A transformer winding mechanical state detection method based on non-negative tensor decomposition

Abstract

The present application discloses a transformer winding mechanical state detection method based on non-negative tensor decomposition, the method comprising: obtaining a short-circuit current signal and a vibration signal during a short-circuit impact of the transformer winding. According to the short-circuit current signal and the vibration signal, the time domain envelope of the short-circuit current signal and the time domain envelope of the vibration signal are calculated. According to the time domain envelope of the short-circuit current signal and the time domain envelope of the vibration signal, a third-order non-negative tensor is constructed. The core tensor of the third-order non-negative tensor is calculated to obtain a first core tensor. The statistics of the first core tensor are calculated to obtain a first statistic. According to the first statistic, a first control limit is calculated. If the first statistic belongs to the interval of the first control limit, the mechanical state of the transformer winding is normal. The present application can effectively identify the initial faults of the transformer winding, and then adopt an effective operation and maintenance strategy to avoid major faults.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment state detection, and in particular to a transformer winding mechanical state detection method based on non-negative tensor decomposition. Background Art

[0002] Nowadays, large power transformers are the most important electrical equipment in the power grid. Their performance directly affects the safe and stable operation of the power system. For power transformers that operate for a long time, their reliability is crucial. With the continuous increase in the capacity of my country's power grid, the short-circuit capacity increases accordingly. The huge electromagnetic force generated by the impact current formed by the short-circuit at the transformer outlet will pose a serious threat to the mechanical strength and stability of the transformer winding. If the faulty transformer is not repaired in time, it will not only damage the transformer, but also affect the normal operation of the power grid and even cause the collapse of the power system.

[0003] After a transformer in operation is subjected to multiple short-circuit shocks, its windings are deformed by force, which hides insulation defects. Once it encounters voltage fluctuations, it may cause insulation breakdown. Years of practical experience have shown that after the transformer windings are deformed, it is difficult to accurately judge through general insulation tests and oil tests, and the fault manifests as a latent fault. Therefore, during operation, when the transformer experiences an external short-circuit accident or undergoes routine maintenance, how to effectively diagnose whether the transformer windings are loose and then determine whether the transformer needs to be repaired is an important measure to ensure the safe operation of the transformer.

[0004] Transformer winding deformation detection is one of the current routine test items for transformers. There are two most commonly used detection methods: one is the short-circuit impedance method, which measures and analyzes the short-circuit impedance or leakage reactance of the transformer winding under power frequency voltage, and determines whether the transformer winding has changed in a way that affects safe operation by the change in impedance or leakage reactance. The other is the frequency response analysis method, which measures the transfer function of the transformer winding and describes it in the frequency domain, thereby diagnosing whether the transformer winding has changed.

[0005] However, the above methods have the following problems: First, the short-circuit impedance method has low sensitivity and low fault detection rate, and can only obtain relatively accurate diagnostic results when the overall deformation of the transformer coil is serious. Second, the frequency response analysis method has a relatively complex frequency response waveform, and it requires more tests to judge the winding condition, which makes it difficult to form a clear quantitative judgment, which is not conducive to the judgment of the staff. Summary of the invention

[0006] The present invention provides a transformer winding mechanical state detection method based on non-negative tensor decomposition to solve the problem of how to effectively diagnose whether the transformer winding is abnormal when the transformer experiences an external short circuit accident or undergoes routine maintenance during operation. The present invention diagnoses the mechanical working state of the transformer winding and effectively and accurately detects whether the mechanical state of the transformer winding is abnormal, so that effective measures can be taken in time for the transformer winding to improve its operating reliability.

