Active power distribution network fault feature library construction method and system and fault diagnosis method

Abstract

The application discloses an active power distribution network fault feature library construction method and system and a fault diagnosis method, and comprises the following steps: acquiring historical normal recording wave and historical fault recording wave data as sample data, and extracting sample characteristic quantities representing fault characteristics from the sample data; and constructing an active power distribution network fault data feature library based on the correlation of the sample characteristic quantities. According to the application, the characteristic quantities of faults are acquired through wavelet base function attributes and decomposition layers, the correlation between the fault characteristic quantities is checked, the fault data feature library is constructed, all faults can be recorded through the fault data feature library, the fault type can be predicted, and the faults can be diagnosed.

Description

TECHNICAL FIELD

[0001] The application relates to a fault feature library, in particular to a method and system for constructing a fault feature library of an active power distribution network and a fault diagnosis method. BACKGROUND

[0002] As the end of the power grid, the power distribution network is directly connected with users and can sensitively respond to the requirements of users on power supply safety and quality. The operation safety, reliability and economy of the power distribution network are directly related to social production and people's life. Once a fault occurs in the power distribution network, great loss of social production and great inconvenience of people's life will be caused. According to statistics, more than 80% of faults in the power system are caused by the power distribution network. Therefore, the fault diagnosis and prediction of the highly intelligent and automatic power distribution network have great social and economic benefits.

[0003] The extraction of fault signal feature quantities is the key to the fault diagnosis and prediction of the power distribution network. However, with the power electronicization of electrical equipment in the active power distribution network, the access of a large number of distributed power sources, the change of power flow from unidirectional to bidirectional and other factors, the fault features of the power distribution network are greatly different from those of the traditional power distribution network. The weak features and high-frequency transient characteristics of the power distribution network faults are increasingly obvious. The method of fault diagnosis and prediction based on the historical database of the traditional power distribution network is no longer applicable. A fault feature library for the intelligent fault diagnosis and prediction of the active power distribution network is constructed. The fault feature library can be used to judge the faults of the active power distribution network. However, there is no fault feature library at present. SUMMARY

[0004] In view of the fact that the weak features and high-frequency transient characteristics of the power distribution network faults are increasingly obvious and the existing technology cannot perform fault diagnosis and prediction according to the historical database of the traditional power distribution network, the application provides a method for constructing a fault feature library of an active power distribution network, which comprises the following steps.

[0005] Obtain historical normal recording wave data and historical fault recording wave data as sample data;

[0006] Extract sample feature quantities representing fault characteristics from the sample data;

[0007] Construct a fault data feature library of the active power distribution network based on the correlation of the sample feature quantities.

[0008] Preferably, the step of extracting sample feature quantities representing fault characteristics from the sample data comprises the following steps.

[0009] Based on a pre-determined wavelet basis function and a decomposition layer number, analyze the historical normal recording wave data and the historical fault recording wave data on a set scale domain to extract a plurality of sample feature quantities representing fault characteristics;

[0010] The feature quantities comprise energy features, variance features and entropy features.

[0011] Preferably, the determination of the wavelet basis function and the decomposition level comprises:

[0012] Based on the wavelet basis function attribute of the transient signal compression and the pre-selected index, the wavelet transform function is screened to obtain the wavelet basis function of the sample data.

[0013] The wavelet basis function is calculated by using the method of calculating the wavelet decomposition level to determine the decomposition level.

[0014] The wavelet basis function attribute includes linear phase characteristics, compact support, normality and vanishing moment characteristics.

[0015] The index includes a signal compression ratio index and a reconstruction distortion rate index.

[0016] Preferably, the wavelet basis function based on the wavelet basis function attribute of the transient signal compression and the pre-selected index, the wavelet transform function is screened to obtain the wavelet basis function of the sample data, comprising:

[0017] The wavelet transform function is preliminarily screened based on the wavelet basis function attribute;

[0018] The signal compression ratio index value is calculated by the data amount of the stored signal after the sample data compression and the original signal of the sample data;

[0019] The reconstruction distortion rate index value is calculated by the error root mean square of the sample data and the root mean square of the original signal of the sample data;

[0020] Based on the signal compression ratio index value and the reconstruction distortion rate index value, the wavelet transform function after preliminary screening with low index value is selected.

