A transformer winding looseness fault diagnosis method based on improved MPE and K-medoids algorithm

Abstract

The application discloses a transformer winding looseness fault diagnosis method based on an improved MPE and K-medoids algorithm, which is used for forming a transformer winding looseness MPE value criterion and realizing winding looseness fault diagnosis. The method steps are as follows: 1, measuring points are arranged on a transformer box body, vibration signals of each measuring point are acquired, and a vibration amplitude maximum measuring point D is selected as an optimal measuring point; 2, vibration signals of the transformer winding in different states are measured; 3, a particle swarm optimization algorithm is used to optimize parameters in a traditional MPE algorithm, so that the overall change of the MPE value is stable; 4, the MPE value of the measuring point D is calculated by using the optimized MPE algorithm; 5, the MPE values under two adjacent scale factors are selected as horizontal and vertical coordinates of a clustering coordinate system; 6, the K-medoids algorithm is used in the coordinate system to realize accurate classification of the transformer winding fault type; and 7, the MPE value criterion is summarized and formed, and a database is established. The method reduces the cross aliasing phenomenon of the traditional MPE algorithm, and realizes accurate judgment of the transformer fault type.

Description

Technical Field

[0001] This invention belongs to the field of power equipment fault diagnosis, specifically relating to a method for diagnosing transformer winding loosening faults based on improved MPE and K-medoids algorithms. Background Technology

[0002] Loose windings are a major mechanical fault in transformers, and if not investigated promptly, they can pose a threat to the safe and reliable operation of the power grid. Since there is a causal relationship between loose windings and abnormal transformer winding vibrations, identifying such faults using vibration methods is currently a hot topic. Compared to existing transformer winding deformation testing techniques, the biggest advantage of vibration signal analysis is that it uses sensors attached to the surface of the transformer tank for online monitoring. This method has no electrical connection to the entire power system and has no impact on its normal operation, achieving online monitoring safely and reliably. Currently, there is a large amount of research on frequency domain analysis of transformer vibration signals, such as EMD decomposition and VMD decomposition. Most of these methods decompose non-stationary vibration signals into multiple stationary signal components using time-frequency analysis methods, and then calculate the time-frequency characteristics of each component. However, these methods inevitably lose some characteristic information of the original signal during the time-frequency conversion process, such as mode mixing and endpoint effects, thus having certain limitations. Multiscale permutation entropy (MPE) is a method for characterizing the complexity of time series data and can reflect the characteristics of the original time series. However, since the parameters are generally set based on experience, the MPE values ​​calculated from time series of different states often show overlapping and mixed phenomena, thus losing the ability to detect fault types. Summary of the Invention

[0003] To address the aforementioned problems, this invention proposes a method for diagnosing transformer winding loosening faults based on an improved MPE and K-medoids algorithm. This method can accurately detect transformer winding loosening faults.

[0004] The technical solution of this invention is as follows:

[0005] Step 1: Set up P measuring points on the transformer tank to obtain the vibration signals of each measuring point on the transformer, and select the measuring point D with the largest vibration amplitude as the optimal vibration signal of the transformer.

[0006] Step 2: Measure the vibration signals of the transformer windings under different degrees of looseness through measuring point D;

[0007] Step 3: Based on the shortcomings of traditional MPE algorithm in setting parameters based on experience, particle swarm optimization algorithm is used to optimize its parameter selection. The square of the MPE skewness S of transformer vibration signal under multi-scale factor is used as the fitness function to reduce the cross-aliasing phenomenon of MPE value.

[0008] Step 4: Substitute the optimized MPE parameters into the MPE algorithm to calculate the MPE value of the vibration signal at transformer measuring point D;

[0009] Step 5: In a two-dimensional coordinate system, select the MPE values ​​of two adjacent scale factors calculated in Step 4 as the horizontal and vertical coordinates to form a clustering coordinate system of the MPE values ​​of the vibration signal under different states of the transformer winding.

[0010] Step 6: In the coordinate system generated in step 5, the K-medoids algorithm is used to perform cluster analysis on the MPE values ​​of the transformer winding under different states to achieve accurate classification of winding looseness.

