CN-121980300-A - Dry-type transformer detection method based on porous medium acceleration algorithm

CN121980300ACN 121980300 ACN121980300 ACN 121980300ACN-121980300-A

Abstract

The invention discloses a dry-type transformer detection method based on a porous medium acceleration algorithm, which belongs to the technical field of transformer detection and comprises the steps of establishing an equivalent electromagnetic porous medium model based on a periodic structure of winding layered arrangement, obtaining a spatial distribution reference model through calibration, arranging a time domain antenna array, directionally transmitting broadband electromagnetic pulses to a winding area, forming a field data set by collecting full-channel time domain echo signals, constructing a rapid time domain finite difference forward model based on the equivalent electromagnetic porous medium model, reconstructing a three-dimensional distribution image of equivalent electromagnetic parameters of a winding through iterative optimization based on a backscatter inversion algorithm, extracting the difference between the three-dimensional distribution image and the spatial distribution reference model, and extracting local equivalent conductivity abnormal characteristics caused by turn-to-turn short circuits, wherein the local equivalent conductivity abnormal characteristics are mapped into quantitative detection results of fault positions, fault ranges and severity. The invention solves the problems of huge calculated amount and incapability of rapidly detecting internal defects of the transformer caused by complex winding structure in the traditional electromagnetic detection method.

Inventors

ZHANG ZHIHAO
XIE ZHICHENG
YAN YINGJIE
LIU YADONG
LI YUHANG
LV YUHUI
BAI JIN
Gu Chuanyu
DENG JUN
ZHOU HAIBIN

Assignees

上海交通大学

Dates

Publication Date: 20260505
Application Date: 20251205

Claims (10)

