Power distribution network fault detection method and device based on variational mode decomposition

Abstract

The embodiment of the invention discloses a power distribution network fault detection method and device based on variational mode decomposition, and the method comprises the steps: carrying out the collection and calculation of a transient part of the zero-sequence voltage of a power distribution network according to a control strategy for suppressing the negative-sequence current, and carrying out the decomposition of a transient signal according to a variational mode decomposition algorithm, thereby achieving the fault detection of the zero-sequence voltage of the power distribution network. Calculating the energy distribution proportion of the modal component in the transient signal, calculating the energy entropy according to the calculation result, and calculating the amplitude mean value of the zero sequence voltage time domain signal; constructing a feature vector by using a result obtained by calculation, and evaluating the support vector machine model; and identifying the characteristics of the zero-sequence voltage by using the evaluated support vector machine model to complete the fault detection of the power distribution network. According to the invention, the method achieves the recognition of the high-resistance fault through the zero-sequence voltage signal, guarantees the recognition accuracy of the fault type, and improves the adaptability and interpretability of a system.

Description

Technical Field

[0001] The present disclosure relates to the field of distribution network fault detection, and in particular, to a distribution network fault detection method and device based on variational mode decomposition. Background Art

[0002] With the wide application of distributed power sources in the distribution network, the scale of the distribution network system is becoming larger and the operating environment is becoming more complex, which makes the detection of high-resistance faults an increasingly important issue. High-resistance faults are usually caused by non-conductive media such as branches and soil. Due to the high transition resistance, the amplitudes of the fault current and voltage signals are small, resulting in very weak fault characteristics and being easily affected by harmonics, noise, etc. Therefore, high-resistance ground faults are more difficult to detect, leading to the continuous existence of faults, and ultimately causing serious accidents such as electric shock and fire.

[0003] Traditional fault detection methods, such as time-domain, frequency-domain, and time-frequency domain methods, all determine by extracting the characteristic quantities of the fault signal and setting thresholds. However, these methods have certain limitations. Especially in high-resistance faults, due to the small signal amplitude and being easily affected by external interference, the anti-interference ability and adaptability of the above methods are poor, and false negatives or false positives are likely to occur. In addition, other studies have combined deep learning methods such as convolutional neural networks (CNNs) to further improve the accuracy of feature extraction. However, although the neural network-based algorithm can improve the accuracy of the model, it requires a large amount of training data, which may lead to a decline in the performance of the model in a small-sample environment for this relatively rare fault type of high-resistance faults.

[0004] Therefore, one or more methods are needed to solve the above problems.

[0005] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of the present disclosure, and thus may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present disclosure is to provide a distribution network fault detection method and device based on variational mode decomposition, thereby at least to a certain extent overcoming one or more problems caused by the limitations and defects of the related art.

[0007] According to one aspect of the present disclosure, a distribution network fault detection method based on variational mode decomposition is provided, including:

[0008] According to the negative-sequence current suppression control strategy, by collecting and calculating the transient part of the zero-sequence voltage of the distribution network, a transient signal is generated;

[0009] According to the variational mode decomposition algorithm, by performing variational mode decomposition on the transient signal, modal components are generated;

[0010] According to the orthogonality of variational mode decomposition, by calculating the proportion of the energy distribution of the modal components in the transient signal, a variational mode energy distribution is generated;

[0011] According to the modal energy distribution, by calculating the energy entropy value of variational mode decomposition, a variational mode energy entropy is generated;

[0012] According to the transient signal, by calculating the amplitude mean value of the time-domain signal of the zero-sequence voltage, a transient signal amplitude mean value is generated;

[0013] Using the variational mode energy distribution, variational mode energy entropy, and transient signal amplitude mean value to construct a feature vector, and evaluating a support vector machine model through the feature vector;

[0014] Using the evaluated support vector machine model to identify the transient characteristics of the zero-sequence voltage, and completing the detection of distribution network faults.

[0015] In an exemplary embodiment of the present disclosure, by collecting and calculating the transient part of the zero-sequence voltage of the distribution network, it includes:

[0016] Based on the negative-sequence current suppression control strategy, by calculating the component reference values of the positive-sequence current on the direct axis and cross axis of the inverter when a single-phase high-resistance grounding fault occurs in the distribution network, the circuit topology analysis of the zero-sequence component during the fault is completed;

[0017] Based on the circuit topology analysis of the zero-sequence component during the fault, by equivalently setting the fault point as a virtual power source, a high-resistance fault equivalent circuit is constructed;

[0018] Based on the high-resistance fault equivalent circuit, by collecting and calculating the zero-sequence voltage in the circuit, a zero-sequence voltage signal is generated;

[0019] Based on the zero-sequence voltage signal, by collecting the transient part of the zero-sequence voltage through a preset time window, a transient signal is generated.

