A method for intelligent state monitoring of a main transformer based on multi-source data

Abstract

The present application relates to the technical field of power equipment state monitoring and fault diagnosis, and particularly relates to a kind of intelligent state monitoring method of main transformer based on multi-source data, comprising: synchronously obtaining three-dimensional vibration acceleration time domain sequence, ultra-high frequency partial discharge pulse sequence, reference phase voltage and load current signal;The displacement response obtained by acceleration twice integration is input into viscoelastic mechanics model calculated by alternating Lorentz electric power of load current, and the steady-state residual error is extracted as irreversible plastic cavity index;Extract the phase angle of fundamental frequency to map the discharge pulse sequence, and calculate the discharge characteristic ratio in the maximum mechanical stretching interval as the electromechanical modulation depth index;Determine the dynamic air gap distance based on the cavity index, input the gas breakdown model to obtain the minimum theoretical breakdown voltage, and verify whether the discharge transient phase voltage coincides with the minimum value;When both indexes are out of limit and the coincidence is verified, an early warning is output.The present application establishes a cross-physical-domain causal verification mechanism and accurately locates early latent damage of winding.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring and fault diagnosis technology, and in particular to a method for intelligent condition monitoring of main transformers based on multi-source data. Background Technology

[0002] With the construction and operation of ultra-high voltage AC / DC power grids, the main transformer, as a key piece of equipment in the power grid, is of paramount importance in terms of operational reliability. When the main transformer is subjected to external short-circuit current impacts or long-term severe load fluctuations, the windings and insulation structure are prone to microscopic irreversible plastic damage under the action of huge alternating Lorentz electrodynamic forces, which in turn forms tiny tear cavities inside the insulating paperboard.

[0003] Currently, most conventional main transformer condition monitoring systems employ a multi-parameter, physically isolated analysis framework, independently extracting the variation trend of mechanical vibration amplitude and the phase spectrum of electrical partial discharge. However, when microscopic cavities form within the insulation, these cavities exhibit a periodic opening and closing "breathing" effect under forced winding vibration. When the cavity opens to its maximum and the internal gas pressure suddenly drops, the gas breakdown condition is met, inducing partial discharge. Existing monitoring technologies lack underlying causal criteria across physical domains, relying solely on a single electrical power frequency phase for discharge spectrum statistics, thus severing the underlying physical laws governing the modulation of high-frequency discharge pulses by mechanical vibration displacement. This results in the system's inability to isolate the actual breathing-type discharge controlled by mechanical damage from massive amounts of ionized noise in the extremely complex electromagnetic interference environment of substations. Consequently, in the very early, latent stage before macroscopic damage to the main transformer insulation manifests, partial discharges accompanying microscopic plastic damage are easily missed. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a smart condition monitoring method for main transformers based on multi-source data. It aims to improve the problem that traditional main transformer monitoring mostly uses isolated analysis of electromechanical signals. Due to the lack of cross-physical field coupling criteria, noise can easily mask micro-damage discharge, leading to early missed detection.

[0005] In a first aspect, the present invention provides the following technical solution: a method for intelligent condition monitoring of a main transformer based on multi-source data, comprising the following steps: S1. Synchronously acquire three-dimensional vibration acceleration time-domain sequence, ultra-high frequency partial discharge pulse sequence, reference phase voltage signal and load current signal; S2. The three-dimensional vibration acceleration time-domain sequence is integrated twice to obtain the winding vibration displacement response, which is then input into the viscoelastic mechanical model together with the alternating Lorentz electrodynamic force calculated based on the load current signal. The steady-state residual between the predicted and measured displacements is used as the irreversible plastic cavity index. S3. Extract the fundamental frequency phase angle based on the winding vibration displacement response and map it to the ultra-high frequency partial discharge pulse sequence; calculate the ratio of the probability or average amplitude of discharge pulses inside and outside the maximum mechanical stretch preset phase interval as the electromechanical modulation depth index. S4. Determine the dynamic air gap distance based on the irreversible plastic cavity index, input the gas breakdown model to obtain the minimum value of the theoretical breakdown voltage threshold; determine whether the transient phase voltage amplitude corresponding to the reference phase voltage signal at the time of occurrence of the ultra-high frequency partial discharge pulse sequence coincides with the minimum value point within the preset deviation tolerance. S5. When both the irreversible plastic cavity index and the electromechanical modulation depth index are greater than the preset threshold, and the judgment results are the same, an early warning command is output.

[0006] By adopting the above technical solution, combining the plastic cavity index and electromechanical modulation depth and using the breakdown model for phase verification, cross-domain causal closed-loop verification is achieved, thereby improving the problem that traditional main transformer monitoring mostly uses isolated analysis of electromechanical signals, which lacks cross-physical field coupling criteria, resulting in noise easily masking micro-damage discharge and causing early missed detection.

[0007] The present invention has the following beneficial effects: 1. In this invention, by combining the plastic cavity index and the electromechanical modulation depth and using the breakdown model for phase verification, cross-domain causal closed-loop verification is achieved, thereby improving the problem that traditional main transformer monitoring mostly uses isolated analysis of electromechanical signals, which, due to the lack of cross-physical field coupling criteria, causes noise to easily mask micro-damage discharge and lead to early missed detection.

[0008] 2. In this invention, by using a unified high-frequency clock source to assign the same timestamp to each sensor for alignment, a synchronous data matrix is ​​constructed. This improves the problem that traditional data acquisition mostly uses independent crystal oscillator sampling of modules, which causes serious phase misalignment between high-frequency pulses and low-frequency vibrations due to cumulative drift during clock operation.

[0009] 3. In this invention, steady-state residuals are extracted by observing the state of the viscoelastic mechanical model, and then the pure plastic cavity index is accurately decoupled. This improves the problem that traditional deformation assessment mostly uses vibration amplitude analysis, which cannot separate the reversible elastic components of the insulating material, thus causing normal transformer under load micro-vibration to be misjudged as real structural damage.