1. The dry-type transformer detection method based on the porous medium acceleration algorithm is characterized by comprising the following steps of: Establishing an equivalent electromagnetic porous medium model based on a periodic structure of winding layered arrangement, and obtaining a spatial distribution reference model of equivalent dielectric constant and equivalent conductivity under a healthy state through calibration; arranging a time domain antenna array on a dry type transformer shell, directionally transmitting broadband electromagnetic pulses to a winding area, and acquiring full-channel time domain echo signals to form a field data set; According to the field data set, constructing a fast time domain finite difference forward model based on an equivalent electromagnetic porous medium model, and reconstructing a three-dimensional distribution image of equivalent electromagnetic parameters of the winding through iterative optimization based on a back scattering inversion algorithm; Based on the difference between the three-dimensional distribution image and the spatial distribution reference model, the local equivalent conductivity abnormal characteristics caused by turn-to-turn short circuits are extracted, and the local equivalent conductivity abnormal characteristics are mapped into quantitative detection results of fault positions, fault ranges and severity according to an electromagnetic loss mechanism.
2. The method for detecting a dry-type transformer based on a porous medium acceleration algorithm according to claim 1, wherein the step of establishing an equivalent electromagnetic porous medium model based on a periodic structure of winding layer arrangement, and obtaining a spatial distribution reference model of equivalent dielectric constant and equivalent conductivity in a healthy state by calibration, comprises the steps of: On the basis of the periodic structure of the winding layered arrangement, defining a single repeating unit comprising a wire and an outer insulating layer thereof as a wrapping characterization unit, establishing an equivalent electromagnetic porous medium model based on a Floquet periodic boundary condition on the scale of the wrapping characterization unit; repeatedly executing operations of directionally transmitting broadband electromagnetic pulse to a winding area and collecting full-channel time domain echo signals for each calibration sample to form a health standard field data set corresponding to each calibration sample; based on an equivalent electromagnetic porous medium model, performing iterative fitting on a health standard field data set corresponding to each calibration sample by using a reverse optimization algorithm to respectively obtain an equivalent dielectric constant spatial distribution parameter and an equivalent conductivity spatial distribution parameter corresponding to each calibration sample; Obtaining statistical distribution characteristics of equivalent dielectric constant space distribution parameters and equivalent conductivity space distribution parameters of all calibration samples through statistical analysis; and generating a spatial distribution reference model of equivalent dielectric constant and equivalent conductivity in a healthy state based on the Floquet periodic boundary condition and the statistical distribution characteristic.
3. The dry transformer detection method based on a porous medium acceleration algorithm according to claim 2, wherein establishing an equivalent electromagnetic porous medium model based on Floquet periodic boundary conditions on the scale of the wraparound characterization unit comprises: Establishing a three-dimensional periodic grid model according to the periodic structure of the winding layered arrangement; Based on the Floquet periodic boundary condition, solving macroscopic electromagnetic response of the wrapping characterization unit in a target detection frequency band; Based on the macroscopic electromagnetic response, calculating an equivalent complex dielectric constant spectrum of the wraparound characterization unit by using an electromagnetic homogenization method, wherein the imaginary part of the equivalent complex dielectric constant spectrum is used for characterizing the equivalent conductivity; and endowing the three-dimensional periodic grid model with parameters of an equivalent complex dielectric constant spectrum to generate an equivalent electromagnetic porous medium model.
4. The method for detecting the dry-type transformer based on the porous medium acceleration algorithm according to claim 2, wherein the iterative fitting is performed on the health standard field data set corresponding to each calibration sample by using a reverse optimization algorithm based on the equivalent electromagnetic porous medium model, so as to obtain the equivalent dielectric constant spatial distribution parameter and the equivalent conductivity spatial distribution parameter corresponding to each calibration sample respectively, and the method comprises the following steps: based on the equivalent complex dielectric constant spectrum parameter of the wrapping characterization unit, setting initial values of the equivalent dielectric constant and the equivalent conductivity spatial distribution parameter, and determining equivalent electromagnetic parameter spatial distribution; Calling a fast time domain finite difference forward model based on equivalent electromagnetic parameter spatial distribution, and performing simulation to generate a corresponding prediction time domain echo signal; calculating the difference between the predicted time domain echo signal and the health standard field data set, and constructing an optimized objective function aiming at minimizing the difference; And iteratively solving the optimization objective function by using a gradient optimization algorithm, synchronously updating the values of the spatial distribution parameters of the equivalent dielectric constant and the equivalent conductivity on all spatial units of the three-dimensional periodic grid model until a preset convergence condition is met, and outputting the spatial distribution parameters of the equivalent dielectric constant and the spatial distribution parameters of the equivalent conductivity corresponding to each calibration sample.
5. The method for detecting a dry-type transformer based on a porous medium acceleration algorithm according to claim 2, wherein obtaining the statistical distribution characteristics of the equivalent dielectric constant spatial distribution parameters and the equivalent conductivity spatial distribution parameters of all the calibration samples by statistical analysis comprises: the equivalent dielectric constant space distribution parameters and the equivalent conductivity space distribution parameters of all the calibration samples are aligned and registered according to the space positions of the three-dimensional periodic grid model; Respectively constructing parameter value sets of equivalent dielectric constants and equivalent conductivities of each space unit in all healthy calibration samples aiming at each space unit of the three-dimensional periodic grid model; Respectively carrying out normal distribution fitting on parameter value sets of equivalent dielectric constants and equivalent conductivities of each space unit, and extracting first statistical distribution characteristics representing numerical centers and fluctuation ranges of normal distribution; based on the layered arrangement rule of the winding structure, carrying out space correlation analysis along the layers in the winding layer on the first statistical distribution characteristics of all the space units to generate second statistical distribution characteristics representing the parameter space distribution rule; And fusing the first statistical distribution characteristics and the second statistical distribution characteristics of all the space units to form statistical distribution characteristics of the equivalent dielectric constant spatial distribution parameters and the equivalent conductivity spatial distribution parameters of all the calibration samples.
