A method for fault classification and pole selection of a flexible direct current power transmission system

Abstract

The application relates to a flexible direct current power transmission system fault classification and pole selection method, which realizes fault classification and pole selection according to the difference of high-frequency components of bus and line fault voltages when faults occur, and comprises the following steps: S1, when a fault occurs in the system, collecting bus and line fault voltage data of a single converter station, constructing a data set and dividing the data set into a training set and a test set; S2, adopting discrete wavelet transform to decompose the fault voltage and extracting bus and line fault voltage features; S3, calculating a fine composite multi-scale discrete entropy; S4, adopting an extreme random tree model as a data mining model, taking the calculated RCMDE value as the input of the ET classifier; for the data in the training set and the test set, after the data are processed through steps S2-S3, the extreme random tree model is trained and evaluated, and a trained extreme random tree model is obtained, which can be used for fault classification and pole selection. The method can improve the accuracy and reliability of fault detection, realizes accurate classification and pole selection of different fault types.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of direct current transmission line operation fault detection, and in particular to a flexible direct current transmission system fault classification and pole selection method based on wavelet diffusion entropy and extreme random tree fusion. BACKGROUND

[0002] High-voltage direct current transmission has the advantages of flexible power control, larger power supply capacity, lower transmission line loss, and easier interaction with renewable energy. However, overhead transmission lines are easily affected by extreme weather events and have poor operating environments. The complex structure of multi-terminal high-voltage direct current architecture and power flow makes it difficult to apply traditional alternating current system fault classification methods. When a fault occurs on the direct current side, the rapid rise of fault current threatens the stability of the direct current transmission line and damages the health of the equipment. Due to the high cost of high-voltage direct current breakers, it is not possible to deploy a large number of breakers in long-distance high-voltage direct current transmission. Therefore, it is necessary to use a fast and accurate fault detection method to reduce the requirements for direct current breakers and damage to the transmission line.

[0003] Currently, the research on fault detection of flexible direct current transmission lines can be divided into the following two aspects: one is based on the amplitude change and increment of fault voltage or current for fault detection, which sets a threshold relying on expert experience, resulting in a lower accuracy of the fault detection method; the other is to use corresponding mathematical methods to extract and analyze the amplitude, polarity and duration of the transient traveling wave of the fault part to detect the fault. However, this method still needs to manually determine the threshold after signal analysis, and the accuracy is also affected by subjective factors. SUMMARY

[0004] The purpose of the present application is to provide a flexible direct current transmission system fault classification and pole selection method, which can improve the accuracy and reliability of fault detection and achieve accurate classification and pole selection of different fault types.

[0005] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows: a flexible direct current transmission system fault classification and pole selection method, which realizes fault classification and pole selection according to the difference of high-frequency components of bus and line fault voltage when a fault occurs, comprising the following steps:

[0006] S1, when a fault occurs in the system, collect the bus and line fault voltage data of a single converter station, including bus voltage and line voltage, construct a data set and divide it into a training set and a test set;

[0007] S2, decompose the fault voltage using discrete wavelet transform (DWT) to extract bus and line fault voltage features;

[0008] S3, based on the feature quantity extracted by the discrete wavelet transform, calculate the refined composite multi-scale discrete entropy (RCMDE);

[0009] S4, using the extreme random tree model as a data mining model, taking the RCMDE value calculated in step S3 as the input of the ET classifier; for the data in the training set and the test set, after processing by steps S2-S3, inputting the ET classifier, training and evaluating the extreme random tree model, obtaining the trained extreme random tree model, which can be used for fault classification and pole selection.

[0010] Further, in step S1, the busbar and line fault voltage data within 3ms after the start of the fault are collected.

[0011] Further, in step S2, the discrete wavelet function is represented as:

[0012]

[0013] Wherein, a is a scale factor, b is a displacement coefficient, j and k are integers, and t is time; the DWT of the fault voltage u(t) is:

[0014]

[0015] Wherein, represents the conjugate of

[0016] The db4 wavelet is used for 2-layer DWT decomposition of the fault voltage, and the fault voltage signal is output by the low-pass filter and the high-pass filter, and is subjected to down-sampling to obtain the approximation coefficient A1 and the detail coefficient D1. The approximation coefficient is further subjected to 2-layer DWT decomposition to obtain the detail coefficient D j .