[0010] 4. In this invention, by calculating the ratio of discharge pulses inside and outside the stretching interval, the discharge characteristics controlled by the cavity breathing effect are locked, thereby improving the problem that traditional partial discharge diagnosis mostly uses electrical power frequency phase statistics, which cannot reflect the mechanical opening and closing law of the cavity, thus making it difficult to accurately identify the real breathing type discharge in a large amount of background noise. Attached Figure Description

[0011] Figure 1 This is a flowchart of a smart condition monitoring method for main transformers based on multi-source data proposed in this invention; Figure 2 This is a flowchart of the viscoelastic mechanical model decoupling and irreversible plastic cavity index extraction of a smart condition monitoring method for main transformers based on multi-source data proposed in this invention. Figure 3 This is a flowchart of the electromechanical phase modulation mapping and electromechanical modulation depth index extraction of a main transformer intelligent condition monitoring method based on multi-source data proposed in this invention. Figure 4 This is a flowchart of the gas breakdown physical model verification and overlap determination of a smart condition monitoring method for main transformers based on multi-source data proposed in this invention. Figure 5 This is a flowchart of the cross-physical domain multidimensional parameter logic decision and early warning output of a smart state monitoring method for main transformers based on multi-source data proposed in this invention. Detailed Implementation

[0012] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Example 1: In a first embodiment of the present invention, the present invention provides a method for intelligent condition monitoring of main transformers based on multi-source data, such as... Figure 1 As shown, it includes the following steps: S1. Synchronously acquire three-dimensional vibration acceleration time-domain sequence, ultra-high frequency partial discharge pulse sequence, reference phase voltage signal and load current signal; Furthermore, in S1, the synchronous acquisition of the three-dimensional vibration acceleration time-domain sequence, the ultra-high frequency partial discharge pulse sequence, the reference phase voltage signal, and the load current signal includes: The same timestamp is assigned to the three-dimensional vibration sensor array, ultra-high frequency partial discharge sensor, voltage transformer and current transformer by a unified high frequency clock source. By using the same timestamp, the collected three-dimensional vibration acceleration time-domain sequence, ultra-high frequency partial discharge pulse sequence, reference phase voltage signal and load current signal are time-axis aligned to construct a multi-source heterogeneous data matrix.

[0014] Specifically, step S1 discloses the underlying acquisition and interaction architecture for multi-physics field signals, constructing a closed-loop input-output data flow from the analog end to the digital end. The system input receives raw physical signals acquired from a 3D vibration sensor array, a UHF partial discharge sensor, a voltage transformer, and a current transformer. The processing platform adopts a low-level hardware clock synchronization architecture based on a field-programmable gate array (FPGA). By executing the IEEE 1588 precise time synchronization protocol, it sends absolute time reference pulses to the 3D vibration sensor array, the UHF partial discharge sensor, the voltage transformer, and the current transformer. Based on the sent absolute time reference pulses, the data acquisition channels of each heterogeneous sensor overcome nanosecond-level hardware group delays, forcibly attaching strictly aligned timestamps during analog-to-digital conversion. The system output, based on the same timestamp, extracts and splices time-stamped 3D vibration acceleration time-domain sequences, UHF partial discharge pulse sequences, reference phase voltage signals, and load current signals, outputting a multi-source heterogeneous data matrix.

[0015] Constructing a multi-source heterogeneous data matrix using functions The matrix expression is as follows: ; in The first allocation of a unified high-frequency clock source A discrete timestamp. For discrete timestamps The transient amplitude variables of the three-dimensional vibration acceleration time-domain sequence collected at the location. For discrete timestamps The transient discharge amplitude variable of the ultra-high frequency partial discharge pulse sequence collected at the site. For discrete timestamps The transient phase voltage amplitude variable of the reference phase voltage signal collected at the location. For discrete timestamps The transient current amplitude variable of the load current signal collected at the location. (Superscript) This represents the matrix transpose operation.

[0016] To support the cross-physical domain phase alignment requirements discussed later, the time synchronization error of each signal acquisition unit is... Specific rigid constraint equations need to be satisfied: ; in This represents the actual physical moment when any first sensor data channel is stamped with the same timestamp. This represents the actual physical moment when any second sensor data channel is stamped with the same timestamp. This represents the maximum permissible synchronization time error threshold defined by the system. To ensure the capture of the true physical phase correspondence between ultra-high frequency partial discharge pulse sequences above megahertz and the 100 Hz mechanical vibration frequency, the maximum permissible synchronization time error threshold... The specific value is set to 1 microsecond.

[0017] Step S1 plays a crucial supporting role in establishing the underlying physical scale throughout the entire technical solution. Conventional condition monitoring systems employ separate sampling modules, with voltage and current signals acquired on the high-voltage side and vibration partial discharge signals acquired on the low-voltage side each having their own independent, freely operating crystal oscillators, inevitably leading to cumulative clock drift. By relying on the absolute time reference alignment operation of the aforementioned unified high-frequency clock source, the phase misalignment distortion caused by independent sampling is eliminated. Without the synchronous constraint of a multi-source heterogeneous data matrix, the ultra-high frequency partial discharge pulse sequence will experience severe phase angle deviation when mapped to the subsequently calculated real-time mechanical vibration phase angle sequence. This results in the extracted maximum mechanical stretching preset phase interval failing to accurately align with the maximum opening moment of the internal plastic cavity, thus destroying the causal verification mechanism between the theoretical breakdown voltage threshold minimum point and the transient phase voltage amplitude from the underlying physical link.

