Converter transformer winding state evaluation method and device

Abstract

This invention discloses a method and apparatus for assessing the condition of converter transformer windings based on electromagnetic vibration and stiffness identification. The method includes: acquiring comprehensive monitoring data of the valve-side winding of the converter transformer to be assessed, the data including: multi-band electrical excitation vectors extracted based on load current signals, multi-band mechanical response vectors extracted based on vibration signals, and environmental condition vectors; constructing an electromagnetic structure coupling mechanism model including the stiffness parameters to be identified, and obtaining a theoretical vibration vector; inputting the electrical excitation vector and the environmental condition vector into a trained neural network model to obtain a prediction residual vector of the mechanism model; constructing an objective function, minimizing the objective function, solving the objective function, and inverting to obtain the optimal value of the stiffness parameters to be identified; and determining the winding condition level based on the optimal value of the stiffness parameters to be identified, which significantly improves the accuracy and reliability of converter transformer winding condition assessment under complex operating conditions.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring and fault diagnosis technology, and in particular to a method and device for assessing the condition of converter transformer windings based on electromagnetic vibration and stiffness identification. Background Technology

[0002] As a core component of high-voltage direct current (HVDC) transmission systems, converter transformers undertake the crucial tasks of power transmission and voltage transformation between AC and DC systems. Unlike conventional power transformers, the valve-side windings of converter transformers are directly connected to the converter valves. During operation, they not only bear the fundamental voltage and current but also the load current excitation containing DC components and abundant high-order harmonic components. Under the influence of the winding leakage magnetic field, these complex harmonic currents generate axial electrodynamic forces with complex frequency components, namely Lorentz forces, which in turn induce continuous mechanical vibrations in the converter transformer windings and tank structure.

[0003] In the mechanical structure of a converter transformer, the winding is a composite structure formed by alternating layers of copper conductor discs and insulating pads, axially compressed. The insulating pads, as the main axial load-bearing components, exhibit significant nonlinear mechanical properties; that is, their equivalent stiffness coefficient is positively correlated with the axial compressive force. During long-term operation, the axial compressive force on the winding gradually decreases due to the combined effects of electromagnetic vibration, thermal expansion and contraction, and aging and creep of the insulating materials. This causes the insulating pads to spring back and soften, leading to a decrease in the overall equivalent stiffness of the winding. Once the winding becomes loose, its short-circuit withstand capability will be significantly reduced, making it highly susceptible to permanent deformation or even collapse under external short-circuit current impacts, seriously threatening the safe and stable operation of the power grid.

[0004] Currently, the monitoring and evaluation of converter transformer winding conditions mainly rely on two methods: offline detection and online monitoring. Offline detection methods, such as Frequency Response Analysis (FRA), while highly accurate, must be performed when the equipment is offline and powered off, making it unable to capture gradual faults during operation in real time. Furthermore, the long detection cycle makes it difficult to meet the requirement of immediacy. Existing online monitoring technologies primarily rely on the amplitude characteristics of vibration signals, judging the equipment status by monitoring whether the vibration amplitude on the tank surface exceeds a set threshold. However, this method has significant limitations: the vibration amplitude of a converter transformer essentially depends on the combined effect of the electrical excitation source (current magnitude) and mechanical structural characteristics (winding stiffness). Under heavy loads or severe harmonic fluctuations, even if the windings are in a healthy and tight state, the vibration amplitude will naturally increase with the increase of the excitation force, easily causing false alarms. Conversely, under light load conditions, even if the winding stiffness decreases and becomes loose, the vibration amplitude may not exceed the standard, leading to missed detections. Summary of the Invention

[0005] This invention provides a method and apparatus for evaluating the condition of converter transformer windings based on electromagnetic vibration and stiffness identification. This method addresses the shortcomings of existing technologies, which suffer from low accuracy, high false alarm and false alarm rates in converter transformer condition evaluation due to the failure to isolate the influence of load current and harmonic fluctuations on vibration signals during operation. This invention achieves accurate quantitative evaluation of the mechanical condition of the valve-side windings of converter transformers.

