reference line and the comparison reference line + 1).

[0075] The value of the preset decomposition interactivity can be determined by the user based on the actual application scenario. The smaller the value of the preset decomposition interactivity, the greater the user's demand for direct determination based on the historical difference coefficient and the iteration fluctuation value. A preset decomposition interactivity value is provided, and the historical records directly determined by the user based on the historical difference coefficient and the iteration fluctuation value are detected. The average value of the decomposition interactivity corresponding to the historical records that can meet the user's needs is recorded as the preset decomposition interactivity;

[0076] The number of EMD is the number of iterations of decomposing the original signal during the EMD process. An intrinsic mode function (IMF) is extracted in each iteration.

[0077] When the number of empirical mode decompositions is directly determined based on the historical difference coefficient and the iteration fluctuation value, the number of empirical mode decompositions is positively correlated with the evaluation threshold, where evaluation threshold = historical difference coefficient × iteration fluctuation value.

[0078] The comparison reference line corresponding to the maximum value of the sub-interaction coefficients between the target reference line and each comparison reference line is recorded as the target comparison reference line. The historical difference coefficient = 1 / the sub-interaction coefficient between the target reference line and the target comparison reference line. The iterative fluctuation value is the standard deviation of the fluctuation threshold corresponding to each iteration of the zero-sequence current signal corresponding to the target comparison reference line.

[0079] The fluctuation threshold corresponding to the h-th iteration of the zero-sequence current signal corresponding to the target comparison reference line = the Euclidean norm of the residual signal corresponding to the h-th iteration of the zero-sequence current signal corresponding to the target comparison reference line / the Euclidean norm of the initial zero-sequence current signal corresponding to the target comparison reference line;

[0080] The residual signal corresponding to the h-th iteration of the zero-sequence current signal corresponding to the target comparison reference line = the initial zero-sequence current signal corresponding to the target comparison reference line - the IMF corresponding to the 1st iteration - ... - the IMF corresponding to the h-th iteration;

[0081] When the number of empirical mode decompositions is determined successively according to the amplitude threshold and the attenuation coefficient, the iteration is stopped when the amplitude threshold is less than the preset amplitude threshold and the attenuation coefficient is less than the preset attenuation coefficient. The number of empirical mode decompositions is the number of iterations corresponding to the amplitude threshold being less than the preset amplitude threshold and the attenuation coefficient being less than the preset attenuation coefficient.

[0082] The amplitude threshold corresponding to the h-th iteration is the maximum value of the residual signal corresponding to the h-th iteration of the zero-sequence current signal corresponding to the target comparison reference line at each time point. The attenuation coefficient = the amplitude threshold corresponding to the h-th iteration / the amplitude threshold corresponding to the h-1-th iteration;

[0083] The values ​​of the preset amplitude threshold and the preset attenuation coefficient can be determined by the user according to the actual application scenario. The greater the user's demand for reducing the reconstruction error, the smaller the values ​​of the preset amplitude threshold and the preset attenuation coefficient. A value of the preset amplitude threshold and the preset attenuation coefficient is provided, and the historical records determined by the user based on the amplitude threshold and the attenuation coefficient are detected. The average value of the amplitude threshold of the last iteration corresponding to the historical records that can meet the user's needs is recorded as the preset amplitude threshold, and the average value of the attenuation coefficient of the last iteration corresponding to the historical records that can meet the user's needs is recorded as the preset attenuation coefficient.

[0084] Specifically, the calculation formula of sample entropy is: SampEn(m, t) = lnB m (t)-lnB m+1 (t).

[0085] Among them, in S1, let the original signal be X = {X1, X2, ……, XN}, and construct an m-dimensional vector: Yi = {Xi

[0086] , Xi+1, ……, Xi+m-1}, i = 1, 2, ……, N-m+1;

[0087] In S2, define the distance d[Yi, Yj] between vectors Yi and Yj as the largest one among the differences of their corresponding elements, that is, d[Yi, Yj] = max(|X(i+k)-X(j+k)|), i, j = 1, 2, ……, N-m+1; i≠j; k = 0, 1, ……, m-1;

[0088] In S3, given a threshold r (r>0), count the number of vector pairs that satisfy d<r, and calculate the ratio B of it to the total number of pairs i m The calculation formula is:

[0089] In S4, calculate the average value B of each m (t),

[0090] In S5, increase the vector dimension by 1, that is, form an m+1-dimensional vector and repeat steps S2 to S4 to calculate and obtain B m+1 (t).