[0018] like Figure 2 As shown, S2, the winding vibration displacement response is obtained by performing a second integral on the three-dimensional vibration acceleration time-domain sequence, and is input into the viscoelastic mechanical model together with the alternating Lorentz electrodynamic force calculated based on the load current signal. The steady-state residual between the predicted and measured displacements is used as the irreversible plastic cavity index. Furthermore, in S2, the winding vibration displacement response is obtained by performing a quadratic integration on the three-dimensional vibration acceleration time-domain sequence. This response, along with the alternating Lorentz electrodynamics calculated based on the load current signal, is used as input to the viscoelastic mechanical model, which includes: The three-dimensional vibration acceleration time-domain sequence is processed by time-domain quadratic integration and the trend term is eliminated to obtain the winding vibration displacement response; The amplitude of the load current signal is extracted and squared. The result of the amplitude squared operation is multiplied by a pre-calibrated electromagnetic excitation force conversion coefficient to obtain the alternating Lorentz electrodynamic force. The winding vibration displacement response and alternating Lorentz electrodynamic force are synchronously input into a pre-constructed viscoelastic mechanical model containing fractional derivative operators.

[0019] In S2, the steady-state residual between the predicted and measured displacements is used as the irreversible plastic cavity index, including: An extended Kalman filter algorithm is used to perform state observation on a viscoelastic mechanical model and output theoretically predicted displacements. Calculate the displacement difference sequence between the theoretically predicted displacement and the winding vibration displacement response; When the displacement difference sequence reaches the convergence state, the corresponding displacement difference is extracted as the steady-state residual, and the steady-state residual is used as the irreversible plastic cavity index.

[0020] Specifically, the system data channel extracts the three-dimensional vibration acceleration time-domain sequence and first performs a time-domain quadratic integration operation on it. To eliminate baseline drift caused by low-frequency temperature drift of the sensor and cumulative integration error, a polynomial fitting algorithm is applied to remove extremely low-frequency trend terms from the integrated sequence, restoring the pure winding vibration displacement response. Regarding the force mechanism of the transformer winding, the mechanical force borne by the winding coil in the leakage magnetic field is directly proportional to the square of the current amplitude flowing through the coil. The transient amplitude of the load current signal is extracted and squared, then converted using a pre-calibrated electromagnetic excitation force conversion coefficient to calculate the alternating Lorentz electrodynamic force applied to the winding.

[0021] The corresponding equation for calculating the alternating Lorentz electrodynamic force is: ; in The alternating Lorentz electrodynamic variables obtained from the solution. This represents the transient amplitude variable of the load current signal. The specific method for obtaining the pre-calibrated electromagnetic excitation force conversion coefficient is as follows: A three-dimensional electromagnetic and structural coupling model of the target main transformer is established using multiphysics finite element simulation software. Rated current excitation is applied, and the nodal electrodynamic forces on the windings are extracted. The conversion coefficient is then calibrated using multiple linear regression fitting. And it is stored in the system's main control unit.

[0022] After obtaining the winding vibration displacement response and alternating Lorentz electrodynamic force, these two are used as synchronous input sources and injected into a pre-constructed viscoelastic mechanical model containing fractional derivative operators. Transformer oil-paper insulation is a polymeric viscoelastic composite medium. Conventional integer-order spring-damping models cannot characterize the long-range history memory decay characteristics exhibited by polymeric materials under long-term alternating stress. Introducing fractional derivative operators can accurately quantify the stress relaxation behavior of this material.

[0023] The ideal viscoelastic constitutive mechanical equations under the condition of no microscopic damage are as follows: ; in This represents the theoretically predicted displacement output by the model under alternating Lorentz electrodynamic drive. It is the equivalent elastic stiffness constant of the insulation structure. The viscous damping coefficient is used to characterize the viscous dissipation properties of oil-paper insulation. It is the Caputo fractional derivative operator. It is a fractional order, and its specific range of values ​​is limited to 1. Equivalent elastic stiffness constant Viscous damping coefficient and fractional order The engineering acquisition path is as follows: under the factory test or initial undamaged healthy state of the main transformer, a known wideband sweep frequency excitation force signal is applied to the winding, and the reference vibration displacement response sequence of the winding is acquired simultaneously; then, the nonlinear least squares algorithm is used to identify the system parameters of the applied measured excitation force and the reference vibration displacement response sequence, and the specific values ​​of the above three viscoelastic constitutive parameters are inverted and fitted, and these values ​​are pre-fixed into the system memory as known constants.

[0024] The system's main control unit invokes the Extended Kalman Filter (EKF) algorithm to perform state-space iterative observations of the aforementioned viscoelastic constitutive mechanical equations. To ensure the accuracy of the initial filtering values, during the no-load steady-state test phase of the main transformer before its factory delivery, steady-state vibration data under partial discharge-free conditions are extracted, and the initial state vector and initial covariance matrix of the EKF algorithm are calculated and fixed with zero bias. When implementing online monitoring, since the standard EKF algorithm cannot directly handle fractional derivatives containing global memory, the processor pre-applies the Oustaloup continuous-frequency filter approximation algorithm to discretize the equations containing Caputo fractional derivative operators, transforming them into finite-dimensional integer-order state and observation equations.

[0025] The constructed system state vector is: ; The state transition equation is: ; The observation equation is: ;in Here is the state transition matrix. To control the input matrix, For the observation matrix, and ; these are the system process noise and measurement noise sequences, respectively.

[0026] Assume system process noise The covariance matrix is Measurement noise sequence The covariance matrix is Measurement noise covariance matrix The engineering method for obtaining the values ​​is as follows: Under the static condition of main transformer de-energization, extract the background white noise data of the three-dimensional vibration sensor array and calculate its statistical variance matrix; obtain the process noise covariance matrix. The engineering parameter selection method is as follows: During the healthy service life of the main transformer, the residual sequence of the predicted output of the ideal viscoelastic constitutive mechanical model and the measured winding vibration displacement are compared. The maximum likelihood estimation method is used for offline parameter identification and acquisition, and the calibrated covariance matrix is ​​then used. and Pre-installed into the system program.

