Transformer anti-short-circuit impact method and system based on adaptive adjustment of winding inter-turn dynamic pre-tightening force

Abstract

The application discloses a transformer anti-short-circuit impact method and system based on winding inter-turn dynamic pre-tightening force self-adaptive adjustment. The method comprises the following steps: dividing the winding into multiple regulation and control sub-zones along the axial direction, collecting the electrical operation parameters and mechanical response parameters of each sub-zone in real time; identifying the short-circuit fault type by using the symmetrical component method, and matching the stress distribution mode in combination with an offline calibration library; and identifying the stiffness and damping state of the winding on line based on the Bouc-Wen nonlinear model. The optimal pre-tightening force command of each sub-zone is calculated by using the sequential quadratic programming algorithm with the maximum system energy dissipation as the target, and the dynamic pre-tightening force is applied to the winding through the hydraulic servo system with an accumulator. The pre-tightening strategy can be adjusted in real time according to the short-circuit working condition and the winding aging state, the winding vibration instability is effectively inhibited, and the anti-short-circuit capability of the transformer is improved.

Description

Technical Field

[0001] This invention relates to the field of power transformer safety protection technology, and in particular to a method and system for transformer short-circuit impact protection based on adaptive adjustment of dynamic preload between winding turns. Background Technology

[0002] Power transformers are core equipment in power grids, and their short-circuit withstand capability is directly related to the safe and stable operation of the power system. With the continuous increase in the short-circuit capacity of the power grid, the electromagnetic force impact that transformer windings are subjected to at the moment of a short-circuit fault is becoming increasingly severe, which can easily lead to serious accidents such as radial instability, axial collapse, or end insulation damage of the windings.

[0003] Currently, to improve the short-circuit withstand capability of transformers, constant force preload devices or simple spring clamping mechanisms are commonly used. These devices apply a fixed axial preload during the transformer manufacturing stage, using mechanical limits to resist electromagnetic shocks. Alternatively, simple feedback control based on current amplitude can be employed, where a hydraulic mechanism outputs pressure proportionally when a large current is detected.

[0004] However, existing technologies face multiple challenges in practical applications. Fixed preload is difficult to adapt to the dimensional shrinkage of insulation materials over time due to aging, leading to preload decay and loss of protective capability in the later stages of operation. Simple control based on current amplitude ignores the differences in short-circuit fault types, making it difficult to cope with different fault modes, such as single-phase grounding versus three-phase short circuits, and the significant spatial differences in stress distribution between the winding ends and the middle, resulting in localized overvoltage or undervoltage. Existing solutions treat the winding as a linear elastic body, ignoring the nonlinear hysteresis characteristics of oil-paper insulation under high short-circuit stress, making it difficult to achieve active energy dissipation. Instead, rigid resistance may exacerbate mechanical vibration and trigger structural resonance. Therefore, there is an urgent need for an active protection technology that can sense the winding's mechanical state in real time and perform adaptive zoned control. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for transformer short-circuit impact resistance based on adaptive adjustment of dynamic preload between winding turns (also known as winding zoning), in order to solve at least one of the above-mentioned problems existing in the prior art.

[0006] According to one aspect of this application, a method for transformer short-circuit impact protection based on adaptive adjustment of dynamic preload between winding turns includes:

[0007] Real-time acquisition of electrical operating parameters of transformer windings and mechanical response parameters of multiple control zones distributed along the axial direction;

[0008] Identify short-circuit fault characteristics based on electrical operating parameters and assess the current mechanical state of the winding based on mechanical response parameters;

[0009] Based on the characteristics of short-circuit faults and mechanical states, the target preload command for each control zone is calculated using a preset control strategy.

[0010] The actuators of each control zone are driven to operate according to the target preload command, and dynamic preload is applied to the winding.

[0011] According to another aspect of this application, a transformer short-circuit impact protection system based on adaptive adjustment of dynamic preload between winding turns includes:

[0012] The sensing module includes a current transformer for collecting electrical operating parameters, as well as displacement sensors and pressure sensors arranged in each control zone of the winding.

[0013] The controller is configured to execute any of the method steps described above in the transformer short-circuit impact resistance method based on adaptive adjustment of inter-turn dynamic preload, and to calculate the target preload command for each control zone based on the collected parameters.

[0014] The hydraulic drive unit, including an accumulator group, a multi-channel electro-hydraulic servo valve, and a hydraulic cylinder array, is used to respond to a target preload command and apply a zoned, independent dynamic preload to the winding using the energy released by the accumulator group.

[0015] Beneficial effects: This invention can adjust the pre-tightening strategy in real time according to short-circuit conditions and winding aging status, effectively suppressing winding vibration instability and improving the transformer's short-circuit withstand capability. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall process of the transformer short-circuit impact resistance method based on adaptive adjustment of dynamic preload between winding turns provided in the embodiments of this application.

[0017] Figure 2 This is a schematic diagram of the parameter identification process for a nonlinear hysteresis dynamics model using a recursive filtering algorithm, provided in an embodiment of this application.

[0018] Figure 3 This is a schematic diagram of the process for evaluating the current mechanical state of the winding based on mechanical response parameters, provided in an embodiment of this application.

[0019] Figure 4 This is a schematic diagram of the process for generating a stress distribution pattern library provided in the embodiments of this application. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0021] Example 1 provides an overall implementation framework for a transformer short-circuit impact resistance method based on adaptive adjustment of dynamic preload between winding turns, such as... Figure 1 As shown, by completing closed-loop control from sensing, analysis to decision-making and execution within an extremely short time (milliseconds) of short circuit occurrence, the technical problem of traditional constant force preload being unable to adapt to the time-varying characteristics of winding insulation aging and short circuit impact is solved.

[0022] Step 101: Real-time acquisition of electrical operating parameters of transformer windings and mechanical response parameters of multiple control zones distributed along the axial direction.

[0023] Specifically, electrical operating parameters mainly include the three-phase currents and three-phase voltages on the high-voltage, medium-voltage, and low-voltage sides of the transformer. These are typically obtained using current transformers (CTs) and voltage transformers (PTs) installed at the transformer bushings. The sampling frequency is preferably set to 10kHz or higher to meet the requirements for capturing short-circuit transient processes. Mechanical response parameters include the axial displacement and axial pressure experienced by each section of the winding.

[0024] In this embodiment, considering the significant spatial non-uniformity of the force distribution on the transformer windings under short-circuit impact, the concept of control zones is introduced. A control zone divides the continuous windings along the axial direction (height) into several independent control units. Each control zone is independently equipped with a displacement sensor and a pressure sensor. The displacement sensor can specifically be a fiber optic grating sensor embedded in the winding pad or an eddy current sensor installed at the winding end, used to monitor in real time the minute compression deformation of the winding coil relative to the yoke or adjacent coils in that zone. The pressure sensor can specifically be a thin-film pressure sensor or a piezoelectric force sensor, installed at the pressure pins or plates of each zone, used to directly measure the mechanical preload force on the windings.

[0025] The axial distribution density of the control zones is determined based on the leakage magnetic field distribution gradient of the winding. The control zone density at the winding ends is greater than that in the middle of the winding. The physical basis for the non-uniform zone design is that the transverse component of the leakage magnetic field at the ends of the transformer winding is the largest, resulting in the largest axial electromagnetic force and its gradient, making it a high-risk area for short-circuit damage. In contrast, the leakage magnetic field in the middle of the winding is mainly longitudinal, with smaller and more evenly distributed axial forces. For example, for a 220kV transformer winding with a height of 2000mm, control zones can be set every 50mm within a 200mm range at the top and bottom (4 zones at the ends); and every 200mm within a 1600mm range in the middle (8 zones in the middle). This results in a total of 16 zones across the entire winding, ensuring precise control of the high-risk areas at the ends while avoiding excessive redundancy of sensors and actuators in the middle area, achieving a balance between system cost and protection performance.

[0026] Step 102: Identify short-circuit fault characteristics based on electrical operating parameters and assess the current mechanical state of the winding based on mechanical response parameters.

[0027] In this step, the short-circuit fault characteristics refer not only to the amplitude of the short-circuit current, but also to the type of short-circuit fault, such as three-phase short circuit, single-phase ground fault, etc., and the corresponding stress mode. The system quickly determines the nature of the fault by analyzing the instantaneous values, amplitudes, and phase relationships of the three-phase currents, and by calculating the zero-sequence, positive-sequence, and negative-sequence components using the symmetrical component method. The specific identification logic will be described in detail in the subsequent embodiment 2.

[0028] Mechanical state refers to the set of physical parameters describing the mechanical characteristics of the winding at the current moment. In traditional passive defense schemes, the winding is usually considered as a linear spring with constant stiffness. However, actual oil-paper insulation systems exhibit significant nonlinearity, hysteresis, and aging characteristics over time. In this embodiment, evaluating the mechanical state refers to identifying the nonlinear stiffness, damping coefficient, and other dynamic parameters of the winding in real time. This process is achieved by establishing a dynamic model that includes nonlinear terms and using real-time collected pressure and displacement data for parameter estimation. The specific identification algorithm will be described in detail in subsequent embodiment 3.

[0029] Step 103: Based on the characteristics of the short-circuit fault and the mechanical state, calculate the target preload command for each control zone using a preset control strategy.

[0030] The control strategy is not a simple table lookup or proportional amplification, but a multi-objective optimization solution process. Its input includes two parts of information: first, the external load distribution determined by the characteristics of short-circuit faults, that is, predicting which part will be under the greatest stress based on the fault type; second, the internal bearing capacity determined by the mechanical state, that is, judging how much deformation the winding can still withstand based on the current stiffness.

[0031] Based on the two inputs, the controller calculates the optimal preload that should be applied to each control zone at the current moment, i.e., the target preload command. The calculation principle is to absorb short-circuit impact energy and reduce the vibration transmitted to the core and tank by utilizing the viscoelastic damping of the insulating pads as much as possible, while ensuring the axial stability of the winding, i.e., preventing instability and collapse. In some implementations, this is achieved by solving a sequential quadratic programming (SQP) problem with the goal of maximizing energy dissipation. The specific optimization model will be described in detail in the subsequent embodiment 4.

[0032] Step 104: Drive the actuators of each control zone to operate according to the target preload command, and apply dynamic preload to the winding.