[0091] Specifically, according to the entropy shift threshold, determine the construction method of the fault feature matrix, including:

[0092] If the entropy shift threshold is greater than or equal to the preset entropy shift threshold, the construction method of the fault feature matrix is to construct the fault feature matrix according to the effective sample entropy;

[0093] If the entropy shift threshold is less than the preset entropy shift threshold, the construction method of the fault feature matrix is to construct the fault feature matrix according to the IMF influence characterization value.

[0094] Among them, the entropy shift threshold is the standard deviation of the sample entropy corresponding to each IMF. The value of the preset entropy shift threshold can be determined by the user according to the actual application scenario. The smaller the value of the preset entropy shift threshold, the greater the user's need to construct the fault feature matrix according to the effective sample entropy. Provide a value of the preset entropy shift threshold, detect the historical records of the user constructing the fault feature matrix according to the effective sample entropy, and record the average value of the entropy shift threshold corresponding to the historical records that can meet the user's needs as the preset entropy shift threshold;

[0095] When constructing the fault feature matrix according to the effective sample entropy, use the IMF with sample entropy greater than the preset sample entropy to construct the fault feature matrix;

[0096] When constructing the fault feature matrix based on the IMF impact representation value, the IMF with the largest sample entropy is recorded as the target IMF, and the other IMFs outside the target IMF are recorded as reference IMFs. The reference IMFs whose IMF impact representation values ​​with the target IMF are greater than the preset IMF impact representation values ​​and the target IMF are used to construct the fault feature matrix;

[0097] The IMF impact characterization value is confirmed as follows: for any two IMFs, the IMF impact characterization value = (the larger of the number of iterations corresponding to the two IMFs - the smaller of the number of iterations corresponding to the two IMFs) / (the absolute value of the difference in sample entropy corresponding to the two IMFs + 1);

[0098] The values ​​of the preset sample entropy and the preset IMF influence characterization value can be determined by the user according to the actual application scenario. The greater the user's demand for enhancing the accuracy of fault feature expression, the greater the values ​​of the preset sample entropy and the preset IMF influence characterization value. A value of the preset sample entropy and the preset IMF influence characterization value is provided. The historical record of the user constructing the fault feature matrix based on the effective sample entropy is detected, and the average value of the sample entropy corresponding to each IMF in the fault feature matrix corresponding to the historical record that can meet the user's needs is recorded as the preset sample entropy. The historical record of the user constructing the fault feature matrix based on the IMF influence characterization value is detected, and the average value of the reference IMF influence characterization value corresponding to each fault feature matrix in the historical record that can meet the user's needs is recorded as the preset IMF influence characterization value; the reference IMF influence characterization value corresponding to a single fault feature matrix is ​​the average value of the IMF influence characterization values ​​corresponding to each reference IMF and the target IMF in the fault feature matrix.

[0099] Specifically, the basic classification method is determined according to the matrix characteristic coefficient and the matrix attribution coefficient, including:

[0100] If the matrix characteristic coefficient is greater than or equal to the preset matrix characteristic coefficient or the matrix attribution coefficient is less than the preset matrix attribution coefficient, the base classification method is to select a base classifier according to the conversion correlation degree;

[0101] If the matrix characteristic coefficient is less than the preset matrix characteristic coefficient and the matrix attribution coefficient is greater than or equal to the preset matrix attribution coefficient, the base classification method is to select a base classifier according to the simulated matching coefficient.