[0011] Step 7: Based on Step 6, summarize the MPE value criteria for different states of transformer windings, establish a fault database to diagnose winding loosening.

[0012] The following is a detailed explanation of each step in the technical solution:

[0013] In step 1:

[0014] The data acquisition instrument is a DHDAS model, and the sensor is an IEPE piezoelectric accelerometer, model 1A212E. P vibration sensors are placed at the top and bottom of the transformer housing. When acquiring signals, the vibration amplitude at each measuring point is observed, and the measuring point with the largest amplitude is selected as the optimal measuring point.

[0015] In step 2:

[0016] A simulated fault experiment needs to be conducted on the transformer to control the degree of looseness of the transformer windings and to measure the vibration signal at measuring point D under different winding conditions.

[0017] In step 3:

[0018] First, for a time series {x(i)}, i = 1, 2, ..., N, where N represents the number of sampling points, after reconstructing the phase space using coordinate delay, we get...

[0019]

[0020] In equation (1): m is the embedding dimension; t is the delay time; K = N - (m - 1)t.

[0021] Each row vector of the reconstructed matrix is ​​called a reconstructed component, resulting in K reconstructed components. Elements within each reconstructed component are numbered from left to right to form indices 1, 2, 3, ..., m. These indices are then sorted in ascending order of element value. Each element's value is replaced by its index, resulting in K symbol sequences. When the embedding dimension is m, there are m! possible symbol sequences. All K symbol sequences obtained must be one of m! possible sequences. Therefore, the number of symbol sequences formed by a time signal of length N is k ≤ m!.

[0022] For each symbol sequence, its probability of occurrence after each phase space reconstruction can be calculated as follows:

[0023]

[0024] In equation (2): T i Let be the number of occurrences of the i-th symbol sequence.

[0025] Therefore, the permutation entropy of this time series can be expressed as:

[0026]

[0027] After normalization, we get

[0028]

[0029] Next, by introducing a scaling factor s, we can calculate the permutation entropy of the signal at multiple scaling factors:

[0030] For the original time series x i Coarse-graining was performed to obtain coarse-grained sequences.

[0031]

[0032] In equation (5): s is the scale factor. Indicates to Round down. If s = 1, it degenerates into the original time series.

[0033] The permutation entropy values ​​under different scale factors s are calculated by replacing the original time series with coarse-grained sequences.

[0034] Furthermore, the particle swarm optimization algorithm is used to optimize the parameters N, m, s, and t.

[0035] The position X of particle i in n-dimensional space i = (x1, x2, ..., x n ), speed V i = (v1, v2, ..., v n The optimal position P traversed during the motion. best = (p1, p2, ..., p nThe overall optimal position G traversed by all particles in the swarm. best = (g1, g2, ..., g n Each time, particle i adjusts its speed and position for the next movement using equations (6) and (7).

[0036]

[0037] X i ′=X i +V i ′ (7)

[0038] In equations (6) and (7): ω is the inertia factor; c1 and c2 are the learning factors; r1 and r2 are random numbers in (0, 1).

[0039] In the multi-scale permutation entropy algorithm based on particle swarm optimization, the fitness function is the square of the MPE skewness S under the multi-scale factor.

[0040] S=E[(H PE (X)-H μPE ) 3 ] / [H σPE ] 3 (8)

[0041] F(X) = S 2 (9)

[0042] In equation (8): H PE (X)={H PE (1), H PE (2), ..., H PE (s)} represents the MPE sequence under s scaling factors, H μPE and H σPE These are the mean and standard deviation of the MPE series, respectively.

[0043] The fitness function is chosen to be the square of the MPE skewness, with the goal of minimizing it. A smaller skewness in a dataset indicates more stable numerical changes and a more representative mean. Therefore, the goal of particle swarm optimization is to stabilize the values ​​of each element in the MPE sequence after setting specific parameters, ensuring that the MPE value at any scale factor can characterize the time series.

[0044] In step 4:

[0045] First, the optimized parameters N, m, s, and t obtained in step 3 are used as input parameters for the MPE algorithm. Then, the MPE values ​​of the vibration signals of the transformer winding under different states are calculated respectively.

