Transformer partial discharge positioning detection method based on ultrahigh frequency signal

Abstract

A transformer partial discharge positioning detection method based on ultrahigh frequency signals is characterized by comprising the following steps: S1, estimating time delay of signals obtained by an ultrahigh frequency (UHF) sensor array based on a generalized cross-correlation method; and S2, arranging and combining signals of the UHF sensor array, and solving coordinates of a positioning point based on an improved multi-moth optimization (MFO) algorithm according to the time delay obtained in the step S1. And S3, clustering the coordinates of the plurality of positioning points obtained in the step S2 based on a density clustering algorithm (DBSCAN), and selecting the geometric center position of the category with the maximum sample number as the final local discharge source coordinate.

Description

Technical Field

[0001] This invention belongs to the field of equipment insulation defect detection, and relates to a method for locating and detecting partial discharge in transformers based on ultra-high frequency signals. Background Technology

[0002] The degree of insulation degradation in transformers has always been a crucial indicator of their health. When a transformer develops air bubbles within its insulation structure due to manufacturing defects or exposure to harsh environments during transportation, leading to insulation degradation, partial discharge can easily occur inside the transformer. Extensive previous research has shown a direct relationship between the degree of insulation degradation and the location of the partial discharge source. Furthermore, once on-site maintenance personnel detect partial discharge within the transformer, accurately locating the discharge source is critical, as it directly impacts the efficiency of subsequent maintenance strategy development and implementation. Therefore, research on transformer partial discharge location technology is of great significance for transformer equipment condition monitoring and fault diagnosis.

[0003] When partial discharge occurs, it generates various physical signals such as light, ultrasound, and ultra-high frequency (UHF) electromagnetic waves. Correspondingly, there are partial discharge detection methods such as optical methods, ultrasonic methods, and UHF methods. Among them, the UHF method, with its excellent anti-interference ability and high sensitivity, has stood out in the field of transformer partial discharge detection and has become a current research hotspot. This method achieves high-precision location of the partial discharge source by capturing UHF electromagnetic wave signals in the range of 300 to 3000 MHz generated inside the transformer.

[0004] Therefore, transformer partial discharge localization technology based on ultra-high frequency signals is a key component of partial discharge detection. Summary of the Invention

[0005] The purpose of this invention is to improve the accuracy of UHF partial discharge location in transformers. It provides a transformer partial discharge location detection method based on UHF signals. By combining three algorithms—Generalized Secondary Cross-Correlation (GSCC), Improved Moth to a Flame (MFO) algorithm, and Density Clustering (DBSCAN) algorithm—it achieves time delay estimation, location equation solving, and location point correction for UHF partial discharge signals in transformers.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for locating and detecting partial discharge in transformers based on ultra-high frequency signals, characterized by the following steps:

[0008] S1. Estimating the time delay of the signal obtained from the UHF sensor array based on the generalized cross-correlation method, specifically including:

[0009] S1.1 Calculate the autocorrelation power spectrum of the first signal x1(t);

[0010] S1.2 Find the cross-correlation function of the first signal x1(t) and the second signal x2(t);

[0011] S1.3 Perform a second cross-correlation on the autocorrelation power spectrum and cross-correlation function and then perform weighted processing to obtain a weighted cross-power spectrum function;

[0012] S1.4 Perform an inverse Fourier transform on the weighted cross-power spectrum function to obtain the time delay function, and take the peak value of the function as the desired time delay;

[0013] S2. Arrange and combine the signals of the UHF sensor array, and solve for the location point coordinates based on the time delay obtained in step S1, using the improved Multiple Moth Optimization (MFO) algorithm. Specifically, this includes:

[0014] S2.1 Arrange and combine the signals of M-1 UHF sensor arrays to obtain L combination results;

[0015] S2.2 For each combination result, the improved MFO algorithm is used to calculate the coordinates of the positioning point;

[0016] S3. Based on the density clustering algorithm (DBSCAN), cluster the coordinates of the multiple positioning points obtained in step S2, and select the geometric center of the category with the largest number of samples as the final local discharge source coordinates.

[0017] Furthermore, in step S1.1, when obtaining the autocorrelation power spectrum, the frequency domain signal obtained after the Fourier transform of the first signal x1(t) and its conjugate are used; in step S1.2, when obtaining the cross-correlation function, the conjugate of the frequency domain signal obtained after the Fourier transform of the second signal x2(t) is used.