[0102] Wherein, matrix characteristic coefficient = (number of IMFs contained in a single fault feature matrix × average value of sample entropy corresponding to each IMF contained in a single fault feature matrix) / average value of matching coefficient corresponding to each IMF in a single fault feature matrix; the matching coefficient corresponding to a single IMF is the average value of the IMF influence characterization value corresponding to the IMF and other IMFs in the fault feature matrix;

[0103] For a single fault feature matrix, record it as the target fault feature matrix. The matrix attribution coefficient corresponding to the target fault feature matrix is ​​the standard deviation of the sub-attribution coefficients corresponding to each fault feature matrix in the reference historical records. Record the historical records that can meet user needs as the reference historical records. The sub-attribution coefficient corresponding to a single fault feature matrix in the reference historical records = 1 / [|the matrix characteristic coefficient corresponding to the fault feature matrix in the reference historical records - the matrix characteristic coefficient corresponding to the target fault feature matrix| + 1];

[0104] The values ​​of the preset matrix characteristic coefficient and the preset matrix attribution coefficient can be determined by the user according to the actual application scenario. The smaller the value of the preset matrix characteristic coefficient and the larger the value of the preset matrix attribution coefficient, the greater the user's need to select a base classifier according to the conversion correlation. A value of a preset matrix characteristic coefficient and a preset matrix attribution coefficient is provided, and the historical records of the user selecting the base classifier according to the conversion correlation are detected. The average value of the matrix characteristic coefficients corresponding to the historical records that can meet the user's needs is recorded as the preset matrix characteristic coefficient, and the average value of the matrix attribution coefficients corresponding to the historical records that can meet the user's needs is recorded as the preset matrix attribution coefficient.

[0105] When selecting a base classifier according to the conversion correlation, a base classifier having a conversion correlation greater than a preset conversion correlation is selected as the selected base classifier;

[0106] The conversion correlation degree corresponding to a single base classifier = the standard deviation of the sub-attribution coefficients corresponding to each fault feature matrix selected as the selected base classifier in the reference history / [|the matrix feature coefficient corresponding to the target fault feature matrix - the average value of the matrix feature coefficients corresponding to each fault feature matrix selected as the selected base classifier in the reference history| + 1];

[0107] When selecting a base classifier according to the simulation matching coefficient, a base classifier having a simulation matching coefficient greater than a preset simulation matching coefficient is selected as the selected base classifier;

[0108] The base classifier is selected from the detection reference history as the fault feature matrix of the selected base classifier and recorded as the analysis fault feature matrix. The simulation matching coefficient corresponding to a single base classifier = (the number of analysis fault feature matrices / the number of fault feature matrices in all historical records that select the base classifier as the selected base classifier) ​​× conversion correlation degree;

[0109] Base classifiers include but are not limited to Extra Trees Classifier, Random Forest, GradientBoosting Classifier, and Adaptive Boosting;

[0110] The values ​​of the preset conversion correlation degree and the preset simulation matching coefficient can be determined by the user according to the actual application scenario. The greater the user's demand for improving the preliminary classification accuracy of the base classifier, the larger the values ​​of the preset conversion correlation degree and the preset simulation matching coefficient. A value of a preset conversion correlation degree and a preset simulation matching coefficient is provided. The historical record of the user selecting the base classifier according to the conversion correlation degree is detected, and the average value of the conversion correlation degree corresponding to each selected base classifier in the historical record that can meet the user's needs is recorded as the preset conversion correlation degree. The historical record of the user selecting the base classifier according to the simulation matching coefficient is detected, and the average value of the simulation matching coefficient corresponding to each selected base classifier in the historical record that can meet the user's needs is recorded as the preset simulation matching coefficient.

[0111] Specifically, if the base classification defect coefficient is greater than or equal to the preset base classification defect coefficient and the evaluation instability threshold is greater than or equal to the preset evaluation instability threshold, the optimization method is to increase the number of cross-validation adjustments;

[0112] The increase in the number of cross-validation times is positively correlated with the comprehensive evaluation coefficient.