[0046] In step 5:

[0047] Based on step 4, the MPE values ​​of the transformer winding under various states after parameter optimization were obtained. This series of MPE values ​​showed a stable distribution under each scale factor s, indicating that the MPE value under any scale factor can be used as a characteristic value to characterize the looseness of the winding.

[0048] Using the MPE values ​​of scale factors s=1 and s=2 as the horizontal and vertical coordinates of a two-dimensional coordinate system, a clustering coordinate system is generated. The horizontal coordinate is called the length, and the vertical coordinate is called the width. Thus, a series of cluster points are formed distributed in the clustering coordinate system, and because these cluster points change smoothly under each scale factor, they all cluster near the diagonal of the coordinate system.

[0049] In step 6:

[0050] Based on step 5, the K-medoids algorithm is used to perform cluster analysis on the cluster points in the clustering coordinate system. The algorithm steps are as follows:

[0051] 1) Determine the number of clusters to be K;

[0052] 2) Randomly select K cluster centers from all cluster points;

[0053] 3) Calculate the distance from each point to each center; if the distance from the center of a certain class is smaller, then the point is classified into that class.

[0054] 4) Calculate the distance from each point in each category to other points, and take the point with the smallest sum of distances as the new center;

[0055] 5) Iterate through steps 2, 3, and 4 until the new center is essentially unchanged from the original.

[0056] Thus, accurate classification of MPE values ​​for windings under various fault types was achieved.

[0057] In step 7:

[0058] Further verification in step 6 proved that each MPE value has the ability to characterize the degree of transformer winding looseness, thus forming a set of criterion intervals for judging the degree of transformer winding looseness faults. Subsequent calculations showed that the MPE values ​​of a large number of sample data all fell within the specified intervals, providing a reference for diagnosing the winding looseness of similar transformers. For different transformer models, it is necessary to extract the corresponding MPE feature values ​​according to the proposed method and summarize them into criteria to improve the applicability of the method.

[0059] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention overcomes the phenomenon of MPE values ​​of various types of signals being cross-aliased due to the empirical selection of parameters in traditional MPE algorithms. This phenomenon can significantly affect the accuracy of fault diagnosis models and the establishment of fault criteria. The research results provide a new approach for judging transformer winding loosening faults. Attached Figure Description

[0060] Figure 1 This is a flowchart of the present invention;

[0061] Figure 2 It refers to the arrangement of transformer measuring points;

[0062] Figure 3 This is a diagram of a transformer simulated fault experiment;

[0063] Figure 4 This is a comparison chart of MPE values ​​of the winding under different states before parameter optimization;

[0064] Figure 5 This is a comparison chart of MPE values ​​of the winding under different states after parameter optimization;

[0065] Figure 6 This is a graph showing the clustering results before parameter optimization;

[0066] Figure 7 This is a graph showing the clustering results after parameter optimization. Detailed Implementation

[0067] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0068] The present invention describes a method for diagnosing transformer winding loosening faults based on an improved MPE and K-medoids algorithm. The specific process is detailed in [link to documentation]. Figure 1 The method includes the following steps:

[0069] Step 1: Set up 3 measuring points on the top of the transformer tank. The data acquisition instrument is a DHDAS model, and the sensor is an IEPE piezoelectric accelerometer, model 1A212E. Acquire the vibration signals from each measuring point on the transformer, and select the measuring point 1 with the largest vibration amplitude as the optimal vibration signal for the transformer. Figure 2 As shown.

[0070] Step 2: To simulate a winding looseness fault, the transformer needs to be emptied of oil, the transformer cover removed, and the preload of the faulty phase nuts adjusted via the hydraulic system to determine the degree of winding looseness. The transformer cover is then replaced and oil is added, followed by a power-on test. Therefore, two faults were simulated at phase A: a 40% reduction in winding preload (corresponding to slight winding looseness) and an 80% reduction in winding preload (corresponding to severe winding looseness). Figure 3 As shown.

[0071] Step 3: For the vibration signal at measuring point 1 under each winding state measured in Step 2, perform coordinate delay phase space reconstruction as follows:

[0072]

[0073] In equation (1): m is the embedding dimension; t is the delay time; K = N - (m - 1)t.