[0021] Extracting the characteristic

[0022] Preferably, the correlation based on the sample characteristic quantity is used to construct the active power distribution network fault data feature library, comprising:

[0023] Based on each of the sample data and the sample characteristic quantity of the plurality of fault characteristics, a sample characteristic quantity matrix is constructed, and the correlation coefficient between each sample characteristic quantity in the sample characteristic quantity matrix is calculated by a correlation coefficient formula;

[0024] Based on the correlation coefficient, the correlation coefficient between each row in the sample characteristic quantity matrix is calculated to obtain a correlation coefficient matrix of the sample characteristic quantity matrix.

[0025] Based on the correlation coefficient in the correlation coefficient matrix except the diagonal line and the set threshold, the sample characteristic quantity matrix is reduced to obtain the active power distribution network fault data feature library.

[0026] Preferably, the sample feature matrix is ​​reduced based on the correlation coefficients (excluding the diagonal lines) in the correlation coefficient matrix and a set threshold to obtain an active distribution network fault data feature library, including...

[0027] When the correlation coefficient is greater than the set threshold, delete the feature row in the feature matrix; otherwise, retain the feature row in the feature matrix to obtain the active distribution network fault data feature library.

[0028] Preferably, the signal compression ratio is as shown in the following formula:

[0029]

[0030]

[0031] In the formula, A is the number of original signal data; A c This represents the number of valid data items after compression. A represents the number of wavelet coefficients and their location parameters at each scale after quantization; j A represents the number of non-zero data points in each level of the compressed wavelet coefficients; J This represents the number of data points for the approximate coefficients obtained after the final level of decomposition.

[0032] The reconstruction distortion rate is shown in the following formula:

[0033]

[0034] In the formula, c0(n) is the original signal before compression; This is the reconstructed signal after compression.

[0035] Preferably, the energy characteristic is as shown in the following formula:

[0036]

[0037] In the formula, E represents the energy characteristic; E j Define the energy at scale j;

[0038] The variance feature is shown in the following formula:

[0039]

[0040] In the formula, V j 2 The variance characteristic on scale j; Variance features at sample scale j; N is the number of wavelet decomposition levels; W j,n Let n be the wavelet coefficients of the nth layer at scale j; The mean of the wavelet coefficients at scale j;

[0041] The entropy features are shown below:

[0042]

[0043] In the formula, H(p1,p2,...,p...) J ) represents wavelet entropy features; p j This refers to the energy characteristics of the fault signal.

[0044] Based on the same inventive concept, this invention provides an active distribution network fault feature database construction system, including an acquisition module, a feature quantity module, and a fault data feature database construction module;

[0045] The acquisition module acquires historical normal waveform recordings and historical fault waveform recordings as sample data.

[0046] The feature quantity module extracts sample feature quantities that characterize the fault characteristics from the sample data;

[0047] The module for constructing a fault data feature library: Based on the correlation of the sample feature quantities, constructs an active power distribution network fault data feature library.

[0048] Based on the same inventive concept, the present invention provides a fault diagnosis method, including:

[0049] Acquire waveform recording information of active power distribution network and extract feature quantities from waveform recording information;

[0050] The feature values ​​are input into a pre-trained support vector machine to obtain the association relationship;

[0051] Based on the aforementioned correlation, fault diagnosis is performed on the active power distribution network;

[0052] The support vector machine is trained using a sample set constructed from the fault data feature library according to any one of claims 1-9.

[0053] Preferably, the training of the support vector machine includes:

[0054] Fault features from the fault data feature library are used as training samples for the support vector machine.

[0055] The support vector machine is trained using the fault features in the training samples to obtain the correlation between fault features and faults.