[0113] The base classification defect coefficient is the standard deviation of the sub-defect coefficients corresponding to each base classifier. The sub-defect coefficient corresponding to a single base classifier = the average defect probability values ​​corresponding to each pre-fault line obtained using the base classifier / the standard deviation of the defect probability values ​​corresponding to each pre-fault line obtained using the base classifier. The defect probability value corresponding to a single pre-fault line obtained using the base classifier is the probability value determined by the base classifier that the pre-fault line is a faulty line.

[0114] The assessment instability threshold is the maximum value of the sub-instability thresholds corresponding to each pre-fault line. The sub-instability threshold corresponding to a single pre-fault line = |(defect probability value corresponding to the pre-fault line / average of defect probability values ​​corresponding to each pre-fault line) - (transient reference value corresponding to the pre-fault line / average of transient reference values ​​corresponding to each pre-fault line)|.

[0115] The values ​​of the preset base classification defect coefficient and the preset evaluation instability threshold can be determined by the user according to the actual application scenario. The smaller the values ​​of the preset base classification defect coefficient and the preset evaluation instability threshold, the greater the user's need to increase the number of cross-validation adjustments. A preset base classification defect coefficient and a preset evaluation instability threshold are provided, and the historical records of the user increasing the number of cross-validation adjustments are detected. The average value of the base classification defect coefficient corresponding to the historical records that can meet the user's needs is recorded as the preset base classification defect coefficient, and the average value of the evaluation instability threshold corresponding to the historical records that can meet the user's needs is recorded as the preset evaluation instability threshold;

[0116] The number of cross-validation times n is the number of non-overlapping subsets into which the entire dataset is randomly and evenly divided. In each training, (n-1) subsets are used as training sets, and the remaining subset is used as validation sets. Each base classifier is trained using the training set and predicted on the validation set to generate a defect probability value.

[0117] The increase in the number of cross-validation times is the smallest integer greater than or equal to M1, where M1 = [comprehensive evaluation coefficient / (preset base classification defect coefficient × preset evaluation instability threshold + 1)] + 1; comprehensive evaluation coefficient = base classification defect coefficient × evaluation instability threshold; the initial number of cross-validation times in the present invention is 5;

[0118] It should be noted that when the assessment instability threshold is less than the preset assessment instability threshold, no adjustment is required.

[0119] Specifically, if the base classification defect coefficient is less than the preset base classification defect coefficient and the assessment instability threshold is greater than or equal to the preset assessment instability threshold, the optimization method is to increase the dimension of the fault feature matrix;

[0120] The increase in the dimension of the fault feature matrix is ​​positively correlated with the assessment instability threshold.

[0121] Among them, the fault feature matrix dimension is the number of IMF components selected in a single fault feature matrix;

[0122] The increase value of the fault feature matrix dimension is the smallest integer greater than or equal to M2, where M2=(assessed instability threshold / preset assessed instability threshold)+1.

[0123] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

[0124] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A 10KV distribution network ground fault line selection method based on ICEEMDAN and Stacking ensemble learning, characterized by: include: When the bus side zero sequence voltage is greater than the preset bus side zero sequence voltage, the distribution network state is determined according to the line risk factor and the fault transient threshold. The line selection method is determined according to the distribution network state to obtain the pre-fault line. The line selection method is feature selection based on the line influence coefficient or association selection based on the line interaction degree. The zero-sequence current signal corresponding to the pre-fault line is collected, the number of empirical mode decompositions is determined according to the decomposition interaction degree, and the zero-sequence current signal is decomposed based on the number of empirical mode decompositions to obtain IMF; Detect the sample entropy corresponding to each IMF, and determine the fault feature matrix construction method based on the entropy value deviation threshold: construct the fault feature matrix based on the effective sample entropy or IMF impact representation value; The fault feature matrix is ​​input into the pre-trained Stacking ensemble learning model. The base classification method is determined based on the matrix feature coefficient and the matrix attribution coefficient. The base classifier is selected based on the conversion correlation degree or the simulation matching coefficient. According to the base classification defect coefficient and the evaluation instability threshold, the optimization method is determined to adjust the fault feature matrix dimension or the number of cross-validation times; The output results of the base classifier are concatenated as the input features of the meta-classifier to select the faulty lines.