[0027] Extended Kalman filter algorithm with alternating Lorentz electrodynamics As an external stimulus, combined with the fixed covariance matrix and Iterative calculation theory predicts displacement Further analysis of theoretically predicted displacements The winding vibration displacement response, which is a measured physical quantity Subtraction is performed to generate a displacement difference sequence that progresses with the time series. When the relative rate of change of the trace of the posterior estimated covariance matrix output by the extended Kalman filter algorithm is continuously less than a set convergence threshold within a set number of consecutive observation steps, the system determines that the displacement difference sequence has reached a steady-state convergence state. At this state node, the corresponding displacement difference is extracted and confirmed as the steady-state residual, and the steady-state residual is directly assigned to the irreversible plastic cavity exponent.

[0028] The equation for extracting the displacement difference sequence is: ; in The generated displacement difference sequence variable. This is the extracted winding vibration displacement response.

[0029] When the gain matrix and covariance matrix of the extended Kalman filter algorithm converge to near the boundary threshold, the system determines that the displacement difference sequence has reached a convergent state. At this state node, the corresponding displacement difference is extracted and confirmed as the steady-state residual, and the steady-state residual is directly assigned to the irreversible plastic cavity exponent.

[0030] The role of step S2 is to cut off the masking effect of macroscopic electrical fluctuations on microscopic defect characteristics. The aforementioned steady-state residuals, within the physical domain, strip away the forced reversible elastic deformation component and the viscous hysteresis component of the insulating material. The remaining value absolutely corresponds to the volume of the microscopic cavities formed inside the insulating paperboard due to the cumulative tearing caused by Lorentz forces. Establishing this parameter as the irreversible plastic cavity index provides a quantitative criterion for characterizing purely mechanical micro-damage in subsequent cross-domain verification, preventing the misjudgment of normal transformer-loaded micro-vibrations as structural damage faults.

[0031] like Figure 3 As shown, S3, extract the fundamental frequency phase angle based on the winding vibration displacement response and map the UHF partial discharge pulse sequence; calculate the ratio of the probability or average amplitude of discharge pulses inside and outside the maximum mechanical stretch preset phase interval as the electromechanical modulation depth index; Furthermore, in S3, the extraction of the fundamental frequency phase angle based on the winding vibration displacement response and the mapping of the ultra-high frequency partial discharge pulse sequence include: Hilbert transform is applied to the winding vibration displacement response to obtain an analytical signal. The real-time mechanical vibration phase angle sequence is calculated based on the ratio of the analytical signal to the winding vibration displacement response as the fundamental frequency phase angle. Extract the timestamp of each discharge pulse in the ultra-high frequency partial discharge pulse sequence; Based on the occurrence timestamp, the corresponding transient phase point is matched in the real-time mechanical vibration phase angle sequence to construct a three-dimensional data matrix of phase resolution, mechanical vibration, and partial discharge.

[0032] In S3, the ratio of the probability or mean amplitude of discharge pulses inside and outside the preset phase interval of maximum mechanical stretching is calculated as the electromechanical modulation depth index, including: In the fundamental frequency phase angle, locate the center phase where the winding vibration displacement response is in the maximum mechanical tension state; The window width is expanded to both sides based on the center phase to define the maximum mechanical stretching preset phase range; The total number of first partial discharge pulses within the preset phase interval of maximum mechanical stretching is counted, and the total number of second partial discharge pulses within the global phase period is counted. Calculate the ratio of the total number of first partial discharge pulses to the total number of second partial discharge pulses, and output the ratio as the electromechanical modulation depth index.

[0033] Specifically, step S3 uses the winding vibration displacement response extracted in step S2 and the UHF partial discharge pulse sequence obtained in step S1 as pre-input parameters. By extracting the mechanical phase to reconstruct the partial discharge spectrum, it outputs an electromechanical modulation depth index for quantifying the cross-domain correlation characteristics of micro-damage.

[0034] The system's digital processor reads the winding vibration displacement response data sequence and applies a continuous-time Hilbert transform to it. The Hilbert transform maps the original one-dimensional real signal to a complex-plane spatial signal, stripping the real part of the original signal and generating an analytic signal carrying transient imaginary part characteristics. Based on the tangent relationship between the acquired analytic signal and the original winding vibration displacement response, the processor calculates the real-time mechanical vibration phase angle sequence measured in radians. The system directly defines the calculated real-time mechanical vibration phase angle sequence as the fundamental frequency phase angle.

[0035] The corresponding equations for the Hilbert transform and the fundamental frequency phase angle are as follows: ; ; in These are the analytic signal variables obtained through the Hilbert transform; The extracted winding vibration displacement response variables; The sampling time variable; This is the integral shift parameter in the Hilbert integral transform; The real-time mechanical vibration phase angle sequence variable is calculated, i.e., the fundamental frequency phase angle.

[0036] After acquiring the phase scale in the mechanical dimension, the processing platform calls upon a data channel with a microsecond-level absolute time reference to extract the precise occurrence timestamp of each individual discharge pulse in the UHF partial discharge pulse sequence. Using the occurrence timestamp as the index key for data association, time axis alignment and comparison are performed in the real-time mechanical vibration phase angle sequence to match the transient phase point corresponding to the occurrence time of each discharge pulse. Based on the matched transient phase points and discharge pulse amplitude parameters, the system constructs a three-dimensional data matrix encompassing phase resolution, mechanical vibration, and partial discharge elements.