[0056] Preferably, the fault diagnosis of the active distribution network based on the correlation includes:

[0057] The feature values ​​in the recorded waveform information are input into a support vector machine for matching to determine the correlation and to determine whether a fault has occurred.

[0058] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0059] 1. A method for constructing an active distribution network fault feature library, comprising: acquiring historical normal waveform recordings and historical fault waveform recordings as sample data; extracting sample feature quantities characterizing fault characteristics from the sample data; and constructing an active distribution network fault data feature library based on the correlation of the sample feature quantities. This invention obtains fault feature quantities through wavelet basis function attributes and decomposition level, and checks the correlation between fault feature quantities to construct the fault data feature library.

[0060] 2. This invention extracts the features of faults, performs correlation analysis on the features of all faults, and eliminates identical fault features to obtain a typical fault feature library. Based on the fault feature library, faults are predicted and diagnosed when they occur.

[0061] 3. This invention can diagnose faults through a fault data feature database. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the active power distribution network fault feature database construction method of the present invention;

[0063] Figure 2 This is a flowchart of the active power distribution network fault feature database construction method of the present invention;

[0064] Figure 3 This is a schematic diagram of the original three-phase current signal of the waveform recording data of the present invention;

[0065] Figure 4 This is a graph showing the relationship between the Daubichies5 wavelet compression ratio and distortion rate and the decomposition series in this invention.

[0066] Figure 5 This is a graph showing the relationship between the Daubichies10 wavelet compression ratio and distortion rate and the decomposition series in this invention.

[0067] Figure 6 This is a graph showing the relationship between the wavelet compression ratio and distortion rate of the Bior 3.1 wavelet and the decomposition series in this invention.

[0068] Figure 7 This is a graph showing the relationship between the wavelet compression ratio and distortion rate of the Bior 3.3 wavelet and the decomposition series in this invention.

[0069] Figure 8 This is a graph showing the relationship between the Bior3.5 wavelet compression ratio and distortion rate and the decomposition series in this invention.

[0070] Figure 9 This is a graph showing the relationship between the wavelet compression ratio and distortion rate of Bior 4.4 and the decomposition series in this invention.

[0071] Figure 10 This is a graph showing the fault diagnosis identification rate of the present invention. Detailed Implementation

[0072] The embodiments of the present invention will be further described with reference to the accompanying drawings.

[0073] Example 1

[0074] Combination Figure 1 This invention provides a method for constructing a fault feature database for active power distribution networks, including:

[0075] Step 1: Obtain historical normal waveform recordings and historical fault waveform recordings as sample data;

[0076] Step 2: Extract sample features that characterize the fault characteristics from the sample data;

[0077] Step 3: Construct an active distribution network fault data feature library based on the correlation of sample feature quantities.

[0078] Step two involves extracting sample features characterizing the fault features from the sample data, including:

[0079] Based on the predetermined wavelet basis function and decomposition level, historical normal waveform data and historical fault waveform data are analyzed in the set scale domain to extract sample feature quantities of multiple fault characteristics.

[0080] Among them, the features include: energy features, variance features, and entropy features.

[0081] The determination of wavelet basis functions and the number of decomposition levels includes:

[0082] Based on the wavelet basis function properties of transient signal compression and pre-selected indices, the wavelet transform functions are screened to obtain the wavelet basis functions of the sample data.

[0083] The wavelet basis functions are calculated using a method that calculates the number of wavelet decomposition levels to determine the number of decomposition levels;

[0084] Among them, the properties of wavelet basis functions include: linear phase characteristics, compact support, normality, and vanishing moment characteristics;

[0085] The metrics include: signal compression ratio and reconstruction distortion rate.

[0086] Based on the wavelet basis function properties of transient signal compression and pre-selected indices, the wavelet transform basis functions are screened to obtain the wavelet basis functions of the sample data, including:

[0087] The basis functions of wavelet transform are initially screened based on the properties of wavelet basis functions;

[0088] The signal compression ratio is calculated by comparing the amount of data stored after compressing the sample data with the original signal of the sample data.