[0006] This invention provides a method for evaluating the condition of converter transformer windings based on electromagnetic vibration and stiffness identification, comprising: Acquire comprehensive monitoring data of the valve-side winding of the converter transformer to be evaluated. The data includes: multi-band electrical excitation vector extracted based on load current signal, multi-band mechanical response vector extracted based on vibration signal, and environmental condition vector. Based on the electrical excitation vector and the environmental condition vector, an electromagnetic structure coupling mechanism model containing the stiffness parameters to be identified is constructed to obtain the theoretical vibration vector. The electrical excitation vector and the environmental condition vector are input into the trained neural network model to obtain the prediction residual vector of the mechanism model. Based on the mechanical response vector and the theoretical vibration vector corrected by the predicted residual vector, an objective function is constructed. The objective function is minimized, and the optimal value of the stiffness parameter to be identified is obtained by inversion. Based on the optimal value of the stiffness parameter to be identified, the state level of the winding is determined.

[0007] According to the present invention, a method for evaluating the state of converter transformer windings based on electromagnetic vibration and stiffness identification is provided. The method involves obtaining a theoretical vibration vector based on the electrical excitation vector and the environmental operating condition vector, using an electromagnetic structure coupling mechanism model that includes the stiffness parameter to be identified. Specifically, this includes: Based on the electrical excitation vector, the electromagnetic excitation force at each frequency point is calculated by frequency domain discrete convolution operation; Based on the temperature parameters in the environmental condition vector, the stiffness parameters and damping coefficients to be identified are corrected for temperature. Based on the temperature-corrected stiffness parameters to be identified, the damping coefficient, the electromagnetic excitation force, and the equivalent mass of the winding, the frequency domain response transfer function of the mechanism model is constructed, and the theoretical vibration amplitude at each frequency point is calculated to form the theoretical vibration vector.

[0008] According to the present invention, a method for evaluating the winding condition of a converter transformer based on electromagnetic vibration and stiffness identification is provided. The step of performing temperature correction on the stiffness parameter and damping coefficient to be identified based on the temperature parameter in the environmental operating condition vector specifically includes: The corrected stiffness parameter to be identified is obtained using the following formula. :

[0009] The corrected damping coefficient is obtained using the following formula. :

[0010] Where T represents the temperature value; Indicates reference temperature The following stiffness parameters need to be identified; It represents the thermal softening coefficient of insulating materials, characterizing the rate at which stiffness decreases with increasing temperature; Indicates the average oil temperature; Indicates the foundation damping coefficient. This indicates the reference temperature corresponding to the base damping. This represents the viscosity-temperature constant of the oil.

[0011] According to the present invention, a method for evaluating the winding state of a converter transformer based on electromagnetic vibration and stiffness identification is provided, wherein the frequency domain response transfer function of the mechanism model is expressed by the following formula: ; in, This represents the theoretical vibration amplitude at the nth harmonic frequency; This represents the angular frequency of the nth harmonic. Represents the equivalent mass constant of the winding; electromagnetic excitation force ; Indicates the electromechanical conversion factor; The loop variable for summing the convolutions; and These are the complex spectra of the current components at the corresponding frequencies.

[0012] According to the present invention, a method for evaluating the state of a converter transformer winding based on electromagnetic vibration and stiffness identification is provided. The neural network model includes an input fusion layer, a feature extraction layer, an attention mechanism layer, and an output layer. Correspondingly, the step of inputting the electrical excitation vector and the environmental condition vector into the trained neural network model to obtain the prediction residual vector of the mechanism model specifically includes: The electrical excitation vector and the environmental condition vector are input into the input fusion layer to obtain normalized input features; The normalized features are input into the feature extraction layer, and its forward hidden state and backward hidden state are calculated respectively. The two are then concatenated to obtain the comprehensive hidden layer state vector. The integrated hidden layer state vector is input into the attention mechanism layer, the attention weights of each feature dimension of the integrated hidden layer state vector are calculated, and the hidden layer state vector is weighted and summed to obtain the context feature vector. The context feature vector is input into the output layer, and the prediction residual vector of the mechanism model is obtained by mapping.

[0013] According to the present invention, a method for evaluating the winding state of a converter transformer based on electromagnetic vibration and stiffness identification is provided, wherein the objective function is as follows: ; in, Indicates the stiffness parameter to be identified; This represents the weighting coefficient; M is the upper limit of the harmonic order; This represents the measured vibration amplitude at the nth harmonic frequency; This represents the theoretical vibration amplitude at the nth harmonic frequency; This represents the prediction residual vector output by the neural network.