[0074] For each symbol sequence of the vibration signal measured at measuring point 1, the probability of its occurrence after each phase space reconstruction can be calculated as follows:

[0075]

[0076] In equation (2): T i Let be the number of occurrences of the i-th symbol sequence.

[0077] Therefore, the arrangement entropy of the vibration signal at measuring point 1 can be expressed as:

[0078]

[0079] After normalization, we get

[0080]

[0081] Next, by introducing a scale factor s, the permutation entropy of the vibration signal at measuring point 1 can be calculated under multiple scale factors:

[0082] For the original time series x i Coarse-graining was performed to obtain coarse-grained sequences.

[0083]

[0084] In equation (5): s is the scale factor. Indicates to Round down. If s = 1, it degenerates into the original time series.

[0085] Furthermore, the particle swarm optimization algorithm is used to optimize the parameters N, m, s, and t of the vibration signal at measuring point 1 when the winding is in each state.

[0086] The position X of particle i in n-dimensional space i = (x1, x2, ..., x n ), speed V i = (v1, v2, ..., v n The optimal position P traversed during the motion. best = (p1, p2, ..., p n The overall optimal position G traversed by all particles in the swarm. best = (g1, g2, ..., g n Each time, particle i adjusts its speed and position for the next movement using equations (6) and (7).

[0087]

[0088] X i ′=X i +V i ′ (7)

[0089] In equations (6) and (7): ω is the inertia factor; c1 and c2 are the learning factors; r1 and r2 are random numbers in (0, 1).

[0090] In the PSO-based multi-scale permutation entropy algorithm, the fitness function is the square of the MPE skewness S under the multi-scale factor.

[0091] S=E[(H PE (X)-H μPE ) 3 ] / [H σPE ] 3 (8)

[0092] F(X) = S 2 (9)

[0093] In equation (8): H PE (X)={H PE (1), H PE (2), ..., H PE (s)} represents the MPE sequence under s scaling factors, H μPE and H σPE These are the mean and standard deviation of the MPE series, respectively.

[0094] Therefore, the optimized parameters for the vibration signal of measuring point 1 under various winding states are formed, as shown in Table 1.

[0095] Table 1. Optimization results of particle swarm optimization parameters

[0096]

[0097] Step 4: Input the MPE parameter optimization results of the vibration signal under each winding state in Step 3 into the MPE algorithm, calculate each MPE value, and compare the optimization effect. Figure 4 , 5 As shown.

[0098] Step 5: Place the MPE values ​​obtained in Step 4 into a clustering coordinate system, and select the MPE values ​​with scale factors s=1 and s=2 as the horizontal and vertical coordinates respectively. The resulting cluster points are clustered near the diagonal.

[0099] Step 6: Perform cluster analysis on the cluster points in the clustering coordinate system using the K-medoids algorithm, such as... Figure 6 ,7 As shown.

[0100] Step 7: Further verification in Step 6 proved that each MPE value has the ability to characterize the degree of transformer winding looseness, thus forming a set of criterion intervals for judging the degree of transformer winding looseness faults. Subsequent calculations showed that the MPE values ​​of a large number of sample data all fell within the specified intervals, providing a reference for diagnosing the winding looseness of similar transformers, as shown in Table 2.

[0101] Table 2. Reference for MPE characteristic value diagnosis under various winding conditions.

[0102]