[0007] The present invention is achieved through the following technical solutions:

[0008] The present application provides a transformer winding mechanical state detection method based on non-negative tensor decomposition, the method comprising:

[0009] Obtain short-circuit current signal and vibration signal when transformer winding is short-circuited;

[0010] According to the short-circuit current signal and the vibration signal, a time domain envelope of the short-circuit current signal and a time domain envelope of the vibration signal are calculated;

[0011] Constructing according to the time domain envelope of the short-circuit current signal and the time domain envelope of the vibration signal to obtain a third-order non-negative tensor;

[0012] Calculate the core tensor of the third-order non-negative tensor to obtain a first core tensor;

[0013] Calculating a statistic of the first core tensor to obtain a first statistic;

[0014] Calculate a first control limit based on the first statistic;

[0015] Determine the first statistic, if it belongs to the first control limit interval, then the transformer winding mechanical state is normal;

[0016] If the first statistic does not belong to the interval of the first control limit, the mechanical state of the transformer winding is abnormal.

[0017] Furthermore, the calculation formula of the short-circuit current signal time domain envelope is:

[0018]

[0019]

[0020] Among them, i(t) is the short-circuit current signal, i h (t) is the Hilbert transform result of the short-circuit current signal.

[0021] Furthermore, the method for calculating the time domain envelope of the vibration signal includes:

[0022] Find all local maximum points in the vibration signal;

[0023] Representing all positive local maxima and all negative local maxima in the local maximum points as a data sequence;

[0024] Construct a rectangular structure element according to the data sequence to obtain a first structure element;

[0025] According to the first structural element, performing multi-scale morphological wavelet decomposition on the vibration signal to obtain a first scale signal;

[0026] Calculating the vibration signal and the first scale signal to obtain a first correlation coefficient, and selecting a scale signal corresponding to a maximum value of the first correlation coefficient to obtain a second scale signal;

[0027] The time domain envelope of the vibration signal is calculated according to the second scale signal.

[0028] Furthermore, the third-order non-negative tensor is constructed by the transpose of the time series of the vibration signal time domain envelope and the time series of the vibration signal time domain envelope, and the short-circuit current signal time domain envelope.

[0029] Furthermore, the calculation method of the first core tensor includes:

[0030] Establishing a solution model for the first core tensor;

[0031] Use the gradient descent method to solve and obtain the first objective function;

[0032] Iteratively calculating the first objective function, and if a convergence condition is met, outputting the first core tensor;

[0033] If the convergence condition is not met, continue the iterative calculation.

[0034] Furthermore, the convergence condition is a preset iterative convergence threshold.

[0035] Furthermore, the calculation method of the first statistic includes:

[0036] Performing modular expansion on the first core tensor to obtain a first matrix and a second matrix;

[0037] Calculating the eigenvalues ​​of the first matrix to obtain first eigenvalues;

[0038] Calculating the eigenvalues ​​of the second matrix to obtain second eigenvalues;

[0039] The first statistic is constructed according to the first eigenvalue and the second eigenvalue.

[0040] Furthermore, the first control limit is determined by applying the 3σ criterion to the first statistic.

[0041] This application calculates and analyzes the vibration signal and current signal under the short-circuit impact of the transformer, extracts the time domain envelope of the short-circuit current signal through amplitude demodulation using Hilbert transform, calculates the time domain envelope of the vibration signal using adaptive non-sampling morphological wavelet transform, constructs a third-order non-negative tensor using the time domain envelope of the short-circuit current signal and the time domain envelope of the vibration signal, calculates the core tensor of the third-order non-negative tensor under sparse constraints based on the gradient descent method, constructs the statistics and control limits of the core tensor, and discriminates the mechanical state of the transformer winding through the distribution of the control limits. Accurate detection of the mechanical state of the transformer is achieved, so that the hidden dangers of the initial faults of the transformer winding can be effectively identified, and then effective operation and maintenance strategies can be adopted to avoid major faults and improve the reliability of transformer operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0043] Figure 1 A schematic flow chart of a transformer winding mechanical state detection method based on non-negative tensor decomposition provided in this application;

[0044] Figure 2 A schematic diagram of a vibration signal for detecting the mechanical state of a transformer winding provided in this application;