2. The 10KV distribution network ground fault line selection method based on ICEEMDAN and Stacking integrated learning according to claim 1 is characterized in that: If the distribution network status is that the line risk coefficient is greater than or equal to the preset line risk coefficient or the fault transient threshold is greater than or equal to the preset fault transient threshold, the line selection method is to perform associated line selection based on the line interaction degree.

3. The 10KV distribution network ground fault line selection method based on ICEEMDAN and Stacking integrated learning according to claim 2 is characterized in that: If the distribution network status is that the line risk coefficient is less than the preset line risk coefficient and the fault transient threshold is less than the preset fault transient threshold, the line selection method is to perform feature line selection based on the line influence coefficient.

4. The 10KV distribution network ground fault line selection method based on ICEEMDAN and Stacking integrated learning according to claim 3 is characterized in that: The fault transient threshold is determined according to the line matching threshold and the change impact threshold; The fault transient threshold is positively correlated with the line matching threshold and the change impact threshold.

5. The 10KV distribution network ground fault line selection method based on ICEEMDAN and Stacking integrated learning according to claim 4 is characterized in that: The number of empirical mode decompositions is determined based on the decomposition interaction degree, including: If the decomposition interaction degree is greater than or equal to the preset decomposition interaction degree, the number of empirical mode decompositions is directly determined based on the historical difference coefficient and the iterative fluctuation value; If the decomposition interaction degree is less than the preset decomposition interaction degree, the number of empirical mode decompositions is determined successively according to the amplitude threshold and the attenuation coefficient.

6. The 10KV distribution network ground fault line selection method based on ICEEMDAN and Stacking integrated learning according to claim 5 is characterized in that: The calculation formula of sample entropy is: SampEn(m, t) = lnB m (t)-lnB m+1 (t).

7. The 10KV distribution network ground fault line selection method based on ICEEMDAN and Stacking integrated learning according to claim 6 is characterized in that: The fault feature matrix construction method is determined based on the entropy value deviation threshold, including: If the entropy value deviation threshold is greater than or equal to the preset entropy value deviation threshold, the fault feature matrix is ​​constructed in the manner of constructing the fault feature matrix according to the effective sample entropy; If the entropy value deviation threshold is less than the preset entropy value deviation threshold, the fault feature matrix is ​​constructed in a manner of constructing the fault feature matrix according to the IMF impact characterization value.

8. The 10KV distribution network ground fault line selection method based on ICEEMDAN and Stacking integrated learning according to claim 7 is characterized in that: The basic classification method is determined according to the matrix characteristic coefficient and the matrix attribution coefficient, including: If the matrix characteristic coefficient is greater than or equal to the preset matrix characteristic coefficient or the matrix attribution coefficient is less than the preset matrix attribution coefficient, the base classification method is to select a base classifier according to the conversion correlation degree; If the matrix characteristic coefficient is less than the preset matrix characteristic coefficient and the matrix attribution coefficient is greater than or equal to the preset matrix attribution coefficient, the base classification method is to select a base classifier according to the simulated matching coefficient.

9. The 10KV distribution network ground fault line selection method based on ICEEMDAN and Stacking integrated learning according to claim 8 is characterized in that: If the base classification defect coefficient is greater than or equal to the preset base classification defect coefficient and the evaluation instability threshold is greater than or equal to the preset evaluation instability threshold, the optimization method is to increase the number of cross-validation adjustments; The increase in the number of cross-validation times is positively correlated with the comprehensive evaluation coefficient.

10. The 10KV distribution network ground fault line selection method based on ICEEMDAN and Stacking integrated learning according to claim 9 is characterized in that: If the base classification defect coefficient is less than the preset base classification defect coefficient and the assessment instability threshold is greater than or equal to the preset assessment instability threshold, the optimization method is to increase the dimension of the fault feature matrix; The increase in the dimension of the fault feature matrix is ​​positively correlated with the assessment instability threshold.

Images