[0089] The reconstruction distortion rate index is calculated using the root mean square error of the sample data and the root mean square signal of the original sample data.

[0090] Based on the signal compression ratio and reconstruction distortion rate, the basis functions of the wavelet transform after initial screening are selected with low index values.

[0091] Extraction and characterization

[0092] Step 3 involves constructing an active distribution network fault data feature library based on the correlation of sample feature quantities, including:

[0093] Based on the sample features of each sample data and multiple fault characteristics, a sample feature matrix is ​​constructed, and the correlation coefficient between each sample feature in the sample feature matrix is ​​calculated using the correlation coefficient formula.

[0094] The correlation coefficient matrix of the sample feature matrix is ​​obtained by calculating the correlation coefficient between each row of the sample feature matrix.

[0095] Based on the correlation coefficients (excluding the diagonal lines) in the correlation coefficient matrix and a set threshold, the sample feature matrix is ​​reduced to obtain an active power distribution network fault data feature library.

[0096] Based on the correlation coefficients (excluding those on the diagonal) in the correlation coefficient matrix and a set threshold, the sample feature matrix is ​​reduced to obtain an active distribution network fault data feature library, including...

[0097] When the correlation coefficient is greater than the set threshold, delete the feature rows in the feature matrix; otherwise, retain the feature rows in the feature matrix to obtain the active distribution network fault data feature library.

[0098] The signal compression ratio is shown in the following formula:

[0099]

[0100]

[0101] In the formula, A is the number of original signal data; A c This represents the number of valid data items after compression. A represents the number of wavelet coefficients and their location parameters at each scale after quantization; j A represents the number of non-zero data points in each level of the compressed wavelet coefficients; J This represents the number of data points for the approximate coefficients obtained after the final level of decomposition.

[0102] The reconstruction distortion rate is shown in the following formula:

[0103]

[0104] In the formula, c0(n) is the original signal before compression; This is the reconstructed signal after compression.

[0105] The energy characteristics are shown in the following formula:

[0106]

[0107] In the formula, E represents the energy characteristic; E j Define the energy at scale j;

[0108] The variance characteristics are shown in the following formula:

[0109]

[0110] In the formula, V j 2 The variance characteristic on scale j; Variance features at sample scale j; N is the number of wavelet decomposition levels; W j,n Let n be the wavelet coefficients of the nth layer at scale j; The mean of the wavelet coefficients at scale j;

[0111] Entropy features are shown below:

[0112]

[0113] In the formula, H(p1,p2,...,p...) J ) represents wavelet entropy features; p j This refers to the energy characteristics of the fault signal.

[0114] Example 2

[0115] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0116] A method for constructing a fault feature database for active distribution networks, combined with Figure 2 This includes the following steps:

[0117] First, the definition of continuous wavelet and its discretization method are given. Second, considering that the fault transient signal needs to be accurately restored to the original signal after wavelet transform decomposition, the following are proposed: wavelet basis function properties for transient signal compression, two comprehensive performance evaluation indicators of wavelet basis functions (signal compression ratio and reconstruction distortion rate), and a method for selecting the number of wavelet decomposition layers. The wavelet basis functions and decomposition layers suitable for fault waveform data processing are selected.

[0118] Second: After selecting the wavelet basis function and decomposition level suitable for fault recording data processing, the characteristics of the line fault waveform signal collected by the monitoring device are analyzed in the scale domain, and three feature quantities that can effectively describe the fault characteristics are extracted, namely energy feature, variance feature and entropy feature.

[0119] Third: Use the correlation coefficient matrix method to perform correlation analysis on the obtained sample features, and construct a fault data feature library of typical samples by reducing the feature quantity of each type of fault data sample.

[0120] The first step specifically includes the following steps:

[0121] 1-1: Discretize the continuous wavelet.

[0122] Let the wavelet function ψ(t)∈L 2 (R)(L 2 (R) represents the square-integrable real space. After scaling and translation, a wavelet sequence can be obtained:

[0123]

[0124] In the formula, a is the scaling factor; b is the translation factor.