[0020] In an exemplary embodiment of the present disclosure, by performing variational mode decomposition on the transient signal, it includes:

[0021] Based on the variational mode decomposition algorithm, by constructing the constrained variational problem of the transient signal, a transient signal constrained variational problem is generated;

[0022] Based on the preset distortion degree limit, by introducing a noise tolerance to solve the Lagrangian operator, a Lagrangian operator is generated;

[0023] In an exemplary embodiment of the present disclosure, performing variational mode decomposition on the transient signal includes:

[0024] Based on the Lagrangian algorithm, by introducing the Lagrangian operator, transforming the constrained variational problem of the transient signal, and generating an unconstrained variational problem;

[0025] Using an iterative algorithm, by introducing a quadratic penalty factor, performing a Fourier isometric transform on the unconstrained variational problem, and starting to iterate on the components of the modal signal decomposition;

[0026] When the components of the modal signal decomposition are not less than the preset accuracy, it is determined that the iteration does not terminate;

[0027] When the components of the modal signal decomposition are less than the preset accuracy, it is determined that the iteration terminates, outputting the iteration result and generating modal components.

[0028] In an exemplary embodiment of the present disclosure, constructing a feature vector by using the variational mode energy distribution, variational mode energy entropy, and transient signal amplitude mean includes:

[0029] Using the variational mode energy distribution, variational mode energy entropy, and transient signal amplitude mean, constructing the features of the input samples of the support vector machine to generate a feature vector;

[0030] Based on the feature vector, preparing the training set of the support vector machine model to generate a training data set;

[0031] Based on the feature vector, preparing the test set of the support vector machine model to generate a test data set;

[0032] The ratio of the training data set to the test data set is 2:1.

[0033] In an exemplary embodiment of the present disclosure, evaluating the support vector machine model by using the feature vector includes:

[0034] Based on the support vector machine model, mapping the training data set to a high-dimensional space, and using a hyperplane to classify and separate the training data set in the mapped high-dimensional space to generate a classification interval constraint problem;

[0035] Based on the Lagrangian algorithm, by introducing Lagrangian multipliers, transforming the classification interval constraint problem to generate a classification interval optimization problem;

[0036] Based on the classification interval optimization problem, taking the partial derivative of the Lagrangian function and setting the result to zero to obtain a dual problem.

[0037] In an exemplary embodiment of the present disclosure, evaluating the support vector machine model through the feature vector includes:

[0038] Calculating the parameters of the support vector machine model by solving the dual problem to generate the support vector machine model parameters;

[0039] Based on the support vector machine model parameters, constructing a decision function, introducing the test data set into the decision function, and solving the decision function to complete the evaluation of the support vector machine model.

[0040] In an exemplary embodiment of the present disclosure, identifying the transient characteristics of the zero-sequence voltage by using the evaluated support vector machine model to complete the detection of distribution network faults includes:

[0041] Taking the transient characteristics of the zero-sequence voltage as input data, inputting them into the evaluated support vector machine model, and solving the decision function to generate a decision result;

[0042] When the decision result falls into the preset high-resistance fault decision interval, it is determined that a high-resistance grounding fault has occurred in the distribution network;

[0043] When the decision result falls into the preset line interference decision interval, it is determined that the distribution network is interfered;

[0044] When the decision result falls into the preset line normal decision interval, it is determined that the line of the distribution network is operating normally.

[0045] In an exemplary embodiment of the present disclosure, it further includes:

[0046] Using the local interpretability analysis algorithm to perform interpretability analysis on the trained support vector machine model to generate a preset variable analysis report of the working condition identification for the support vector machine model;

[0047] Using the local interpretability analysis algorithm to perform adaptability analysis on the fault scenario to generate an analysis report on the impact of the fault scenario on the detection effect.

[0048] In one aspect of the present disclosure, a distribution network fault detection device based on variational mode decomposition is provided, including:

[0049] A transient signal acquisition module, configured to collect and calculate the transient part of the zero-sequence voltage of the distribution network according to the negative-sequence current suppression control strategy to generate a transient signal;

[0050] A variational mode decomposition module, configured to perform variational mode decomposition on the transient signal according to the variational mode decomposition algorithm to generate modal components;

[0051] An energy distribution ratio calculation module, configured to calculate the energy distribution ratio of the modal components in the transient signal according to the orthogonality of variational mode decomposition, and generate a variational mode energy distribution;

[0052] An energy entropy value calculation module, configured to calculate the energy entropy value of variational mode decomposition according to the modal energy distribution, and generate a variational mode energy entropy;

[0053] A time-domain signal amplitude mean value calculation module, configured to calculate the amplitude mean value of the time-domain signal of the zero-sequence voltage according to the transient signal, and generate a transient signal amplitude mean value;

[0054] A feature vector construction module, configured to construct a feature vector by using the variational mode energy distribution, the variational mode energy entropy, and the transient signal amplitude mean value, and evaluate a support vector machine model through the feature vector;

[0055] A fault detection module, configured to identify the transient characteristics of the zero-sequence voltage by using the evaluated support vector machine model, and complete the detection of the distribution network fault.

