Transformer winding monitoring and diagnosis data acquisition method, system, equipment and medium

Abstract

The invention discloses a transformer winding monitoring and diagnosis data acquisition method, system and device and a medium, and belongs to the technical field of power equipment state monitoring, and the method comprises the steps: constructing a multi-scale wavelet transform and improved adaptive Kalman filtering fusion architecture, the multi-scale wavelet transform and improved adaptive Kalman filtering fusion architecture is used for acquiring a monitoring signal of a transformer winding; performing frequency division processing on the monitoring signal by using the multi-scale wavelet transform, and determining each scale coefficient; adjusting a noise covariance matrix based on improved adaptive Kalman filtering, and carrying out filtering processing on each scale coefficient to determine a filtering result; and reconstructing the filtering result into a reconstructed signal, and performing signal quality evaluation on the reconstructed signal to determine the signal quality of the transformer winding. Through multi-scale decomposition and adaptive filtering, the limitation of a traditional method under non-stationary signal processing and strong noise interference is solved, and the retention degree and the signal-to-noise ratio of signal features are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment condition monitoring, in particular to a method, system, device and medium for collecting monitoring and diagnostic data of transformer windings. Background Art

[0002] In a power system, as a core device, the winding state of a transformer is directly related to the reliability of power grid operation; with the continuous growth of power load and the increasing complexity of the power grid topology, the electromagnetic-thermal coupling effect on the transformer winding is significantly enhanced, resulting in an increasing annual incidence of faults such as local overheating and insulation deterioration.

[0003] However, the traditional periodic maintenance mode has the problem of lagging response when dealing with sudden faults, and in a complex electromagnetic interference environment, the signal-to-noise ratio of the data acquisition system in the existing on-line monitoring technology is generally low, making it difficult to meet the requirements of accurate state assessment; therefore, there is an urgent need for a method that can improve the quality of on-line monitoring data acquisition of transformer windings in a complex power environment to support more accurate and reliable fault diagnosis and early warning. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: how to propose a method for collecting on-line monitoring and diagnostic data of transformer windings based on the fusion of multi-scale wavelet transform and improved adaptive Kalman filter. Aiming at the problem that the data acquisition quality of a single signal processing method in the existing technology is not high in a complex power environment, which in turn affects the accuracy and reliability of the diagnostic results, the multi-resolution analysis ability of multi-scale wavelet transform and the dynamic filtering advantage of the improved adaptive Kalman filter are used to achieve high-precision noise reduction and feature extraction of the monitoring signal of the transformer winding.

[0006] To solve the above technical problem, the present invention provides the following technical solution: a method for collecting monitoring and diagnostic data of transformer windings, which includes the following steps. Construct a fusion architecture of multi-scale wavelet transform and improved adaptive Kalman filter, and the fusion architecture of multi-scale wavelet transform and improved adaptive Kalman filter is used to obtain the monitoring signal of the transformer winding; use the multi-scale wavelet transform to perform frequency division processing on the monitoring signal to determine each scale coefficient; adjust the noise covariance matrix based on the improved adaptive Kalman filter, and perform filtering processing on each scale coefficient to determine the filtering result; reconstruct the filtering result into a reconstructed signal, and perform signal quality assessment on the reconstructed signal to determine the signal quality of the transformer winding.

[0007] As a preferred solution of the method for collecting monitoring and diagnostic data of transformer windings according to the present invention, the method includes: performing frequency division processing on the monitoring signal by using the multi-scale wavelet transform, and determining each scale coefficient, including selecting a basis function that meets the vibration characteristics of the transformer winding, and introducing an adaptive threshold function considering the distribution characteristics of the winding temperature field. The specific formula is as follows: ; Wherein, represents the wavelet denoising threshold at scale ; represents the standard deviation of the wavelet coefficients at scale ; represents the number of wavelet coefficients at the current scale; represents the scale energy adjustment coefficient; represents the -th layer of wavelet coefficients norm; represents the -th layer of filtered wavelet coefficients norm; represents the temperature-vibration coupling coefficient; represents the maximum temperature gradient of the winding; represents the temperature adjustment index factor. The beneficial effect of this preferred technical solution is to introduce a dedicated basis function that meets the vibration characteristics of the transformer winding and combine an adaptive threshold function considering the distribution characteristics of the winding temperature field, solving the problem that the traditional wavelet transform is difficult to adapt to the specific vibration mode and temperature influence of the transformer.

[0008] As a preferred solution of the method for collecting monitoring and diagnostic data of transformer windings according to the present invention, the method includes: adjusting the noise covariance matrix based on the improved adaptive Kalman filter, including: constructing a state prediction equation and an observation equation. The specific formula is as follows: ; ; Determining a state vector including displacement, velocity, elastic modulus, and damping coefficient. The specific formula is as follows: ; Updating the noise covariance matrix based on the prediction equation and the observation equation. The specific formula is as follows: ; ; Wherein, represents the state prediction value at time ; represents the state estimation value at time ; represents displacement; represents the state estimate at the current time k, i.e., the optimal estimate of the system state after fusing the observed values; represents the previous time of the state prediction value, which is used for the current state prediction or error correction; represents the velocity; represents the elastic modulus; represents the damping coefficient; represents the state transition matrix, that is, from time to time of the dynamic mapping; represents the control input matrix; represents the process noise, assumed to be Gaussian white noise; represents the observed value at the current time; represents the observation matrix, which is used to map the state to the measurement space; represents the measurement noise; represents the time of the process noise covariance matrix; represents the time of the process noise covariance matrix; represents the time of the measurement noise covariance matrix; represents the time of the measurement noise covariance matrix; represents the forgetting factor; represents the measurement noise update forgetting factor; represents the temperature gradient coupling coefficient, which characterizes the influence degree of temperature change on the process noise; represents the square norm of the temperature gradient, indicating the severity of the temperature field change; represents the identity matrix. The beneficial effect of this preferred technical solution is that by adjusting the noise covariance matrix based on the improved adaptive Kalman filter and filtering the scale coefficients, accurate filtering of each frequency component in the transformer winding monitoring signal is achieved.

[0009] As a preferred scheme of the transformer winding monitoring and diagnosis data acquisition method described in the present invention, wherein: filtering the scale coefficients to determine the filtering result, including: using the wavelet coefficients as the observed values of the improved adaptive Kalman filter; filtering the scale coefficients through the calculation of the filtering gain matrix, state update, and posterior error covariance update to obtain the filtered scale coefficients; the specific formulas for the calculation of the filtering gain matrix, state update, and posterior error covariance update are as follows: ; where, Indicating the moment The Kalman gain matrix for state update; Indicating the state prediction error covariance matrix before update; Indicating the observation matrix for mapping the state to the observation space; Indicating the observation noise covariance matrix, reflecting measurement uncertainty.

[0010] As a preferred solution of the transformer winding monitoring and diagnosis data acquisition method described in the present invention, wherein: determining the signal quality of the transformer winding includes: reconstructing each scale coefficient after the filtering process to obtain a reconstructed signal; extracting the transformer winding state characteristics of the reconstructed signal, determining the signal quality evaluation result of the reconstructed signal according to the transformer winding state characteristics; and feedback-optimizing the feature extraction error according to the signal quality evaluation result of the reconstructed signal.