[0018] Furthermore, the improved MFO algorithm in step S2.2 specifically includes:

[0019] S2.2.1 Randomly generate an initial moth population, wherein the number of moths and the dimension of the problem variables are determined according to the distance range of the inner wall of the transformer;

[0020] S2.2.2 Calculate the fitness of each moth population to obtain the fitness matrix of the moth population;

[0021] S2.2.3 Sort the moth population according to fitness to form the flame population, and calculate the fitness matrix of the flame population;

[0022] S2.2.4 During the iteration process, the moth moves towards the flame along a logarithmic spiral curve in order to find a better solution;

[0023] In S2.2.5, the flame population adaptively decreases during the iteration process.

[0024] Furthermore, the DBSCAN algorithm in step S3 specifically includes:

[0025] S3.1 Initialization parameters, including input sample set, neighborhood radius ε and neighborhood density threshold MinPts;

[0026] S3.2 Find the core object, which is the sample point whose number of samples in the neighborhood is greater than or equal to MinPts;

[0027] S3.3 starts with the core object and expands the clusters until the clustering termination condition is met;

[0028] S3.4 outputs the clustering results, selecting the geometric center of the cluster with the largest number of samples as the final local discharge source coordinates.

[0029] Furthermore, it also includes optimizing the layout of the sensor array to improve positioning accuracy.

[0030] Furthermore, it also includes preprocessing the UHF signal to reduce noise interference.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0032] 1) This invention integrates the Generalized Cross-Correlation (GSCC) method, the improved Multi-Moth Optimization (MFO) algorithm, and the Density Clustering (DBSCAN) algorithm into the field of transformer partial discharge location detection. This cross-domain algorithm integration not only improves the location accuracy but also optimizes the solution process, which is a major innovation in transformer partial discharge location detection technology.

[0033] 2) Traditional time delay estimation methods may be affected by noise, signal attenuation, and other factors, leading to inaccurate estimation results. This invention uses the GSCC method, which utilizes the autocorrelation power spectrum and cross-correlation function of the signal to perform secondary cross-correlation and weighted processing, effectively improving the accuracy of time delay estimation.

[0034] 3) Traditional optimization algorithms may get stuck in local optima, leading to inaccurate localization results. This invention improves the MFO algorithm, achieving an organic combination of global search and local optimization, effectively avoiding the problem of local optima and improving solution efficiency. During the iteration process, by adaptively reducing the number of flames, the algorithm can converge to the optimal solution more quickly while maintaining its global search capability.

[0035] 4) Traditional clustering methods are susceptible to noise, leading to inaccurate clustering results. This invention utilizes the DBSCAN algorithm to automatically identify and eliminate noisy data, ensuring the reliability and accuracy of the clustering results. By clustering multiple location point coordinates using the DBSCAN algorithm and selecting the geometric center of the category with the largest number of samples as the final local discharge source coordinates, the positioning accuracy is effectively improved.

[0036] 5) This invention also considers the layout optimization of the sensor array and the preprocessing of ultra-high frequency signals, further improving the overall performance and positioning accuracy of the system.

[0037] This invention not only improves positioning accuracy and solution efficiency, but also enhances the system's adaptability and robustness, providing new ideas and methods for the development of transformer partial discharge location and detection technology. Attached Figure Description

[0038] Figure 1 Schematic diagram of the generalized quadratic cross-correlation algorithm;

[0039] Figure 2 A schematic diagram of the improved MFO method. Detailed Implementation

[0040] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments, but this should not be construed as limiting the scope of protection of the present invention.

[0041] This embodiment discloses a transformer partial discharge (PD) location detection method based on ultra-high frequency (UHF) signals. The method mainly includes three core steps: estimating the signal delay obtained from the UHF sensor array using the GSCC method; solving for the location point coordinates using an improved multi-moth optimization (MFO) algorithm; and clustering the obtained multiple coordinate points using a density clustering algorithm (DBSCAN) to correct the location results. The details are as follows.