[0056] A distribution network fault detection method based on variational mode decomposition in an exemplary embodiment of the present disclosure: First, according to the negative-sequence current suppression control strategy, the transient part of the zero-sequence voltage of the distribution network is collected and calculated. Then, according to the variational mode decomposition algorithm, the transient signal is decomposed, and then the energy distribution ratio of the modal components in the transient signal is calculated, and the energy entropy value is calculated according to the calculation result. At the same time, the amplitude mean value of the zero-sequence voltage time-domain signal is calculated. Then, a feature vector is constructed by using the calculated results, and the support vector machine model is evaluated. Finally, the characteristics of the zero-sequence voltage are identified by using the evaluated support vector machine model, and the detection of the distribution network fault is completed. The embodiment of the present disclosure provides a solution for decomposing the zero-sequence voltage based on the variational mode decomposition algorithm, and using the decomposed parameters as feature vectors to evaluate the support vector machine model for fault detection. This solution realizes the identification of high-resistance faults by using the zero-sequence voltage signal, ensures the accuracy of identifying the fault type, and improves the adaptability and interpretability of the system.

[0057] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. Brief Description of the Drawings

[0058] By referring to the accompanying drawings to describe its exemplary embodiments in detail, the above and other features and advantages of the present disclosure will become more obvious.

[0059] Figure 1Shows a flowchart of a distribution network fault detection method based on variational mode decomposition according to an exemplary embodiment of the present disclosure;

[0060] Figure 2 Shows a fault detection flowchart of a distribution network fault detection method based on variational mode decomposition according to an exemplary embodiment of the present disclosure;

[0061] Figure 3 Shows an equivalent circuit of high - resistance fault of a distribution network fault detection method based on variational mode decomposition according to an exemplary embodiment of the present disclosure;

[0062] Figure 4 Shows a flowchart of variational mode decomposition algorithm of a distribution network fault detection method based on variational mode decomposition according to an exemplary embodiment of the present disclosure;

[0063] Figure 5 Shows a distribution diagram of VMD energy under various working conditions of a distribution network fault detection method based on variational mode decomposition according to an exemplary embodiment of the present disclosure;

[0064] Figure 6 Shows a comparison diagram of characteristic types of a distribution network fault detection method based on variational mode decomposition according to an exemplary embodiment of the present disclosure;

[0065] Figure 7 Shows a schematic diagram of the principle of support vector machine of a distribution network fault detection method based on variational mode decomposition according to an exemplary embodiment of the present disclosure;

[0066] Figure 8 Shows a topology diagram of a non - grounded neutral distribution network of a distribution network fault detection method based on variational mode decomposition according to an exemplary embodiment of the present disclosure;

[0067] Figure 9 Shows a confusion matrix diagram of fault determination by support vector machine model simulation of a distribution network fault detection method based on variational mode decomposition according to an exemplary embodiment of the present disclosure;

[0068] Figure 10 Shows a schematic block diagram of a distribution network fault detection device based on variational mode decomposition according to an exemplary embodiment of the present disclosure. Detailed implementation manners

[0069] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art. Like reference numerals in the figures denote like or similar parts, and thus their repetitive description will be omitted.

[0070] In addition, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of the present disclosure. However, those skilled in the art will realize that the technical solutions of the present disclosure may be practiced without one or more of the specific details, or other methods, components, materials, devices, steps, etc. may be employed. In other cases, well-known structures, methods, devices, implementations, materials, or operations are not shown or described in detail to avoid obscuring aspects of the present disclosure.

[0071] The block diagrams shown in the drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more software-hardened modules, or in different networks and / or processor devices and / or microcontroller devices.

[0072] In the present example embodiment, first, a method for detecting faults in a distribution network based on variational mode decomposition is provided; as shown in Figure 1 this method for detecting faults in a distribution network based on variational mode decomposition may include the following steps:

[0073] Step S110, according to the negative-sequence current suppression control strategy, generate a transient signal by collecting and calculating the transient part of the zero-sequence voltage of the distribution network;

[0074] Step S12, according to the variational mode decomposition algorithm, generate modal components by performing variational mode decomposition on the transient signal;

[0075] Step S130, according to the orthogonality of variational mode decomposition, generate the variational mode energy distribution by calculating the energy distribution ratio of the modal components in the transient signal;

[0076] Step S140, according to the modal energy distribution, generate the variational mode energy entropy by calculating the energy entropy value of variational mode decomposition;

[0077] Step S150, according to the transient signal, generate the transient signal amplitude mean by calculating the amplitude mean of the time-domain signal of the zero-sequence voltage;

[0078] Step S160: Construct a feature vector using the variational mode energy distribution, variational mode energy entropy, and transient signal amplitude mean, and evaluate the support vector machine model using the feature vector.

[0079] Step S170: Identify the transient characteristics of the zero-sequence voltage using the evaluated support vector machine model to complete the detection of distribution network faults.