[0033] This step is executed by the hydraulic drive system. The actuators specifically include hydraulic cylinders corresponding to each control zone. The controller converts the calculated target preload command in digital form into an analog voltage signal or a pulse width modulation (PWM) signal, which is then sent to the electro-hydraulic servo valve. The electro-hydraulic servo valve precisely adjusts the flow and pressure of the hydraulic oil entering the hydraulic cylinder according to the command, pushing the piston rod of the hydraulic cylinder to extend and retract, thus changing the preload applied to the end or middle of the winding.

[0034] To meet the millisecond-level response speed requirement of short-circuit impact, peak force is typically required to be reached within 10ms-20ms. Therefore, the drive system in this embodiment preferably adopts an accumulator + electro-hydraulic servo valve architecture. Under normal operating conditions, the hydraulic pump pre-charges the accumulator, storing high-pressure hydraulic fluid. When a short circuit occurs, the accumulator acts as an instantaneous burst energy source, releasing the stored energy to ensure the actuator can provide sufficient instantaneous power. The specific hydraulic control circuit and logic will be described in detail in subsequent embodiment 5.

[0035] This embodiment constructs an adaptive defense system that can be adjusted in real time according to short-circuit conditions and winding status through four stages: partition perception, state assessment, optimized decision-making, and rapid execution, effectively improving the transformer's short-circuit resistance capability.

[0036] Example 2 describes in detail how to identify specific short-circuit fault types by analyzing electrical operating parameters, and accordingly match the corresponding stress distribution pattern from a pre-set library. This solves the problem that traditional solutions only rely on current amplitude for general control, ignoring the huge differences in spatial stress distribution caused by different types of short-circuit faults to the windings.

[0037] Step 201: Identify short-circuit fault characteristics based on electrical operating parameters.

[0038] This step utilizes real-time acquired three-phase current data to analyze the electrical characteristics of the fault using the symmetrical component method. The specific electrical operating parameters are the instantaneous values ​​of the three-phase current, i. A(t), i B (t), i C (t). In a digital signal processor, the fundamental amplitude I of the three-phase current is first extracted using a full-wave Fourier algorithm. A I B I C And the phase angle.

[0039] Step 202: Calculate the amplitude and sequence component of the three-phase current based on the three-phase current in the electrical operating parameters, and identify the short circuit type based on the amplitude and sequence component of the three-phase current.

[0040] The system first calculates the zero-sequence component I0 and the positive-sequence component I of the three-phase current. + and negative order component I - The specific calculation formula is as follows:

[0041] I0=(1 / 3)*(I A +I B +I C );

[0042] I + =(1 / 3)*(I A +α'*I B +α' 2 *I C );

[0043] I - =(1 / 3)*(I A +α' 2 *I B +α'*I C );

[0044] Among them, I A I B I C Here, α' is the three-phase current phasor, and α' is the rotation factor, with a value of e. (j*120°) , where j is the imaginary unit.

[0045] Based on the calculated sequence components and amplitude characteristics, the fault type is determined according to the following logic:

[0046] When the zero-sequence component is detected to exceed the preset zero-sequence start-up threshold and the three-phase current amplitude is asymmetrical, it is determined to be a single-phase ground fault.

[0047] Specifically, the system compares the zero-order component I0 with a preset zero-order start threshold I. 0_set If |I0|>I 0_set Furthermore, the amplitude of one phase of the three-phase current is significantly greater than that of the other two phases, for example, I. A >1.2*I B And I A>1.2*I C If so, it is determined to be a single-phase ground fault. This situation is commonly seen in transformer winding insulation breakdown to ground.

[0048] When the zero-sequence component is below the zero-sequence start-up threshold, the negative-sequence component exceeds the preset negative-sequence start-up threshold, and the amplitude of the two-phase current increases, it is determined to be a two-phase short circuit.

[0049] Specifically, if |I0| 0_set , but |I - |>I neg_set (Negative sequence start threshold) If the current amplitude of two phases, such as phases B and C, is detected to exceed the multiple threshold of the rated current, while the current of the other phase (phase A) is relatively small, then it is determined to be a two-phase short circuit. This situation is common in phase-to-phase short circuits.

[0050] When both the zero-sequence component and the negative-sequence component are below the corresponding preset threshold and the amplitudes of the three-phase currents increase synchronously, it is determined to be a three-phase short circuit.

[0051] Specifically, if |I0| 0_set And|I - | neg_set However, the three-phase current amplitude I A I B I C If all values ​​exceed the short-circuit protection setting, it is determined to be a three-phase symmetrical short circuit. This is the most severe operating condition that tests the mechanical strength of the transformer.

[0052] In practical applications, the zero-order initiation threshold I 0_set The threshold value is typically set to 0.05 to 0.15 times the rated current, with a negative sequence starting threshold I. neg_set It is generally set to 0.1 to 0.2 times the rated current. The specific value is determined based on the rated parameters of the transformer and the background unbalance of the power grid.

[0053] Based on the above judgment logic, at a sampling frequency of 10kHz, the system can output a reliable fault type judgment result after completing a full fundamental frequency cycle (20ms) of data acquisition. Considering that the full-wave Fourier algorithm requires at least one full cycle of data, the fault type identification delay is approximately 20ms to 25ms.

[0054] Step 203: Match the corresponding stress distribution pattern from the preset pattern library according to the short circuit type, and use the stress distribution pattern as the short circuit fault feature.

[0055] Once the short-circuit type is determined, the system immediately calls the stress distribution pattern library pre-stored in the controller's memory. The pattern library is a lookup table indexed by the fault type and with the normalized stress weight vector as the value.

[0056] ​​​Specifically, the stress distribution pattern can be represented by the vector W. stress This indicates that the dimension of the vector is the same as the number N of the control partitions in Example 1.

[0057] W stress =[w1,w2,...,w N ];

[0058] Among them, w i This represents the relative stress weight of the i-th control zone under a predetermined fault type, and typically satisfies the normalization condition max(w i )=1 or ∑(w i =1.

[0059] For example, if a three-phase short circuit is determined, the system retrieves the corresponding vector W from the database. 3ph For a typical two-winding transformer, during a three-phase short circuit, the axial force mainly manifests as an inward compressive force, with a larger force in the middle. At this time, W... 3ph It may exhibit a distribution that is small at both ends and large in the middle, such as [0.4, 0.6, 0.8, 1.0, 1.0, 0.8, 0.6, 0.4].

[0060] If the fault is determined to be a single-phase ground fault, assuming it is phase A, due to the asymmetrical stress on the windings, the corresponding end region of the fault may experience a large local electromagnetic force. In this case, the retrieved vector W... 1ph It may exhibit a distribution with maxima at the ends and smaller values ​​at the rest, such as [1.0, 0.3, 0.2, ..., 0.2, 1.0].

[0061] Through the above matching mechanism, the system can transform abstract fault types into specific spatial force maps. Vector W stress This will be used as a weighting coefficient and directly input into the energy optimization algorithm in subsequent embodiments to ensure that the strongest pre-tightening control is applied in the area with the most severe stress, while maintaining appropriate relaxation in the area with less stress to maintain the damping effect.

[0062] In some optional implementations, the stress distribution pattern library can further include a current amplitude range dimension. That is, for the same three-phase short circuit, the axial force distribution may slightly change due to differences in the degree of magnetic circuit saturation at different short-circuit current multiples, such as 5 times the rated current and 10 times the rated current. In this case, the lookup table will be upgraded to a two-dimensional interpolation table, with the fault type and current amplitude as inputs and the stress weight vector as output, further improving the matching accuracy.

[0063] Example 3: A detailed description of how to use a recursive filtering algorithm incorporating a nonlinear dynamic model to identify the mechanical state of a transformer winding in real time, such as... Figure 3As shown. This scheme breaks through the limitations of traditional methods that treat the winding as a linear spring, and can accurately capture the stiffness degradation and hysteresis characteristics of the insulating pad under aging, oil immersion and short-circuit impact.

[0064] Step 301: Establish a nonlinear hysteresis dynamic model describing the relationship between axial force and deformation of the winding, wherein the model parameters of the nonlinear hysteresis dynamic model include at least the equivalent stiffness coefficient.

[0065] The nonlinear hysteresis dynamics model adopts the Bouc-Wen model. The restoring force of the Bouc-Wen model is composed of the superposition of linear elastic force components, viscous damping force components, and hysteresis restoring force components controlled by the first-order differential equation.

[0066] The model parameters also include damping coefficients, hysteresis variables, and shape parameters that control the shape of the hysteresis loop.

[0067] In other words, the model parameters also include damping coefficients and shape parameters that control the shape of the hysteresis loop; the nonlinear hysteresis dynamics model also includes hysteresis displacement variables as internal state variables describing the hysteresis restoring force.

[0068] In this embodiment, to accurately describe the complex mechanical behavior of the winding pad, a viscoelastic material, the Bouc-Wen model is selected to construct the dynamic equations. The restoring force of this model is composed of the superposition of three parts: linear elastic force, viscous damping force, and hysteretic restoring force.

[0069] The specific dynamic equations are expressed as follows:

[0070] F total (t)=m*a(t)+c*v(t)+F hys (t);

[0071] Among them, F total (t) represents the total external force acting on the winding unit, i.e., the sum of the electromagnetic force and the preload; m is the equivalent mass of the winding unit; a(t) is the acceleration; c is the viscous damping coefficient; v(t) is the velocity; F hys (t) represents the nonlinear hysteresis restoring force.

[0072] Furthermore, according to the Bouc-Wen theory, the hysteresis restoring force F hys (t) is specifically expanded as follows:

[0073] F hys (t)=α*k*x(t)+(1-α)*k*z(t);

[0074] Where x(t) is the relative displacement, k is the initial elastic stiffness, i.e. the equivalent stiffness coefficient, α is the ratio of the stiffness after yielding to the stiffness before yielding, with a value ranging from 0 to 1, and z(t) is the hysteresis displacement variable.

[0075] The evolution of the hysteresis displacement variable z(t) follows the following first-order nonlinear differential equation:

[0076] dz(t) / dt=v(t)-β*|v(t)|*|z(t)| (n-1) *z(t)-γ*v(t)*|z(t)| n ;

[0077] Where dz(t) / dt represents the derivative of z(t) with respect to time, β and γ are shape parameters that control the shape of the hysteresis loop, and n is an exponential parameter that controls the smoothness.

[0078] The above system of equations establishes a highly nonlinear mapping relationship between input (force) and output (displacement, velocity). The model parameter set θ specifically includes:

[0079] θ=[k,c,α,β,γ];

[0080] The parameters mentioned above directly reflect the current tightness and aging condition of the winding. For example, a significant decrease in stiffness k usually indicates that the insulation pads have undergone permanent deformation or loosening.