[0042] Step 1: Estimate the signal delay of the UHF sensor array based on the GSCC method. The purpose is to estimate the time delay between the two signals x1(t) and x2(t) received by the two UHF sensors using the GSCC (Generalized Cross-Correlation) method. The specific steps include:

[0043] Step 1.1 After performing a Fourier transform on the signal x1(t), calculate its autocorrelation power spectrum G. 11 (ω):

[0044]

[0045] In the formula, This represents the conjugate of the frequency domain signal X1(ω) obtained after the Fourier transform of signal x1(t);

[0046] After performing a Fourier transform on the signal x2(t), calculate its autocorrelation power spectrum G. 22 (ω):

[0047]

[0048] In the formula, This represents the conjugate of the frequency domain signal X2(ω) obtained after the Fourier transform of the signal x2(t);

[0049] Step 1.2 Calculate the cross-correlation function G of signals x1(t) and x2(t). 12 (ω);

[0050]

[0051] In the formula, X1(ω) represents the frequency domain signal obtained after the Fourier transform of signal x1(t). It represents the conjugate of the frequency domain signal X2(ω) obtained after the Fourier transform of the signal x2(t).

[0052] Step 1.3 Autocorrelation power spectrum G of signal x1(t) 11 The cross-correlation function G between (ω) and signals x1(t) and x2(t) 12 (ω) is subjected to a second cross-correlation and weighting process to obtain the weighted cross-power spectrum function.

[0053]

[0054] In the formula, G RR (ω) is G 11 (ω) and G 11 The quadratic cross-correlation function of (ω), ψ 12 (ω) is the weight function.

[0055] Step 1.4 Calculate the weighted cross-power spectral function Performing an inverse Fourier transform yields the time delay function, the peak value of which is the desired time delay τ. 12 .

[0056] Step 2: Solve for the coordinates of the positioning point based on the improved MFO algorithm

[0057] Transformer partial discharge localization algorithms inevitably contain errors. Furthermore, due to the dispersion of partial discharge signals and the randomness of interference, the localization results often also have some error. Therefore, the DBSCAN algorithm is subsequently used to correct the localization points. To avoid the influence of large time delay errors from individual sensors on the localization results, and to expand the number of clustered objects and improve localization accuracy, the obtained sensor signals are first arranged and combined, and then an improved MFO algorithm is used to solve for the localization point coordinates. The specific steps include:

[0058] (1) The time delays of the M-1 signals obtained after one measurement of the M sensor array (with one reference sensor) are arranged and combined to obtain L combination results. Since partial discharge localization requires at least 4 sensors, each combination contains at least 4 time delay signals.

[0059]

[0060] This indicates that at least four time-delayed signals are involved in solving the positioning equation.

[0061] (2) After N measurements, a total of L×N time delay groups are obtained. The improved MFO method is used to solve the positioning points of each group.

[0062] The specific method for improving the MFO algorithm is as follows:

[0063] ① Randomly generate an initial population to obtain a moth population M, which represents the location of the local release point;

[0064]

[0065]

[0066] In the formula, and respectively, variable m i，j The upper and lower limits are the distance ranges of the inner wall of the transformer on the x-axis, y-axis, and z-axis; n is the number of moths in the population, and d is the dimension of the problem to be solved, d = 3.

[0067] ② Calculate M for each moth population i fitness OM i The fitness matrix OM of the moth population is obtained as follows:

[0068] OM i =(ΔR-R+R0) T (ΔR-R+R0)

[0069]

[0070] In the formula, r iLet ri represent the distance between the i-th sensor in the M-sensor array and the partial discharge point (i = 1, 2, ..., M-1), r0 represent the distance between the 0th reference sensor and the partial discharge point, and Δr0 represent the distance between the reference sensor and the partial discharge point. i0 =r i -r0=vΔt i0 , Δt i0 Let Δt be the time delay between the i-th sensor and the 0th reference sensor. i0 =t i -t0, v is the propagation speed of ultra-high frequency electromagnetic waves in the medium;

[0071] ③ The moth populations were sorted according to their fitness to form the flame population:

[0072] If the current iteration is the first generation, then according to OM, the moth population is arranged from high to low fitness to become the first generation flame population F, and the fitness matrix OF of the flame population is calculated.

[0073]

[0074] If the iteration is not the first generation, the flame population is updated and reordered when the moth's fitness value is higher than the corresponding flame's fitness value.

[0075] ④ The moth moves towards the flame along a logarithmic spiral curve in search of a better solution:

[0076] In each iteration, the moth's optimal path is a logarithmic spiral curve S with the corresponding flame as the endpoint:

[0077] S(M i F i ) = D i ·e bt ·cos(2πt)+F i

[0078]

[0079] In the formula, D i The distance between the i-th moth and the flame is given by b, which is a logarithmic spiral shape constant (b = 1 in this embodiment), t is a random number representing the proximity of the moth to the flame, and l represents the current iteration number. max M represents the maximum number of iterations. r1 and M r2 All individuals are randomly selected, and σ is a random number of 0 or 1.