[0080] A distribution network fault detection method based on variational mode decomposition and support vector machine in an exemplary embodiment of the present disclosure: First, according to the negative-sequence current suppression control strategy, collect and calculate the transient part of the zero-sequence voltage of the distribution network. Then, according to the variational mode decomposition algorithm, decompose the transient signal, calculate the energy distribution ratio of the modal components in the transient signal, and calculate the energy entropy value based on the calculation result. At the same time, calculate the amplitude mean of the zero-sequence voltage time-domain signal. Then, construct a feature vector using the calculated results and evaluate the support vector machine model. Finally, identify the characteristics of the zero-sequence voltage using the evaluated support vector machine model to complete the detection of distribution network faults. The embodiment of the present disclosure realizes the identification of high-resistance faults using the zero-sequence voltage signal to ensure the accuracy of fault type identification and improve the adaptability and interpretability of the system.

[0081] Next, as Figure 2 shown, a distribution network fault detection method based on variational mode decomposition in this exemplary embodiment will be further described.

[0082] In the template configuration step S110, according to the negative-sequence current suppression control strategy, the transient part of the zero-sequence voltage of the distribution network can be collected and calculated to generate a transient signal.

[0083] Currently, the voltage transformers installed in the distribution network are relatively sensitive. When a single-phase high-resistance grounding fault occurs on the line, the voltage transformer can collect obvious transient signals of the zero-sequence voltage. Therefore, the transient characteristics of the zero-sequence voltage after a single-phase high-resistance grounding fault in a distribution network with distributed power sources can be analyzed, and this characteristic can lay a theoretical foundation for the high-resistance fault detection method.

[0084] In the embodiment of this example, distributed power sources using PQ control (where "P" and "Q" in PQ control refer to active power and reactive power respectively) usually adopt the negative-sequence current suppression control strategy during grid faults. Therefore, based on this control strategy, it is considered that the inverter-type power source only outputs positive-sequence current during faults.

[0085] Then, the current differential equations of the inverter dq axis (direct axis and cross axis) during the fault can be expressed as:

[0086]

[0087] Solving the differential equation for Equation (1), the direct-axis and cross-axis current expressions can be obtained:

[0088]

[0089] where A 1 , A 2 , B 1 , B 2 are undetermined coefficients, and i d * and i q * are the reference values of the direct-axis and cross-axis components of the positive-sequence current.

[0090] Thus, the circuit topology analysis of the zero-sequence component during a fault is completed. That is, when a single-phase high-resistance grounding fault occurs in the distribution network, the distributed voltage in the microgrid will not output a zero-sequence component to the power grid.

[0091] As Figure 3 shown, based on the results of the above circuit topology analysis, an equivalent circuit for high-resistance faults in an ungrounded neutral system is constructed. To simplify the ungrounded neutral system model, the fault point can be equivalent to a virtual power source where U m is the phase voltage amplitude of the fault phase during normal operation, ω is the power frequency angular frequency, is the initial fault phase angle. And in Figure 3 , R h is the resistance between the neutral point and the ground; I 0n is the zero-sequence current flowing through the outlet of the fault line, I 0f is the zero-sequence current flowing through the fault point; C j (j = 1, 2,... n) is the zero-sequence distributed capacitance to the ground of the jth feeder; I cj represents the current flowing through the zero-sequence distributed capacitance Cj of the jth feeder to the ground; u c is the zero-sequence voltage of the bus, u 0f is the zero-sequence voltage of the fault point.

[0092] The acquisition of the zero-sequence voltage signal can be completed from this equivalent circuit, and the specific steps are as follows:

[0093] First step, the first-order differential equation expression of the virtual power source is obtained as:

[0094] u f = u c + R(i cΣ + i 0z ) (3)

[0095] where, i c∑ is the sum of the currents of all the feeder-to-ground capacitances, i 0zThe zero-sequence current flowing into the earth through the neutral point.

[0096] Second, for simplicity of calculation, assume that the resistance R of the neutral point to the ground in the ungrounded neutral system h is approximately equal to 0, then i 0z tends to 0, and we can get:

[0097]

[0098] where C ∑ is the sum of the zero-sequence distributed capacitances of all feeders to the ground.

[0099] Third, according to Equation (4), transform Equation (3):

[0100]

[0101] Fourth, solve Equation (5), and the zero-sequence voltage u of the bus can be obtained c :

[0102]

[0103] where e is the natural logarithm, ω is the power frequency angular frequency, and the parameters θ and δ in Equation (6) are respectively:

[0104]

[0105] Finally, through the above steps of analysis, it is concluded that: when a high-resistance fault occurs in an ungrounded neutral system, its zero-sequence voltage and zero-sequence current are both composed of the superposition of a DC component and a steady-state sinusoidal component. And as the transition resistance R increases, the steady-state voltage, transient voltage, and the initial amplitude of the current component will all decrease accordingly. According to Equation (5), it can be seen that as the transition resistance increases, the attenuation factor δ will continuously decrease, thereby making the attenuation speed slower and slower.

[0106] Then in this example, by setting up a 500 us time window to obtain the transient signal of the zero-sequence voltage signal and extracting and comparing its time-domain characteristics and frequency-domain characteristics, the accurate identification of high-resistance faults can be achieved.