[0037] To capture the breathing discharge characteristics of the insulating cavity caused by forced vibration, the main control logic searches the fundamental frequency phase angle data column for the position where the winding vibration displacement response is at the peak of the positive displacement, and positions the absolute phase point where the corresponding winding deformation is in the maximum mechanical tension state as the center phase. Starting from the center phase as the centerline, a pre-defined window width is symmetrically extended to both sides to delineate the maximum mechanical tension preset phase interval covering the fully opened state of the micro-cavity. To cover the effective discharge duration of the dynamically opened cavity, the specific value range of the set window width is defined as a mechanical phase angle offset of 15 to 45 degrees to both sides. This specific value range of the set window width is calibrated using the physical statistical laws of the alternating tensile micropore imaging experiment of the oil-paper insulation sample.

[0038] When a microscopic insulating cavity experiences a sudden drop in internal pressure within a preset phase range of maximum mechanical tension, the probability of partial discharge increases dramatically. The data statistics module counts pulse events falling within the preset phase range of maximum mechanical tension to obtain the first total number of partial discharge pulses; and performs a complete count of all discharge events occurring throughout the entire global phase cycle to obtain the second total number of partial discharge pulses. The processing unit performs a division operation to calculate the ratio of the first to the second total number of partial discharge pulses. The processor outputs the calculated ratio and uses it as an electromechanical modulation depth index, characterizing the strength of the electromechanical-physical correlation.

[0039] The formula for the electromechanical modulation depth index is as follows: ; in The output is the electromechanical modulation depth index variable; This represents the set of predefined phase intervals for maximum mechanical stretching. The index number of the discretized phase window; In the first The cumulative count of detected discharge pulses within each phase window; This represents the total number of the first partial discharge pulses obtained from statistics; This represents the total number of second partial discharge pulses obtained from statistics, among which The total window count value set for global phase period division.

[0040] Step S3 breaks away from the existing monitoring equipment's reliance solely on power frequency electrical phase to determine partial discharge. The equivalent volume of the microscopic plastic cavity inside the transformer changes with the mechanical vibration of the windings, forming a localized low-pressure air gap within a specific mechanical displacement extremity range, thus satisfying the ionization breakdown condition. A three-dimensional data matrix is ​​constructed, and the electromechanical modulation depth index is extracted, eliminating external background corona interference from the substation and random electrical noise unrelated to mechanical displacement. This step pinpoints the discharge characteristics derived from the breathing effect of the microscopic mechanical cavity, establishing a cross-domain causal verification criterion for the evolution from the mechanical damage domain to the electromagnetic discharge domain.

[0041] like Figure 4 As shown, S4 determines the dynamic air gap distance based on the irreversible plastic cavity index, inputs the gas breakdown model to obtain the minimum value of the theoretical breakdown voltage threshold, and judges whether the transient phase voltage amplitude corresponding to the reference phase voltage signal at the moment of occurrence of the UHF partial discharge pulse sequence coincides with the minimum value point within the preset deviation tolerance. Furthermore, in S4, determining the dynamic air gap distance based on the irreversible plastic cavity index includes: The initial gap constant and deformation-coupled expansion coefficient of the microplastic damage of the main transformer winding were obtained. The dynamic expansion modulation term is obtained by multiplying the deformation coupling expansion coefficient by the irreversible plastic cavity exponent. The initial gap constant of microplastic damage is amplified by using a dynamic expansion modulation term to generate a dynamic air gap distance that varies with the phase of mechanical vibration.

[0042] In S4, the minimum values ​​of the theoretical breakdown voltage threshold obtained by inputting the gas breakdown model include: The dynamic air gap distance is input into the breakdown verification equation based on Paschen's law. Combined with the preset gas pressure parameter and the vaporization collision ionization constant of insulating oil, the fluctuation curve of the theoretical breakdown voltage threshold with the phase angle of real-time mechanical vibration is calculated. By performing an extreme value search on the fluctuation curve, the minimum value of the theoretical breakdown voltage threshold corresponding to the maximum value of the dynamic air gap distance is located.

[0043] In S4, determining whether the transient phase voltage amplitude corresponding to the moment the ultra-high frequency partial discharge pulse sequence occurs in the reference phase voltage signal coincides with the minimum point within the preset deviation tolerance includes: Extract the transient phase voltage amplitude corresponding to the moment of occurrence of the ultra-high frequency partial discharge pulse sequence from the reference phase voltage signal; Determine the actual breakdown phase point where the transient phase voltage amplitude exceeds the discharge initiation voltage, and determine the theoretical breakdown phase point where the theoretical breakdown voltage threshold reaches its minimum value; Calculate the absolute value of the phase difference between the actual breakdown phase point and the theoretical breakdown phase point; Determine whether the absolute value of the phase difference is less than the preset deviation tolerance.

[0044] Specifically, the arithmetic unit reads the initial gap constant of the microplastic damage in the main transformer winding and the pre-calibrated deformation coupling expansion coefficient. It then performs a multiplication operation between the deformation coupling expansion coefficient and the irreversible plastic cavity index to obtain a dynamic expansion modulation term. Using this obtained dynamic expansion modulation term, it proportionally amplifies and modulates the initial gap constant of the microplastic damage, calculating the dynamic air gap distance that undergoes periodic opening and closing deformation with the mechanical vibration phase.

[0045] The equation for solving the dynamic air gap distance is: ; in This refers to the generated dynamic air gap distance variable. This represents the phase angle of real-time mechanical vibration. The initial gap constant for microplastic damage of the main transformer winding is read. The pre-calibrated deformation coupling expansion coefficient. The irreversible plastic cavity index is extracted. The engineering approach to obtain the initial gap constant of microplastic damage and the deformation coupled expansion coefficient is as follows: before leaving the factory, offline tensile destructive mechanical tests are conducted on oil-paper insulation samples of the same batch of transformers. The internal pore volume change rate is measured by scanning electron microscopy, and the data is input into the system after parameter identification and calibration.