[0081] Step 302: The electromagnetic force correction coefficient is calibrated in advance through offline simulation. The electromagnetic force correction coefficient is used to correct the electromagnetic force calculation deviation caused by magnetic circuit saturation effect and end leakage magnetic effect.

[0082] In the process of parameter identification of the nonlinear hysteresis dynamic model using the recursive filtering algorithm, the electromagnetic force acting on the winding is calculated in real time using electrical operating parameters and electromagnetic force correction coefficients, and the electromagnetic force is used as the input excitation of the nonlinear hysteresis dynamic model.

[0083] Before online identification, the excitation force acting on the winding, i.e., the electromagnetic force, must be accurately determined. Due to the enormous short-circuit current, the iron core often enters a deep saturation region, rendering the simple law of proportionality to the square of the current inapplicable. Therefore, a correction factor is introduced in this step.

[0084] Specifically, electromagnetic force F em The real-time calculation formula is:

[0085] F em (t)=K sat (I)*K geo *i(t) 2 ;

[0086] Where i(t) is the instantaneous value of the real-time current, K geo K is a theoretical coefficient determined by the winding geometry. sat (I) is a nonlinear correction coefficient that varies with the current amplitude I.

[0087] Correction factor K sat (I) Obtained through offline finite element simulation. For example, the actual electromagnetic force at rated current, 5 times rated current, ..., 20 times rated current is calculated in the simulation software, and compared with the theoretical value to fit K. sat (I) is a polynomial function or lookup table.

[0088] Correction factor K sat (I) is a positive function that decreases monotonically with the current amplitude I, reflecting the physical law that the efficiency of electromagnetic force generated per unit current decreases after the magnetic circuit of the iron core enters the saturation region as the current multiplier increases. K sat (I) The specific functional form was determined through offline finite element simulation calibration: the actual electromagnetic force at different current multiples was calculated in the simulation software and compared with the theoretical value to fit K. sat The function relationship (I) is provided for real-time calls by the online controller.

[0089] Step 303: Using a recursive filtering algorithm, the mechanical response parameters are taken as observed values, and the winding electromagnetic force determined based on the electrical operating parameters is taken as the input excitation for the nonlinear hysteresis dynamic model. Parameter identification is performed on the nonlinear hysteresis dynamic model, and the model parameters are updated in real time. Figure 2 As shown.

[0090] In this step, to address the nonlinear parameter estimation problem, the Unscented Kalman Filter (UKF) algorithm is preferred. This algorithm approximates the nonlinear distribution through deterministic sampling (Sigma points), and has higher accuracy and stability than the Extended Kalman Filter (EKF).

[0091] Construct an augmented state vector that includes the motion state variables of the nonlinear hysteresis dynamic system and the model parameters to be identified.

[0092] Define augmented state vector X A For: X A =[x,v,z,k,c] T ;

[0093] Here, the parameters k and c to be identified are expanded into state variables, and their dynamic rate of change is assumed to be zero, or to have small random walk noise. The remaining model parameters (α, β, γ, n) are determined through offline material testing before the system is put into operation. Specifically, an insulating pad sample of the same specifications as that used in the transformer winding is taken, and cyclic compression loads of different amplitudes and frequencies are applied on a hydraulic servo testing machine. The force-displacement hysteresis loop data are recorded. The results are obtained using a nonlinear least squares fitting method. These material parameters are considered known constants during online operation and only need to be recalibrated when the winding undergoes major overhaul and the pads are replaced.

[0094] The augmented state vector is sampled using the unscented Kalman filter algorithm to obtain the Sigma point set.

[0095] Based on the current estimated mean X hat Given the covariance matrix P, calculate 2L+1 Sigma points (where L is the vector dimension):

[0096] χ0=X hat ;

[0097] χ i =X hat +(sqrt((L+λ)*P)) i where i = 1, 2, ..., L;

[0098] χ i =X hat -(sqrt((L+λ)*P)) (i-L) Where i = L+1, L+2, ..., 2L;

[0099] Where λ is the scale parameter, and the subscript i represents the i-th column of the matrix.

[0100] Time updates are performed based on the Sigma point set and the nonlinear hysteresis dynamics model to obtain the state prediction value and the prediction covariance matrix.

[0101] Substitute each Sigma point into the system of differential equations (discrete form) from step 301 for one-step prediction to obtain the predicted point set, and then sum them by weight to obtain the prior estimate X. - and prior covariance P - .

[0102] In this process, the corrected electromagnetic force F calculated in step 302 em u(t) is used as the control input u(t) of the system in the calculation.

[0103] The predicted state values ​​are updated by measuring the mechanical response parameters collected in real time, and the posterior estimate of the augmented state vector is output. The updated model parameters are then extracted from this estimate.

[0104] The system's observation equation is: Y(t) = h(X) A (t))+V(t);

[0105] Where h(·) is the nonlinear observation function, Y(t) is the observation vector, and V(t) is the observation noise. The observation vector Y(t) contains the displacement x measured by the sensor. meas and pressure F measSince the measured pressure includes a product of stiffness k and displacement x, the observation equation is a nonlinear function. The unscented Kalman filter algorithm can directly process this nonlinear observation without the need for linearization approximation.

[0106] By calculating the Kalman gain K, the prior estimate is corrected using the measured value Y(t):

[0107] X hat_new =X - +K*(Y(t)-Y pred );

[0108] Among them, Y pred To predict the observations, the updated vector X hat_new The 4th and 5th elements are the stiffness k and damping c identified at the current moment.

[0109] In practical implementation, the settings of the process noise covariance matrix Q and the observation noise covariance matrix R of the UKF algorithm have a significant impact on the identification accuracy. In the diagonal elements of Q, the elements corresponding to motion state variables (displacement, velocity, hysteresis variables) can be set to 10. -8 Up to 10 -4 The magnitude reflects the modeling error of the system dynamics; the elements corresponding to the model parameters (stiffness, damping) can be taken as 10. 6 Up to 10 10 The magnitude reflects the expected drift rate of the parameter—a larger value indicates a faster allowable parameter change and a higher identification and tracking speed, but it also increases the estimation fluctuation. R is set according to the sensor's accuracy; when the displacement sensor's accuracy is on the order of micrometers, the corresponding element can be 10. -10 Up to 10 -8 .

[0110] Besides the unscented Kalman filter algorithm mentioned above, recursive filtering algorithms can also include the Extended Kalman Filter (EKF), the Volumetric Kalman Filter (CKF), or the Particle Filter. Among these, the EKF uses a first-order Taylor expansion for linearization, resulting in lower computational cost but slightly lower accuracy for strongly nonlinear systems; the Particle Filter is not limited by the Gaussian distribution assumption, making it suitable for highly nonlinear scenarios, but it requires more computation. Those skilled in the art can select an appropriate filtering algorithm based on the controller's computing power and required identification accuracy.

[0111] Step 304: Determine the updated model parameters as the current mechanical state of the winding.

[0112] The system compares the newly identified k value with the preset reference stiffness k0. For example, the stiffness degradation rate D is defined. k =(k0-k) / k0. If D kIf the value exceeds a certain threshold, such as 20%, it indicates that the winding in that area is severely loose, requiring a larger preload to compensate in subsequent control strategies; conversely, if the k value is normal, the normal preload can be maintained. This parameter will be used as a core state variable and fed to the subsequent energy optimization controller.

[0113] According to one aspect of this application, in some embodiments, the differential equation with hysteresis displacement variables can also be:

[0114] The evolution of the hysteresis displacement variable z(t) follows the following first-order nonlinear differential equation:

[0115] dz(t) / dt=A'*v(t)-β*|v(t)|*|z(t)| (n-1) *z(t)-γ*v(t)*|z(t)| n ;

[0116] Where dz(t) / dt represents the derivative of z(t) with respect to time, v(t) is the velocity, A' is the initial slope control parameter of the hysteresis loop, β and γ are shape parameters that control the shape and width of the hysteresis loop, and n is an exponential parameter that controls the smoothness of the loading-unloading transition.

[0117] To ensure the thermodynamic consistency of the Bouc-Wen model, i.e., to ensure that the energy dissipation of the hysteresis loop is non-negative under any loading path, the above shape parameters must satisfy the following constraints:

[0118] β+γ>0, and β-γ>=0;

[0119] In the subsequent online identification process, this constraint is executed as a necessary boundary condition for the feasible region of parameters.

[0120] The model parameter set θ specifically includes:

[0121] θ=[k,c,α,A',β,γ,n];

[0122] Where k is the initial elastic stiffness, i.e., the equivalent stiffness coefficient, c is the viscous damping coefficient, α is the ratio of the stiffness after yielding to the stiffness before yielding, and the meanings of A', β, γ, and n are as above. n is an exponential parameter that controls the smoothness of the loading-unloading transition.

[0123] To further illustrate the above identification process, a specific numerical example will be used for demonstration. Assume that the design reference stiffness k0 of a certain control zone of a 220kV transformer winding is 5 × 10⁻⁶. 8 N / m, equivalent mass m is 200kg, initial damping coefficient c0 is 1×10 5 N·s / m. Five years after commissioning, the insulation pads in this section underwent slight shrinkage due to aging.

[0124] During a short-circuit impact, the peak displacement value collected by the sensor in this zone was 1.2 mm, and the peak pressure value was 180 kN. Inputting this data into the UKF algorithm, the diagonal elements of the process noise covariance matrix Q were set as follows:

[0125] [1×10 -8 1×10 -6 1×10 -4 1×10 10 1×10 6 The diagonal elements of the observation noise covariance matrix R are:

[0126] [1×10 -10 1×10 2 After approximately 50 sampling steps, with a sampling period of 0.1ms and a total sampling time of 5ms, the filter converges and outputs the identification result: k = 3.8 × 10⁻⁶. 8 N / m, c=0.85×10 5 N·s / m.

[0127] Calculate the stiffness degradation rate D k =(5×10 8 -3.8×10 8 ) / 5×10 8 =24%, exceeding the 20% warning threshold, indicating significant loosening of the winding in this section. Subsequent control strategies require increasing the preload compensation for this section. This result is consistent with the pad shrinkage phenomenon observed during actual maintenance, verifying the effectiveness of the identification algorithm.