[0011] As a preferred solution of the transformer winding monitoring and diagnosis data acquisition method described in the present invention, wherein: determining the signal quality evaluation result of the reconstructed signal includes: constructing a fault feature vector and energy features of the reconstructed signal based on the thermal-mechanical coupling effect model; combining the fault feature vector and energy features, introducing a signal quality evaluation function based on the thermal-mechanical coupling model to evaluate the signal quality of the reconstructed signal, and obtaining the signal quality evaluation result of the reconstructed signal. The beneficial effect of this preferred technical solution is that a fault feature vector and energy features are constructed based on the thermal-mechanical coupling effect model, and a signal quality evaluation function considering the thermal-mechanical coupling model is introduced, solving the problems that traditional monitoring methods are difficult to comprehensively evaluate signal quality and difficult to adapt to the complex working conditions of transformers.

[0012] As a preferred solution of the transformer winding monitoring and diagnosis data acquisition method described in the present invention, wherein: the specific formula of the signal quality evaluation function is as follows: ; wherein, Indicating the signal quality evaluation factor at the current moment ; Indicating the observed value at the current moment; Indicating the moment The state predicted value; Indicating the observation matrix for mapping the state to the observation space; Indicating the second norm of the state prediction error, i.e., the residual square; Indicating the trace of the observation noise covariance matrix, i.e., the sum of diagonal elements; Indicating the stress gradient tensor obtained based on the thermal-mechanical coupling model; It represents the stress gradient weight coefficient, which is used to measure the influence intensity of thermo-mechanical coupling on signal quality assessment.

[0013] Another object of the present invention is to provide a transformer winding monitoring and diagnostic data acquisition system.

[0014] To solve the above technical problems, the present invention provides the following technical solutions: A transformer winding monitoring and diagnostic data acquisition system includes: a signal acquisition module, which is used to construct a multi-scale wavelet transform and improved adaptive Kalman filter fusion architecture, and the multi-scale wavelet transform and improved adaptive Kalman filter fusion architecture is used to obtain the monitoring signals of the transformer winding; a multi-scale wavelet transform module, which is used to perform frequency division processing on the monitoring signals by using multi-scale wavelet transform to determine the scale coefficients of each scale; an improved adaptive Kalman filter module, which is used to adjust the noise covariance matrix based on the improved adaptive Kalman filter and perform filtering processing on the scale coefficients of each scale to determine the filtering result; a feature fusion diagnosis module, which is used to reconstruct the filtering result into a reconstructed signal and perform signal quality assessment on the reconstructed signal to determine the signal quality of the transformer winding.

[0015] The present invention provides a computer device, including a memory and a processor, where: when the processor executes the computer program, the steps of the transformer winding monitoring and diagnostic data acquisition method are implemented.

[0016] The present invention provides a computer-readable storage medium, on which a computer program is stored, where: when the computer program is executed by the processor, the steps of the transformer winding monitoring and diagnostic data acquisition method are implemented.

[0017] The beneficial effects of the present invention: The data acquisition method based on the fusion of multi-scale wavelet transform and improved adaptive Kalman filter proposed by the present invention effectively combines the advantages of multi-scale wavelet transform in time-frequency analysis and the ability of improved adaptive Kalman filter in noise suppression, and can improve the accuracy and reliability of transformer winding monitoring data acquisition in the power environment; through multi-scale decomposition and adaptive filtering, it effectively solves the limitations of traditional methods in non-stationary signal processing and strong noise interference, and improves the signal feature retention and signal-to-noise ratio; it can provide high-quality data support for transformer winding state monitoring, provide a more accurate basis for subsequent fault diagnosis and early warning, help improve the reliability of transformer operation, reduce maintenance costs, and is of great significance to the construction of smart grids. Description of the Drawings

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0019] Figure 1 It is a schematic flow chart of the on-line monitoring and diagnosis method for transformer windings provided by an embodiment of the present invention.

[0020] Figure 2 It is a schematic structural diagram of a computer device provided by an embodiment of the present invention.

[0021] Figure 3 It is a comparison chart of the average accuracy rates of the methods provided by an embodiment of the present invention under the same test data set.

[0022] Figure 4 It is a trend chart of the accuracy rate changes of three methods under different noise levels provided by an embodiment of the present invention.

[0023] Figure 5 It is a schematic diagram of the noise reduction effect of different methods on noisy signals provided by an embodiment of the present invention.

[0024] Figure 6 It is a schematic diagram of the residual scatter distribution provided by an embodiment of the present invention.

[0025] Figure 7 It is a comparison chart of the three methods in terms of feature peak retention provided by an embodiment of the present invention.

[0026] Figure 8 It is a schematic diagram of the performance of each method on five key performance indicators provided by an embodiment of the present invention.

[0027] Figure 9 It is a distribution diagram of vibration characteristic parameters of a transformer in normal and faulty states provided by an embodiment of the present invention.

[0028] Figure 10 It is a distribution diagram of temperature characteristic parameters of a transformer in normal and faulty states provided by an embodiment of the present invention.

[0029] Figure 11 It is a distribution diagram of electrical characteristic parameters of a transformer in normal and faulty states provided by an embodiment of the present invention.

[0030] Figure 12 It is a comparison chart of the diagnostic performance of three methods in terms of vibration, temperature, and electrical characteristics provided by an embodiment of the present invention. Specific embodiments

[0031] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the following detailed description will be given to the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification.

[0032] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0033] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure or characteristic that may be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an embodiment that is separate or selectively exclusive of other embodiments.

[0034] Example 1, referring to Figure 1 , which is the first embodiment of the present invention. This embodiment provides a method for collecting monitoring and diagnostic data of transformer windings, including: S100: Construct a multi-scale wavelet transform and improved adaptive Kalman filter fusion architecture, which is used to obtain the monitoring signals of the transformer windings.

[0035] S200: Use multi-scale wavelet transform to perform frequency division processing on the monitoring signals to determine the scale coefficients.

[0036] S300: Adjust the noise covariance matrix based on the improved adaptive Kalman filter and perform filtering processing on the scale coefficients to determine the filtering result.

[0037] S400: Reconstruct the filtering result into a reconstructed signal and perform signal quality evaluation on the reconstructed signal to determine the signal quality of the transformer windings.

[0038] It should be noted that during the operation of the transformer, it is subjected to complex electromagnetic-thermal coupling effects, and there are multiple technical challenges in winding state monitoring. For example, there is severe aliasing between high-frequency discharge pulses and low-frequency mechanical vibration characteristics in the time-frequency domain, and traditional filtering methods are difficult to effectively separate them; the gradient change of the temperature field will significantly affect the material parameters and signal propagation characteristics, resulting in the failure of the fixed threshold denoising method; the noise covariance matrix of the existing Kalman filter algorithm lacks an adjustment mechanism and cannot adapt to the time-varying characteristics of the winding state parameters; these problems result in insufficient monitoring signal quality, limited sensitivity of partial discharge detection, and significant errors in mechanical deformation feature extraction.

[0039] Therefore, to address the above problems of multi-physical-field coupling interference, time-varying noise suppression, and insufficient feature extraction accuracy, through steps S100 - S400, a composite processing framework integrating wavelet packet decomposition and improved Kalman filtering is established to achieve multi-level frequency band separation; an adaptive threshold function for the temperature field is introduced to accurately extract high-frequency discharge pulses and low-frequency vibration characteristics under a specific wavelet basis; the noise suppression ability is enhanced through a dynamic covariance update algorithm that couples the forgetting factor and the temperature gradient coefficient; the feature extraction error is reduced by combining the closed-loop optimization mechanism of the bidirectional long short-term memory network diagnostic network; through the synergistic effect of multi-scale decomposition and adaptive filtering, the signal-to-noise ratio and feature retention degree of monitoring signals in a complex electromagnetic environment are effectively improved, providing more reliable data support for the state assessment of power equipment and enhancing the accuracy of fault warning.