[0107] In the template configuration step S120, according to the variational mode decomposition algorithm, by performing variational mode decomposition on the transient signal, modal components are generated.

[0108] In the embodiment of this example, as Figure 4As shown, to extract the time-domain and frequency-domain features of transient signals, it is necessary to perform variational mode decomposition (VMD, Variational Mode Decomposition) on the original transient signals and decompose them into K intrinsic mode function (IMF) components.

[0109] First step, the constrained variational problem can be constructed as shown in Equation (8):

[0110]

[0111] where K is the total number of modes to be decomposed, is the partial derivative with respect to time t, {u k} is the k-th component after decomposition, {ω k} is the central frequency of the k-th component, and s.t. is the constraint condition that the sum of all {u k} needs to be f.

[0112] Second step, to solve Equation (8), the Lagrange operator λ is introduced to transform it into an unconstrained variational problem, and we get:

[0113]

[0114] where, in order to limit the distortion degree of the decomposed signal within a certain range, the noise tolerance γ needs to be introduced. Thus, the solution process of the Lagrange operator λ is as shown in Equation (10):

[0115]

[0116] Third step, a quadratic penalty factor α is introduced in Equation (9) to reduce the Gaussian noise interference in the signal. Then, Fourier isometric transformation is performed on u k (t), ω(t) and λ(t) (that is, at time t, Fourier isometric transformation is performed on the k-th component, the k + 1-th component, the power frequency angular frequency, and the Lagrange operator). Subsequently, the optimal components and central frequencies of the modal signal decomposition are calculated using the iterative algorithm, and the solution processes are as shown in Equations (11) and (12):

[0117]

[0118]

[0119] Fourth step, a preset accuracy ε is set, as shown in Equation (13):

[0120]

[0121] In the fifth step, compare and determine the calculation result after the third - step iteration with the preset precision ε. If the precision requirement of Equation (13) is met, determine that the iteration terminates, output the result for plotting, and generate modal components. If the precision requirement is not met, return to step (11) to continue the next - iteration calculation.

[0122] In template configuration steps S130 and S140, according to the orthogonality of variational mode decomposition, by calculating the proportion of the energy distribution of the modal components in the transient signal, a variational - mode energy distribution is generated. And, according to the modal - energy distribution, by calculating the energy - entropy value of variational mode decomposition, a variational - mode energy entropy is generated.

[0123] In the embodiment of this example, when a line fails or is disturbed, the frequency distribution of the zero - sequence voltage changes, so the energy distribution of its signal also changes accordingly. Therefore, the concept of energy entropy can be introduced to study the energy distribution of each IMF obtained after VMD decomposition.

[0124] Based on the K IMF components obtained by performing variational mode decomposition on the original transient signal as described above, the energy of the i - th IMF component can be expressed as:

[0125]

[0126] K is the total number of sampling points of the i - th IMF signal, and k(j) is the j - th sampling point of the i - th IMF signal.

[0127] Since VMD decomposition has orthogonality, the sum of the energies of the k IMFs is equal to the total energy E of the original transient signal. Thus, the definition of the VMD energy - entropy value can be obtained:

[0128]

[0129] In the formula, p i represents the proportion of the energy of the i - th IMF signal in the total energy E, that is, the energy distribution of the i - th modal component in the total signal after VMD decomposition, which can be expressed as:

[0130]

[0131] In template configuration step S150, according to the transient signal, by calculating the amplitude mean value of the time - domain signal of the zero - sequence voltage, a transient - signal amplitude mean value is generated.

[0132] In the embodiment of this example, as Figure 5 shown, when the line is operating normally, under high - impedance faults, noise and harmonic interference, asymmetric load switching, etc., the VMD modal - energy distribution of the zero - sequence voltage of the distribution network is different.

[0133] Among them, the VMD energy distribution characteristics of high-resistance faults, unbalanced loads, and normal operations are relatively similar. In particular, the VMD energy distribution characteristics of high-resistance faults and unbalanced loads are extremely similar. Thus, using only the VMD energy distribution as a feature vector may result in a relatively low classification and recognition accuracy of the SVM.

[0134] At the same time, since the zero-sequence voltage amplitudes during high-resistance faults and unbalanced load switching are much larger than those during normal operations, in order to better distinguish the transient characteristics of the above conditions, the mean value of the transient signal amplitudes of the zero-sequence voltage time-domain signal can be calculated:

[0135]

[0136] where K is the total number of sampling points within a 500 μs time window, and v 0 (k) is the k-th sampling point. At the same time, the transient process of high-resistance faults has more capacitive factors, while the transient process of unbalanced loads has more inductive factors. Therefore, there are also significant differences in the VMD energy entropy characteristics of their transient processes.

[0137] In the template configuration step S160, a feature vector is constructed using the variational mode energy distribution, variational mode energy entropy, and mean value of the transient signal amplitude, and the support vector machine model is evaluated using the feature vector.