[0128] According to one aspect of this application, in the measurement update step of the unscented Kalman filter described above, due to physical constraints on the Bouc-Wen model parameters, such as the aforementioned thermodynamic consistency condition and the positive value requirements for stiffness and damping, the identification results may exceed the physically permissible range due to noise interference. Therefore, after each measurement update, projection constraint processing is performed on the parameter components in the augmented state vector. Specifically, for the parameter θ to be identified... i Updated value θ i_updated Limit it to the preset upper and lower bounds [θ] i_min ,θ i_max ]Inside:

[0129] θ i_constrained =max(θ i_min min(θ) i_max ,θ i_updated ));

[0130] Where, θ i_constrained For the constrained parameter values, θ i_updated θ is the parameter value directly output by the filter. i_minand θ i_max These are the physical lower and upper limits for the parameter, respectively. For example, the lower limit of stiffness k is set as a certain percentage of the design value, such as 0.3 times, and the upper limit is set as a certain percentage of the design value, such as 2.0 times.

[0131] In another embodiment of this application, the above-mentioned mechanical state identification step may further be:

[0132] According to one aspect of this application, based on the above-mentioned mechanical state identification, an adaptive mapping coefficient update step can be further performed to convert the identified real-time stiffness parameters into stress prediction mapping coefficients, so that the prediction relationship from short-circuit current to winding stress can be automatically updated as the winding state changes. In conventional schemes, the prediction from short-circuit current to winding stress uses fixed mapping coefficients, which remain unchanged after calibration during the design phase. However, winding stiffness changes over time due to temperature variations, insulation aging, and accumulated short-circuit impacts, leading to a gradual deterioration in the prediction accuracy of fixed coefficients. This embodiment utilizes the real-time stiffness parameters identified in Embodiment 3 to automatically update the mapping coefficients, ensuring that the prediction model always matches the actual winding state.

[0133] Step 3A01: ​​Establish the dynamic response model of the winding section.

[0134] The i-th control zone of the winding is simplified into a single-degree-of-freedom spring-mass-damped system, and its axial motion equation is:

[0135] m i *a i (t)+c i *v i (t)+k i *x i (t)=F em_i (t);

[0136] Where, m i For the equivalent quality of the i-th partition, a i (t) is the axial acceleration, c i v is the equivalent damping coefficient. i (t) is the axial velocity, k i For equivalent stiffness, x i (t) represents the axial displacement, F em_i (t) represents the short-circuit electromagnetic force acting on this partition.

[0137] Define the natural angular frequency and damping ratio of this partition:

[0138] ω n_i =sqrt(k i / m i );

[0139] Where, ωn_i Let k be the natural angular frequency of the i-th partition. i For equivalent stiffness, m i For equivalent quality.

[0140] ζ i =c i / (2*sqrt(k i *m i ));

[0141] Where, ζ i Let c be the damping ratio of the i-th partition. i This is the equivalent damping coefficient.

[0142] Step 3A02: Calculate the dynamic amplification factor.

[0143] The winding is not a static load-bearing structure; short-circuit impacts are dynamic excitations primarily at power frequency. When electromagnetic force excites the winding at an angular frequency ω, the displacement response amplitude of the winding exhibits an amplification effect compared to the displacement under static force. The dynamic amplification factor H is defined as follows. i (ω) is used to describe this effect:

[0144] H i (ω)=1 / sqrt((1-(ω / ω n_i ) 2 ) 2 +(2*ζ i *ω / ω n_i ) 2 );

[0145] Among them, H i (ω) is the dynamic amplification factor of the i-th partition at the excitation frequency ω, where ω is the excitation angular frequency, i.e., the angular frequency of the short-circuit current. n_i ζ is the natural angular frequency. i The damping ratio is denoted as .

[0146] The physical meaning of the above equation is that when the excitation frequency ω is much smaller than the natural frequency ω n_i At that time, H i When ω approaches 1, the winding exhibits a quasi-static response; when ω approaches ω0... n_i At that time, H i The stiffness k increases significantly, producing a resonance amplification effect. i The change will be achieved by altering ω n_i And change H i The value of leads to different stress responses of varying amplitudes under the same electromagnetic force excitation. This mechanism is the physical basis for the adaptive prediction update in this invention.

[0147] Step 3A03: Calculate the rate of change of stiffness and the stiffness correction factor.

[0148] Define the rate of change of stiffness γi The ratio of the current identified stiffness to the design reference stiffness:

[0149] γ i =k hat_i / k i_0 ;

[0150] Where, γ i Let k be the rate of change of stiffness in the i-th partition. hat_i For the current equivalent stiffness identified in Example 3, k i_0 This is the baseline stiffness value used in the design phase. γ i A value greater than 1 indicates increased stiffness, while a value less than 1 indicates decreased stiffness.

[0151] When the stiffness changes, the natural frequency becomes:

[0152] ω n_i_new =ω n_i_0 *sqrt(γ i );

[0153] Where, ω n_i_new ω is the current natural angular frequency. n_i_0 This is the natural angular frequency under design conditions.

[0154] Define the frequency ratio r0 under the design conditions as:

[0155] r0=ω0 / ω n_i_0 ;

[0156] Where ω0 is the power frequency angular frequency of the short-circuit current, ω n_i_0 This is the natural angular frequency under design conditions.

[0157] Based on this, the stiffness correction factor Γ is defined. i This is the ratio of the dynamic amplification factor under the current stiffness state to the dynamic amplification factor under the design state:

[0158] Γ i (γ i =sqrt((1-r0) 2 ) 2 +4*ζ i 2 *r0 2 ) / sqrt((1-r0 2 / γ i ) 2 +4*ζ i 2 *r0 2 / γ i );

[0159] Among them, Γ i γ is the stiffness correction factor for the i-th partition.i Let r0 be the rate of change of stiffness, r0 be the design frequency ratio, and ζ be the constant. i is the damping ratio.

[0160] When γ i =1, meaning when the stiffness remains unchanged, Γ i =1, prediction requires no correction. When γ i <1 means that when the stiffness decreases, if the system is closer to the resonance region, then Γ i A value greater than 1 indicates that the stress response is amplified, and the prediction coefficient needs to be increased accordingly.

[0161] Step 3A04: Update the adaptive mapping coefficients.

[0162] The predictive relationship from the peak short-circuit current to the peak winding stress is defined as follows:

[0163] σ pred_i =α i *I peak 2 ;

[0164] Where, σ pred_i Let α be the predicted peak stress value for the i-th partition. i For adaptive mapping coefficients, I peak This represents the predicted peak short-circuit current.

[0165] The adaptive mapping coefficient α i Decompose into the product of two factors:

[0166] α i =β i *Γ i ;

[0167] Where, β i This is a geometric factor, related only to the winding geometry and electromagnetic coupling relationship, and remains unchanged during operation. Its initial value was determined through offline finite element simulation calibration in Example 6; Γ i This is the stiffness correction factor, calculated in real time by step 3A03.

[0168] This decomposition method allows online updates to simply recalculate the stiffness correction factor Γ. i This calculation involves only algebraic operations and can be completed in microseconds, ensuring the real-time performance of the system.

[0169] Step 3A05: Perform equivalent linearization on the nonlinear stiffness.

[0170] In Example 3, the nonlinear hysteresis stiffness characteristics identified based on the Bouc-Wen model are used, while the derivation of the dynamic amplification factor in steps 3A02 to 3A04 is based on the linear stiffness assumption. Therefore, a method for converting the nonlinear identification results to equivalent linear stiffness needs to be established.

[0171] Define the equivalent linear stiffness k eq_i Secant stiffness of the hysteresis loop:

[0172] k eq_i =(F r_max_i -F r_min_i ) / (x max_i -x min_i );

[0173] Where, k eq_i Let F be the equivalent linear stiffness of the i-th partition. r_max_i and F r_min_i These represent the maximum and minimum resilience in a load-unload loop, x. max_i and x min_i These represent the corresponding maximum and minimum displacements.

[0174] For small-amplitude vibrations (normal operating conditions), the equivalent stiffness can be directly approximated by the parameters of the Bouc-Wen model:

[0175] k eq_i ≈k i *(α+(1-α)*A');

[0176] Where, k i Let α be the initial stiffness of the Bouc-Wen model, α be the stiffness ratio parameter, and A' be the initial slope parameter of the hysteresis loop.

[0177] For large-amplitude vibrations (short-circuit impact conditions), the hysteresis variable tends to saturate, and the equivalent stiffness is related to the displacement amplitude. In practical applications, a lookup table relationship between the equivalent stiffness and the displacement amplitude is established in advance, and the equivalent stiffness value is determined online based on the estimated impact displacement.

[0178] The above equivalent linear stiffness k eq_i Substitute the stiffness change rate γ from step 3A03 i Calculation can bridge the gap between nonlinear identification and linear predictive analysis.

[0179] Step 3A06: Perform adaptive update.

[0180] Adaptive update settings enable two collaborative update modes.

[0181] The first method is event-triggered updating. After each short-circuit impact occurs and is cleared by the protection device, the system uses the current, displacement, and pressure time history data recorded during the impact to fully identify the winding stiffness parameters using the UKF algorithm in Example 3, and obtains the updated equivalent stiffness k. hat_i Calculate the stiffness variation rate γ of each zone according to steps 3A03 to 3A05. iand stiffness correction factor Γ i Update the mapping coefficient α i The updated parameters are used for stress prediction in the next short-circuit event.

[0182] The second method is periodic verification and updating. During normal operation, the system uses small disturbance signals such as load fluctuations and inrush currents at fixed intervals to slowly track and identify the winding stiffness parameters, identify the stiffness drift trend caused by temperature changes and insulation aging, and slowly correct the stiffness reference value accordingly.

[0183] The two modes work together: periodic verification updates maintain the stiffness benchmark, while event-triggered updates perform fine corrections during short-circuit events.

[0184] Step 3A07: Input the updated results into the control strategy.

[0185] Updated adaptive mapping coefficients α i This is directly used in the stress prediction stage of the subsequent control strategy in Example 4. Specifically, in the energy optimization calculation of Example 4, the predicted stress σ of each partition... pred_i Determined by the following relationship:

[0186] σ pred_i =α i *I peak_pred 2 ;

[0187] Where, α i For the adaptive mapping coefficients updated in step 3A04, I peak_pred This is the peak current predicted based on the rate of rise of the short-circuit current.

[0188] The predicted stress values ​​for each zone are used to construct the weighting coefficients for each zone in the energy optimization objective function of Example 4 and the maximum displacement estimate in the displacement safety constraint. This forms a complete data loop: the identification module outputs stiffness parameters, the adaptive update module calculates correction factors, the prediction module generates stress predictions, the optimization module calculates the optimal preload, the execution module completes the control, and the new response data is fed back to the identification module.