[0017] Further, in step S3, the detail coefficient extracted by the discrete wavelet transform is taken as a characteristic quantity, which is a single-variable signal with a length of L: x={x1,x2,…,x L}, and the x is subjected to multi-scale coarse-grained processing, and the f-th coarse-grained sequence is:

[0018]

[0019] Wherein, f=1,2,...,τ,τ is a scale factor;

[0020] The normal distribution function is used to map to y={y i, |i=1,2…,F}:

[0021]

[0022] Wherein, y∈(0,1), y i is an element in y, and xi For the i-th sequence of input univariate signal, σ and μ are the standard deviation and mean of normal distribution function, respectively;

[0023] The linear algorithm is applied to map y into the range of [1, 2, …, c]:

[0024]

[0025] where c is the number of classes, and round() represents the integer;

[0026] Under the given embedding dimension m and time delay d, the phase space reconstruction is performed on the coarse-grained sequence to obtain the embedding vector

[0027]

[0028] where m represents the embedding dimension, and d is the time delay;

[0029] The wandering pattern is defined as The corresponding wandering pattern is Each pattern is composed of c data, and each value has m values, so the corresponding wandering entropy pattern has c m ;

[0030] For each scaling factor, the RCMDE obtained from the fault voltage is represented as:

[0031]

[0032] where represents the embedding vector mapped to the number of wandering patterns , is the probability value of the coarse-grained sequence ; τ max is the maximum scaling factor.

[0033] ​Compared with the prior art, the application has the following beneficial effects: the application provides a flexible HVDC system fault classification and pole selection method based on wavelet scatter entropy and extreme random tree fusion, so as to solve the problems of low accuracy, poor sensitivity to high resistance faults and poor pole selection performance of the existing fault detection method in the MTDC system, improve the accuracy and reliability of fault detection, realize accurate classification and pole selection of different fault types, and have strong anti-interference ability, anti-transition resistance ability and low dependence on sampling frequency. The method introduces RCMDE to extract features of the detail coefficients obtained by DWT, the method can quantify the nonlinearity and uncertainty of the fault signal, has small influence on the mutation signal, considers the amplitude relationship between amplitudes, has better stability and faster calculation speed, and thus can more accurately reflect the differences between different fault types. In addition, the method uses ET intelligent classifier for fault classification and pole selection, ET has good performance and strong generalization ability, and only a small amount of samples are needed to obtain high diagnostic accuracy, and through the combination of DWT and RCMDE, accurate classification and pole selection of bus and line faults in the MTDC system are realized. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 is a method implementation flowchart of an embodiment of the application;

[0035] Figure 2 is a schematic diagram of a four-terminal MMC HVDC power transmission system in an embodiment of the application;

[0036] Figure 3 is a schematic diagram of the propagation process of a traveling wave in an embodiment of the application;

[0037] Figure 4 is a fault transient equivalent circuit diagram when a bipolar short-circuit fault occurs in the OHL24 line in an embodiment of the application;

[0038] Figure 5 is a frequency response diagram of the voltage between the current limiting reactors in an embodiment of the application;

[0039] Figure 6 is a schematic diagram of the discrete wavelet decomposition feature extraction process in an embodiment of the application. DETAILED DESCRIPTION

[0040] The application will be further described below in combination with the drawings and embodiments.

[0041] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the present application belongs.

[0042] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, components, and / or groups thereof, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.

[0043] As shown in Figure 1 The embodiment provides a flexible direct current power transmission system fault classification and pole selection method. According to the difference of high-frequency components of bus and line fault voltages when the fault occurs, the fault classification and pole selection are realized. The method adopts a discrete wavelet transform (DWT) and a refine composite multiscale dispersion entropy (RCMDE) and an extremely randomized tree (ET) fusion fault classification method for fault classification and pole selection of a multi-terminal high-voltage direct current (MTDC) system, including identification of bus faults. The wavelet transform is used to analyze the detail coefficients of the fault voltage in the time-frequency domain. In order to extract multiscale fault features from the fault voltage detail coefficients, the RCMDE is introduced into the signal analysis of the MTDC system for the first time. The entropy is used to measure the complexity and uncertainty of the signal as the feature input of the ET, which greatly reduces the dimension of the data training. The method learns from the mature experience of fault signal analysis, and the fault features can well reflect the differences between different fault types. At the same time, combined with the machine learning algorithm, the classification and pole selection of bus faults and line faults of the four-terminal high-voltage direct current transmission architecture can be considered. The method specifically includes the following steps:

[0044] S1, when the system fails, the bus and line fault voltage data of a single converter station are collected, including bus voltage and line voltage, a data set is constructed and divided into a training set and a test set.