[0125] Then for any function signal f(t)∈L 2 (R), whose continuous wavelet transform is as follows:

[0126]

[0127] In the formula, for .

[0128] make Where J is called the scale number of the wavelet transform, and the scale parameter is discretized; let Where N is called the decomposition level of the wavelet transform, and the translation parameters are discretized; then the continuous wavelet sequence is transformed into a discrete wavelet sequence of the following form:

[0129]

[0130] Then for any function signal f(t)∈L 2 (R), whose discrete wavelet transform is as follows:

[0131]

[0132] In the formula, W f (j,n) are called discrete wavelet coefficients.

[0133] 1-2: Considering that the fault transient signal needs to be accurately restored to the original signal after wavelet transform decomposition, a preliminary screening of the wavelet transform basis functions is proposed based on the wavelet basis function properties for transient signal compression.

[0134] (1) Linear phase characteristic: This characteristic can reduce the distortion rate of the reconstructed signal at the edges;

[0135] (2) Compact support: This property indicates the complexity of wavelet transform. The shorter the support, the faster the wavelet transform can be implemented.

[0136] (3) Regularity: that is, to maximize the smoothness or continuous differentiability of wavelets;

[0137] (4) Vanishing moment characteristic: Even if as many wavelet coefficients as possible are zero, or as few non-zero wavelet coefficients as possible are generated.

[0138] 1-3: Two evaluation indicators, signal compression ratio and reconstruction distortion rate, are proposed. Under the premise that the quality of the reconstructed signal is guaranteed, the wavelet basis functions after initial screening are selected based on the principle that the lower the signal compression ratio and reconstruction distortion rate, the better.

[0139] (1) Signal compression ratio:

[0140]

[0141]

[0142] In the formula, N is the number of original signal data; A c This represents the number of valid data items after compression. A represents the number of wavelet coefficients and their location parameters at each scale after quantization; j A represents the number of non-zero data points in each level of the compressed wavelet coefficients; J This represents the number of data points for the approximate coefficients obtained after the final level of decomposition.

[0143] (2) Reconstruction distortion rate:

[0144]

[0145] In the formula, c0(n) is the original signal before compression; This is the reconstructed signal after compression.

[0146] 1-4: Determine the number of decomposition layers.

[0147] Let the sampling frequency of the recorded waveform data be f. s According to theoretical analysis, the frequency range of the recorded waveform data is [0, f]. s / 2]. Let the scaling coefficients be the original sampling frequencies. After one wavelet decomposition, the scale space and wavelet space are obtained, and their frequency ranges are [0, f]. s / 4],[f s / 4,f s / 2]. After n-level decomposition, the frequency range of the scale space is [0, f]. s / 2 n+1 The frequency ranges of each wavelet space are [f], ... s / 2 n+1 ,f s / 2 n ], [f s / 2 n ,f s / 2 n-1 ],…,[f s / 4,f s / 2].

[0148] Considering that during normal operation of a power system, voltage and current mainly consist of power frequency components, after a fault, voltage and current are primarily composed of two parts: low-frequency components near the power frequency and abrupt changes in the time domain. To prevent the power frequency during normal operation from being decomposed into the wavelet frequency space, thus increasing the computational load, the maximum decomposition frequency in the scale space is obtained:

[0149] f s / 2 n+1 =Power frequency

[0150] Assume the frequency center of the fault electrical quantity is located around the power frequency, that is, let f s / 2 n+1 =50Hz, then the number of wavelet decomposition levels is:

[0151]

[0152] In the formula, int represents integer rounding. The calculated value is N = 4. That is, when the frequency center of the fault electrical quantity is around the power frequency, a wavelet decomposition level of 4 is optimal.