[0014] According to the present invention, a method for evaluating the state of converter transformer windings based on electromagnetic vibration and stiffness identification is provided. The method involves minimizing the objective function, solving the objective function, and inverting to obtain the optimal value of the stiffness parameter to be identified. Specifically, this includes: The objective function is solved iteratively using the sparrow search algorithm, and the solution that minimizes the objective function value is output as the optimal value of the stiffness parameter to be identified.

[0015] According to the present invention, a method for assessing the condition of a converter transformer winding based on electromagnetic vibration and stiffness identification is provided. The step of determining the condition level of the winding based on the optimal value of the stiffness parameter to be identified specifically includes: Based on the optimal value of the stiffness parameter to be identified and the nominal stiffness of the winding, a stiffness degradation index is obtained. Based on the stiffness degradation index, the state level of the winding is determined by comparing it with the preset grading evaluation standard.

[0016] According to the present invention, a method for evaluating the condition of converter transformer windings based on electromagnetic vibration and stiffness identification is provided, wherein the stiffness degradation index... It can be obtained through the following formula:

[0017] in, Indicates nominal stiffness; This represents the optimal value of the stiffness parameter to be identified obtained through inversion.

[0018] The present invention also provides a converter transformer winding condition assessment device based on electromagnetic vibration and stiffness identification, comprising: The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the converter transformer winding state assessment method based on electromagnetic vibration and stiffness identification as described above.

[0019] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the converter transformer winding state assessment method based on electromagnetic vibration and stiffness identification as described above.

[0020] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the converter transformer winding state assessment method based on electromagnetic vibration and stiffness identification as described above.

[0021] Compared with the prior art, the beneficial effects of the present invention are: (1) By constructing a hybrid architecture that combines an electromagnetic-structural coupling mechanism model containing stiffness parameters with neural network residual correction, this invention can more accurately simulate the nonlinear vibration response of the valve-side winding of the converter transformer under DC bias and rich high-order harmonic conditions. By introducing multi-frequency domain electromagnetic excitation force calculation and residual prediction, the interference of load current fluctuation on vibration signal is effectively eliminated, overcoming the defects of traditional online monitoring technology that rely only on vibration amplitude, resulting in "false alarm under heavy load and missed detection under light load", and significantly improving the accuracy and reliability of converter transformer winding condition assessment under complex conditions.

[0022] (2) This invention adopts a parameter identification strategy based on intelligent global optimization to transform qualitative state monitoring into quantitative parameter inversion. By taking measured data as the target and comprehensively considering the theoretical values ​​of the physical model and the prediction residuals of the neural network, the real-time equivalent stiffness parameters of the winding are inverted online, realizing the quantitative inversion of the real mechanical parameters of the winding. This allows the evaluation results to directly reflect the tightness of the winding's mechanical structure, rather than the apparent vibration magnitude, thereby providing maintenance personnel with quantitative evaluation indicators with clear physical meaning.

[0023] (3) By introducing temperature correction functions for stiffness and damping, this invention can automatically adjust model parameters based on real-time temperature data, dynamically compensate for system errors caused by thermal softening of insulating materials and changes in oil viscosity, thus possessing environmental adaptability and dynamic model correction capabilities. This ensures that the monitoring system can maintain the stability of the evaluation benchmark even when there are large temperature differences between winter and summer or when the load causes drastic fluctuations in oil temperature, avoiding model distortion caused by changes in environmental factors in traditional methods, and achieving high-fidelity monitoring around the clock. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0025] Figure 1 This is one of the flowcharts of a converter transformer winding condition assessment method based on electromagnetic vibration and stiffness identification according to an embodiment of the present invention.

[0026] Figure 2 This is a flowchart illustrating a method for constructing an electromagnetic structure coupling mechanism model containing stiffness parameters to be identified according to an embodiment of the present invention.

[0027] Figure 3 This is a flowchart illustrating a method for obtaining a predicted residual vector using a neural network model according to an embodiment of the present invention.

[0028] Figure 4 This is the second flowchart of the converter transformer winding condition assessment method based on electromagnetic vibration and stiffness identification according to an embodiment of the present invention.