[0045] S2, the discrete wavelet transform DWT is used to decompose the fault voltage, and the bus and line fault voltage features are extracted.

[0046] S3, based on the feature quantity extracted by the discrete wavelet transform, the refine composite multiscale dispersion entropy RCMDE is calculated.

[0047] S4, using an extreme random tree model as a data mining model, taking the RCMDE value calculated in step S3 as an input of the ET classifier; for the data in the training set and the test set, after processing through steps S2-S3, inputting the ET classifier, training and evaluating the extreme random tree model, obtaining the trained extreme random tree model, which can be used for fault classification and pole selection.

[0048] The related content involved in the method is further described below.

[0049] A four-port MMC-HVDC system model corresponding to the MTDC system is constructed, the model including four 300km overhead transmission lines (OHL 12 , OHL 13 , OHL 24 , OHL 34 ), current limiting reactors (L xy ) being installed at both ends of each line for limiting fault current, and a hybrid HVDC circuit breaker being installed in the system to realize fault isolation. The parameter settings of each converter station (MMC1-MMC4) are shown in Figure 2 .

[0050] When a fault occurs in a transmission line or a bus, a fault traveling wave is generated and propagated along with the sharp voltage drop at the fault point. When the traveling wave encounters an impedance change point, refraction and reflection of the wave occur. The propagation process of the traveling wave is shown in Figure 3 .

[0051] The natural frequency of the fault voltage traveling wave can be expressed as:

[0052]

[0053] wherein, represents the speed of the nth traveling wave, x fi represents the distance between the measurement point and the fault point.

[0054] The line voltage at the frequency of 2ωf can be expressed as:

[0055]

[0056] wherein, F1(ω) and F2(ω) are impedance coefficients determined by the line boundary conditions, β(ω) is a line attenuation coefficient, and σ(ω) is a line phase shift coefficient. The propagation speed of the traveling wave is defined by the moving speed of the phase point, i.e.:

[0057] ωt+σ(ω)x fi =const (3)

[0058] Differentiating equation (3) to obtain the traveling wave speed:

[0059]

[0060] According to equation (1), the smaller the distance between the fault point and the measurement point, the greater the natural frequency of the traveling wave. According to equation (4), the traveling wave components of different frequencies have different propagation speeds. Therefore, when a fault occurs, the original fault voltage contains many components of different frequencies. These different components are superimposed on the reflected traveling wave, resulting in a high-frequency transient component at the measurement point.

[0061] The main types of faults in MTDC systems are single-pole grounding faults and double-pole short-circuit faults.

[0062] A bipolar short-circuit fault is the most serious type of fault. Taking a bipolar short-circuit fault as an example, the transient equivalent circuit of a bipolar short-circuit fault in an OHL24 line is as follows: Figure 4 As shown.

[0063] When F 24 When a bipolar short-circuit fault occurs in the converter, the equivalent inductance and capacitance are discharged, which is equivalent to superimposing a negative voltage source -V at the fault point. f .

[0064] MMC i L in the station eqi and C eqi It can be represented as:

[0065]

[0066] Where i = 1, 2, 3, 4; L0 is MMC i The bridge arm reactor of the station; C0 is the submodule capacitor; N is the number of submodules in each MMC bridge arm; such as Figure 3 As shown, the current-limiting reactor L dc24 The ratio of the DC voltage on the line side to the DC voltage on the bus side is:

[0067]

[0068] From equation (6) and Figure 5 It can be seen that as the frequency increases, the value of lg(H(jω)) approaches 2, and the current-limiting reactor L... dc24 It can effectively reduce high-frequency fault voltage components.

[0069] Based on the above analysis, fault classification and polarity selection can be achieved by considering the differences in the high-frequency components of the fault voltage on the bus and line when the fault occurs.