[0189] This closed-loop data system enables the system to learn itself. With each short-circuit impact, the system's understanding of the winding state becomes more accurate, and the precision of prediction and control also improves.

[0190] Example 4 details the working mechanism of transformer decision-making, namely, how to calculate the optimal preload command based on fault characteristics and mechanical state. This scheme proposes to maximize system energy dissipation by utilizing the damping characteristics of insulating pads to actively dissipate short-circuit energy and suppress dynamic instability of the windings.

[0191] Step 401: The preset control strategy includes an energy optimization strategy aimed at maximizing system energy dissipation.

[0192] Traditional preload strategies typically aim to maintain constant pressure or track maximum electromagnetic force. This rigid approach often leads to excessive actuator load or winding stress. The energy optimization strategy proposed in this embodiment utilizes the viscoelastic damping characteristics of the oil-paper insulation system. By dynamically adjusting the preload, it controls the winding's trajectory, maximizing the negative work done by the damping force within one cycle of a short-circuit impact. This converts more mechanical kinetic energy into heat energy, reducing the system's vibration amplitude.

[0193] Step 402: Determine the energy absorption terms of each control zone based on the mechanical state, construct a global objective function that includes the energy absorption terms of each control zone, and set the weight coefficients of each control zone in combination with the short-circuit fault characteristics.

[0194] In this step, the controller establishes a mathematical optimization model. To achieve coordinated control of the entire winding, the global objective function J is defined as the weighted sum of energy dissipation across all control zones.

[0195] The discretized mathematical expression of the objective function J is as follows:

[0196] minJ=-∑ i=1 N (w i *E diss_i );

[0197] Where N is the total number of control zones, ∑ represents the summation operation, and E diss_i For the i-th partition in the future prediction time domain T p The energy dissipated internally, w i This represents the weighting coefficient for that partition.

[0198] Among them, the energy dissipation E of a single partition diss_i This can be expressed as the integral of the product of damping force and velocity:

[0199] ;

[0200] In discrete control, it can be approximated as:

[0201] E diss_i ≈∑ k'=1 M (c i *v i (t+k') 2 *Δt);

[0202] Among them, c i It is the damping coefficient identified in Example 3, v iτ is the axial vibration velocity of the winding, τ is an integral variable (dummy variable) representing the instantaneous time in the prediction time domain; Δt is the sampling period, and k' is the summation index.

[0203] Weighting coefficient w i The setting directly depends on the characteristics of short-circuit faults. Specifically, w i The value is directly taken as the stress distribution mode vector W. stress The corresponding element in.

[0204] For example, if a single-phase ground fault occurs, W stress If the objective function is set to [1.0, 0.3, ..., 0.2], then the objective function will focus on the energy dissipation of the first partition (end), forcing the controller at that location to output more aggressive control actions, while applying weaker optimization weights or allowing larger state deviation tolerances to other partitions.

[0205] Step 403: Construct constraints describing the dynamic response of the winding based on the mechanical state; solve the optimal solution of the global objective function under the constraints to obtain the target preload command for each control zone.

[0206] Solving the optimization problem must be done within the physical limits. This requires constructing a series of constraint equations based on the identified stiffness k and damping c.

[0207] The target preload command is used as an optimization variable, and the upper limit of the axial displacement amplitude of the winding and the upper limit of the output force of the actuator are set as inequality constraints.

[0208] Displacement constraint: to prevent excessive compression of the windings from causing insulation damage or excessive elongation from causing collapse.

[0209] |x i (t+k')|<=x max ;

[0210] Where, x i (t+k') represents the axial displacement of the i-th control partition at the k'-th step in the prediction time domain, x max The maximum allowable displacement threshold, for example, 2 mm.

[0211] Output constraints: Limited by the capabilities of the hydraulic cylinder and electro-hydraulic servo valve.

[0212] 0<=F pre_i (t+k')<=F act_max ;

[0213] Among them, F pre_i The preload variable to be solved is F. act_max It is the maximum output of the implementing agency.

[0214] Preferably, the inequality constraint also includes the preload difference constraint between adjacent control zones.

[0215] When using the sequential quadratic programming algorithm to solve the problem, the preload difference constraint is added to the constraint set of the optimization problem to limit the stress gradient between adjacent control zones.

[0216] To prevent localized shear stress caused by excessive differences in preload between adjacent zones, which could lead to tearing of the insulation paper, a smooth constraint must be applied.

[0217] |F pre_i -F pre_i+1 |<=ΔF smooth ;

[0218] Where, ΔF smooth The maximum allowable force difference between adjacent partitions, for example, 50kN.

[0219] Step 404: Using a sequential quadratic programming algorithm, the nonlinear programming subproblem is approximated as a quadratic programming subproblem in each iteration step and solved until the convergence condition is met. The optimal preload value of each control zone is then output as the target preload command.

[0220] Since the aforementioned optimization problem involves nonlinear dynamic constraints (Bouc-Wen model), direct solutions are difficult and time-consuming. This embodiment preferably employs the Sequential Quadratic Programming (SQP) algorithm, an efficient mathematical tool for handling nonlinear constraint optimization.

[0221] The specific execution flow of the SQP algorithm is as follows:

[0222] At the current working point (preload F) l State X l (where l represents the number of iterations in the SQP optimization algorithm) In the vicinity of the area, the nonlinear objective function is approximated as a quadratic function by Taylor expansion, and the nonlinear constraints are approximated as linear constraints, thus constructing a quadratic programming QP subproblem.

[0223] Using an efficient QP solver, such as the effective set method or the interior point method, solve the subproblem to obtain the search direction d. l .

[0224] Along the search direction d l Perform a line search to determine the step size α. l Update the preload command:

[0225] F l+1 =F l +α l *d l ;

[0226] Check if the residuals of the Karush-Kuhn-Tucker (KKT) conditions are less than the preset tolerance ε. If satisfied, output F. l+1 The final target preload command is used; otherwise, return to sub-step 1 to proceed to the next iteration.

[0227] Based on the above calculations, the controller can output an optimal preload sequence F within each control cycle, for example, 1 ms, that maximizes the dissipation of impact energy while satisfying all safety constraints. opt =[F pre_1 ,...,F pre_N ].

[0228] As an alternative implementation, if the controller's computing resources are limited and it is difficult to complete the SQP solution within milliseconds, a simplified lookup table + feedforward strategy can be adopted. That is, the optimal preload curves for different fault types are pre-calculated offline and stored as templates. During online operation, the preload is calculated based on the identified fault type and the real-time current amplitude I. mag The reference command is obtained by directly looking up the table, and then a feedforward term proportional to the square of the current is added:

[0229] F cmd_i =Table LookUp_i (Type,I mag )+K ff *I mag 2 ;

[0230] Among them, F cmd_i For the target preload command of the i-th control zone, Table LookUp_i (.) represents an offline pre-stored lookup function, Type is the short-circuit fault type, and K ff This is the feedforward gain coefficient.

[0231] While this method offers slightly lower control precision than online SQP optimization, it boasts a faster response time and is suitable for cost-sensitive or computationally limited applications.

[0232] The convergence tolerance of the SQP algorithm is generally set to 10. -4 Up to 10 -6 The maximum number of iterations is set to 10 to 20. On the millisecond timescale of actual short-circuit impacts, with a control period of 1 ms, SQP typically converges within 3 to 5 iterations. If it fails to converge within the maximum number of iterations, the current best feasible solution is output, and optimization continues in the next control period.

[0233] In addition to the SQP-based online optimization strategy and the table-based simplified strategy mentioned above, the preset control strategy can also be implemented using other optimization or control methods, such as Model Predictive Control (MPC), dynamic programming-based optimal control, or fuzzy logic-based expert rule systems. These methods can all solve for parameters under objectives such as maximizing energy dissipation or minimizing winding displacement; therefore, this invention is not limited to a specific type of optimization algorithm.

[0234] According to one aspect of this application, the aforementioned damping dissipation is only a part of the energy dissipation. For oil-paper insulated winding systems, the insulating pad exhibits a significant hysteresis effect under cyclic loading. The area enclosed by the hysteresis loop represents the energy dissipated in each cycle through the micro-friction and viscoelastic effects within the material. This hysteresis dissipation energy has a non-monotonic relationship with the preload, the physical mechanism of which is as follows: the preload determines the initial compression state of the pad and the range of stress-strain cycles experienced by the pad during short-circuit impact. A moderate preload keeps the pad in a stable operating region with sufficient deformable space, and the hysteresis dissipation capability increases with increasing preload; however, when the preload is too large, the pad is pushed into a compacted region, the deformable space shrinks sharply, and the hysteresis dissipation capability decreases instead.

[0235] Based on the experimental characteristics of oil-paper insulation materials, the unit volume hysteresis dissipation energy of the pad block under initial strain ε0 and subjected to impact incremental strain Δε is defined as:

[0236] e hyst =η*E p *Δε 2 *(1+a1*ε0-a2*ε0 2 );

[0237] Among them, e hyst E represents the energy dissipated per unit volume due to hysteresis, where η is the hysteresis loss factor of the material. p ε is the compressive modulus of the pad, Δε is the incremental strain caused by short-circuit impact, ε0 is the initial strain under preload, and a1 and a2 are material constants, determined by cyclic loading tests.

[0238] Initial strain ε0 and preload F pre_i The relationship is:

[0239] ε0=F pre_i / (E p *A p_i );

[0240] Among them, A p_i Let be the pressure area of ​​the pad in the i-th partition.

[0241] Substitute the initial strain expression into the hysteresis dissipation formula and multiply by the total volume V of the pad. p_iThe total hysteresis dissipation energy E of the i-th partition is obtained. hyst_i An explicit function of preload:

[0242] E hyst_i (F pre_i )=η*V p_i *E p *Δε 2 *(1+a1*F pre_i / (E p *A p_i )-a2*F pre_i 2 / (E p *A p_i ) 2 );

[0243] Among them, V p_i Let represent the total volume of the pad block in the i-th partition; the meanings of the other symbols are the same as before.

[0244] This function is a quadratic function of the preload. Since the coefficient of the quadratic term a2 is negative, this function has a unique maximum point. The corresponding optimal preload is:

[0245] F pre_i_opt =a1*E p *A p_i / (2*a2);

[0246] Among them, F pre_i_opt The optimal preload value is determined to maximize the energy dissipation due to hysteresis.