[0045] Figure 3 A schematic diagram of a time domain envelope of a vibration signal used for detecting the mechanical state of a transformer winding is provided for this application;

[0046] Figure 4 This is a flow chart of the method for calculating the time domain envelope of the vibration signal described in this application;

[0047] Figure 5 This is a flowchart of the calculation method of the first core tensor of this application;

[0048] Figure 6 This is a flowchart of the method for calculating the first statistic in this application. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] After a transformer in operation is subjected to multiple short-circuit shocks, its windings are deformed by stress, hiding insulation defects, which may lead to insulation breakdown once encountering voltage fluctuations. Years of practical experience have shown that after the transformer windings are deformed, it is difficult to accurately judge through general insulation tests and oil tests, and the fault manifests as a latent fault. Therefore, the present application provides a transformer winding mechanical state detection method based on non-negative tensor decomposition, which can diagnose the mechanical working state of the transformer winding, effectively and accurately detect whether the mechanical state of the transformer winding is abnormal, so that effective measures can be taken in time for the transformer winding to improve its operating reliability. The following is a detailed description of the technical solution of the present application:

[0051] A short-circuit impulse test was carried out on a 110kV transformer. The vibration signal and current signal in the short-circuit impulse test were tested to illustrate the detection method and process of the mechanical state of the transformer winding.

[0052] See also Figure 1 , is a schematic flow chart of a transformer winding mechanical state detection method based on non-negative tensor decomposition provided in an embodiment of the present application;

[0053] from Figure 1 It can be seen that the transformer winding mechanical state detection method based on non-negative tensor decomposition provided in this application includes:

[0054] S1: Obtain the short-circuit current signal and vibration signal of the transformer winding during short-circuit impact; perform a short-circuit impact test on the transformer winding to obtain the short-circuit current signal i(t) and vibration signal s of the transformer winding during short-circuit impact 0 (t), such as Figure 2 As shown in FIG. 1 , a vibration signal for detecting the mechanical state of a transformer winding provided in an embodiment of the present application is provided. Among them, the length of the short-circuit current signal is N 0 , the sampling frequency is f 0 , the vibration signal length is N 1 , the sampling frequency is f 1 ; Here, f 0 =10kHz, f 1 =51200kHz, N 0 =3000, N 1 =30720.

[0055] S2: According to the short-circuit current signal i(t) and the vibration signal s 0 (t), calculate and obtain the time domain envelope of the short-circuit current signal and the time domain envelope of the vibration signal;

[0056] The calculation formula of the short-circuit current signal time domain envelope is:

[0057]

[0058]

[0059] Among them, i(t) is the short-circuit current signal, i h (t) is the Hilbert transform result of the short-circuit current signal.

[0060] Depend on Figure 3 It can be seen that the present application provides a schematic diagram of the time domain envelope of the vibration signal used for detecting the mechanical state of the transformer winding;

[0061] The embodiment of the present application uses adaptive non-decimated morphological wavelet transform to calculate the time domain envelope of the vibration signal. The specific steps are as follows:

[0062] See also Figure 4 , is a schematic flow chart of a method for calculating a time domain envelope of a vibration signal described in this application;

[0063] S21: Find all local maximum points in the vibration signal; record all local maximum points as s 1 (M), where M is the number of all local maximum points in the vibration signal. The method for confirming the local maximum point is as follows: take the derivative of the reduced vibration signal to obtain the sequence s d (t), and then calculate the product of two adjacent points in the sequence and record it as cs di (t), cs di (t) = s di (t)×s d(i-1) (t), i=1,2,L,N 1 -1, according to the cs obtained after multiplication di (t) and the sequence s d The positive and negative values ​​of (t) are used to find all the local maximum points of the vibration signal in turn. The specific method is:

[0064] When cs di (t)<0, if cs di (t)<0 and s d(i-1) (t)>0, then s (i-1) (t) is the local maximum point;

[0065] When cs di When (t)>0, s(i-1) (t) is a non-extreme point;

[0066] When cs di (t) = 0, if s d(i-1) (t) = 0, calculate the two points s i (t) and s i-2 (t) = 0, let cs di (t)'=s i (t)×s i-2 (t), if cs di (t)'<0 and s d(i-2) (t)>0, then s i-1 (t) is a local maximum point, if s d(i-2) (t)=0, then s i-1 (t) is a non-extreme point.