[0029] Figure 5 This is a schematic diagram of the structure of a converter transformer winding condition assessment device based on electromagnetic vibration and stiffness identification according to an embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0031] Existing online monitoring technologies struggle to effectively isolate the interference of load current fluctuations on vibration signals during operation, lacking an evaluation method that can directly identify the internal mechanical parameters of the windings online, beyond the surface appearance of vibration signals. Therefore, this invention provides a solution by constructing an electromagnetic-structural coupling model containing the stiffness parameters to be identified, using a deep learning network to correct the residuals of this model, and then employing an intelligent optimization algorithm to online invert the real-time equivalent stiffness coefficients of the windings. Thus, using the stiffness change rate as the core indicator, a precise quantitative evaluation of the mechanical state of the converter transformer windings is achieved.

[0032] The following is a detailed description of a converter transformer winding condition assessment method based on electromagnetic vibration and stiffness identification disclosed in an embodiment of the present invention.

[0033] like Figure 1 As shown, this invention provides a method for evaluating the winding condition of a converter transformer based on electromagnetic vibration and stiffness identification. The method includes the following steps: Step 100: Obtain comprehensive monitoring data of the valve-side winding of the converter transformer to be evaluated. The data includes: multi-band electrical excitation vector extracted based on load current signal, multi-band mechanical response vector extracted based on vibration signal, and environmental condition vector.

[0034] Specifically, the total monitoring data of the valve-side winding of the converter transformer to be evaluated can be acquired. Considering the operating characteristics of the converter transformer valve-side winding subjected to DC bias and non-sinusoidal current excitation, a multi-source data acquisition system encompassing electrical, mechanical, and environmental dimensions can be established. (1) Acquisition of electrical excitation signal: The load current signal of the valve-side winding is acquired in real time using a broadband Rogowski coil or a zero-flux Hall current sensor installed at the lead-out line of the end screen of the valve-side bushing of the converter transformer. Considering that the load current of the converter transformer contains abundant high-order harmonic components, in order to satisfy the Nyquist sampling theorem and avoid aliasing, the sampling frequency is set to... The frequency response bandwidth of the current transformer should cover DC to 5kHz to ensure accurate capture of characteristic harmonic components above 2kHz.

[0035] (2) Mechanical vibration signal acquisition: Piezoelectric accelerometer arrays are arranged at key locations on the surface of the converter transformer tank (including the high-voltage side bushing riser, the center of the tank side wall reinforcing rib, and near the core grounding lead) to acquire vibration acceleration signals. The sensor's sensitivity is no less than 100 mV / g, its frequency response range covers 0.1 Hz to 10 kHz, and its linearity error is less than 1%.

[0036] (3) Acquisition of thermal state and operating parameters: The top oil temperature is collected by pre-embedded fiber optic grating sensors or top / bottom thermocouples of the oil tank. and bottom oil temperature Calculate the average oil temperature Simultaneously, the current on-load tap changer position is synchronously obtained through the Supervisory Control and Data Acquisition (SCADA) system interface. .

[0037] After completing the acquisition of multi-source data in electrical, mechanical, and environmental dimensions, it is also necessary to perform spatiotemporal synchronization and noise reduction of the signals.

[0038] (1) Timing Alignment: Using the Global Positioning System (GPS) / BeiDou unified timing signal as the trigger source, the timing is aligned with the time sequence. and Synchronous sampling is performed to ensure that the phase error between electrical excitation and mechanical response on the time axis is less than 1% (based on the fundamental frequency), thus guaranteeing the causal correspondence between input and output data.

[0039] (2) Wavelet denoising: The original vibration signal is processed by wavelet threshold denoising algorithm. The Daubechies (db4) wavelet basis is selected for 5-level decomposition. The soft threshold function is used to filter out the high-frequency electromagnetic interference noise in the field and retain the effective vibration components that reflect the mechanical characteristics of the winding.

[0040] After completing the spatiotemporal synchronization and denoising preprocessing of the signal, in order to quantify the nonlinear coupling effect of the full harmonic current component on structural vibration, the time-domain signal needs to be transformed into a frequency-domain feature vector.

[0041] (1) Signal discretization and Fast Fourier Transform (FFT): The collected continuous analog signals and Discrete time series obtained by analog-to-digital conversion and Where p is the sampling point index. A Fast Fourier Transform (FFT) is performed on the data sequence within the current monitoring time window. Taking a current signal as an example, the transform formula is: ; (1) Where N is the number of sampling points, and k is the discrete frequency index. It represents the complex spectrum.