[0138] In the embodiment of this example, as Figure 6 shown, the transient process of high-resistance faults has more capacitive factors, while the transient process of unbalanced loads has more inductive factors. Therefore, there are also significant differences in the VMD energy entropy characteristics of their transient processes.

[0139] Therefore, the variational mode energy entropy H of the transient signal is calculated using equations (15), (16), and (17) respectively EVMD , the variational mode energy distribution p of each mode in the signal i and the mean value of the transient signal amplitude of the time-domain waveform and a feature vector T is constructed based on these values.

[0140] After that, the feature vector T is used to form a training data set and a test data set. The SVM (Support Vector Machine) is trained using the training data set, and the accuracy of the trained model is verified using the test data set. The ratio of the training data set to the test data set is 2:1.

[0141] In a specific example, as Figure 7As shown in the figure, the support vector machine model divides the samples into two types of samples. H is the optimal hyperplane for classification. H1 and H2 are the hyperplanes passing through the samples closest to the classification line in each class and parallel to H. The so-called optimal classification hyperplane requires that the classification surface can not only correctly separate the two classes (the training error rate is 0), but also maximize the classification margin.

[0142] If the sample set (x i , y i ), i = 1, 2,... l is linearly separable and satisfies:

[0143] y i [(ω·x i ) + b] - 1 ≥ 0 (18)

[0144] Then the classification margin of the samples can be set to 2 / |ω|. In order to obtain the maximum classification margin, it is necessary to minimize the value of |ω|. Therefore, the optimal classification hyperplane can be found by solving the classification margin constraint problem (19). 2 The value of |ω| is minimized, so the optimal classification hyperplane can be found by solving the classification margin constraint problem (19).

[0145]

[0146] x.t.y i (ω·x i + b) ≥ 1, i = 1, 2,..., l (19)

[0147] After that, in order to solve the above constraint problem, the Lagrange multiplier ai can be introduced to transform the classification margin constraint problem, construct the Lagrangian function, and establish the classification margin optimization problem:

[0148]

[0149] To solve for ω and b, by taking the partial derivatives of the Lagrangian function and setting the results to zero, the optimization problem is transformed into a dual problem:

[0150]

[0151] By solving the dual problem, after obtaining the Lagrange multiplier a i , the optimal ω and b can be further solved:

[0152]

[0153] Then, through the obtained optimal ω and b, the final decision function is constructed:

[0154]

[0155] Finally, the evaluation of the support vector machine model is completed by calculating the constructed decision function using the test data set (i.e., verifying the accuracy of the trained support vector machine model).

[0156] In a specific example, through calculations in MATLAB (Matrix Laboratory), it can be known that when y = 1, the fault that occurs in the distribution network is a high-resistance fault, and when y = 0, the distribution network operates normally. For other operating modes, multiple decision functions are required for comprehensive judgment.

[0157] In the template configuration step S170, the transient characteristics of the zero-sequence voltage are identified using the evaluated support vector machine model to complete the detection of distribution network faults.

[0158] In the embodiment of this example, a high-resistance fault simulation model of an ungrounded neutral distribution network as shown in Figure 8 is built based on the PSCAD / EMTDC platform, and distributed power sources are connected to the ends of feeder 1 and feeder 3. In this simulation model, the distribution network is composed of overhead lines and power cables, and 8 fault points are set at typical lines such as the line start, overhead line, power cable, line end, and branch feeder to collect high-resistance fault characteristics at different line positions. At the same time, the Emanuel high-resistance fault model is used for high-resistance fault simulation.

[0159] In a specific example, 900 groups of feature vectors for each working condition are collected within a 500 μs time window. These feature vectors are used as input data and input into the evaluated support vector machine model for fault determination. When the decision result calculated by the decision function falls within the preset high-resistance fault decision interval, it is determined that a high-resistance grounding fault has occurred in the distribution network; when the decision result falls within the preset line interference decision interval, it is determined that the distribution network is interfered; when the decision result falls within the preset line normal decision interval, it is determined that the lines of the distribution network are operating normally.

[0160] From Figure 9 it can be seen that whether the collected feature vectors are put into the training set or the test set, the recognition rate of this method for high-resistance faults reaches 100%, and among 3600 samples, only 5 are misrecognized, and the total accuracy of the model reaches 99.86%, indicating that the evaluated support vector machine model has a high high-resistance fault recognition accuracy.

[0161] In the embodiment of this example, since the evaluated support vector machine model is regarded as a complex classification model or black box model, for a black box model, it is usually difficult to fit it globally with an interpretable simple model, but some local samples in the model can be fitted with a simple model. Therefore, the local part of the black box model can be explained according to the local interpretability analysis algorithm.

[0162] First, sample the original sample features to be explained after N perturbations.

[0163] Second, perform weighting according to the distances between the sampling points and the points to be explained.

[0164] Third, construct a simple classification model to classify the samples obtained from the N perturbations.

[0165] Fourth, then use the object function to display the estimated predictor variable importance in the simple model.