[0046] After obtaining the dynamic air gap distance, the system uses it as a core boundary condition and substitutes it into the gas breakdown verification equation based on Paschen's law to perform cross-physics domain derivation. Considering that the main transformer's interior is an oil-immersed insulation environment, and the microscopic cavity is actually a two-phase fluid of oil and gas, the gas breakdown behavior differs physically from that of a pure air gap. Therefore, an oil-gas two-phase correction coefficient based on the saturation of dissolved gas in the oil is introduced into the classical Paschen equation. The processing platform retrieves the extracted gas pressure parameters inside the microscopic cavity and the collisional ionization constant characterizing the vaporization properties of the insulating oil. Combined with the input dynamic air gap distance, the system derives and calculates the mapping curve of the theoretical breakdown voltage threshold as a function of the real-time mechanical vibration phase angle.

[0047] The corrected verification equation for the theoretical breakdown voltage threshold is as follows: ; in This is to derive the theoretical breakdown voltage threshold variable for the corresponding mechanical phase. The oil-gas two-phase correction coefficient is introduced, and its specific value is obtained by fitting the destructive test data of the insulating oil insulation withstand voltage standard. The gas pressure parameters inside the micro-cavity are extracted. These parameters are calculated by reading the steady-state pressure monitoring value from the static pressure sensor at the top of the transformer tank and superimposing it with the static head pressure of the insulating oil at the physical location of the winding where the defect is located. and Both are the collisional ionization constants of insulating oil vaporization, which depend on the inherent physical properties of the collisional ionization cross-section of insulating material molecules under a strong electric field. The secondary electron emission coefficient is denoted as . On the generated wave curve, the system locates the coordinate point where the derivative approaches 0 and the amplitude is lowest, and extracts the corresponding phase abscissa as the theoretical breakdown phase point.

[0048] After completing the theoretical minimum value location, the system performs causal closed-loop verification at the actual discharge occurrence time. In the data sequence of the reference phase voltage signal, based on the timestamp index of the UHF partial discharge pulse sequence occurrence time, the transient phase voltage amplitude at the corresponding moment is extracted. The system calibrates the electromagnetic transient phase where the transient phase voltage amplitude exceeds the discharge initiation voltage as the actual breakdown phase point. The arithmetic unit calculates the difference between the extracted actual breakdown phase point and the derived theoretical breakdown phase point, obtaining the absolute value of the phase difference. Finally, the judgment module compares the calculated absolute value of the phase difference with a pre-set deviation tolerance to determine whether the absolute value of the phase difference falls within the limited tolerance range. Due to the inherent effects of the sampling delay of the field sensors and the delay of the hardware filter group, the specific value of the preset deviation tolerance is limited to a mechanical phase angle of 3 to 5 degrees. This value range is determined by the statistical results of the normal distribution of the error probability from multiple main transformer field-loaded excitation simulation experiments.

[0049] Step S4 plays a crucial role in the overall condition monitoring method, determining the mechanism and eliminating false signals. Substation sites are filled with massive amounts of ionized electromagnetic interference, including spatial corona, loose contacts, and external lightning strikes. Relying solely on threshold over-limit alarms or conventional phase distribution spectrum statistics can easily misjudge external background electromagnetic noise as severe, destructive discharges inside the transformer. This step introduces a gas breakdown physical model to establish a rigid criterion, mandating that the timing of the collected electrical discharge strictly coincides with the physical extreme point where mechanical stretching causes the insulation cavity breakdown threshold to drop to its lowest value. Utilizing this underlying cross-causal verification mechanism, all irrelevant pulses that violate the mechanical breathing modulation pattern on the time axis are precisely eliminated, establishing a unique and exclusive fault characteristic defense line against discharges associated with microscopic plastic damage in the windings.

[0050] like Figure 5As shown, S5, when both the irreversible plastic cavity index and the electromechanical modulation depth index are greater than the preset threshold, and the judgment results are the same, an early warning command is output; Furthermore, in S5, when both the irreversible plastic cavity index and the electromechanical modulation depth index are greater than preset thresholds, and the judgment results coincide, the output warning command includes: Read the pre-set irreversible plastic damage boundary threshold and the critical value of strong correlation electromechanical modulation; The irreversible plastic cavity index is compared with the irreversible plastic damage boundary threshold. The electromechanical modulation depth index is compared with the critical value of strongly correlated electromechanical modulation; When the irreversible plastic cavity index is greater than the irreversible plastic damage boundary threshold, and the electromechanical modulation depth index is greater than the strongly correlated electromechanical modulation critical value, and the judgment results are coincident, the closed-loop output includes the warning command containing the characteristic result of the microscopic plastic damage of the main transformer winding accompanied by the breathing-type cavity discharge.

[0051] Specifically, the system logic judgment module reads the irreversible plastic cavity index extracted in step S2 and the electromechanical modulation depth index calculated in step S3. In synchronous operation, the processor retrieves two pre-set benchmark criteria from the system memory: the pre-defined irreversible plastic damage boundary threshold and the strongly correlated electromechanical modulation critical value. Subsequently, the comparison unit compares the irreversible plastic cavity index with the irreversible plastic damage boundary threshold and compares the electromechanical modulation depth index with the strongly correlated electromechanical modulation critical value.

[0052] The triggering logic for the joint judgment of the above multi-dimensional parameters is characterized by Boolean control equations: ; in This is a status variable for the early warning command generated by the system. A value of 1 indicates that the control loop is triggered and an early warning is issued, while a value of 0 indicates that no action is taken and the current monitoring status is maintained. The irreversible plastic cavity index variable is input to the decision module. This is a pre-defined threshold constant for irreversible plastic damage. The electromechanical modulation depth index variable is input to the decision module. It is a pre-set critical value constant for strong correlation electromechanical modulation. The physical mechanism verification state variable passed in step S4 is assigned a value of 1 if the absolute value of the phase difference in step S4 is strictly less than a preset deviation tolerance, and otherwise assigned a value of 0. (Symbol) It represents the execution of logic and operations.