[0042] (2) Extraction of total harmonic features Considering that converter transformers generate significant even-order harmonics (2nd, 4th, etc.) under DC bias conditions, and that the coupling effect between the DC component and each harmonic produces a complex vibration spectrum, this embodiment of the invention adopts a full-band extraction strategy to avoid feature omission: extracting the DC component. , fundamental wave and second harmonic Up to the Mth harmonic All continuous harmonics (i.e., 100Hz, 150Hz, ..., M∙50Hz).

[0043] (3) Constructing an electromagnetic-vibration monitoring vector set: Based on the above analysis, a comprehensive monitoring vector set for the current monitoring time t is constructed. : (2) Among them, the full harmonic electrical excitation vector : ; (3) Mechanical response vector corresponding to frequency

[0044] ; (4) The root mean square value in the time domain of the vibration signal reflects the total vibration energy and is used to capture broadband random loosening noise; the remaining components are the vibration amplitudes at the corresponding frequencies, which reflect the response characteristics of the structure to electromagnetic forces at specific frequencies.

[0045] Environmental condition vector : (5) The above method obtains a characteristic dataset that can reflect the vibration response of the valve-side winding of the converter transformer without distortion and with complete characteristics under the special excitation of DC bias and rich harmonics. This provides a high-quality data input foundation for subsequent implementation of electrical interference removal and accurate inversion of mechanical stiffness.

[0046] Step 101: Based on the electrical excitation vector and the environmental condition vector, the theoretical vibration vector is obtained through an electromagnetic structure coupling mechanism model that includes the stiffness parameter to be identified.

[0047] Specifically, the traditional vibration amplitude method cannot distinguish whether the increase in vibration is due to an increase in current (load change) or a decrease in stiffness (mechanical loosening). Therefore, this invention establishes an electromagnetic structure coupling mechanism model that decouples electrical excitation from mechanical characteristics, transforming the state assessment problem into a parameter identification problem. This model is used to calculate the theoretical vibration response of the winding under given electrical excitation and operating conditions.

[0048] Optionally, such as Figure 2 As shown, the electromagnetic structure coupling mechanism model can be constructed through the following steps: Step 1011: Based on the electrical excitation vector, calculate the electromagnetic excitation force at each frequency point through frequency domain discrete convolution operation.

[0049] Specifically, an electromagnetic excitation force model based on the square law of current is constructed, wherein the nonlinear excitation force calculation method is as follows: based on the Lorentz force being proportional to the square of the current ( The physical criteria are used to calculate the excitation force amplitude at the nth harmonic frequency through frequency domain discrete convolution. : ; (6) in, The electromechanical conversion factor depends on the number of winding turns, leakage flux geometric coefficient, and effective coil area. The loop variable for summing the convolutions covers all frequency combinations involved in the operation. and These are the complex spectra of the corresponding frequency current components. This calculation model, based on the principle of frequency-domain self-convolution, can fully characterize the unique nonlinear frequency coupling effect of converter transformers. Specifically, under DC bias conditions, the DC component of the excitation current in the converter transformer... With fundamental current The cross term directly excites vibrations at the fundamental frequency (50Hz); while the coupling between the fundamental and the second harmonic will produce vibrations at 150Hz. The above formula can automatically cover these complex excitation frequency components caused by DC bias and asymmetric triggering.

[0050] Step 1012: Based on the temperature parameters in the environmental condition vector, perform temperature correction on the stiffness parameters and damping coefficients to be identified.

[0051] Specifically, stiffness is a direct and essential physical representation of the mechanical fastening state of the winding. Temperature correction of the stiffness parameters and damping coefficients to be identified aims to enable the mechanistic model to dynamically adapt to changes in the operating environment, thereby achieving more accurate state inversion. Depending on the characteristics of the insulating material and transformer oil, and the model's accuracy requirements, other functions or methods capable of describing the temperature-dependent changes in the stiffness parameters and damping coefficients can be used. Examples include polynomial fitting functions based on experimental calibration data, piecewise linear interpolation, or lookup table methods, or using historical operating data to train a data-driven model (such as a neural network) to directly predict stiffness and damping coefficients. This embodiment of the invention does not impose any limitations on these methods.