[0153] The second part specifically includes the following steps:

[0154] 2-1: After selecting the wavelet basis function and decomposition level suitable for fault waveform data processing, the characteristics of the line fault waveform signal acquired by the monitoring device are analyzed in the scale domain. Three features that can effectively describe the fault characteristics are extracted: energy feature, variance feature, and entropy feature. The extraction method is as follows:

[0155] (1) Energy characteristics

[0156] The wavelet coefficient vector at scale j is defined as:

[0157] W j =[w j,1 ,w j,2 ,…,w j,N ]

[0158] In the formula, w j,n The wavelet coefficients of the nth layer at scale j are calculated by equation (4).

[0159] The coefficient vectors at each scale can then form a vector sequence {W}:

[0160] {W} = {W1, W2, ..., W} J}

[0161] The energy at scale j is defined as:

[0162]

[0163] The energy characteristics of the fault signal are expressed as follows:

[0164] p j =E j / E

[0165] In the formula,

[0166] (2) Variance characteristics

[0167] Wavelet variance can be used as a substitute for the sample variance of a stochastic process. Its characteristic is that it decomposes the sample variance of a time series according to the scale. The variance characteristics at scale j are expressed as follows:

[0168]

[0169] In the formula, Let be the mean of the wavelet coefficients at scale j.

[0170] (3) Entropy characteristics

[0171] Let the distribution of the energy sequence at each wavelet scale be:

[0172] {P} = {p1, p2, ..., p J}

[0173] The wavelet entropy feature is then expressed as follows:

[0174]

[0175] The third step specifically includes the following steps:

[0176] 3-1: Considering that the feature extraction in step 2 includes feature information from all fault data, the redundancy is high and the features of each sample may have strong correlations, directly performing fault diagnosis and prediction based on existing features would not only result in redundant data but also affect the fault diagnosis and prediction effect, leading to a decrease in accuracy. This patent utilizes the correlation coefficient matrix analysis method to perform correlation analysis on the three features of the samples.

[0177] First, construct the eigenvalue matrix M:

[0178]

[0179] In the formula, m is the number of samples, x m1 x m2 x m3 These represent the three features of the m-th sample.

[0180] The correlation between the features of each sample is obtained by calculating the correlation coefficient:

[0181]

[0182] In the formula, cov(x) i x j ) is a vector x i x j The covariance represents the different vectors x i x j The more samples m there are, the more reliable the covariance becomes, and the more adequate the correlation between the vectors is. For vector x i The standard deviation is given by μ, where μ is the sample mean and E is the expected value. Generally, the mean is used instead of the expected value.

[0183] Based on the correlation coefficient, calculate the correlation coefficient between each row of the feature matrix M to obtain the correlation coefficient matrix P of the feature matrix M:

[0184]

[0185] Due to the correlation r between the two vectors ij =r ji The correlation coefficient matrix P is a symmetric square matrix, and the correlation between two identical vectors is 1, so all diagonal elements are 1. Therefore, the correlation coefficient matrix is ​​symmetric, and analyzing the correlation between features only requires analyzing the upper triangular matrix of P.

[0186] Set a threshold T and determine the element value r in the correlation coefficient matrix P. ij The relationship between r and T (excluding diagonal elements) is as follows: ijIf >T, it means that elements Xi and Xj in the feature matrix M are strongly correlated, then the i-th row of matrix M should be deleted.

[0187] Therefore, the feature count of each type of fault data sample is reduced to construct a fault data feature library of typical samples:

[0188]

[0189] In the formula, n is the number of extracted sample features.

[0190] Example 3

[0191] To verify the effectiveness of the above method, the present invention combines... Figure 3 The original signals of the fault recording data are analyzed.

[0192] A total of 1200 data points were collected from the two cycles before the fault and the ten cycles after the fault, and wavelet transform was performed. The A-phase current signal of the faulty line was compressed using the dbN wavelet and the biorNr.Nd wavelet respectively. Considering the performance of computational speed, experiments were conducted using bior3.1-3.5 and bior4.4 wavelet bases. The experimental results were analyzed and evaluated to select the wavelet basis function and the number of decomposition levels.