[0070] Step S1: Fault Data Acquisition and Processing

[0071] When a system fault occurs, bus and line fault voltage data for a single converter station are collected, including bus voltage (V). dc2, E1, E2) and line voltage (V dc24 The sampling frequency is 100 kHz. The data within 3 ms after the fault occurs is collected to form the data set, which is divided into training set (70%) and test set (30%).

[0072] Step S2: Discrete Wavelet Transform (DWT) decomposition of fault voltage

[0073] When a fault occurs, the line voltage and bus voltage will change dramatically. The voltage drop amplitude and voltage oscillation intensity caused by different fault types are different. Since the fault voltage contains rich transient high-frequency components, wavelet transform is a suitable solution to enhance the high-frequency characteristics of the burst signal in time and frequency domain. In addition, compared with continuous wavelet function, DWT method significantly reduces the amount of data. Therefore, this method uses discrete wavelet transform to decompose the fault voltage and extract the bus and line fault voltage features.

[0074] The discrete wavelet function can be expressed as:

[0075]

[0076] where a is the scale factor, b is the displacement coefficient, j and k are integers, and t is time. Then the DWT of the fault voltage u(t) is:

[0077]

[0078] where, represents the conjugate of

[0079] The db4 wavelet is used to decompose the fault voltage by 2-layer DWT. The fault voltage signal is passed through a low-pass filter and a high-pass filter, and the output is down-sampled to obtain the approximation coefficient A1 and the detail coefficient D1. Then the approximation coefficient is decomposed by the same step, i.e. further 2-layer DWT decomposition, to obtain the detail coefficient D j (D1 and D2 are taken in this paper), as shown in Figure 6 .

[0080] The original signal x hf (kT) can extract high-frequency components D1 and D2 from the fault traveling wave.

[0081] In order to obtain the time-frequency characteristics of the fault voltage in each frequency band. The db4 wavelet is more suitable for power transient analysis than other wavelet decomposition. In this paper, the db4 wavelet of discrete wavelet transform is used to decompose the fault voltage by multiple layers, and the time-frequency analysis is carried out in different frequency ranges.

[0082] Step S3: Calculate the refined composite multiscale discrete entropy (RCMDE)

[0083] To analyze the characteristics of high-frequency components under MTDC fault conditions, entropy is a good solution to quantify the nonlinearity and uncertainty of fault signals, which has been widely used in feature extraction and fault diagnosis. RCMDE is an entropy calculation method that can measure the complexity and uncertainty of input signals while reducing the dimension of training data. The detail coefficients D1 and D2 obtained by DWT are taken as feature quantities, which are set as single-variable signals with length L.

[0084] First, the feature quantity is the series of detail coefficients D1 and D2 obtained by DWT transformation of fault voltage. It is a single-variable signal with length L: x = {x1, x2, …, xL}. L} After multi-scale coarse-graining processing of x, the f-th coarse-grained sequence is:

[0085]

[0086] where, f = 1, 2, …, τ, τ is the scale factor.

[0087] Map to y = {y1, y2, …, yF} using the normal distribution function as formula (10): i,

[0088]

[0089] where, y ∈ (0, 1), y i is the element in y, x i is the i-th sequence of input single-variable signal, σ and μ are the standard deviation and mean value of the normal distribution function, respectively.

[0090] Map y to the range of [1, 2, …, c] using linear algorithm:

[0091]

[0092] where, c is the number of classes, round() represents the integer. This step can also be realized by other linear mapping and nonlinear mapping techniques. Under the given embedding dimension m and time delay d, the embedding vector

[0093]

[0094] where, m represents the embedding dimension, d is the time delay.

[0095] Define the wandering pattern, if then The corresponding wandering pattern is ​Each pattern is composed of c data, each data has m values, so the corresponding entropy pattern has c m

[0096] For each scale factor, the RCMDE obtained from the fault voltage can be expressed as:

[0097]

[0098] where, represents the embedding vector mapped to the entropy pattern , the number of is the probability value of the coarse-grained sequence ; τ max is the maximum scale factor.