[0080] ⑤ Adaptive reduction of flame population: During the iteration process, the number of flames is gradually reduced to improve the global search capability of the algorithm.

[0081] Because there is a lack of correlation between different moth individuals during the search process, the algorithm's global search capability is not high. Therefore, the number of flames is gradually reduced during the iteration process, and the number of flame populations after each iteration is:

[0082]

[0083] In the formula, n represents the current number of flames. Each iteration removes flames with poor fitness. When the number of moths exceeds the number of flames, the remaining moths will update their positions towards the flame with the lowest current fitness value, such as... Figure 2 As shown.

[0084] (6) Repeat steps (2)-(5) until the iteration conditions are met, and output the optimal solution.

[0085] Step 3: Perform clustering based on DBSCAN to select the final partial discharge source coordinates.

[0086] DBSCAN was used to correct and cluster the obtained L×N partial discharge location results. The specific method is as follows:

[0087] (1) Initialize parameters

[0088] Determine the input sample set D = (x1, x2, ..., x...) m ), neighborhood radius ε and neighborhood density threshold MinPts. Initialize the core object set. Cluster number k = 0, unvisited sample set r = D, cluster partitioning

[0089] (2) Find the core object

[0090] For j = 1, 2, ..., m, calculate the relationship between all other points and sample x. j The Euclidean distance is used to determine x. j The ε-neighborhood is N ε (x j )={x i ∈D|distance(x i x j If the number of samples in the sample set satisfies |N ≤ ε}. ε (x j If x ≥ MinPts, then x j Add to the core sample set Ω=Ω∪{x j}

[0091] (3) If The algorithm ends if the condition is met; otherwise, proceed to step (4).

[0092] (4) Update parameters

[0093] In the core object set Ω, randomly select a core object o and initialize the current cluster core object queue Ω. cur ={o}, category index k = k+1, current cluster sample set C k ={o}, the unvisited sample set r = r - {o}.

[0094] (5) Update the current cluster core object queue

[0095] In the current cluster core object queue Ω cur Select a core object o′ and find the ε-neighborhood subsamples N of o′. ε (o′), let Δ=N ε (o′)∩r, update the current cluster sample set C k =C k ∪Δ, the unvisited sample set r = r - Δ, and the current cluster core object queue Ω. cur =Ω cur ∪(Δ∩Ω)-o′.

[0096] (6) Jump to the next cluster

[0097] If the current cluster core object queue Then the current cluster C k Once generated, update the cluster partition C = {C1, C2, ..., C}. k}, the core object set Ω = Ω - C k Proceed to step (3).

[0098] (7) Output of Results

[0099] The final output is a cluster partition C = {C1, C2, ..., C}. k The geometric center of the cluster with the largest number of samples is selected as the final coordinates of the partial discharge source.

[0100] Test results:

[0101] To verify the feasibility and accuracy of the positioning technology proposed in this invention, a simulated defect was set in a full-scale transformer. An external voltage was applied to generate partial discharge and detect the UHF signal. The coordinates of the partial discharge source inside the transformer were (0.65, 0.5, 1.28) m. The dimensions of the transformer in the x-axis, y-axis, and z-axis directions were 0.72 m, 1 m, and 1.48 m, respectively, and the sampling rate was 5 Gsps.

[0102] The signal was detected 7 times, with 10 discharge signals collected consecutively each time. Without initial correction of the positioning results, the positioning results for the 7 sets of data were calculated using the GSCC-improved MFO method, the energy accumulation-GA algorithm, and the feature point-PSO algorithm proposed in this paper. The GSCC algorithm used the Roth weighting function. The results are shown in Table 1.

[0103] Table 1 Comparison of Partial Discharge Localization Algorithm Results

[0104]

[0105]

[0106] According to Table 1, the average positioning errors of the algorithm of this invention, the energy accumulation-GA algorithm, and the feature point-PSO algorithm are 13.44cm, 18.15cm, and 27.42cm, respectively. Compared with the latter two, the positioning error accuracy (error reduction / error before correction) is improved by 35.04% and 40.04%, respectively.