[0040] Example 2, referring to Figure 1 , which is the second embodiment of the present invention. Based on the above embodiment, a method for collecting monitoring and diagnostic data of transformer windings is provided.

[0041] In the embodiment of the present invention, in step S100, a multi-scale wavelet transform and improved adaptive Kalman filter fusion architecture is constructed, including: The monitoring signal is frequency-divided using a five-layer wavelet packet decomposition framework. Among them, the Daubechies 8th-order wavelet basis function is used to extract 0.1 - 5 MHz discharge pulses in the high-frequency band, and the symmetric 6th-order wavelet function is applied in the low-frequency band to retain 10 - 500 Hz mechanical vibration characteristics.

[0042] A 16-dimensional feature matrix including temperature, vibration, and electrical parameters is established, and the observation weight of the improved adaptive Kalman filter is dynamically adjusted based on the sliding window mutual information entropy to adapt to the dynamic changes of the signal.

[0043] Through the closed-loop optimization formed by the bidirectional long short-term memory network diagnostic network and the signal processing module, real-time correction of the decomposition layer and the forgetting factor is achieved, further enhancing the adaptability and accuracy of data collection.

[0044] It should be noted that in view of the inherent limitations of traditional methods between the time-frequency analysis accuracy and the dynamic tracking ability, the present invention proposes a collaborative processing architecture of multi-scale wavelet transform and improved adaptive Kalman filter. The specific formula of the multi-scale wavelet transform and improved adaptive Kalman filter fusion architecture is as follows: ; Among them, represents the wavelet coefficient after the -th layer of decomposition, that is, the feature quantity; represents the -th layer of adaptive wavelet decomposition kernel function, reflecting the feature extraction weight of this scale; represents the The input signal or wavelet coefficients of the layer; Denotes the time-domain sampling point index; Denotes the set of integers, representing all possible convolution kernel indices; Denotes the Sampling value at the th position after downsampling the signal of the Denotes the index of the convolution kernel, used to represent the shift index range corresponding to the filter when calculating the current wavelet coefficient.

[0045] Furthermore, the design criterion of the multi-scale wavelet transform and improved adaptive Kalman filter fusion architecture is determined by the following variational optimization model: ; ; Wherein, Denotes the spectrum of the signal in the frequency domain ; Denotes the expression of the th layer kernel function in the frequency domain; Denotes the representation of the th layer wavelet coefficients in the frequency domain; Denotes the regularization weight factor in variational optimization; Denotes the variation regularization term, that is, the total norm of the change rate of the frequency domain kernel function; Denotes the gradient of the kernel function in the frequency domain; Denotes the complex conjugate form of the th layer frequency domain kernel function; J represents the total number of layers of the multi-scale wavelet transform; j (ω) denotes the gradient of the

[0046] Furthermore, in the dynamic filtering stage, a state space equation with scale adaptability is constructed, and the specific formula is as follows: ; Wherein, Denotes the unupdated predicted state vector; Denotes the th state transition matrix at scale; Denotes the state input at scale ; Denotes the process noise term, which satisfies the Gaussian distribution; Denotes the process noise variance at scale ;

[0047] Furthermore, the process noise covariance matrix The time-varying parameters are dynamically calculated as follows: ; Wherein, represents the estimated value of the process noise covariance of the scale ; represents the total length of the wavelet coefficients of the th layer; represents the th layer and the th wavelet coefficient.

[0048] Furthermore, the iterative update formula of the Kalman gain matrix is as follows: ; Wherein, represents the Kalman gain under the scale ; represents the predicted error covariance matrix; represents the observation matrix (mapping the state to the observation space); represents the observation noise covariance matrix, which is dynamically updated with the innovation term ; represents the innovation term of the th layer, that is, the residual between the actual observation value and the predicted observation value; represents the transpose of the observation matrix , which is used to reflect the error information in the observation space back to the state space to correct the state estimate value.

[0049] It should be noted that the present invention realizes the efficient acquisition and preprocessing of the monitoring signals of the transformer winding by constructing a multi-scale wavelet transform and improved adaptive Kalman filter fusion architecture, wherein by combining the advantages of multi-scale wavelet transform in signal decomposition and the expertise of improved adaptive Kalman filter in noise suppression, the problem that traditional single signal processing methods are difficult to cope with multi-source interference in the working environment of the transformer winding is solved; it can effectively filter out external interference and measurement noise while retaining key feature information, improving the reliability of subsequent analysis; compared with the common single filtering methods in the prior art, the present invention captures the feature changes in different frequency domains through multi-scale analysis and combines the adjustment ability of adaptive Kalman filter, improving the perception accuracy of the transformer winding state in the power grid environment, providing a more solid data basis for early fault warning, and finally achieving the beneficial effects of improving the operation safety of the power grid and reducing the maintenance cost.

[0050] In the embodiment of the present invention, in step S200, the multi-scale wavelet transform is used to perform frequency division processing on the monitoring signal, and determining each scale coefficient includes selecting a basis function that meets the vibration characteristics of the transformer winding and introducing an adaptive threshold function considering the distribution characteristics of the winding temperature field.

[0051] In the embodiment of the present invention, a basis function that satisfies the vibration characteristics of the transformer winding is selected, and the specific formula of the adaptive threshold function considering the distribution characteristics of the winding temperature field is as follows: ; Wherein, represents the wavelet denoising threshold at scale , is the set of positive real numbers; represents the standard deviation of the wavelet coefficients at scale ; represents the number of wavelet coefficients at the current scale; represents the scale energy adjustment coefficient; represents the -norm of the wavelet coefficients of the th layer; represents the -norm of the wavelet coefficients after filtering of the th layer; represents the temperature-vibration coupling coefficient; represents the maximum temperature gradient of the winding; represents the temperature adjustment index factor, is the set of real numbers.

[0052] It should be noted that by using multi-scale wavelet transform to perform frequency division processing on the monitoring signal and determining the coefficients of each scale, the precise decomposition and analysis of the complex monitoring signal of the transformer winding are realized; the present invention introduces a dedicated basis function that satisfies the vibration characteristics of the transformer winding and combines an adaptive threshold function considering the distribution characteristics of the winding temperature field, solving the technical problem that the traditional wavelet transform is difficult to adapt to the specific vibration mode and temperature influence of the transformer; it can adaptively extract the key features at different frequency scales for the unique electromagnetic-thermal-mechanical coupling characteristics of the transformer winding, making the signal decomposition more in line with the physical characteristics of the transformer winding; the traditional method often only focuses on the vibration signal itself and ignores the influence of the temperature field on the vibration characteristics, while the present invention constructs a coupling relationship model between temperature and vibration by introducing the temperature-vibration coupling coefficient and the temperature gradient parameter, so as to be able to more accurately identify the real fault features and the changes caused by environmental factors, improving the accuracy of feature extraction and the anti-interference ability, and finally providing a more reliable feature basis for the state evaluation of the transformer winding.