[0067] S22: Represent all positive local maxima and all negative local maxima in the local maximum points as data sequences; the data sequences are respectively denoted as ps(M 1 ) and ns(M 2 ), where M1 and M2 are the numbers of positive local maxima and negative local maxima, respectively.

[0068] S23: According to the data sequence ps(M 1 ) and ns(M 2 ) constructs a rectangular structure element, and based on this, constructs a rectangular structure element with a length and a height of L and H respectively, to obtain a first structure element B;

[0069] The first structure element B is constructed by a matrix of length L and height H:

[0070] L=2×ceil(0.5×min(L(i)))i=1,2,L,min(M 1 ,M 2 );

[0071] H=0.5×min(H(i))i=1,2,L,min(M 1 ,M 2 );

[0072] L(i) = ps(i) - ns(i);

[0073]

[0074] Where: abs means taking the absolute value, ns(1) and ps(1) respectively represent the first positive local maximum and the first negative local maximum, and H(i) means there are two cases: one is when the first positive local maximum ns(1) is greater than the first negative local maximum ps(1), the minimum value after taking the absolute value is taken. The other is when the first positive local maximum ns(1) is less than the first negative local maximum ps(1), the minimum value after taking the absolute value is taken.

[0075] S24: performing multi-scale morphological wavelet decomposition on the vibration signal according to the first structural element to obtain a first scale signal and a detail signal;

[0076]

[0077] y j+1 =(id-T)(x j );

[0078] Where: x j+1 and j+1 are the scale signal and detail signal at the j+1th scale respectively; id is the identity operator; T(·) represents the mathematical morphology operator, where the average value of erosion and dilation is used; ε is the erosion operator; δ is the dilation operator.

[0079] S25: Calculate the vibration signal and the first scale signal to obtain a first correlation coefficient, select the scale signal corresponding to the maximum value of the first correlation coefficient, and obtain a second scale signal;

[0080] S26: Demodulate and calculate the vibration signal time domain envelope according to the second scale signal.

[0081] The vibration signal time domain envelope calculation formula is:

[0082]

[0083] Where: ys R and ys I are respectively the real part and imaginary part of the selected second-scale signal.

[0084] The third-order non-negative tensor is constructed based on the time-domain envelope of the short-circuit current signal and the time-domain envelope of the vibration signal. Specifically, the third-order non-negative tensor The first order length is I 3 The transpose of the time series of the vibration signal time domain envelope and the second order length is I 3 The time series of the vibration signal time domain envelope, and the third order length is I 3 The time domain envelope of the short-circuit current signal is constructed.

[0085] Depend on Figure 5 It can be seen that the flowchart of the calculation method of the first core tensor of this application is as follows;

[0086] S5: Calculate the third-order non-negative tensor The core tensor of , obtain the first core tensor;

[0087] S51: Establishing a solution model for the first core tensor;

[0088]

[0089] Us (1) ≥0,U (2) ≥0,U (3) ≥0;

[0090] Where: is the objective function, represents the Frobenius norm of the matrix; U (1) , U (2) and U (3) are the 1st, 2nd and 3rd order projection tensors respectively; ||U (3) || 1 represents the 1-norm of the 3rd-order projection tensor; λ||U (3) || 1 Represents a sparse constraint term, which is used to eliminate redundant terms in the 3rd-order projection tensor.