[0166] Among them, it can be seen from the analysis report of the preset variables of the support vector machine model for different working condition identifications that whether the black box model identifies high-resistance faults or various disturbance working conditions, the time-domain features (the mean value of the transient signal amplitude ) and the frequency-domain features (the variational mode energy entropy H EVMD , the variational mode energy distribution p i ) are all predictor variables with relatively high importance, and the prediction results of the simple decision tree model and the black box model are both consistent. It can be seen from the analysis report of the influence of the fault scenario on the detection effect that when high-resistance faults occur at different positions on the power cable and the overhead line, the importance of each predictor variable in the SVM model for high-resistance fault identification does not change significantly, indicating that different fault positions on different lines have no significant influence on the detection effect.

[0167] It should be noted that although the steps of the method in the present disclosure are described in a specific order in the accompanying drawings, this does not require or imply that these steps must be executed in that specific order, or that all the steps shown must be executed to achieve the desired result. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step for execution, and / or one step may be decomposed into multiple steps for execution, etc.

[0168] In addition, in the present exemplary embodiment, a distribution network fault detection device based on variational mode decomposition is also provided. Referring to Figure 10 as shown, the distribution network fault detection device 400 based on variational mode decomposition may include: a transient signal acquisition module 410, a variational mode decomposition module 420, an energy distribution ratio calculation module 430, an energy entropy value calculation module 440, a time-domain signal amplitude mean value calculation module 450, a feature vector construction module 460, and a fault detection module 470. Among them:

[0169] The transient signal acquisition module 410 is configured to generate a transient signal by collecting and calculating the transient part of the zero-sequence voltage of the distribution network according to the negative-sequence current suppression control strategy;

[0170] A variational mode decomposition module 420, which is used to generate modal components by performing variational mode decomposition on the transient signal according to the variational mode decomposition algorithm;

[0171] An energy distribution ratio calculation module 430, which is used to calculate the energy distribution ratio of the modal components in the transient signal according to the orthogonality of the variational mode decomposition to generate a variational mode energy distribution;

[0172] An energy entropy value calculation module 440, which is used to calculate the energy entropy value of the variational mode decomposition according to the modal energy distribution to generate a variational mode energy entropy;

[0173] A time-domain signal amplitude mean calculation module 450, which is used to calculate the amplitude mean of the time-domain signal of the zero-sequence voltage according to the transient signal to generate a transient signal amplitude mean;

[0174] A feature vector construction module 460, which is used to construct a feature vector by using the variational mode energy distribution, the variational mode energy entropy, and the transient signal amplitude mean, and evaluate a support vector machine model through the feature vector;

[0175] A fault detection module 470, which is used to identify the transient characteristics of the zero-sequence voltage by using the evaluated support vector machine model to complete the detection of the distribution network fault.

[0176] The specific details of each of the above-mentioned modules of a distribution network fault detection device based on variational mode decomposition have been described in detail in the corresponding distribution network fault detection method based on variational mode decomposition, so they will not be repeated here.

[0177] It should be noted that although several modules or units of a distribution network fault detection device 400 based on variational mode decomposition are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of the two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0178] In addition, the above drawings are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present invention, rather than for limiting purposes. It is easy to understand that the processes shown in the above drawings do not indicate or limit the time sequence of these processes. Additionally, it is also easy to understand that these processes can be executed, for example, synchronously or asynchronously in multiple modules.

[0179] Other embodiments of the present disclosure will be readily apparent to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include known common general knowledge or conventional technical means in the technical field not disclosed herein. The specification and examples are only to be considered as exemplary, and the true scope and spirit of the present disclosure are pointed out by the claims.

[0180] It should be understood that the present disclosure is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present disclosure is only limited by the appended claims.

Claims

1. A distribution network fault detection method based on variational mode decomposition, characterized in that: include: According to the negative sequence current suppression control strategy, the transient signal is generated by collecting and calculating the transient part of the zero sequence voltage of the distribution network; According to a variational modal decomposition algorithm, a modal component is generated by performing variational modal decomposition on the transient signal; Generate variational modal energy distribution by calculating the energy distribution proportion of the modal component in the transient signal according to the orthogonality of variational modal decomposition; According to the modal energy distribution, the variational modal energy entropy is generated by calculating the energy entropy value of the variational modal decomposition; According to the transient signal, a transient signal amplitude mean is generated by calculating the amplitude mean of the time domain signal of the zero-sequence voltage; constructing a feature vector using the variational modal energy distribution, the variational modal energy entropy, and the transient signal amplitude mean, and evaluating the support vector machine model using the feature vector; The evaluated support vector machine model is used to identify the transient characteristics of the zero-sequence voltage and complete the detection of distribution network faults.