[0053] It should be noted that the analysis of the vibration characteristics of the transformer winding includes: The vibration response of the transformer winding under the action of electromagnetic-mechanical coupling constitutes the key physical feature of fault diagnosis. Among them, the axial helical symmetry of the winding structure makes its dynamic characteristics subject to the dual constraints of geometric parameters and material parameters, and its natural frequency distribution can be derived from the three-dimensional elastic body wave equation. The specific formula is as follows: ; wherein, represents the divergence of the stress tensor; represents the displacement field vector, i.e., the displacement change of each point in the winding over time; represents the material density, i.e., the mass per unit volume; represents the Young's modulus, i.e., the index of the elastic properties of the material; represents time; represents the body force term, i.e., the external excitation force per unit volume, such as the electromagnetic Lorentz force; represents the second-order time derivative of the displacement, corresponding to the acceleration distribution; represents the introduction of the geometric nonlinear term, used to consider the influence of structural deformation.

[0054] Furthermore, when the transformer winding is subjected to the Lorentz force excitation generated by the leakage magnetic field, the axial vibration acceleration obeys a nonlinear second-order differential equation, and the specific formula is as follows: ; wherein, represents the equivalent mass, i.e., the mass of the winding segment to be analyzed; represents the damping coefficient, i.e., the attenuation ability of the system to vibration; represents the axial vibration displacement; represents the velocity, i.e., the first-order derivative of the displacement with respect to time; represents the second-order derivative of the displacement with respect to time; represents the nonlinear stiffness factor, characterizing the change of the system stiffness when the amplitude increases; represents the total number of modes or harmonic components; represents the amplitude of the magnetic induction intensity of the th harmonic component of the current; represents the magnetic-force coupling coefficient, used to represent the influence of the harmonic frequency on the system response; represents the th represents the initial phase of the th

[0055] It should be noted that for the analysis of the distribution characteristics of the winding temperature field, including: The temperature field of the transformer winding presents three-dimensional non-steady state distribution characteristics, and its spatio-temporal evolution law is described by the modified Fourier-Kirchhoff equation as follows: ; wherein, denotes the temperature, which is a function of position r and time t; denotes the rate of change of temperature with respect to time, i.e., the partial derivative of the temperature field with respect to time, with the unit of K / s; is the time partial derivative operator, representing the operation of taking the partial derivative with respect to the time variable t; α represents the thermal diffusivity, with the unit of m² / s; denotes the Laplace operator, which is used to calculate the second-order spatial derivative of the temperature field; r represents the position, usually in three-dimensional coordinates (x, y, z), used to describe the position dependence of the field variable; denotes the volume heat source intensity per unit volume, with the unit of W / m³, such as the core eddy current loss, etc.; denotes the specific heat capacity of the material, with the unit of J / (kg·K), i.e., the heat required for the temperature of a unit mass of the material to increase by 1K; denotes the material density, with the unit of kg / m³; denotes the current density vector, with the unit of A / m², i.e., the current intensity per unit area; denotes the electrical conductivity of the material, with the unit of S / m; denotes the square of the magnitude of the current density vector, which is the core term of the Joule heat power density.

[0056] Under rated operating conditions, the axial and radial temperature gradients of the winding follow a two-scale distribution law, and the specific formula is as follows: ; where, and denote the axial and radial temperature gradients respectively; denotes the winding operating current, with the unit of A; denotes the AC resistance, with the unit of Ω; denotes the thermal conductivity of the insulating oil, with the unit of W / (m·K); denotes the equivalent cross-sectional area of the cooling channel, with the unit of m²; denotes the eddy current loss power density, with the unit of W / m³; denotes the thermal conductivity of the turn insulation material, with the unit of W / (m·K); denotes the characteristic length of the winding structure, with the unit of m; and denote the axial and radial normalized coordinates respectively; denotes the error function.

[0057] When an inter-turn short-circuit fault occurs, the heat source intensity at the fault point shows an exponential growth characteristic, and the specific formula is as follows: ; where, denotes the time of the fault heat source intensity; Represents the initial value of the normal heat source intensity; Represents the current time; Represents the starting moment of fault development; Represents the heat source growth time constant.

[0058] The aging process of the insulating material leads to a non - linear degradation of the equivalent thermal conductivity. The specific formula is as follows: ; Among them, Represents the thermal conductivity of the insulating material at time ; Represents the initial thermal conductivity; Represents the material aging rate coefficient; Represents the current time; Represents the aging time threshold.

[0059] The mechanical stress distribution caused by the thermo - electric coupling effect can be expressed as: ; Among them, Represents the axial stress; Represents Young's modulus; , and Represent the axial, radial and circumferential strains respectively; Represents the coefficient of thermal expansion; Represents the change in temperature rise; Represents the Poisson - related coupling factor.

[0060] The temperature gradient variance matrix The dynamic relationship equation with the Kalman filter parameters is as follows: ; ; Among them, Represents the process noise covariance matrix at the corrected time ; Represents the sampling period; Represents the thermal relaxation time constant; Represents the diagonal matrix form of the temperature gradient variance matrix; Represents the temperature gradient coupling coefficient; Represents the squared norm of the temperature gradient; Represents the identity matrix; Represents the characteristic length (in m) of the heat diffusion path in the z - axis direction (or the specified direction); α represents the thermal diffusivity (in m² / s).

[0061] It should be noted that when the on-line monitoring system of transformer windings is under the coupling action of power frequency harmonics, random pulses and broadband noise, the traditional threshold denoising method is prone to losing the edge information of the characteristic waveform; to break through the contradictory relationship between the time-frequency characteristics and the dynamic noise, a hybrid architecture of multi-scale wavelet transform and improved adaptive Kalman filter is constructed, and its mathematical representation is defined by the following key equations, including: The discrete wavelet transform realizes signal feature extraction through multi-scale orthogonal decomposition, and its mathematical representation is as follows: ; Among them, x represents the original input signal, in the form of a discrete time series; x[n] represents the value of the original signal x at the time domain sampling point n, usually the amplitude value of the nth sampling point in a one-dimensional discrete sequence; n represents the discrete sampling index in the time domain; represents the order scale space at the displacement position of the wavelet coefficient; represents the scale factor and the displacement factor generated orthogonal wavelet basis function; represents the mother wavelet function that satisfies the admissibility condition, satisfying .

[0062] In the design of the improved Kalman filter, the dynamic estimation mechanism of the noise covariance matrix, and its state prediction and measurement update processes satisfy: ; Among them, represents the prior state estimation vector at time represents the linearized state transition matrix of the nonlinear system; represents the prior error covariance matrix, characterizing the statistical characteristics of the prediction error; represents the dynamic weighting matrix; represents the prior state estimation vector at the current time k, that is, the predicted value; represents the state prediction value at the previous time ; represents the control input matrix, used to introduce the control input vector into the state update; represents the control input vector, which can be an external excitation, disturbance or command input; represents the state estimation error covariance matrix at the previous time, measuring the statistical uncertainty of the prediction error; represents the linearized state transition matrix of the nonlinear system transposed for covariance propagation; represents the corrected time Process noise covariance matrix.

[0063] In the embodiment of the present invention, the specific formula for constructing the time-varying covariance scaling factor is as follows: ; Wherein, represents the dynamic weighting matrix; W j represents the j-th layer wavelet coefficient; k represents the time domain sampling point index; L represents the half-width of the sliding window; ∥⋅∥ 2 represents the square of the L2 norm, which is used to calculate the energy; SNR k represents the signal-to-noise ratio at time k, which is used to dynamically adjust the covariance weighting coefficient; the numerator term calculates the local wavelet energy through a sliding window, reflecting the transient characteristics of the signal; the denominator term establishes a parameterized adjustment channel for the signal-to-noise ratio, and automatically enhances the covariance constraint when the signal-to-noise ratio at time k decreases; the diagonal matrix construction ensures independent adjustment of each state component, realizing fine control of noise suppression.