[0091] S52: Using the gradient descent method to solve, obtain the first objective function;

[0092]

[0093] S53: Perform iterative calculation on the first objective function, and if a convergence condition is met, output the first core tensor.

[0094] Initialize the number of iterations k=0, λ=0.01, ε=0.001, where ε is the preset iteration convergence threshold and is also the convergence condition in this application;

[0095] The 1st, 2nd and 3rd order projection tensors are randomly initialized and recorded as and Ensure non-negative;

[0096] Let k = k + 1, and update the 1st, 2nd, and 3rd order projection tensors according to the following formulas: and Denoted as:

[0097]

[0098]

[0099]

[0100] Where: 1 is a 3×3 matrix of all 1s; represents element-wise multiplication; e represents element-wise division.

[0101] If the convergence condition is met Then output the first core tensor and projection matrix U (1) , U (2) and U (3) , where the first core tensor The dimension is C 1 ×C 2 ×C 3 ;

[0102] S54: If the convergence condition is not met, continue the iterative calculation.

[0103] Depend on Figure 6 It can be seen that this is a flowchart of the calculation method of the first statistic in this application.

[0104] S6: Calculate a first control limit based on the first statistic;

[0105] S61: Performing modular expansion on the first core tensor to obtain a first matrix and a second matrix;

[0106] The first core tensor Expand modulo 1 and modulo 2 to get a dimension of C 1 ×C 2 C 3 and C 2 ×C 1 C 3 The first matrix G 1 and the second matrix G 2

[0107] S62: Calculate the eigenvalue of the first matrix to obtain a first eigenvalue;

[0108] Calculate the first matrix G 1 Get the first eigenvalue λ 1 ,λ 2 ,L, Here, b 1 is the number of eigenvalues;

[0109] S63: Calculate the eigenvalue of the second matrix to obtain a second eigenvalue;

[0110] Calculate the second matrix G 2 The second eigenvalue λ is obtained1 ,λ 2 ,L, Here, b 2 is the number of eigenvalues;

[0111] S64: constructing the first statistic Γ according to the first eigenvalue and the second eigenvalue;

[0112]

[0113] Where: L 1 For b 1 The number of zero elements in the diagonal matrix composed of eigenvalues; L 2 For b 2 The number of zero elements in the diagonal matrix composed of eigenvalues; η 1 and η 2 is a constant, where η 1 =η 2 =0.5;

[0114] The first control limit is determined by the 3σ criterion of the first statistic. The 3σ criterion is a commonly used mathematical method and will not be described in detail here. Then the first statistic is judged whether it belongs to the interval range of the first control limit. If it does, it means that the mechanical state of the transformer winding is normal; if it does not, it means that the mechanical state of the transformer winding is abnormal and has changed. At this time, timely maintenance is required to avoid major failures.

[0115] This application calculates and analyzes the vibration signal and current signal under the short-circuit impact of the transformer, extracts the time domain envelope of the short-circuit current signal through amplitude demodulation using Hilbert transform, calculates the time domain envelope of the vibration signal using adaptive non-sampling morphological wavelet transform, constructs a third-order non-negative tensor using the time domain envelope of the short-circuit current signal and the time domain envelope of the vibration signal, calculates the core tensor of the third-order non-negative tensor under sparse constraints based on the gradient descent method, constructs the statistics and control limits of the core tensor, and discriminates the mechanical state of the transformer winding through the distribution of the control limits. Accurate detection of the mechanical state of the transformer winding is achieved, so that the hidden dangers of the initial faults of the transformer winding can be effectively identified, and then effective operation and maintenance strategies can be adopted to avoid major faults and improve the reliability of transformer operation.

[0116] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of the present invention, which follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present invention are indicated by the appended claims.