[0193] Table 1 presents a comparison of the characteristics of six wavelet bases at level 5 decomposition, combined with... Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 and Figure 9 The trends of compression ratio and distortion rate under six wavelet transforms are presented.

[0194] Table 1

[0195]

[0196] Considering both compression ratio and distortion rate, the Bior 3.5 wavelet basis was chosen as the compression tool for the recorded signal, with a signal compression decomposition scale of 5. Based on this, feature extraction was performed on the fault recording data of power lines, generating feature libraries for various typical fault data samples, as shown in Tables 2, 3, 4, and 5.

[0197] Table 2 Feature Library of Wildfire Wave Recording Data

[0198]

[0199] Table 3 Feature library of external force damage waveform data

[0200]

[0201] Table 4 Feature library of icing waveform data

[0202]

[0203] Table 5. Feature library of tree discharge waveform data

[0204]

[0205]

[0206] Example 4

[0207] Based on the same inventive concept, this invention provides an active distribution network fault feature database construction system, including an acquisition module, a feature quantity module, and a fault data feature database construction module;

[0208] Acquisition module: Acquires historical normal waveform recordings and historical fault waveform recordings as sample data;

[0209] Feature module: Extracts sample features that characterize fault properties from sample data;

[0210] Module for constructing a fault data feature library: Based on the correlation of sample feature quantities, a fault data feature library for active distribution networks is constructed.

[0211] Based on the same inventive concept, the present invention provides a fault diagnosis method, including:

[0212] Acquire waveform recording information of active power distribution network and extract feature quantities from waveform recording information;

[0213] The features are input into a pre-trained support vector machine to obtain the correlation.

[0214] Fault diagnosis of active power distribution networks based on correlation relationships;

[0215] Among them, the support vector machine is trained by a sample set constructed from the fault data feature library based on any one of weights 1-9.

[0216] Training a support vector machine includes:

[0217] Fault features from the fault data feature library are used as training samples for the support vector machine;

[0218] The support vector machine is trained using the fault features in the training samples to obtain the correlation between fault features and faults.

[0219] Fault diagnosis of active distribution networks based on correlation includes:

[0220] By inputting the feature values ​​from the waveform recordings into a support vector machine for matching, the correlation is determined, and it is determined whether a fault has occurred.

[0221] Example 5

[0222] Scenario 1 - Fault Diagnosis

[0223] The fault data sample feature library is used as the training sample set to obtain a multi-class SVM model for fault type discrimination. The fault data sample feature library is then used as the test sample and input into the SVM model. The results are then combined... Figure 10 The average identification rate of the diagnostic results is as high as 85%.

[0224] Scenario 2 - Fault Prediction

[0225] The fault data sample feature library was randomly divided into training and test sets in a 7:3 ratio. Using the CNN-MSVM algorithm, a complete convolutional neural network was first built, and then trained using the fault data sample feature library to obtain the convolutional layer weight parameters. The convolutional layer output was converted into an SVM input vector, and then the SVM was trained. The prediction results are shown in Table 5. Levels 1-3 represent mild, moderate, and severe faults, respectively; precision represents the proportion of samples predicted as belonging to a certain class that actually belong to that class; recall represents the proportion of samples that actually belong to a certain class that were correctly predicted; and overall accuracy is the overall prediction accuracy for all three classes. As shown in the table, the overall fault prediction accuracy based on the fault feature library constructed in this patent is 90.38%.

[0226] Table 5

[0227]

[0228] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied 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.

[0229] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.