[0064] Furthermore, the length of the sliding window is adaptively determined by the following formula: ; Wherein, represents the adaptive sliding window length at time ; represents the 2-norm of the current wavelet coefficient, that is, the energy intensity; represents the time interval between adjacent samples; represents the time constant of the system dynamic response.

[0065] In the embodiment of the present invention, adjusting the noise covariance matrix based on the improved adaptive Kalman filter in step S300 includes: Constructing a state prediction equation and an observation equation.

[0066] Determining a state vector including displacement, velocity, elastic modulus, and damping coefficient.

[0067] Updating the noise covariance matrix based on the prediction equation and the observation equation.

[0068] Specifically, in the embodiment of the present invention, the specific formula of the state prediction equation is as follows: ; The specific formula of the observation equation is as follows: ; Wherein, the specific formula of the state vector is as follows: ; Furthermore, the specific formula for updating the noise covariance matrix is as follows: ; ; where represents the predicted state value at time ; represents the estimated state value at time ; represents displacement; represents the estimated state value (posterior state estimate) at the current time k, that is, the optimal estimate of the system state after fusing the observed values; represents the predicted state value at the previous time , which is used for the current state prediction or error correction; represents velocity; represents Young's modulus; represents the damping coefficient; represents the state transition matrix, that is, the dynamic mapping from time to time ; represents the control input matrix; represents the process noise, assumed to be Gaussian white noise; represents the observed value at the current time; represents the observation matrix, which is used to map the state to the measurement space; represents the measurement noise; represents the process noise covariance matrix at time ; represents the process noise covariance matrix at time ; represents the measurement noise covariance matrix at time ; represents the measurement noise covariance matrix at time ; represents the forgetting factor; represents the measurement noise update forgetting factor; represents the temperature gradient coupling coefficient, which characterizes the influence degree of temperature change on the process noise; represents the square norm of the temperature gradient, which represents the severity of the temperature field change; represents the identity matrix.

[0069] It should be noted that in the winding mechanical vibration propagation model, the axial vibration propagation characteristics of the transformer winding form the theoretical basis for fault diagnosis; based on the theory of elastic dynamics of non-uniform media, the vibration propagation process of the winding under electromagnetic excitation is described by the modified d'Alembert principle, and an axial coordinate system is established , the vibration displacement field satisfies the following formula: ; where, represents the winding cross-sectional area function; and respectively represent the spatial distributions of the elastic modulus and density caused by insulation aging; represents the viscous damping coefficient; represents the Lorentz force density term, and this equation reveals the dynamic influence mechanism of material deterioration on the vibration wave propagation speed through a variable coefficient form; represents the current time.

[0070] Furthermore, a state space model is constructed to adapt to the improved adaptive Kalman filter framework, and the specific formula for defining the state vector is as follows: ; The non-linear observation equation is derived, and the specific formula is as follows: ; where, represents the state vector at time , including displacement , velocity , elastic modulus and damping coefficient ; represents the state vector at time ; represents the non-linear state transition function; represents the process noise influence matrix; represents the process noise; represents the observation matrix; represents the observed quantity; represents the observation noise.

[0071] Furthermore, the covariance matrices of the process noise and the observation noise are updated in real time in the improved adaptive Kalman filter, and the specific formula is as follows: ; where, represents the corrected process noise covariance matrix; represents the forgetting factor; represents the Kalman gain matrix, represents the transpose of the Kalman gain matrix; represents the innovation vector, i.e., the residual term; represents the outer product of the innovation, reflecting the covariance estimate update amount.

[0072] In an embodiment of the present invention, in step S300, filtering processing is performed on each scale coefficient to determine a filtering result, including: Taking the wavelet coefficient as the observation value of the improved adaptive Kalman filter.

[0073] By calculating the filtering gain matrix, updating the state, and updating the posterior error covariance, filtering processing is performed on each scale coefficient to obtain each scale coefficient after filtering processing.

[0074] In an embodiment of the present invention, the specific formulas for calculating the filtering gain matrix, updating the state, and updating the posterior error covariance are as follows: ; Wherein, represents the Kalman gain matrix at time for state update; represents the state prediction error covariance matrix before update; represents the observation matrix for mapping the state to the observation space; represents the observation noise covariance matrix reflecting the measurement uncertainty.

[0075] It should be noted that the present invention realizes precise filtering of each frequency component in the transformer winding monitoring signal by adjusting the noise covariance matrix based on the improved adaptive Kalman filter and performing filtering processing on each scale coefficient; among them, by constructing a complete state vector including displacement, velocity, elastic modulus, and damping coefficient, and introducing a temperature gradient coupling coefficient, the technical problem of fixed parameters and insufficient adaptability of the traditional Kalman filter in the face of the complex working environment of the transformer winding is solved; a noise covariance matrix update mechanism based on temperature gradient is proposed, and through the forgetting factor and temperature gradient coupling coefficient, the filtering parameters can be automatically adjusted according to the real-time temperature field change of the transformer winding, so as to adapt to the change of winding characteristics under different load conditions; compared with the filtering method with static parameter setting in the prior art, the present invention can maintain the stability of filtering performance under extreme working conditions such as severe temperature fluctuations, effectively suppress various non-stationary noise interferences, improve the signal-to-noise ratio of fault characteristics, provide clearer signal characteristics for subsequent winding state diagnosis, and finally achieve high precision and high reliability in transformer winding monitoring and diagnosis.

[0076] It should be noted that the multi-scale wavelet transform and improved adaptive Kalman filter fusion architecture realizes a feature-level fusion model, and its dynamic weight allocation mechanism can be expressed as a non-linear superposition form of multi-source signals, and the specific formula is as follows: ; Wherein, represents the fused feature vector at time Output value; Indicates the number of fused source signals; Indicates the th signal source at time Weight factor; Indicates the wavelet decomposition coefficient of the signal at scale i.e., wavelet feature; Indicates the channel's improved filter response kernel, dynamically adjusted with time and measurement variables; Indicates the Hadamard product, used to represent the element-wise product of corresponding position elements, i.e., matrix or vector multiplied element by element; Indicates the tensor product, used to fuse different modal features.

[0077] Furthermore, the process noise covariance matrix is updated recursively with exponential weighting. The specific formula is as follows: ; where Indicates the noise covariance matrix at time ; Indicates the noise covariance matrix at time ; Indicates the forgetting factor, ; Indicates the innovation, i.e., the observation residual; Indicates the current observation value; Indicates the observation matrix; Indicates the previous step's predicted state.

[0078] Furthermore, the final decision fusion layer adopts the modified D-S evidence synthesis rule. The specific formula is as follows: ; where Indicates the confidence value of the target state ; Indicates the th source signal's support probability for ; Indicates the fusion weight; Indicates the th source signal's support probability for ; Indicates the confidence correction factor, calculating the source confidence credibility based on the Kullback-Leibler divergence distance P k Indicates the th source signal's support probability for the target state, P ref Indicates the probability distribution under standard conditions; K indicates the total number of fused signal sources; Θ jRepresents the j-th candidate target state; this formula is used to calculate the normalized confidence value of each state based on multi-source support and confidence correction factors 。

[0079] In an embodiment of the present invention, determining the signal quality of the transformer winding in step S400 includes: Reconstruct each scale coefficient after filtering processing to obtain a reconstructed signal.