2. The method according to claim 1, characterized in that By collecting and calculating the transient part of the zero-sequence voltage of the distribution network, including: Based on the negative sequence current suppression control strategy, the reference values ​​of the positive sequence current components in the direct axis and cross axis of the inverter are calculated when a single high-resistance grounding fault occurs in the distribution network, and the circuit topology analysis of the zero sequence component during the fault is completed; Based on the circuit topology analysis of the zero-sequence component during the fault, a high-resistance fault equivalent circuit is constructed by setting the fault point as a virtual power supply equivalent; Based on the high-resistance fault equivalent circuit, a zero-sequence voltage signal is generated by collecting and calculating the zero-sequence voltage in the circuit; Based on the zero-sequence voltage signal, the transient part of the zero-sequence voltage is collected through a preset time window to generate a transient signal.

3. The method according to claim 1, characterized in that By performing variational mode decomposition on the transient signal, comprising: Based on the variational mode decomposition algorithm, a transient signal constrained variational problem is generated by constructing a constrained variational problem of the transient signal; Based on the preset distortion degree limit, the Lagrangian operator is solved by introducing noise tolerance to generate the Lagrangian operator.

4. The method according to claim 3, characterized in that By performing variational mode decomposition on the transient signal, comprising: Based on the Lagrangian algorithm, the transient signal constrained variational problem is transformed by introducing the Lagrangian operator to generate an unconstrained variational problem; Using an iterative algorithm, Fourier isometric transformation is performed on the unconstrained variational problem by introducing a quadratic penalty factor, and the decomposed components of the modal signal are iterated; When the decomposed component of the modal signal is not less than a preset accuracy, determining that the iteration is not terminated; When the decomposed component of the modal signal is less than a preset accuracy, it is determined that the iteration is terminated and the iteration result is output to generate the modal component.

5. The method according to claim 1, characterized in that The characteristic vector is constructed by using the variational modal energy distribution, the variational modal energy entropy, and the transient signal amplitude mean, including: Using the variational modal energy distribution, variational modal energy entropy, and transient signal amplitude mean, constructing features of a support vector machine input sample to generate a feature vector; Based on the feature vector, preparing a training set for a support vector machine model to generate a training data set; Based on the feature vector, preparing a test set of the support vector machine model to generate a test data set; The ratio of the training data set to the test data set is 2:

1.

6. The method according to claim 5, characterized in that The support vector machine model is evaluated by the feature vector, including: Based on the support vector machine model, a classification interval constraint problem is generated by mapping the training data set into a high-dimensional space and classifying the training data set using a hyperplane in the mapped high-dimensional space; Based on the Lagrangian algorithm, the classification interval constraint problem is transformed by introducing Lagrangian multipliers to generate a classification interval optimization problem; Based on the classification interval optimization problem, the dual problem is obtained by taking partial derivatives of the Lagrangian function and setting the result to zero.

7. The method according to claim 6, characterized in that The support vector machine model is evaluated by the feature vector, including: Calculating the parameters of the support vector machine model by solving the dual problem to generate the support vector machine model parameters; Based on the support vector machine model parameters, a decision function is constructed, the test data set is introduced into the decision function, and the decision function is solved to complete the evaluation of the support vector machine model.

8. The method according to claim 1, characterized in that The transient characteristics of the zero-sequence voltage are identified by using the evaluated support vector machine model to complete the detection of distribution network faults, including: The transient characteristics of the zero-sequence voltage are used as input data, input into the evaluated support vector machine model, and the decision function is solved to generate a decision result; When the decision result falls into the preset high-resistance fault decision interval, it is determined that a high-resistance grounding fault occurs in the distribution network; When the decision result falls into the preset line interference decision interval, it is determined that the distribution network is disturbed; When the decision result falls within the preset line normal decision interval, it is determined that the line of the distribution network is operating normally.

9. The method according to any one of claims 1 to 8, characterized in that: Also includes: Using a local interpretability analysis algorithm, by performing interpretability analysis on the trained support vector machine model, a preset variable analysis report of the support vector machine model for working condition identification is generated; By using the local interpretability analysis algorithm and adaptively analyzing the fault scenarios, an analysis report on the impact of the fault scenarios on the detection effect is generated.

10. A distribution network fault detection device based on variational mode decomposition, characterized in that: include: A transient signal acquisition module is used to generate a transient signal by collecting and calculating the transient part of the zero-sequence voltage of the distribution network according to the negative-sequence current suppression control strategy; A variational modal decomposition module, used for generating modal components by performing variational modal decomposition on the transient signal according to a variational modal decomposition algorithm; An energy distribution weight calculation module, used for generating a variational modal energy distribution by calculating the energy distribution weight of the modal component in the transient signal according to the orthogonality of the variational modal decomposition; An energy entropy value calculation module is used to generate variational modal energy entropy by calculating the energy entropy value of variational modal decomposition according to the modal energy distribution; A time domain signal amplitude mean value calculation module, used to generate a transient signal amplitude mean value by calculating the amplitude mean value of the time domain signal of the zero-sequence voltage according to the transient signal; A feature vector construction module, used to construct a feature vector using the variational modal energy distribution, the variational modal energy entropy, and the transient signal amplitude mean, and evaluate the support vector machine model through the feature vector; The fault detection module is used to identify the transient characteristics of the zero-sequence voltage by using the evaluated support vector machine model to complete the detection of the distribution network fault.

Images