[0247] Therefore, the energy absorption term E of each partition in the global objective function in step 402 diss_i It should be revised to be the sum of damped dissipation and hysteretic dissipation:

[0248] E diss_i_total =E damp_i +E hyst_i (F pre_i );

[0249] Among them, E diss_i_total This is the total energy dissipation of the i-th partition, E damp_i To dampen energy dissipation, E hyst_i (F pre_i () represents energy dissipated due to stagnation.

[0250] The total energy dissipation function with respect to the preload F pre_i It is no longer a monotonic function, but rather has an optimal value. This means that the optimal control strategy is neither the larger nor the smaller the preload, but rather to achieve the best balance of energy dissipation at a predetermined value. This non-monotonic characteristic is the key physical basis that distinguishes this invention from traditional constant force preload schemes.

[0251] Example 5 describes a multi-loop closed-loop control architecture and hydraulic actuator, detailing how macroscopic target preload commands are transformed into microscopic physical actions. Through a hierarchical closed-loop control architecture and high-power-density hydraulic hardware, millisecond-level accurate control of the transformer windings is ensured.

[0252] Step 501: Drive the actuators of each control zone to move according to the target preload command, specifically based on a multi-loop closed-loop control architecture.

[0253] To address the nonlinearity and external disturbances of the hydraulic system, this embodiment employs a nested three-loop control architecture. From the inside out, this architecture consists of an inner pressure control loop, a middle displacement control loop, and an outer state evaluation loop, with the response bandwidth and control objectives of each loop progressively increasing.

[0254] Step 502, pressure control inner loop, is used to adjust the opening of the electro-hydraulic servo valve according to the feedback of the pressure sensor to track the target preload command.

[0255] This is the fastest response control loop, with a bandwidth typically greater than 100Hz. The pressure control inner loop receives the target preload command F from the SQP solver. cmd As a set value, it receives the actual pressure F from a pressure sensor mounted on the hydraulic cylinder. meas .

[0256] The controller calculates the pressure deviation e p =F cmd -F meas The control voltage u of the electro-hydraulic servo valve is calculated using a PID control algorithm or a state feedback algorithm. v .

[0257] For example, the calculation formula when using a PI controller is:

[0258] u v (t)=K p *e p (t)+K i *∫(e p (τ))dτ;

[0259] Among them, K p and K i These are the proportional and integral gains, respectively. The electro-hydraulic servo valve is based on u... v Adjusting the valve core displacement changes the flow rate of oil into the hydraulic cylinder, quickly building up or unloading pressure.

[0260] The PI parameters of the pressure control inner loop can be tuned using the frequency response characteristics of the hydraulic system. Taking a hydraulic cylinder with a diameter of 50mm and a system pressure of 21MPa as an example, the proportional gain K... pThe typical value range is 0.01 to 0.05 V / kN, and the integral gain K i The typical value range is 0.1 to 0.5 V / (kN·s). K p Increasing K can improve tracking speed but may cause pressure oscillations. i Increasing the value can eliminate steady-state error but may reduce system stability.

[0261] Step 503, the displacement control middle loop is used to monitor the compression displacement of the winding. When the compression displacement exceeds the safety threshold, the setting value of the pressure control inner loop is corrected.

[0262] The displacement control loop, acting as a safety protection layer, typically has a bandwidth between 10Hz and 50Hz. While the primary goal of the system is force tracking, excessive deformation of the windings must be prevented. This loop compares the feedback value x from the displacement sensor in real time. meas With the maximum allowable displacement x limit .

[0263] When x meas <x limit At this time, the middle ring does not intervene, and the output correction amount ΔF=0.

[0264] When x meas >=x limit When the pressure is reduced, the middle ring is activated, outputting a negative pressure correction value to forcibly reduce the set value of the inner pressure ring, thereby limiting further increases in displacement.

[0265] The correction logic can be expressed as:

[0266] F cmd_final =F cmd_original -max(0,K disp *(x meas -x limit ));

[0267] Among them, F cmd_final To ultimately issue the preload command to the inner pressure control loop, F cmd_original K is the original preload control command. disp This is the displacement penalty coefficient.

[0268] Step 504, outer loop of state assessment, is used to perform mechanical state assessment and trigger strategy calculation update based on the assessment results.

[0269] This is the outermost slow-speed loop (relative to the inner loop), and its operating frequency is consistent with the parameter identification frequency (e.g., 1kHz). This loop is responsible for continuously updating the mathematical model parameters (stiffness k, damping c) of the controlled object (winding). When a significant drift in the model parameters is detected, such as a short circuit causing plastic deformation of the insulating pad and a decrease in stiffness k, the outer loop will trigger the SQP algorithm in Example 4 to resolve the problem, generating a new target preload command sequence adapted to the new parameters, thus achieving adaptive control.

[0270] Under the above parameter configuration, the step response rise time of the pressure control inner loop is approximately 3ms to 5ms, the overshoot is less than 10%, and the steady-state error is less than 2% of the target value. The response time of the displacement control middle loop is approximately 20ms to 50ms. The time required for the entire system to reach 95% of the target value from receiving the target preload command is approximately 8ms to 15ms, meeting the millisecond-level response requirements for short-circuit impact protection.

[0271] Preferably, the above-mentioned multi-loop closed-loop control architecture also includes an anomaly detection and fault tolerance mechanism.

[0272] In case of sensor malfunction, the system performs online validity checks on the signals of each sensor, including range checks to determine whether the readings are within the physically reasonable range, and consistency checks to ensure that the displacement and pressure in the same zone meet the expected relationship of the current stiffness model. When a sensor is determined to have failed, the zone switches to open-loop safety mode, applies the preset maximum safety preload, and simultaneously sends an alarm to the upper-level monitoring system.

[0273] For UKF anomaly identification, when the identified model parameters deviate from the preset physical constraint boundaries, such as stiffness values ​​being lower than 30% or higher than 200% of the design values, the parameter projection constraint mechanism will automatically truncate the parameters to the boundary values ​​and increase the corresponding process noise covariance, prompting the filter to accelerate correction in subsequent steps. If parameter truncation occurs for multiple consecutive sampling periods, the system generates a model anomaly alarm.

[0274] In the event of an SQP solution anomaly, if the convergence condition is not met within the maximum number of iterations, the system uses the best feasible solution of the current iteration as the output and continues optimization from that solution as the starting point in the next control cycle. If convergence fails for several consecutive control cycles, the system switches to a simplified control strategy based on a lookup table.

[0275] Step 505, the actuator includes a hydraulic station, an accumulator, an electro-hydraulic servo valve and a hydraulic cylinder.

[0276] This step describes the hardware-level implementation in detail.

[0277] Hydraulic power unit: As a basic power source, it includes an electric motor, hydraulic pump and oil tank, used to generate a constant system pressure (e.g. 21 MPa).

[0278] Hydraulic cylinder: A single-acting plunger-type hydraulic cylinder is preferred, installed at the upper and lower pressure plates of the transformer winding or at the pads distributed in the middle of the winding. It has a compact structure and eliminates the problem of leakage on the piston rod side.

[0279] Electro-hydraulic servo valve: Select a high-frequency response (>200Hz) nozzle-baffle type or voice coil motor type electro-hydraulic servo valve to ensure that it can keep up with the frequency of short-circuit current changes (50Hz / 100Hz).

[0280] Step 506: Control the electro-hydraulic servo valve to open, and use the accumulator to release the pre-stored hydraulic energy to provide an instantaneous high-pressure oil source.

[0281] The flow output of a conventional hydraulic pump is limited, making it difficult to meet the flow requirements of multiple hydraulic cylinders simultaneously completing pressurization within 10ms. In this embodiment, a bladder-type or piston-type accumulator is connected in parallel to the hydraulic circuit of each control zone.

[0282] The working sequence is as follows:

[0283] Phase 1 (Charging): During normal operation of the transformer, the hydraulic pump works to charge the accumulator with oil until the set pressure (e.g., 20MPa) is reached. At this time, the accumulator is in standby mode and has stored a large amount of hydraulic potential energy.

[0284] Phase 2 (Burst): When a short-circuit fault is detected and a preload command is issued, the electro-hydraulic servo valve opens fully instantaneously. The compressed gas (nitrogen) in the accumulator expands rapidly, forcing the stored high-pressure oil into the hydraulic cylinder within milliseconds, providing an instantaneous flow rate several times the rated flow rate of the hydraulic pump.

[0285] Phase 3 (Recovery): After the short circuit is cleared, the electro-hydraulic servo valve closes, the hydraulic pump restarts, replenishes the accumulator with energy, and prepares to cope with the next impact.

[0286] Step 507: Adjust the opening of the electro-hydraulic servo valve according to the target preload command, control the flow rate to the hydraulic cylinders corresponding to each control zone, and drive the hydraulic cylinders to apply preload to the winding.

[0287] With sufficient oil supply provided by the accumulator, the electro-hydraulic servo valve operates according to the control signal u calculated in step 502. v Finely adjust the valve opening.

[0288] The flow equation is: Q load =C d *w valve *x valve *sqrt((P s -PL ) / ρ);

[0289] Among them, Q load It is the load flow, C d It is the flow coefficient, w valve It is the valve orifice gradient, x valve It is the valve core displacement, P s It is the oil supply pressure, maintained by the accumulator, P L ρ is the load pressure, i.e., the hydraulic cylinder pressure, and ρ is the oil density.

[0290] By continuously adjusting x valve The system can precisely control the flow rate Q entering the hydraulic cylinder. load This controls the speed of the piston rod, v=Q. load / A p A p The piston area is used to achieve dynamic closed-loop control of the winding preload.

[0291] According to one aspect of this application, when using only the aforementioned feedback controller, the system suffers from an inherent tracking delay when facing step-type preload commands, making it difficult to meet millisecond-level response requirements. Therefore, this embodiment superimposes a feedforward control component onto the feedback loop, forming a feedforward-feedback composite control strategy.

[0292] The feedforward controller directly calculates the estimated control voltage based on the target preload command, without waiting for the deviation signal to establish, providing near-zero-delay coarse-adjustment capability. The feedforward control output u... ff The calculation formula is:

[0293] u ff =K ff *F cmd +u bias ;

[0294] Among them, u ff For feedforward control of the output voltage, K ff F is the feedforward gain coefficient. cmd For the target preload command, u bias A constant term to compensate for bias factors such as static friction and gravity.

[0295] Feedforward gain K ff It is determined based on the steady-state characteristics of the hydraulic system, that is, the reciprocal of the proportional relationship between the control voltage and the output force when the system is in steady state.

[0296] The total control output is the sum of the feedforward component and the feedback component:

[0297] u total =u ff +u v ;

[0298] Among them, u total The final control voltage output to the electro-hydraulic servo valve, u ff For the feedforward component, u v This refers to the PI feedback component mentioned earlier.