[0230] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0231] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0232] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A method for diagnosing faults in an active power distribution network, characterized in that, include: Acquire waveform recording information of active power distribution network and extract feature quantities from waveform recording information; The feature values ​​are input into a pre-trained support vector machine to obtain the association relationship; Based on the aforementioned correlation, fault diagnosis is performed on the active power distribution network; The support vector machine is trained using a sample set constructed from a fault data feature library obtained according to the following steps. The steps include: Acquire historical normal waveform recordings and historical fault waveform recordings as sample data; Extract sample features that characterize fault properties from sample data; Based on the correlation of the sample features, an active power distribution network fault data feature library is constructed. The extraction of sample feature quantities characterizing fault characteristics from sample data includes: Based on a predetermined wavelet basis function and decomposition level, the historical normal waveform data and the historical fault waveform data are analyzed in a set scale domain to extract sample feature quantities of multiple fault characteristics. The features include: energy features, variance features, and entropy features; The determination of the wavelet basis function and the number of decomposition levels includes: Based on the wavelet basis function properties of transient signal compression and pre-selected indices, the wavelet transform functions are screened to obtain the wavelet basis functions of the sample data. The wavelet basis functions are calculated using a method for calculating the number of wavelet decomposition levels to determine the number of decomposition levels; The wavelet basis function properties include: linear phase characteristics, compact support, normality, and vanishing moment characteristics; The metrics include: signal compression ratio and reconstruction distortion rate; The wavelet basis function attributes based on transient signal compression and pre-selected indices are used to filter the wavelet transform basis functions to obtain the wavelet basis functions of the sample data, including: The basis functions of wavelet transform are initially screened based on the properties of wavelet basis functions; The signal compression ratio is calculated by comparing the amount of data stored after compressing the sample data with the original signal of the sample data. The reconstruction distortion rate index is calculated using the root mean square error of the sample data and the root mean square signal of the original sample data. Based on the signal compression ratio index and reconstruction distortion rate index, the basis functions of the wavelet transform after initial screening with low index values ​​are selected. The construction of an active distribution network fault data feature library based on the correlation of the sample feature quantities includes: Based on each sample data and the sample feature quantities of the multiple fault characteristics, a sample feature quantity matrix is ​​constructed, and the correlation coefficient between each sample feature quantity in the sample feature quantity matrix is ​​calculated using the correlation coefficient formula. Based on the correlation coefficient, the correlation coefficient between each row of the sample feature matrix is ​​calculated to obtain the correlation coefficient matrix of the sample feature matrix; Based on the correlation coefficients (excluding the diagonal lines) in the correlation coefficient matrix and a set threshold, the sample feature matrix is ​​reduced to obtain an active power distribution network fault data feature library. The sample feature matrix is ​​reduced by subtracting the correlation coefficients (excluding the diagonal lines) from the correlation coefficient matrix and a set threshold to obtain an active power distribution network fault data feature library, including... When the correlation coefficient is greater than the set threshold, delete the feature row in the feature matrix; otherwise, retain the feature row in the feature matrix to obtain the active distribution network fault data feature library.

2. The method as described in claim 1, characterized in that, The signal compression ratio is shown in the following formula: In the formula, A is the number of original signal data; This represents the number of valid data items after compression. This represents the number of wavelet coefficients and their location parameters at each scale after quantization. This represents the number of non-zero data points in each level of the wavelet coefficients after compression. This represents the number of data points for the approximate coefficients obtained after the final level of decomposition. The reconstruction distortion rate is shown in the following formula: In the formula, This is the original signal before compression; This is the reconstructed signal after compression.

3. The method as described in claim 1, characterized in that, The energy characteristic is shown in the following formula: In the formula, Energy characteristics; Define the energy at scale j; The variance feature is shown in the following formula: In the formula, The variance characteristic on scale j; The variance feature on sample scale j; The wavelet decomposition level; Let n be the wavelet coefficients of the nth layer at scale j; The mean of the wavelet coefficients at scale j; The entropy features are shown below: In the formula, Wavelet entropy features; This refers to the energy characteristics of the fault signal.

4. The method as described in claim 1, characterized in that, The training of the support vector machine includes: Fault features from the fault data feature library are used as training samples for the support vector machine. The support vector machine is trained using the fault features in the training samples to obtain the correlation between fault features and faults.

5. The method as described in claim 4, characterized in that, The fault diagnosis of the active power distribution network based on the aforementioned correlation includes: The feature values ​​in the recorded waveform information are input into a support vector machine for matching to determine the correlation and to determine whether a fault has occurred.

Images