6. The method for detecting a dry-type transformer based on a porous medium acceleration algorithm according to claim 5, wherein generating a spatial distribution reference model of equivalent dielectric constant and equivalent conductivity in a healthy state based on the Floquet periodic boundary condition and the statistical distribution feature comprises: Constructing a reference grid model corresponding to the three-dimensional periodic grid model space based on the periodic structure of the winding layered arrangement and the Floquet periodic boundary condition; Respectively endowing the numerical average value of the equivalent dielectric constant and the equivalent conductivity of each space unit in the first statistical distribution characteristic to the corresponding space unit of the reference grid model to form nominal distribution of reference parameters; The standard deviation of the equivalent dielectric constant and the equivalent conductivity of each space unit in the first statistical distribution characteristics is respectively endowed to the corresponding space units of the reference grid model to form confidence interval distribution of the reference parameters; Based on the second statistical distribution characteristics, carrying out space consistency constraint and smoothing correction between layers in the winding layer along with the nominal distribution of the reference parameters; and fusing the spatial consistency constraint with the smooth modified standard parameter nominal distribution, the confidence interval distribution and the second statistical distribution characteristic to generate a spatial distribution standard model of equivalent dielectric constant and equivalent conductivity in a health state.
7. The method for detecting a dry transformer based on a porous medium acceleration algorithm according to claim 1, wherein a time domain antenna array is arranged in a dry transformer housing, a wideband electromagnetic pulse is directionally emitted to a winding area, and a field data set is formed by collecting full channel time domain echo signals, comprising: arranging a time domain antenna array on the surface of the dry-type transformer shell, and enabling the arrangement of the time domain antenna array to cover the axial range and the radial range of the winding area; Sequentially selecting single antenna units in the time domain antenna array as a transmitting source, and directionally transmitting broadband electromagnetic pulses to the winding area; Synchronously acquiring all channel time domain echo signals received by all antenna units in the time domain antenna array when wideband electromagnetic pulses are transmitted each time; traversing all antenna units as transmitting sources, repeatedly executing the directional transmitting and full-channel collecting operation, and combining all acquired full-channel time domain echo signals to form a field data set.
8. The method for detecting a dry-type transformer based on a porous medium acceleration algorithm according to claim 1, wherein constructing a fast time domain finite difference forward model based on an equivalent electromagnetic porous medium model according to a field data set, and reconstructing a three-dimensional distribution image of equivalent electromagnetic parameters of windings through iterative optimization based on a backscatter inversion algorithm, comprises: Constructing a time domain finite difference forward grid corresponding to the three-dimensional periodic grid model space based on the equivalent electromagnetic porous medium model; setting initial distribution of the equivalent dielectric constant and the equivalent conductivity on all space units of the finite difference forward grid in the time domain; based on the equivalent electromagnetic parameter distribution under the current iteration, driving a rapid time domain finite difference forward model, and generating a corresponding prediction field data set in a simulation mode; calculating residual errors between the predicted field data set and the field data set, and constructing an inversion objective function with the residual errors as targets; adopting a gradient optimization algorithm to iteratively solve the inversion objective function, and synchronously updating the distribution parameters of the equivalent dielectric constant and the equivalent conductivity on each space unit of the finite difference forward grid of the time domain until a preset convergence criterion is met; And reconstructing the equivalent electromagnetic parameter distribution meeting the convergence criterion into a three-dimensional distribution image of the winding equivalent electromagnetic parameter through mapping.
9. The method for detecting a dry-type transformer based on a porous medium acceleration algorithm according to claim 8, wherein constructing a time-domain finite difference forward grid spatially corresponding to the three-dimensional periodic grid model based on the equivalent electromagnetic porous medium model, comprises: determining the maximum space discrete step length of the time domain finite difference forward grid in the winding area based on the equivalent complex dielectric constant spectrum parameter; Generating cell nodes of the time domain finite difference forward grid corresponding to the space positions of the space cells one by one according to the geometric center coordinates of the space cells in the three-dimensional periodic grid model; Calculating and setting a unified time iteration step length meeting a numerical stability condition according to the variation range of the equivalent complex dielectric constant spectrum parameter in a target detection frequency band; And taking the equivalent complex dielectric constant spectrum parameter as an initial electromagnetic parameter of each grid unit corresponding to the winding area in the time domain finite difference forward grid to obtain a corresponding time domain finite difference forward grid.
10. The method for detecting a dry-type transformer based on a porous medium acceleration algorithm according to claim 1, wherein the extracting of the local equivalent conductivity anomaly characteristic caused by the turn-to-turn short circuit based on the difference between the three-dimensional distribution image and the spatial distribution reference model, the mapping into the quantized detection result of the fault location, the fault range and the severity according to the electromagnetic loss mechanism, comprises: Comparing the equivalent conductivity distribution in the three-dimensional distribution image with the confidence interval distribution of the reference parameters in the spatial distribution reference model space by space unit, and identifying abnormal space units exceeding the upper limit of the confidence interval; Based on an electromagnetic loss mechanism, clustering the abnormal space units with continuous space into local abnormal areas, and calculating the space mass center of each local abnormal area to determine the fault position; Counting the number of abnormal space units contained in each local abnormal region, and determining a fault range; extracting the average abnormal amplitude of the equivalent conductivity in each local abnormal region, and calculating the severity level of the turn-to-turn short circuit according to a preset fault degree mapping model based on the position information of each local abnormal region in the winding structure; And fusing the fault position, the fault range and the severity level to generate the quantitative detection result.