[0299] In this composite control architecture, the feedforward component drives the electro-hydraulic servo valve to actuate instantly upon command issuance, rapidly bringing the hydraulic cylinder pressure close to the target value. The feedback component then eliminates residual deviations caused by system parameter uncertainties and external disturbances, ensuring steady-state accuracy. The synergy of these two components enables the system to possess both rapid response capability and high steady-state accuracy.

[0300] Furthermore, to prevent the PI controller's integral term from accumulating and causing output saturation when the preload deviation is large, a conditional integral anti-saturation strategy is adopted: when the controller output u total If the actuator saturation limit is not reached, or if the deviation direction is opposite to the saturation direction, the integrator accumulates normally; otherwise, the integrator pauses accumulation. This measure effectively avoids overshoot and oscillation problems caused by integral saturation.

[0301] Example 6 provides a data source for the stress distribution pattern library and electromagnetic force correction coefficients, describing how to construct the system's prior knowledge base through high-precision multiphysics simulation before the transformer is put into operation. This process moves the expensive computational costs to the offline stage, ensuring the real-time performance of the online control system.

[0302] Before matching the corresponding stress distribution pattern from the preset pattern library according to the short circuit type, such as Figure 4 As shown, it also includes the following steps:

[0303] Step 601: Construct a finite element simulation model that includes the winding geometry and material properties.

[0304] In this step, a full-size three-dimensional model of the transformer is built using commercial finite element software, such as ANSYS Maxwell and COMSOL Multiphysics.

[0305] The winding geometry includes: the number of coils in the high and low voltage windings, the cross-sectional dimensions of the conductors, the thickness of the interlayer insulation, the distribution of the spacers, and the geometric contours of the core and tank.

[0306] Material properties include: the conductivity of copper wires, the magnetization curve (BH curve) of silicon steel sheets, the elastic modulus and Poisson's ratio of insulating paperboard, and the relative permittivity of transformer oil.

[0307] To improve computational efficiency, field-circuit coupling is often used, which involves connecting external circuits (power supply, load, short-circuit switch) to the ports of the finite element model to simulate the real electrical environment.

[0308] Step 602: Simulate three-phase short circuit, single-phase ground fault, and two-phase short circuit conditions in the finite element simulation model, and calculate the spatial distribution data of the winding axial electromagnetic force under each condition.

[0309] By setting the switching state of the external circuit, different types of short-circuit faults are triggered. The solver, based on Maxwell's equations, calculates the leakage magnetic field distribution B(x,y,z,t) inside the winding at the moment of short circuit.

[0310] Calculate the volume electromagnetic force density using the Lorentz force formula: f em =J×B;

[0311] Where J is the current density vector and B is the magnetic induction intensity vector.

[0312] By performing volume integration on the pancake units contained in each control zone, the total axial electromagnetic force F experienced by that zone at a certain moment, typically the moment of maximum electromagnetic force, is obtained. emi .

[0313] F emi =∫ V (f em_z )dV;

[0314] Where V is the volume domain of the i-th control partition, f em_z This represents the axial component of the electromagnetic force density.

[0315] Through multiple rounds of simulation, spatial distribution data matrices under different operating conditions can be obtained. For example, under the three-phase short-circuit condition, the vector F is obtained. 3ph_raw =[F1,F2,...,F N ], where F1 to F N These correspond to the force values ​​of each section from the top to the bottom of the winding.

[0316] Step 603: Normalize the spatial distribution data, extract the vectors representing the stress weights of each control zone, establish the mapping relationship between short-circuit types and vectors, and generate a stress distribution pattern library.

[0317] To make the data more universal and easier for optimization algorithms to use, the raw force data needs to be normalized.

[0318] The specific formula for normalization is: w i =F i / max(F1,F2,...,F N );

[0319] Alternatively, L2 norm normalization can be used: w i =F i / sqrt(∑(Fj 2 ));

[0320] After processing, the dimensionless weight vector W=[w1,w2,...,w2] is obtained. N ].

[0321] Establish a mapping table:

[0322] Key: Single-Phase Ground Fault -> Value: W_1ph, i.e.: Key: Single-phase ground fault -> Value: W 1ph ;

[0323] Key: Three-Phase Fault -> Value: W_3ph, i.e.: Key: Three-phase fault -> Value: W 3ph ;

[0324] Key: Two-Phase Fault -> Value: W_2ph, i.e.: Key: Two-phase fault -> Value: W 2ph .

[0325] This mapping table constitutes the stress distribution pattern library, which is then burned into the controller's non-volatile memory.

[0326] Preferably, this embodiment also includes a process for calibrating the electromagnetic force correction coefficient. In the simulation model, a certain short-circuit type is fixed, and the short-circuit current amplitude I is gradually increased, for example, from 1 times the rated current to 20 times, and the corresponding total electromagnetic force F is recorded. sim .

[0327] Calculation of theoretical electromagnetic force F theory =K geo *I 2 Among them, K geo This is a theoretical coefficient determined by the winding geometry.

[0328] Calculate the correction factor K sat (I)=F sim / F theory .

[0329] Discrete K sat (I) The data points are fitted as a polynomial function of I, which is then called by the online controller in real time.

[0330] Through the above offline construction process, this embodiment compresses the complex physical laws of electromagnetic fields into simple table lookup and algebraic operations, reducing the computational burden on the online controller and providing a fundamental guarantee for achieving millisecond-level response.

[0331] To more fully illustrate the working process of the method of the present invention, the following description uses a typical scenario of a single-phase ground fault occurring in a 220kV / 150MVA power transformer.

[0332] Scenario: The transformer has been in operation for 8 years. The windings are divided into 16 control zones along the axial direction, 4 at each end and 8 in the middle. At t=0, phase A on the low-voltage side of the transformer experiences insulation breakdown to ground.

[0333] The system response process is as follows:

[0334] From t=0 to t=2ms, during the detection phase, the current transformer, sampling at a frequency of 10kHz, captures a sudden increase in the A-phase current to eight times the rated current, while the B and C-phase currents show slight increases. The controller completes a full-wave Fourier transform within 2ms, extracting the fundamental amplitude and phase of the three-phase currents.

[0335] From t=2ms to t=3ms, during the identification phase, the controller calculates the sequence components and detects that the zero-sequence component is 2.5 times the rated current, far exceeding the zero-sequence start-up threshold (0.1 times the rated current). Furthermore, the amplitude of phase A is 3.2 times that of phases B and C, thus determining a single-phase ground fault. The corresponding stress weight vector is retrieved from the mode library.

[0336] W 1ph =[1.0,0.5,0.25,0.15,...,0.15,0.25,0.5,1.0], where the end partition has the highest weight.

[0337] During the evaluation phase from t=3ms to t=5ms, the UKF algorithm used displacement and pressure sensor data for each zone to identify parameters. The output results showed that the stiffness of the first end zone degraded to 78% of the design value (D). k =22%), triggering a loosening warning; the stiffness of the central section is basically normal (D k <5%).

[0338] From t=5ms to t=7ms, during the decision-making phase, the SQP optimizer determines the optimal preload command within three iterations based on stress weights and identified stiffness. The first end zone is assigned a target preload of 380kN, approximately 60% higher than normal operating conditions, to compensate for stiffness degradation, while the middle zone is assigned 180 to 220kN.

[0339] From t=7ms to t=15ms, during the execution phase, the electro-hydraulic servo valve fully opens, the accumulator releases high-pressure oil, and the hydraulic cylinder increases the end preload from the initial value of 230kN to 380kN within 8ms. The pressure inner loop PI controller continuously adjusts, and at t=15ms, the actual pressure reaches 98% of the target value. The displacement control middle loop synchronously monitors, and the displacement of each zone is within the 2mm safety threshold.

[0340] From t=15ms to short-circuit clearing (approximately 100ms), the system maintains the optimal preload configuration and continuously tracks current changes. The peak axial vibration displacement at the winding end is 1.1mm, which is about 39% lower than the 1.8mm in the simulated scenario using a constant uniform preload (300kN), effectively avoiding excessive compression deformation of the end insulation.

[0341] After the short circuit is cleared, the hydraulic pump recharges the accumulator, and the outer loop assessment results are stored in the historical database for reference in the next short circuit event.

[0342] Example 7 provides a physical device for implementing all the above method embodiments, and describes in detail the hardware structure, connection relationship and key component selection of the transformer anti-short circuit impact system based on adaptive adjustment of dynamic preload between winding turns, to ensure the feasibility of the technical solution in engineering.

[0343] Step 701, the sensing module includes a current transformer for collecting electrical operating parameters and displacement and pressure sensors arranged in each control zone of the winding, for monitoring the electrical operating parameters of the transformer winding and the mechanical response parameters of multiple control zones distributed along the axial direction.

[0344] In this embodiment, the sensing module is a distributed sensor network.

[0345] For electrical operating parameters, a high-precision Rogowski coil or Hall current sensor is used and installed at the high-voltage bushing lead of the transformer. Its frequency response range preferably covers DC to 10kHz to capture the high-frequency component of the short-circuit transient current.

[0346] For mechanical response parameters, the sensing module constructs a multi-point monitoring array inside the winding.

[0347] Specifically, fiber Bragg grating (FBG) strain sensors are embedded in the insulating pads of each control zone. Due to their inherent electromagnetic interference immunity and high voltage resistance, fiber optic sensors are suitable for operation in the high electric field environment inside the transformer tank. The fiber optic leads are led out through a specially designed sealing flange on the tank wall and connected to the fiber optic demodulator.

[0348] Meanwhile, piezoelectric ceramic force sensors or resistive strain gauge force sensors are installed at the upper and lower pressure plates of the winding to directly measure the total pressure applied to the ends. For pressure measurement in the middle of the winding, a thin-film pressure sensor is used, clamped between the pads of adjacent coils.

[0349] Step 702: The controller is configured to execute the method of any one of the present invention, calculating the target preload command for each control zone based on the collected parameters. This is used to identify short-circuit fault characteristics based on electrical operating parameters, evaluate the mechanical state of the windings based on mechanical response parameters, and calculate the target preload command for each control zone based on the principle of maximizing system energy dissipation.

[0350] The controller is the core computing unit of the system and is usually installed in a control cabinet next to the transformer.

[0351] From a hardware architecture perspective, the controller is preferably a dual-core heterogeneous architecture of FPGA+DSP or FPGA+ARM.