[0080] Extract the transformer winding state characteristics of the reconstructed signal, and determine the signal quality evaluation result of the reconstructed signal according to the transformer winding state characteristics.

[0081] Feedback and optimize the feature extraction error according to the signal quality evaluation result of the reconstructed signal.

[0082] In an embodiment of the present invention, determining the signal quality evaluation result of the reconstructed signal in step S400 includes: Based on the thermo-mechanical coupling effect model, construct the fault feature vector and energy feature of the reconstructed signal.

[0083] Combine the fault feature vector and energy feature, and introduce a signal quality evaluation function based on the thermo-mechanical coupling model to evaluate the signal quality of the reconstructed signal, and obtain the signal quality evaluation result of the reconstructed signal.

[0084] In an embodiment of the present invention, the specific formula of the signal quality evaluation function is as follows: ; Wherein, Represents the signal quality evaluation factor at the current moment ; Represents the observed value at the current moment; Represents the state prediction value at time ; Represents the observation matrix, which is used to map the state to the observation space; Represents the second norm of the state prediction error, that is, the residual square; Represents the trace of the observation noise covariance matrix, that is, the sum of the diagonal elements; Represents the stress gradient tensor obtained based on the thermo-mechanical coupling model; Represents the stress gradient weight coefficient, which is used to measure the influence intensity of thermo-mechanical coupling on signal quality evaluation.

[0085] It should be noted that for the thermo-mechanical coupling effect model, the temperature gradient and mechanical stress generated during the operation of the transformer winding form a dynamic coupling system, and its interaction mechanism can be mathematically described through the theoretical framework of thermoelasticity, and a generalized heat conduction equation including an energy conversion mechanism is established. The specific formula is as follows: ; Among them, represents the material density; represents the specific heat capacity of the material; represents the temperature; represents the time; represents the thermal conductivity; represents the plastic stress tensor; represents the plastic strain rate tensor; represents the current density; represents the material resistivity; represents the heat conduction term, reflecting the temperature gradient diffusion behavior.

[0086] Furthermore, considering the basic constitutive properties of anisotropic materials, the strain tensor induced by thermal expansion can be decomposed into elastic strain and thermal strain, and the specific formula is as follows: ; Among them, represents the total strain tensor component; represents the material compliance tensor; represents the stress tensor component; represents the thermal expansion coefficient tensor; represents the current temperature; represents the reference temperature; represents the thermal-plastic coupling coefficient; represents the bulk modulus; represents the temperature excitation function; represents the thermally excited term accumulated over time.

[0087] In the embodiment of the present invention, the dynamic response of the winding structure satisfies the differential form of the law of conservation of momentum, and the specific formula is as follows: ; Among them, represents the stress tensor component; represents the elastic stiffness tensor; represents the total strain tensor; represents the thermal expansion coefficient tensor; represents the temperature change; represents the coupling correction coefficient; represents the unit displacement rate, reflecting the dynamic structure deformation.

[0088] In the embodiment of the present invention, in step S400, the feature extraction error is feedback optimized according to the signal quality evaluation result of the reconstructed signal, including: Using the bidirectional long short-term memory network diagnostic network and the signal processing unit to form a closed-loop optimization to correct the decomposition layer number and forgetting factor.

[0089] It should be noted that the present invention realizes the comprehensive evaluation and feedback optimization of the transformer winding state by reconstructing the filtering result into a reconstructed signal and evaluating the signal quality of the reconstructed signal. Among them, a fault feature vector and energy feature are constructed based on the thermal-mechanical coupling effect model, and a signal quality evaluation function considering the thermal-mechanical coupling model is introduced, solving the technical problems that traditional monitoring methods are difficult to comprehensively evaluate signal quality and difficult to adapt to the complex working conditions of transformers; secondly, a closed-loop mechanism from signal reconstruction to quality evaluation and then to feedback optimization is established, and the influence of thermal-mechanical coupling on signal quality is quantified through the stress gradient tensor and stress gradient weight coefficient, which can automatically identify and correct signal distortion caused by temperature changes; compared with the common one-way signal processing flow in the prior art, the present invention can evaluate the reliability of the reconstructed signal in real time and adjust the pre-processing parameters accordingly, improving the monitoring accuracy; at the same time, this quality evaluation method based on the physical model can also effectively distinguish real fault features from normal operation fluctuations, reduce the false alarm rate, enhance the credibility of the diagnosis result, and provide more scientific and reliable data support for transformer operation and maintenance decision-making.

[0090] Exemplarily, the feedback optimization of the feature extraction error according to the signal quality evaluation result of the reconstructed signal includes: Based on the multi-scale wavelet transform and improved adaptive Kalman filter fusion architecture, a joint parameter optimization model with time-varying characteristics is established. Let the monitoring signal at time be , then the multi-scale decomposition coefficient matrix can be expressed as: Among them, represents the multi-scale decomposition coefficient matrix at time ; represents the wavelet decomposition coefficient of the -th layer; represents the frequency band basis matrix of the -th layer, corresponding to the frequency domain wavelet basis; represents the total number of wavelet decomposition scales; represents the weighted superposition of the output results of all wavelet decomposition layers k = 1, 2,..., N, used to construct the feature fusion matrix at multiple scales, that is, the multi-scale decomposition coefficient matrix at time .

[0091] Furthermore, a time-frequency sensitivity factor is constructed to quantify the response intensity of each scale component to the line group fault. The specific formula is as follows: ; Among them, represents the response sensitivity factor of scale ; denotes the layer wavelet coefficient vector; denotes the typical fault mode template; denotes the Hadamard product (element-wise product); denotes the 2-norm, which is used for energy normalization.

[0092] Furthermore, the response sensitivity factor of scale drives the dynamic correction of the driving noise covariance matrix as follows: ; where denotes the observation noise covariance matrix at scale time ; denotes the band noise energy reference value; denotes the response sensitivity factor of scale ; denotes the time-frequency response degradation time constant; denotes the identity matrix; denotes the adjustment coefficient, which controls the spectral correction degree of covariance estimation; denotes constructing a diagonal matrix by extracting the main diagonal elements of the wavelet coefficient autocorrelation matrix .

[0093] Furthermore, the response sensitivity factor of scale drives the dynamic adjustment of the subsequent covariance matrix and adaptively corrects the prediction covariance along with the change of the response sensitivity factor of scale . The specific formula for adaptively correcting the prediction state covariance matrix in the Kalman filtering process is as follows: ; ; where denotes the prediction state covariance matrix at scale ; denotes the state transition matrix; denotes the prediction state covariance matrix at scale ; denotes the derivative of the covariance with respect to the sensitivity, i.e., the rate of change; denotes the sensitivity change amount, which is used for adaptively correcting the prediction error.