Claims

1. A method for diagnosing transformer winding loosening faults based on improved MPE and K-medoids algorithms, characterized in that: The method includes the following steps: Step 1: Set up measuring points on the transformer tank, acquire the vibration signals of each measuring point, and select the measuring point D with the largest vibration amplitude as the optimal measuring point; Step 2: Measure the vibration signals of the transformer windings under different conditions; Step 3: Optimize the parameters in the traditional MPE algorithm using the particle swarm optimization algorithm to ensure a smooth overall change in the MPE value. The steps are as follows: First, for a time series {x(i)}, i = 1, 2, ..., N, where N represents the number of sampling points, after reconstructing the phase space using coordinate delay, we get... In the formula: m is the embedding dimension; t is the delay time; K = N - (m - 1)t; For each symbol sequence, its probability of occurrence after each phase space reconstruction can be calculated as follows: In the formula: T i The number of occurrences of the i-th symbol sequence; Therefore, the permutation entropy of this time series can be expressed as: After normalization, we get Next, by introducing a scaling factor s, we can calculate the permutation entropy of the signal at multiple scaling factors: For the original time series x i Coarse-graining was performed to obtain coarse-grained sequences. In the formula: s is the scale factor, Indicates to Round off; if s = 1, then degenerate into the original time series; The permutation entropy values ​​under different scale factors s were calculated by replacing the original time series with coarse-grained sequences. Furthermore, the particle swarm optimization algorithm is used to optimize the parameters N, m, s, and t; The position X of particle i in n-dimensional space i = (x1, x2, ..., x n ), speed V i = (v1, v2, ..., v n The optimal position P traversed during the motion. best = (p1, p2, ..., p n The overall optimal position G traversed by all particles in the swarm. best = (g1, g2, ..., g n Each time, particle i adjusts its speed and position for the next movement using equations (6) and (7). X t ′=X i +V i ′ (7) In the formula: ω is the inertia factor; c1 and c2 are learning factors; r1 and r2 are random numbers in (0, 1); In the multi-scale permutation entropy algorithm based on particle swarm optimization, the fitness function is the square of the MPE skewness S under the multi-scale factor. S=E[(H PE (X)-H μPE ) 3 ] / [H σPE ] 3 (8) F(X)=S 2 (9) In the formula: H PE (X)={H PE (1), H PE (2), ..., H PE (s)} represents the MPE sequence under s scaling factors, H μPE and H σPE These are the mean and standard deviation of the MPE series, respectively; Step 4: Calculate the MPE value of measurement point D using the optimized MPE algorithm; Step 5: Select the MPE values ​​under two adjacent scale factors as the horizontal and vertical coordinates of the clustering coordinate system; Step 6: Use the K-medoids algorithm in this coordinate system to accurately classify transformer winding fault types; Step 7: Summarize the criteria for MPE value and establish a database.

2. The method for diagnosing transformer winding loosening faults based on improved MPE and K-medoids algorithms according to claim 1, characterized in that: In step 1, vibration signals from all measuring points are acquired at the transformer housing, and the measuring point D with the largest vibration amplitude is selected as the optimal vibration signal for the transformer.

3. A method for diagnosing transformer winding loosening faults based on an improved MPE and K-medoids algorithm according to claim 1 or 2, characterized in that: In step 2, a simulated fault experiment needs to be conducted on the transformer to control the degree of looseness of the transformer winding and to measure the vibration signal at measuring point D under different winding conditions.

4. The method for diagnosing transformer winding loosening faults based on improved MPE and K-medoids algorithms according to claim 1, characterized in that: In step 4, firstly, the optimized parameters N, m, s, and t obtained in step 3 are used as input parameters for the MPE algorithm, and then the MPE values ​​of the vibration signals of the transformer winding under different states are calculated respectively.

5. The method for diagnosing transformer winding loosening faults based on improved MPE and K-medoids algorithms according to claim 1, characterized in that: In step 5, the MPE values ​​of scale factors s=1 and s=2 are used as the horizontal and vertical coordinates of the two-dimensional coordinate system to generate a clustering coordinate system. The horizontal coordinate is called the length and the vertical coordinate is called the width.

6. The method for diagnosing transformer winding loosening faults based on improved MPE and K-medoids algorithms according to claim 1, characterized in that: In step 6, the K-medoids algorithm is used to perform cluster analysis on the cluster points in the cluster coordinate system to achieve accurate classification of the MPE values ​​of the winding under each fault type.

7. The method for diagnosing transformer winding loosening faults based on improved MPE and K-medoids algorithms according to claim 1, characterized in that: A set of criteria intervals was established to determine the degree of transformer winding loosening faults. Subsequent calculations showed that the MPE values ​​of a large number of sample data all fell within the specified intervals, providing a reference for diagnosing the winding loosening of similar transformers.

Images