[0352] The Field Programmable Gate Array (FPGA) is responsible for high-speed parallel processing, including synchronous acquisition and filtering of data from multiple sensors, as well as pressure inner-loop PID control. Its parallel processing capability ensures microsecond-level inner-loop response.

[0353] Digital signal processors (DSPs) or ARM processors are responsible for complex algorithm calculations, including fault type identification (Fast Fourier Transform, FFT, and order component calculation), UKF parameter identification, and SQP optimization. These algorithms are logically complex and are suitable for running on general-purpose processors.

[0354] The controller communicates with the sensing module and drive unit via industrial fieldbuses, such as EtherCAT Ethernet control automation technology or CAN bus, to ensure the real-time and deterministic nature of data transmission.

[0355] Step 703, the hydraulic drive unit, including an accumulator group, a multi-channel electro-hydraulic servo valve and a hydraulic cylinder array, is used to respond to the target preload command and apply a zoned dynamic preload to the winding using the energy released by the accumulator group.

[0356] The hydraulic drive unit adopts a modular design. Each control zone corresponds to an independent hydraulic sub-module.

[0357] Each submodule contains:

[0358] A miniature single-acting piston hydraulic cylinder, with a cylinder diameter designed according to the required maximum pressure, such as 50mm, and a short stroke, such as 10mm, is made of high-strength non-magnetic stainless steel to reduce eddy current losses. The hydraulic cylinder is directly integrated into the winding's pressure stud structure or pad structure.

[0359] An electro-hydraulic servo valve is installed close to the hydraulic cylinder to reduce the pipeline cavity effect and improve dynamic response.

[0360] A diaphragm accumulator serves as an on-site energy station. Its capacity is calculated based on the fuel consumption during a short circuit, for example, 0.5L. The charging pressure is set to 80% of the system operating pressure.

[0361] The entire hydraulic drive unit is connected to an external hydraulic pump station via insulated oil pipes. Transformer oil or a compatible synthetic insulating oil is used as the hydraulic oil to prevent leakage and contamination of the transformer's main insulation system.

[0362] Through the above hardware integration, this embodiment constructs a complete mechatronics-hydraulic integrated intelligent equipment, upgrading the transformer from a traditional static device into an intelligent machine with active defense capabilities.

[0363] It should be noted that the above embodiments are described in detail using power transformers as a typical application. Those skilled in the art should understand that the adaptive preload method of the present invention is also applicable to other electrical equipment with similar winding structures and axial preload requirements, including but not limited to reactors, instrument transformers, and high-voltage motors. The number of control zones, sensor arrangement, and hydraulic cylinder specifications can be adjusted accordingly to accommodate the structural differences of different equipment, but the core sensing-identification-optimization-execution control framework remains unchanged.

[0364] In some alternative implementations, the actuator is not limited to hydraulic drive; it can also employ electric servo actuators, piezoelectric ceramic actuators, or magnetostrictive actuators. Electric servo actuators are suitable for applications requiring response times in the hundreds of milliseconds range and are relatively low-cost. Piezoelectric ceramic actuators offer response speeds in the microsecond range, making them suitable for higher frequency response requirements, but their stroke and output are limited. Those skilled in the art can select a suitable drive scheme based on the transformer's short-circuit protection level and cost budget.

[0365] This application constructs a nonlinear dynamic model describing the hysteresis characteristics of the winding and identifies the equivalent stiffness and damping coefficient of the winding online in real time through a recursive filtering algorithm. This enables the system to sense the shrinkage and loosening of the insulation pad over time and dynamically update the control parameters accordingly, achieving adaptive tracking of the mechanical state throughout the entire life cycle. This solves the problem that traditional fixed pre-tightening methods cannot adapt to insulation aging, leading to protection failure.

[0366] This application introduces an axial zone control and fault feature identification mechanism. It identifies fault types (such as single-phase grounding or three-phase short circuit) by analyzing the sequence component of the short-circuit current and matches corresponding spatial stress distribution patterns from a pre-set library. This allows the actuator to apply differentiated and accurate preload to the high-risk area at the winding ends and the smooth area in the middle, effectively solving the control blind spot in the spatial dimension. It also addresses the problem of uneven local stress caused by neglecting short-circuit types in existing solutions.

[0367] This application abandons simple constant force control and proposes an optimization strategy aimed at maximizing system energy dissipation. Combined with an accumulator hydraulic servo system possessing instantaneous burst capabilities, the preload is actively adjusted within milliseconds. Utilizing the viscoelasticity of the insulating material itself, the short-circuit impact kinetic energy is absorbed to the maximum extent, suppressing dynamic instability of the windings. This solves the problem of aggravated vibration caused by neglecting energy dissipation in rigid resistance.

[0368] It should be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

Claims

1. A method for transformer short-circuit impact resistance based on adaptive adjustment of dynamic preload between winding turns, characterized in that, include: Real-time acquisition of electrical operating parameters of transformer windings and mechanical response parameters of multiple control zones distributed along the axial direction; Identify short-circuit fault characteristics based on electrical operating parameters and assess the current mechanical state of the winding based on mechanical response parameters; Based on the characteristics of short-circuit faults and mechanical states, the target preload command for each control zone is calculated using a preset control strategy. According to the target preload command, the actuators of each control zone are driven to act, and dynamic preload is applied to the winding; The axial distribution density of the control zone is determined based on the leakage magnetic field distribution gradient of the winding, wherein the control zone density at the end region of the winding is greater than the control zone density in the middle region of the winding. Identifying short-circuit fault characteristics based on electrical operating parameters includes: Calculate the amplitude and sequence component of the three-phase current based on the three-phase current in the electrical operating parameters, and identify the short circuit type accordingly; According to the short circuit type, the corresponding stress distribution pattern is matched from the preset pattern library and used as the short circuit fault feature; The current mechanical state of the winding is assessed based on mechanical response parameters, including: Establish a nonlinear hysteresis dynamic model describing the relationship between axial force and deformation of the winding, and the model parameters shall include at least the equivalent stiffness coefficient; Using a recursive filtering algorithm, mechanical response parameters are taken as observed values, and the winding electromagnetic force determined based on electrical operating parameters is taken as the input excitation of the nonlinear hysteresis dynamic model. The parameters of the nonlinear hysteresis dynamic model are identified and the model parameters are updated in real time. The updated model parameters are determined as the current mechanical state of the winding; The nonlinear hysteresis dynamics model adopts the Bouc-Wen model, in which the restoring force is composed of the superposition of linear elastic force components, viscous damping force components, and hysteresis restoring force components controlled by the first-order differential equation. The model parameters also include damping coefficients, hysteresis variables, and shape parameters that control the shape of the hysteresis loop; The preset control strategies include energy optimization strategies aimed at maximizing system energy dissipation; Based on the characteristics of short-circuit faults and mechanical states, the target preload command for each control zone is calculated using a preset control strategy, including: The energy absorption term of each control zone is determined based on the mechanical state, a global objective function containing the energy absorption term is constructed, and the weight coefficient of each control zone is set in combination with the short-circuit fault characteristics. Construct constraints describing the dynamic response of the winding based on the mechanical state; Solve for the optimal solution of the global objective function under the constraints to obtain the target preload command for each control zone; The actuators of each control zone are driven to move according to the target preload command. This is based on a multi-loop closed-loop control architecture, which includes: The pressure control inner loop is used to adjust the opening of the electro-hydraulic servo valve based on feedback from the pressure sensor to track the target preload command; The displacement control middle loop is used to monitor the compression displacement of the winding. When the compression displacement exceeds the safety threshold, the set value of the pressure control inner loop is corrected. The outer loop of the state assessment is used to perform mechanical state assessment and trigger policy calculation updates based on the assessment results.

2. The method according to claim 1, characterized in that, The recursive filtering algorithm is used to identify parameters of a nonlinear hysteresis dynamics model, including: Construct an augmented state vector that includes the motion state variables of the nonlinear hysteresis dynamics system and the model parameters to be identified; The augmented state vector is sampled using the unscented Kalman filter algorithm to obtain the Sigma point set; Time updates are performed based on the Sigma point set and nonlinear hysteresis dynamics model to obtain state prediction values ​​and prediction covariance matrix; The predicted state values ​​are updated by measuring the mechanical response parameters collected in real time, and the posterior estimate of the augmented state vector is output. The updated model parameters are then extracted from this estimate.

3. The method according to claim 1, characterized in that, Solving for the optimal solution of the global objective function under the constraints includes: The target preload command is used as an optimization variable, and the upper limit of the axial displacement amplitude of the winding and the upper limit of the output force of the actuator are set as inequality constraints. Using a sequential quadratic programming algorithm, the nonlinear programming subproblem is approximated as a quadratic programming subproblem in each iteration step and solved until the convergence condition is met. The optimal preload value of each control zone is then output as the target preload command.

4. The method according to claim 1, characterized in that, The actuators include a hydraulic station, accumulator, electro-hydraulic servo valve, and hydraulic cylinder; they drive the actuators of each control zone to move according to the target preload command, including: The electro-hydraulic servo valve is opened by controlling the accumulator to release the pre-stored hydraulic energy and provide an instantaneous high-pressure oil source. Adjust the opening of the electro-hydraulic servo valve according to the target preload command, control the flow rate to the hydraulic cylinders corresponding to each control zone, and drive the hydraulic cylinders to apply preload to the winding.

5. The method according to claim 1, characterized in that, Before matching the corresponding stress distribution pattern from a preset pattern library based on the short-circuit type, the process also includes: Construct a finite element simulation model that includes the winding geometry and material properties; In the finite element simulation model, three-phase short circuit, single-phase ground fault and two-phase short circuit conditions are simulated respectively, and the spatial distribution data of the axial electromagnetic force of the winding under each condition are calculated. The spatial distribution data is normalized, and vectors representing the stress weights of each control zone are extracted. A mapping relationship between short-circuit types and vectors is established to generate a stress distribution pattern library.

6. A transformer short-circuit impact protection system based on adaptive adjustment of dynamic preload between winding turns, characterized in that, include: The sensing module includes a current transformer for collecting electrical operating parameters, as well as displacement sensors and pressure sensors arranged in each control zone of the winding. The controller is configured to perform the method as described in any one of claims 1 to 5, calculating the target preload command for each control zone based on the collected parameters; The hydraulic drive unit, including an accumulator group, a multi-channel electro-hydraulic servo valve, and a hydraulic cylinder array, is used to respond to a target preload command and apply a zoned, independent dynamic preload to the winding using the energy released by the accumulator group.

Images