[0094] In summary, the data acquisition method based on the fusion of multi-scale wavelet transform and improved adaptive Kalman filter proposed by the present invention effectively combines the advantages of multi-scale wavelet transform in time-frequency analysis and the ability of the improved adaptive Kalman filter in noise suppression, and can improve the accuracy and reliability of the monitoring data acquisition of transformer windings in the power environment; through multi-scale decomposition and adaptive filtering, it effectively solves the limitations of traditional methods in non-stationary signal processing and strong noise interference, and improves the signal feature retention and signal-to-noise ratio; it can provide high-quality data support for the condition monitoring of transformer windings, provide a more accurate basis for subsequent fault diagnosis and early warning, help improve the reliability of transformer operation, reduce maintenance costs, and is of great significance to the construction of smart grids.

[0095] Embodiment 3 is the third embodiment of the present invention. This embodiment provides a monitoring and diagnosis data acquisition system for transformer windings, including: a signal acquisition module for constructing a fusion architecture of multi-scale wavelet transform and improved adaptive Kalman filter, and the fusion architecture of multi-scale wavelet transform and improved adaptive Kalman filter is used to obtain the monitoring signal of the transformer winding; a multi-scale wavelet transform module for performing frequency division processing on the monitoring signal by using multi-scale wavelet transform to determine the scale coefficients of each scale; an improved adaptive Kalman filter module for adjusting the noise covariance matrix based on the improved adaptive Kalman filter and performing filtering processing on the scale coefficients of each scale to determine the filtering result; a feature fusion diagnosis module for reconstructing the filtering result into a reconstructed signal and performing signal quality evaluation on the reconstructed signal to determine the signal quality of the transformer winding.

[0096] Embodiment 4 is the fourth embodiment of the present invention. The difference from the previous three embodiments is as follows: As Figure 2 shown, if the function is implemented in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. And the aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0097] The logic and / or steps represented in the flowchart or otherwise described herein can, for example, be considered as a definitional sequence of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.

[0098] More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection (electronic device) having one or more wirings, a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other suitable processing as necessary, and then stored in a computer memory.

[0099] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above-described embodiments, the multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques well known in the art can be used: discrete logic circuits having logic gates for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), and the like.

[0100] Example 5, referring to Figures 3 to 12 , which is the fifth embodiment of the present invention, provides a method for collecting transformer winding monitoring and diagnostic data. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0101] In this embodiment, in order to verify the application effect of the present invention in the on-line monitoring system of transformer windings, specific tests and verifications were carried out. The test environment included hardware configurations and software tools to ensure the comprehensiveness and accuracy of the tests. As shown in Table 1, the specific parameter settings of the simulation model and algorithm were provided. These parameter settings ensured that the simulation environment could comprehensively test and verify the performance and effect of the present invention under different scenarios, providing reliable data support and theoretical basis for practical applications.

[0102] Table 1 Parameter Settings of Simulation Model and Algorithm , Tests were carried out according to the data in Table 1. As Figure 3 and Figure 4 show the performance comparison and analysis results of three fault diagnosis methods. Specifically, Figure 3 presents the average accuracy of each method under the same test data set. The accuracy of the multi-scale wavelet transform-improved adaptive Kalman filter fusion method reaches 92.36%, which is 13.84% and 8.57% higher than that of the traditional multi-scale wavelet transform method and the traditional improved adaptive Kalman filter method, respectively. At the same time, the standard deviation of the multi-scale wavelet transform-improved adaptive Kalman filter fusion method is 1.93, which is lower than 3.21 of the traditional multi-scale wavelet transform method and 2.87 of the traditional improved adaptive Kalman filter method, indicating that its diagnostic results are more stable and reliable. Figure 4Shows the accuracy change trends of three methods under different noise levels. As the noise level increases from 0.1 to 0.5, the accuracy of the multi-scale wavelet transform-improved adaptive Kalman filter fusion method drops from 91.68% to 80.36%, a decrease of 11.32%; in contrast, the accuracies of the traditional multi-scale wavelet transform method and the traditional improved adaptive Kalman filter method drop from 76.83% to 56.78% and from 82.17% to 64.51% respectively, with decreases of 20.05% and 17.66%; data analysis shows that the multi-scale wavelet transform-improved adaptive Kalman filter fusion method is superior to traditional single methods in all indicators; in a low-noise environment (0.1), the accuracy of the multi-scale wavelet transform-improved adaptive Kalman filter fusion method is 14.85% higher than that of the traditional multi-scale wavelet transform method and 9.51% higher than that of the traditional improved adaptive Kalman filter method. In a high-noise environment (0.5), this advantage is further expanded, and the accuracy of the multi-scale wavelet transform-improved adaptive Kalman filter fusion method is 23.58% and 15.85% higher than that of the traditional multi-scale wavelet transform method and the traditional improved adaptive Kalman filter method respectively; this result fully demonstrates the superiority of the multi-scale wavelet transform-improved adaptive Kalman filter fusion method in transformer winding fault diagnosis, especially showing strong robustness and accuracy in a complex noise environment. This method effectively combines the multi-resolution analysis ability of the multi-scale wavelet transform and the dynamic filtering advantage of the improved adaptive Kalman filter, providing an efficient and reliable technical means for on-line monitoring and fault diagnosis of transformer windings.

[0103] As Figure 5 Shows the noise reduction effects of different methods on noisy signals. Among them, the multi-scale wavelet transform-improved adaptive Kalman filter method effectively suppresses noise interference while maintaining the overall trend of the signal, especially showing outstanding performance in the three characteristic peak regions of 2.3 - 2.7s, 5.1 - 5.5s, and 8.2 - 8.6s; compared with the single methods of the multi-scale wavelet transform and the improved adaptive Kalman filter, the multi-scale wavelet transform-improved adaptive Kalman filter combines the multi-resolution analysis ability of the multi-scale wavelet transform and the dynamic filtering advantage of the improved adaptive Kalman filter, achieving a better balance in feature retention and noise suppression; and as Figure 6 The residual scatter distribution further verifies this. The residual distribution of the multi-scale wavelet transform-improved adaptive Kalman filter is more concentrated and has a smaller amplitude, indicating that its noise reduction effect is more stable; quantitative analysis shows that the average residual amplitude of the multi-scale wavelet transform-improved adaptive Kalman filter is reduced by 0.09 and 0.05 compared with the multi-scale wavelet transform and the improved adaptive Kalman filter respectively, and the noise reduction effect is significantly improved.

[0104] As Figure 7The performance of three methods in terms of characteristic peak retention was compared. Among them, the mean peak error of the multi-scale wavelet transform - improved adaptive Kalman filtering method was the lowest, at 0.037, which was reduced by 28.85% and 17.78% compared with the multi-scale wavelet transform (0.052) and the improved adaptive Kalman filtering (0.045) respectively. At the same time, the standard deviation of the error of the multi-scale wavelet transform - improved adaptive Kalman filtering was also the smallest, only 0.008, indicating that it was more stable in the processing of different characteristic peaks. As Figure 8 The performance of each method in five key performance indicators was comprehensively compared. The multi-scale wavelet transform - improved adaptive Kalman filtering was superior to the single methods in three indicators: signal-to-noise ratio improvement (32.15%), feature retention degree (94.27%), and residual stability (92.35%), which reflected its superiority in the signal processing of transformer windings. Although it was slightly inferior in terms of algorithm complexity (68.24%) and calculation time consumption (79.15%), considering its significantly improved signal quality, the multi-scale wavelet transform - improved adaptive Kalman filtering method was still an efficient and reliable signal processing scheme, providing more accurate data support for the fault diagnosis of transformer windings.