[0053] When the logic unit determines that the irreversible plastic cavity index is substantially greater than the irreversible plastic damage boundary threshold, the electromechanical modulation depth index is substantially greater than the strongly correlated electromechanical modulation critical value, and the physical mechanism judgment results confirm overlap, the three preconditions are fully activated. The control unit immediately activates the alarm link, and the output module sends an early warning command to the station control layer monitoring terminal. The generated early warning command data packet directly encapsulates the code of the characteristic result of the microscopic plastic damage of the main transformer winding accompanied by the breathing-type cavity discharge.

[0054] Step S5 establishes a defense line of joint confirmation across physical fields within the monitoring system. In conventional substation environments, single vibration amplitude exceeding limits or frequent partial discharges often originate from normal load surges or external lightning and switch operation interference. The three-dimensional logical threshold established in this step mandates that the monitored object must simultaneously meet the microscopic damage entity conditions in the mechanical domain, the electromechanical strong correlation statistical conditions in the data domain, and the gas breakdown mechanism conditions at the physical level. This joint decision mechanism cuts off the transmission path of false alarms caused by noise in independent physical domains, ensuring that the fault characterization results ultimately issued by the system have a precise physical causal correspondence, directly eliminating superficial interference for on-site operation and maintenance and accurately pinpointing the evolution chain of latent defects.

[0055] Example 2: This invention is applied to ultra-high voltage and extra-high voltage main transformers subjected to external short-circuit current impacts or long-term severe load fluctuations. In such scenarios, the internal oil-paper insulation of the main transformer is prone to microscopic irreversible plastic damage and the formation of internal micro-tear cavities under the enormous alternating Lorentz electrodynamic impact. A specific technical problem in this scenario is that the micro-cavities generate a periodic opening and closing breathing effect under the forced vibration of the windings, and partial discharge is induced by a sudden drop in internal air pressure under maximum mechanical tension. Existing monitoring methods usually treat mechanical vibration and partial discharge separately, relying only on the power frequency electrical phase for discharge spectrum statistics, lacking a definition of the cross-physical domain coupling mechanism between microscopic mechanical damage and high-frequency discharge pulses; under the masking of complex substation background electromagnetic noise, conventional techniques cannot extract the discharge pulse characteristics modulated by microscopic mechanical displacement, making it difficult for the system to achieve attribution and tracing based on the underlying physical causal chain in the very early latent stage of insulation failure, easily resulting in missed detection of breathing cavity discharges, and thus triggering sudden insulation breakdown accidents of the main transformer. To address the aforementioned problems, this invention provides a method for intelligent condition monitoring of main transformers based on multi-source data, such as... Figure 1 As shown. The specific implementation process of this method is as follows: The acquired synchronous multi-source signals provide a baseline data foundation for state tracing. A viscoelastic mechanical model is used to decouple the purely irreversible plastic cavity index, eliminating elastic deformation interference under normal load and achieving a quantitative characterization of the degree of tearing within the insulation. Furthermore, it breaks through the limitations of conventional partial discharge diagnosis relying on power frequency phase, reconstructing the discharge pulse spectrum using the fundamental frequency phase of mechanical displacement as a scale, extracting the electromechanical modulation depth index, and locking onto abnormal discharges controlled by mechanical stretching. Finally, a gas breakdown physical model is introduced to establish a hard criterion, forcibly verifying the coincidence between the discharge transient voltage and the maximum opening extreme point of the cavity. This series of steps is interconnected, completely blocking the transmission of false alarms caused by background noise in independent physical domains, accurately locking onto latent defects of micro-damage in the winding accompanied by breathing-type discharge in the very early stages before macroscopic faults appear.

[0056] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent condition monitoring of main transformers based on multi-source data, characterized in that, Includes the following steps: S1. Synchronously acquire three-dimensional vibration acceleration time-domain sequence, ultra-high frequency partial discharge pulse sequence, reference phase voltage signal and load current signal; S2. The three-dimensional vibration acceleration time-domain sequence is integrated twice to obtain the winding vibration displacement response, which is then input into the viscoelastic mechanical model together with the alternating Lorentz electrodynamic force calculated based on the load current signal. The steady-state residual between the predicted and measured displacements is used as the irreversible plastic cavity index. S3. Extract the fundamental frequency phase angle based on the winding vibration displacement response and map the ultra-high frequency partial discharge pulse sequence; The ratio of the probability or average amplitude of discharge pulses inside and outside the preset phase interval of maximum mechanical stretching is calculated as the electromechanical modulation depth index. S4. Determine the dynamic air gap distance based on the irreversible plastic cavity index, input the gas breakdown model to obtain the minimum value of the theoretical breakdown voltage threshold; determine whether the transient phase voltage amplitude corresponding to the reference phase voltage signal at the time of occurrence of the ultra-high frequency partial discharge pulse sequence coincides with the minimum value point within the preset deviation tolerance. S5. When both the irreversible plastic cavity index and the electromechanical modulation depth index are greater than the preset threshold, and the judgment results are the same, an early warning command is output.

2. The intelligent condition monitoring method for main transformers based on multi-source data according to claim 1, characterized in that, In S1, the synchronous acquisition of the three-dimensional vibration acceleration time-domain sequence, the ultra-high frequency partial discharge pulse sequence, the reference phase voltage signal, and the load current signal includes: The same timestamp is assigned to the three-dimensional vibration sensor array, ultra-high frequency partial discharge sensor, voltage transformer and current transformer by a unified high frequency clock source. Using the same timestamp, the collected three-dimensional vibration acceleration time-domain sequence, the ultra-high frequency partial discharge pulse sequence, the reference phase voltage signal, and the load current signal are time-axis aligned to construct a multi-source heterogeneous data matrix.