[0099] Step S4: Fault classification and pole selection

[0100] There are many methods of data mining, such as support vector machine (SVM), random forest (RF), decision tree (DT), etc. The extreme random tree (ET) has good performance and strong generalization ability. In addition, high diagnostic accuracy can be obtained by using only a small number of samples. Therefore, the extreme random tree model is used as the data mining model in this method.

[0101] Collect data within 3ms after the start of the fault, and construct two parts of the data set, one is the training set, and the other is the test set. The training set accounts for 70% of the data set, and the test set accounts for 30% of the data set. For the data in the training set and the test set, the steps S2-S3 are processed, that is, the discrete wavelet transform is carried out on the fault voltage waveform to obtain the detail coefficients D1 and D2 with time-frequency characteristics. The RCMDE values of D1 and D2 are used as the input of the ET classifier, the extreme random tree model is trained and evaluated, and the trained extreme random tree model is obtained. Using the trained model, that is, it can be based on new data points for prediction, and the final fault classification and pole selection result is obtained.

[0102] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0103] ​The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0104] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0105] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0106] The above description is only preferred embodiments of the present application, and is not intended to limit the present application to other forms described above. Any person skilled in the art can make modifications or improvements to the above-mentioned disclosed technical content without departing from the technical scope of the present application. However, any simple modification, equivalent change and modification made according to the technical essence of the present application without departing from the technical scope of the present application shall fall within the protection scope of the present application.

Claims

1. A method of fault classification and pole selection for a flexible direct current power transmission system, characterized by, According to the difference between the high-frequency components of the bus and line fault voltages when the fault occurs, fault classification and pole selection are realized, including the following steps: S1, when the system fails, the bus and line fault voltage data of a single converter station are collected, including bus voltage and line voltage, a data set is constructed and divided into a training set and a test set; S2, the discrete wavelet transform (DWT) is used to decompose the fault voltage, and the bus and line fault voltage characteristics are extracted; S3, based on the feature quantity extracted by the discrete wavelet transform, the refined composite multi-scale discrete entropy (RCMDE) is calculated; S4, the extreme random tree model is used as a data mining model, and the RCMDE value calculated in step S3 is used as the input of the ET classifier; for the data in the training set and the test set, after processing by steps S2-S3, the ET classifier is input, the extreme random tree model is trained and evaluated, and the trained extreme random tree model can be used for fault classification and pole selection.

2. The method of claim 1, wherein, In step S1, the bus and line fault voltage data within 3ms after the fault starts are collected.

3. The method of claim 1, wherein the method further comprises: In step S2, the discrete wavelet function is represented as: Where a is the scale factor, b is the displacement coefficient, j and k are integers, and t is time; the DWT of the fault voltage u(t) is: wherein represents the conjugate of The db4 wavelet is used to decompose the fault voltage by 2-layer DWT. The fault voltage signal is filtered by low-pass filter and high-pass filter, and then is down-sampled to obtain the approximation coefficient A1 and the detail coefficient D1. The approximation coefficient is further decomposed by 2-layer DWT to obtain the detail coefficient D of the required scale j .

4. The method of claim 1, wherein, In step S3, the detail coefficients extracted by discrete wavelet transform are used as feature quantities, which is a univariate signal of length L: x = {x1, x2, ..., x...} L If x is subjected to multi-scale coarse-graining, then the f-th coarse-grained sequence is: wherein f = 1, 2,..., τ, τ is a scale factor; is mapped to y = {y using the normal distribution function i, |i = 1, 2, …, F}: where y e (0, 1), y i is an element in y, x i is the ith sequence of the input univariate signal, and σ and μ are the standard deviation and mean of the normal distribution function, respectively. The linear algorithm is applied to map y to the range of [1, 2, …, c]: Where c is the number of categories, and round() represents the integer; The coarse-grained sequence is reconstructed in phase space to obtain an embedding vector under the condition of a given embedding dimension m and time delay d Where m represents the embedding dimension, and d is the time delay; Define the walk pattern if then The corresponding walk pattern is Each pattern consists of c data, each value has m values, so the corresponding walk entropy pattern has c m kind; For each scale factor, the RCMDE obtained from the fault voltage is represented as: wherein, represents an embedding vector mapped to a scatter pattern of the number of, is a probability value of a coarse-grained sequence ; τ max is the maximum scale factor.

Images