[0105] As Figure 9 、 Figure 10 and Figure 11 respectively showed the distribution of vibration, temperature, and electrical characteristic parameters under normal and faulty states of the transformer. In terms of vibration characteristics, the average vibration intensity (3.72 m / s²) in the faulty state was significantly higher than that in the normal state (2.35 m / s²), with an increase of 58.30%. Secondly, the temperature characteristics similarly also showed obvious differences. The average temperature (89.25 °C) in the faulty state was 13.63 °C higher than that in the normal state (75.62 °C). Among the electrical characteristics, the average harmonic distortion rate (3.42%) in the faulty state far exceeded that in the normal state (1.25%), with an increase of up to 173.60%. The significant changes in these characteristic parameters provided a reliable data basis for fault diagnosis, highlighting the advantages of the multi-scale wavelet transform and improved adaptive Kalman filtering fusion framework in feature extraction.

[0106] As Figure 12The following compares the diagnostic performances of three methods, namely SVM, CNN, and multi-scale wavelet transform-improved adaptive Kalman filter, in terms of vibration, temperature, and electrical characteristics; the results show that the multi-scale wavelet transform-improved adaptive Kalman filter method performs best in all feature types, with average diagnostic accuracies of 97.23%, 94.68%, and 92.35% respectively, which are 11.91% and 7.51% higher than those of the SVM and CNN methods respectively; the advantage of the fusion architecture of multi-scale wavelet transform and improved adaptive Kalman filter in vibration feature diagnosis is particularly significant, with an accuracy as high as 97.23%, which is attributed to the fact that this method combines the multi-resolution analysis ability of multi-scale wavelet transform and the dynamic filtering advantage of improved adaptive Kalman filter, and can more effectively extract and process the complex vibration signals of transformer windings; in addition, the fusion architecture of multi-scale wavelet transform and improved adaptive Kalman filter also maintains a high level in the diagnostic accuracies of temperature and electrical characteristics, which are 94.68% and 92.35% respectively, reflecting the stability and adaptability of the present invention in processing multi-type features.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for collecting monitoring and diagnostic data of a transformer winding, characterized in that: including constructing a multi-scale wavelet transform and improved adaptive Kalman filter fusion architecture for obtaining monitoring signals of a transformer winding; performing frequency division processing on the monitoring signals by using the multi-scale wavelet transform to determine scale coefficients; adjusting a noise covariance matrix based on the improved adaptive Kalman filter and performing filtering processing on the scale coefficients to determine a filtering result; reconstructing the filtering result into a reconstructed signal and performing signal quality evaluation on the reconstructed signal to determine the signal quality of the transformer winding.

2. The transformer winding monitoring and diagnostic data acquisition method according to claim 1, wherein: Performing frequency division processing on the monitoring signals by using the multi-scale wavelet transform to determine scale coefficients includes selecting a basis function that meets the vibration characteristics of the transformer winding and introducing an adaptive threshold function considering the distribution characteristics of the winding temperature field. The specific formula is as follows: ； Among them, represents the wavelet denoising threshold at scale ; represents the standard deviation of the wavelet coefficients at scale ; represents the number of wavelet coefficients at the current scale; represents the scale energy adjustment coefficient; represents the -th layer norm of the wavelet coefficients; represents the -th layer norm of the wavelet coefficients after filtering; represents the temperature-vibration coupling coefficient; represents the maximum temperature gradient of the winding; represents the temperature adjustment index factor.

3. The transformer winding monitoring and diagnostic data acquisition method according to claim 2, wherein: The adjusting the noise covariance matrix based on the improved adaptive Kalman filter includes: constructing a state prediction equation and an observation equation. The specific formula is as follows: ； ； determining a state vector including displacement, velocity, elastic modulus, and damping coefficient. The specific formula is as follows: ； updating the noise covariance matrix based on the prediction equation and the observation equation. The specific formula is as follows: ； ； wherein, represents the state prediction value at time ; represents the state estimation value at time ; represents displacement; represents the state estimation value at the current time k; represents the previous time state prediction value; represents velocity; represents elastic modulus; represents damping coefficient; represents state transition matrix; represents control input matrix; represents process noise; represents the observation value at the current time; represents observation matrix; represents measurement noise; represents time process noise covariance matrix; represents time process noise covariance matrix; represents time measurement noise covariance matrix; represents time measurement noise covariance matrix; represents forgetting factor; represents measurement noise update forgetting factor; represents temperature gradient coupling coefficient; represents the square norm of temperature gradient; represents identity matrix.

4. The transformer winding monitoring and diagnostic data acquisition method according to claim 3, characterized in that: Performing filtering processing on the scale coefficients to determine a filtering result includes: using the wavelet coefficients as observation values of the improved adaptive Kalman filter; performing filtering processing on the scale coefficients through calculation of a filtering gain matrix, state update, and posterior error covariance update to obtain the scale coefficients after filtering processing; The specific formulas for the calculation of the filtering gain matrix, state update, and posterior error covariance update are as follows: ； Among them, represents the Kalman gain matrix at time ; represents the state prediction error covariance matrix before update; represents the observation matrix; represents the observation noise covariance matrix.

5. The method for collecting monitoring and diagnostic data of a transformer winding according to claim 4, wherein: The determining the signal quality of the transformer winding includes: reconstructing the scale coefficients after filtering processing to obtain a reconstructed signal; extracting the state characteristics of the transformer winding of the reconstructed signal and determining a signal quality evaluation result of the reconstructed signal according to the state characteristics of the transformer winding; performing feedback optimization on the feature extraction error according to the signal quality evaluation result of the reconstructed signal.

6. The transformer winding monitoring and diagnostic data acquisition method according to claim 5, wherein: The determining the signal quality evaluation result of the reconstructed signal includes: constructing a fault feature vector and energy feature of the reconstructed signal based on a thermal-mechanical coupling effect model; combining the fault feature vector and the energy feature and introducing a signal quality evaluation function based on the thermal-mechanical coupling model to evaluate the signal quality of the reconstructed signal to obtain a signal quality evaluation result of the reconstructed signal.

7. The transformer winding monitoring and diagnostic data acquisition method according to claim 6, characterized in that: The specific formula of the signal quality evaluation function is as follows: ； Among them, represents the signal quality evaluation factor at the current moment; represents the observed value at the current moment; represents the predicted value of the state at the moment represents the moment ; represents the observation matrix; represents the two-norm of the state prediction error; represents the trace of the observation noise covariance matrix; represents the stress gradient tensor obtained based on the thermal-mechanical coupling model; represents the stress gradient weight coefficient.

8. A transformer winding monitoring and diagnostic data acquisition system, which applies the transformer winding monitoring and diagnostic data acquisition method according to any one of claims 1 to 7, characterized in that: including a signal acquisition module for constructing a multi-scale wavelet transform and improved adaptive Kalman filter fusion architecture for obtaining monitoring signals of a transformer winding; a multi-scale wavelet transform module for performing frequency division processing on monitoring signals by using the multi-scale wavelet transform to determine scale coefficients; an improved adaptive Kalman filter module for adjusting a noise covariance matrix based on the improved adaptive Kalman filter and performing filtering processing on scale coefficients to determine a filtering result; The feature fusion diagnosis module is used to reconstruct the filtered result into a reconstructed signal, evaluate the signal quality of the reconstructed signal, and determine the signal quality of the transformer winding.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, the steps of the transformer winding monitoring and diagnosis data acquisition method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of the transformer winding monitoring and diagnosis data acquisition method described in any one of claims 1 to 7 are implemented.

Images