3. The intelligent condition monitoring method for main transformers based on multi-source data according to claim 1, characterized in that, In S2, the winding vibration displacement response obtained by performing a second integral on the three-dimensional vibration acceleration time-domain sequence, together with the alternating Lorentz electrodynamic force calculated based on the load current signal, is input into the viscoelastic mechanical model, including: The three-dimensional vibration acceleration time-domain sequence is subjected to time-domain quadratic integration and the trend term is eliminated to obtain the winding vibration displacement response; The amplitude of the load current signal is extracted and squared. The result of the amplitude squared operation is multiplied by the pre-calibrated electromagnetic excitation force conversion coefficient to obtain the alternating Lorentz electrodynamic force. The winding vibration displacement response and the alternating Lorentz electrodynamic force are synchronously input into the pre-constructed viscoelastic mechanical model containing fractional derivative operators.

4. The intelligent condition monitoring method for main transformers based on multi-source data according to claim 1, characterized in that, In S2, the step of using the steady-state residual between the predicted and measured displacements as the irreversible plastic cavity index includes: The extended Kalman filter algorithm is used to perform state observation on the viscoelastic mechanical model and output the theoretically predicted displacement. Calculate the displacement difference sequence between the theoretically predicted displacement and the winding vibration displacement response; When the displacement difference sequence reaches a convergence state, the corresponding displacement difference is extracted as the steady-state residual, and the steady-state residual is used as the irreversible plastic cavity index.

5. The intelligent condition monitoring method for main transformers based on multi-source data according to claim 1, characterized in that, In S3, the step of extracting the fundamental frequency phase angle based on the winding vibration displacement response and mapping the ultra-high frequency partial discharge pulse sequence includes: Hilbert transform is applied to the winding vibration displacement response to obtain an analytical signal, and the real-time mechanical vibration phase angle sequence is calculated based on the ratio of the analytical signal to the winding vibration displacement response as the fundamental frequency phase angle; Extract the occurrence timestamp of each discharge pulse in the ultra-high frequency partial discharge pulse sequence; Based on the occurrence timestamp, the corresponding transient phase point is matched in the real-time mechanical vibration phase angle sequence to construct a three-dimensional data matrix of phase resolution, mechanical vibration, and partial discharge.

6. The intelligent condition monitoring method for main transformers based on multi-source data according to claim 1, characterized in that, In S3, the ratio of the probability or average amplitude of discharge pulses inside and outside the preset phase interval of the maximum mechanical stretch, as the electromechanical modulation depth index, includes: In the fundamental frequency phase angle, the center phase corresponding to the winding vibration displacement response being in the maximum mechanical tension state is located; The window width is extended to both sides based on the central phase to define the maximum mechanical stretching preset phase range; The total number of first partial discharge pulses within the preset phase interval of the maximum mechanical stretch is counted, and the total number of second partial discharge pulses within the global phase period is counted. Calculate the ratio of the total number of the first partial discharge pulses to the total number of the second partial discharge pulses, and output the ratio as the electromechanical modulation depth index.

7. The intelligent condition monitoring method for main transformers based on multi-source data according to claim 1, characterized in that, In S4, determining the dynamic air gap distance based on the irreversible plastic cavity index includes: The initial gap constant and deformation-coupled expansion coefficient of the microplastic damage of the main transformer winding were obtained. The dynamic expansion modulation term is obtained by multiplying the deformation coupling expansion coefficient by the irreversible plastic cavity index; The initial gap constant of the microplastic damage is amplified using the dynamic expansion modulation term to generate the dynamic air gap distance that varies with the phase of mechanical vibration.

8. The intelligent condition monitoring method for main transformers based on multi-source data according to claim 1, characterized in that, In S4, the input gas breakdown model obtains the minimum value of the theoretical breakdown voltage threshold by including: The dynamic air gap distance is input into the breakdown verification equation constructed based on Paschen's law. Combined with the preset gas pressure parameter and the vaporization collision ionization constant of insulating oil, the fluctuation curve of the theoretical breakdown voltage threshold with the phase angle of real-time mechanical vibration is calculated. An extreme value search is performed on the fluctuation curve to locate the minimum value of the theoretical breakdown voltage threshold corresponding to the maximum value of the dynamic air gap distance.

9. The intelligent condition monitoring method for main transformers based on multi-source data according to claim 1, characterized in that, In S4, determining whether the transient phase voltage amplitude corresponding to the moment when the ultra-high frequency partial discharge pulse sequence occurs coincides with the minimum point within a preset deviation tolerance includes: Extract the transient phase voltage amplitude corresponding to the moment when the ultra-high frequency partial discharge pulse sequence occurs from the reference phase voltage signal; Determine the actual breakdown phase point where the transient phase voltage amplitude exceeds the discharge initiation voltage, and determine the theoretical breakdown phase point where the theoretical breakdown voltage threshold reaches its minimum value; Calculate the absolute value of the phase difference between the actual breakdown phase point and the theoretical breakdown phase point; Determine whether the absolute value of the phase difference is less than the preset deviation tolerance.

10. The intelligent condition monitoring method for main transformers based on multi-source data according to claim 1, characterized in that, In S5, when both the irreversible plastic cavity index and the electromechanical modulation depth index are greater than a preset threshold, and the judgment results coincide, the output warning command includes: Read the pre-set irreversible plastic damage boundary threshold and the critical value of strong correlation electromechanical modulation; The irreversible plastic cavity index is compared with the irreversible plastic damage boundary threshold. The electromechanical modulation depth index is compared with the strongly correlated electromechanical modulation threshold value; When the irreversible plastic cavity index is greater than the irreversible plastic damage boundary threshold, and the electromechanical modulation depth index is greater than the strongly correlated electromechanical modulation critical value, and the judgment results are coincident, the closed-loop output includes a warning command containing the characteristic result of microscopic plastic damage of the main transformer winding accompanied by breathing-type cavity discharge.

Images