Description

Dry-type transformer detection method based on porous medium acceleration algorithm Technical Field The invention relates to the technical field of transformer detection, in particular to a dry-type transformer detection method based on a porous medium acceleration algorithm. Background In recent years, in the field of detection of internal defects of transformers, conventional electromagnetic detection methods have been dominant for a long time. The technology is based on the electromagnetic induction principle, excitation signals are applied to a transformer winding, electromagnetic field distribution data around the winding is collected by using a sensor, and then the winding structure parameters and internal defect characteristics are inverted. In the specific implementation, a mathematical model of the winding is established by adopting finite element analysis or a boundary element method, electromagnetic field distribution is simulated through numerical calculation, and then defect characteristics are extracted by combining a signal processing technology. The method is proved to be capable of effectively identifying typical faults such as winding looseness and turn-to-turn short circuit in early research, is widely applied to power equipment state evaluation and fault diagnosis scenes, and becomes one of detection means of industry standards. However, as the capacity level of transformers increases and the winding structure becomes complicated, limitations of the conventional electromagnetic detection method are increasingly highlighted. The method has the core defects that due to complex winding geometry, a high-precision three-dimensional finite element model needs to be constructed, so that the mesh division needs to reach micron-level precision, an equation set with millions of degrees of freedom needs to be processed in single calculation, the calculation takes a long time of hours or even days, the field rapid detection requirement cannot be met, meanwhile, due to the complex winding structure, electromagnetic field distribution presents strong nonlinear characteristics, the traditional inversion algorithm is easy to fall into a local optimal solution, the identification sensitivity on micro defects is insufficient, the omission ratio is as high as 15% -20%, and the reliability of detection results is seriously affected. The technical bottlenecks make the traditional method difficult to realize efficient and accurate defect positioning in the field detection of a large-scale power transformer, and become a 'neck clamping' problem which restricts the industrial application of electromagnetic detection technology. Disclosure of Invention The invention solves the technical problems that the traditional electromagnetic detection method has huge calculated amount and can not quickly detect the internal defects of the transformer due to the complex winding structure. In order to solve the technical problems, the invention provides the following technical scheme: as a preferable scheme of the dry-type transformer detection method based on the porous medium acceleration algorithm, the invention comprises the following steps: Establishing an equivalent electromagnetic porous medium model based on a periodic structure of winding layered arrangement, and obtaining a spatial distribution reference model of equivalent dielectric constant and equivalent conductivity under a healthy state through calibration; arranging a time domain antenna array on a dry type transformer shell, directionally transmitting broadband electromagnetic pulses to a winding area, and acquiring full-channel time domain echo signals to form a field data set; According to the field data set, constructing a fast time domain finite difference forward model based on an equivalent electromagnetic porous medium model, and reconstructing a three-dimensional distribution image of equivalent electromagnetic parameters of the winding through iterative optimization based on a back scattering inversion algorithm; Based on the difference between the three-dimensional distribution image and the spatial distribution reference model, the local equivalent conductivity abnormal characteristics caused by turn-to-turn short circuits are extracted, and the local equivalent conductivity abnormal characteristics are mapped into quantitative detection results of fault positions, fault ranges and severity according to an electromagnetic loss mechanism. Further, an equivalent electromagnetic porous medium model is established based on a periodic structure of winding layered arrangement, and a spatial distribution reference model of equivalent dielectric constant and equivalent conductivity in a healthy state is obtained through calibration, and the method comprises the following steps: defining a single repeating unit comprising a wire and an outer insulating layer thereof as a wrap-around characterizing unit based on the periodic structure of the winding layer