[0107] To verify the impact of the DBSCAN algorithm on positioning accuracy, the positioning results obtained using the GSCC-improved MFO algorithm were corrected using the DBSCAN method. Simultaneously, other clustering methods, k-means clustering and mean-shift clustering, were used for cross-sectional comparison. For the DBSCAN algorithm, parameters ε = 0.15 and MinPts = 5 were selected. Based on experimental comparisons, the K-means algorithm showed better correction results with 2 or 3 clusters. For the mean-shift algorithm, a bandwidth of 0.5 was selected. The results are shown in Table 2.

[0108] Table 2 Comparison of correction results of different clustering algorithms

[0109]

[0110]

[0111] Table 2 shows that the average positioning error obtained using the GSCC-improved MFO method was originally 13.44 cm. After applying different clustering algorithms for positioning correction, the average error of the final positioning results all decreased. The DBSCAN algorithm reduced the average error by 4.3 cm, the K-means algorithm by approximately 2 cm, and the mean-shift algorithm by 0.97 cm. The DBSCAN clustering algorithm yielded the best results.

[0112] Therefore, the GSCC-improved MFO method of this invention has higher positioning accuracy than other traditional partial discharge positioning techniques. The average positioning error of the partial discharge source using the GSCC-improved MFO-DBSCAN method can reach <10cm.

Claims

1. A method for locating and detecting partial discharge in transformers based on ultra-high frequency signals, characterized in that, Includes the following steps: S1. Estimating the time delay of the signal obtained from the UHF sensor array based on the generalized cross-correlation method, specifically including: S1.1 Calculate the autocorrelation power spectrum of the first signal x1(t); S1.2 Find the cross-correlation function of the first signal x1(t) and the second signal x2(t); S1.3 Perform a second cross-correlation on the autocorrelation power spectrum and cross-correlation function and then perform weighted processing to obtain a weighted cross-power spectrum function; S1.4 Perform an inverse Fourier transform on the weighted cross-power spectrum function to obtain the time delay function, and take the peak value of the function as the desired time delay; S2. Arrange and combine the signals of the UHF sensor array, and solve for the location point coordinates based on the time delay obtained in step S1, using the improved Multiple Moth Optimization (MFO) algorithm. Specifically, this includes: S2.1 Arrange and combine the signals of M-1 UHF sensor arrays to obtain L combination results; S2.2 For each combination result, the improved MFO algorithm is used to calculate the coordinates of the positioning point; S3. Based on the density clustering algorithm (DBSCAN), cluster the coordinates of the multiple positioning points obtained in step S2, and select the geometric center of the category with the largest number of samples as the final local discharge source coordinates.

2. The transformer partial discharge location and detection method based on ultra-high frequency signals according to claim 1, characterized in that, In step S1.1, when obtaining the autocorrelation power spectrum, the frequency domain signal obtained after the Fourier transform of the first signal x1(t) and its conjugate are used; in step S1.2, when obtaining the cross-correlation function, the conjugate of the frequency domain signal obtained after the Fourier transform of the second signal x2(t) is used.

3. The transformer partial discharge location and detection method based on ultra-high frequency signals according to claim 1, characterized in that, The improved MFO algorithm in step S2.2 specifically includes: S2.2.1 Randomly generate an initial moth population, wherein the number of moths and the dimension of the problem variables are determined according to the distance range of the inner wall of the transformer; S2.2.2 Calculate the fitness of each moth population to obtain the fitness matrix of the moth population; S2.2.3 Sort the moth population according to fitness to form the flame population, and calculate the fitness matrix of the flame population; S2.2.4 During the iteration process, the moth moves towards the flame along a logarithmic spiral curve in order to find a better solution; In S2.2.5, the flame population adaptively decreases during the iteration process.

4. The transformer partial discharge location and detection method based on ultra-high frequency signals according to claim 1, characterized in that, The DBSCAN algorithm in step S3 specifically includes: S3.1 Initialization parameters, including input sample set, neighborhood radius ε and neighborhood density threshold MinPts; S3.2 Find the core object, which is the sample point whose number of samples in the neighborhood is greater than or equal to MinPts; S3.3 starts with the core object and expands the clusters until the clustering termination condition is met; S3.4 outputs the clustering results, selecting the geometric center of the cluster with the largest number of samples as the final local discharge source coordinates.

5. The transformer partial discharge location and detection method based on ultra-high frequency signals according to any one of claims 1-4, characterized in that, It also includes optimizing the layout of the sensor array to improve positioning accuracy.

6. The transformer partial discharge location and detection method based on ultra-high frequency signals according to any one of claims 1-4, characterized in that, It also includes preprocessing of UHF signals to reduce noise interference.

Images