[0117] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A transformer winding mechanical state detection method based on non-negative tensor decomposition, characterized in that: The method comprises: Obtain short-circuit current signal and vibration signal when transformer winding is short-circuited; According to the short-circuit current signal and the vibration signal, a time domain envelope of the short-circuit current signal and a time domain envelope of the vibration signal are calculated; Constructing according to the time domain envelope of the short-circuit current signal and the time domain envelope of the vibration signal to obtain a third-order non-negative tensor; The third-order non-negative tensor is constructed by the transposition of the time series of the vibration signal time domain envelope and the time series of the vibration signal time domain envelope, and the short-circuit current signal time domain envelope; Calculate the core tensor of the third-order non-negative tensor to obtain a first core tensor; The calculation method of the first core tensor includes: Establishing a solution model for the first core tensor; stU (1) ≥0,U (2) ≥0,U (3) ≥0; Where: is the objective function, represents the Frobenius norm of the matrix; U (1) , U (2) and U (3) are the 1st, 2nd and 3rd order projection tensors respectively; ||U (3) ||1 represents the 1-norm of the 3rd-order projection tensor; λ||U (3) ||1 represents a sparse constraint term, which is used to eliminate redundant terms in the 3rd-order projection tensor; Use the gradient descent method to solve and obtain the first objective function; Iteratively calculating the first objective function, and if a convergence condition is met, outputting the first core tensor; Initialize the number of iterations k=0, λ=0.01, ε=0.001, where ε is the preset iteration convergence threshold and is also the convergence condition in this application; The 1st, 2nd and 3rd order projection tensors are randomly initialized and recorded as and Ensure non-negative; Let k = k + 1, and update the 1st, 2nd, and 3rd order projection tensors according to the following formulas: and Denoted as: Where: 1 is a 3×3 matrix of all 1s; represents element-wise multiplication; e represents element-wise division; If the convergence condition is met Then output the first core tensor and projection matrix U (1) , U (2) and U (3) , where the first core tensor The dimension is C1×C2×C3; If the convergence condition is not met, continue the iterative calculation; Calculating a statistic of the first core tensor to obtain a first statistic; The calculation method of the first statistic includes: Performing modular expansion on the first core tensor to obtain a first matrix and a second matrix; Calculating the eigenvalues ​​of the first matrix to obtain first eigenvalues; Calculating the eigenvalues ​​of the second matrix to obtain second eigenvalues; constructing the first statistic according to the first eigenvalue and the second eigenvalue; Calculate a first control limit based on the first statistic; If the first statistic belongs to the interval of the first control limit, the mechanical state of the transformer winding is normal; If the first statistic does not belong to the interval of the first control limit, the mechanical state of the transformer winding is abnormal.

2. The transformer winding mechanical state detection method based on non-negative tensor decomposition according to claim 1 is characterized in that: The calculation formula of the short-circuit current signal time domain envelope is: Among them, i(t) is the short-circuit current signal, i h (t) is the Hilbert transform result of the short-circuit current signal.

3. The transformer winding mechanical state detection method based on non-negative tensor decomposition according to claim 1 is characterized in that: The calculation method of the vibration signal time domain envelope includes: Acquire all local maximum points in the vibration signal; Representing all positive local maxima and all negative local maxima in the local maximum points as a data sequence; Construct a rectangular structure element according to the data sequence to obtain a first structure element; According to the first structural element, performing multi-scale morphological wavelet decomposition on the vibration signal to obtain a first scale signal; Calculating the vibration signal and the first scale signal to obtain a first correlation coefficient, and selecting a scale signal corresponding to a maximum value of the first correlation coefficient to obtain a second scale signal; The time domain envelope of the vibration signal is calculated according to the second scale signal.

4. The transformer winding mechanical state detection method based on non-negative tensor decomposition according to claim 1 is characterized in that: The convergence condition is a preset iterative convergence threshold.

5. The transformer winding mechanical state detection method based on non-negative tensor decomposition according to claim 1 is characterized in that: The first control limit is determined by applying the 3σ criterion to the first statistic.

Images