[0052] Optionally, the corrected stiffness parameter to be identified can be obtained using the following formula. : (7) The corrected damping coefficient is obtained using the following formula. : (8) Where T represents the temperature value; Indicates reference temperature The following stiffness parameters need to be identified; It represents the thermal softening coefficient of insulating materials, characterizing the rate at which stiffness decreases with increasing temperature; Indicates the average oil temperature; Indicates the foundation damping coefficient. This indicates the reference temperature corresponding to the base damping. This represents the viscosity-temperature constant of the oil.

[0053] Specifically, a single-degree-of-freedom vibration model of the winding, consisting of mass, damping, and spring, is constructed. Considering the significant impact of transformer operating temperature fluctuations on the mechanical properties of the insulation material and the viscosity of the oil, a dynamic correction model for parameters varying with temperature is established. (1) Temperature correction for equivalent stiffness (stiffness parameter to be identified): The winding stiffness is mainly determined by the elastic modulus of the insulating pad. Given the thermal softening effect of insulating materials (such as laminated paperboard), the following is defined: Reference temperature The equivalent stiffness coefficient to be identified is given below. When the average oil temperature is... At that time, the actual operating stiffness (the corrected stiffness parameter to be identified) See formula (7).

[0054] (2) Correction of damping coefficient: The viscosity of transformer oil decreases exponentially with increasing temperature, resulting in reduced fluid damping. A correction function for the damping coefficient is established. See formula (8).

[0055] The above method performs temperature correction on winding stiffness and damping, eliminating the interference of operating temperature fluctuations on the inversion results. This ensures that the finally identified stiffness parameters accurately reflect the changes in the mechanical fastening state of the winding itself, rather than spurious changes affected by ambient temperature. This significantly improves the accuracy, reliability, and all-weather adaptability of the condition assessment.

[0056] Step 1013: Based on the temperature-corrected stiffness parameters to be identified, the damping coefficient, the electromagnetic excitation force, and the equivalent mass of the winding, construct the frequency domain response transfer function of the mechanism model, calculate the theoretical vibration amplitude at each frequency point, and form the theoretical vibration vector.

[0057] Optionally, the frequency domain response transfer function of the mechanism model is expressed by the following formula: ; (9) in, This represents the theoretical vibration amplitude at the nth harmonic frequency; This represents the angular frequency of the nth harmonic. The equivalent mass constant of the winding is represented by k(T); k(T) is the stiffness parameter to be identified after correction of the winding in step 1012; c(T) is the equivalent damping coefficient of the system after temperature correction in step 1012; electromagnetic excitation force. That is, formula (6); Indicates the electromechanical conversion factor; The loop variable for summing the convolutions; and These are the complex spectra of the current components at the corresponding frequencies.

[0058] Specifically, based on mechanical vibration theory, the frequency domain response function of a single-degree-of-freedom system is established, for the nth harmonic frequency. Its theoretical vibration amplitude The calculation formula is formula (9). Using this formula, all frequency points are traversed, and the theoretical vibration response vector of the entire frequency band is constructed with electrical excitation, environmental conditions and the stiffness to be identified as inputs. : (10) The above method directly establishes a clear and inverse physical relationship between excitation, system parameters (stiffness, mass, damping) and response (vibration amplitude).

[0059] Step 102: Input the electrical excitation vector and the environmental condition vector into the trained neural network model to obtain the prediction residual vector of the mechanism model.

[0060] Specifically, although the electromagnetic-structural coupling mechanism model established in step 101 incorporates temperature correction, it is essentially still an idealized and simplified model, which cannot fully encompass the complex effects of turbulent oil flow inside the converter transformer, minute gaps in structural component assembly, and nonlinear damping. Therefore, this embodiment of the invention aims to construct a data-driven deep neural network to learn and predict the inherent residuals between the physical model and the real device, thereby building a high-fidelity stiffness identification benchmark.

[0061] Optionally, the neural network model includes an input fusion layer, a feature extraction layer, an attention mechanism layer, and an output layer; Correspondingly, such as Figure 3 As shown, the step of inputting the electrical excitation vector and the environmental condition vector into the trained neural network model to obtain the prediction residual vector of the mechanism model specifically includes the following steps: Step 1021: Input the electrical excitation vector and the environmental condition vector into the input fusion layer to obtain normalized input features.

[0062] Step 1022: Input the normalized features into the feature extraction layer, calculate its forward hidden state and backward hidden state respectively, and concatenate the two to obtain the comprehensive hidden layer state vector.

[0063] Step 1023: Input the integrated hidden layer state vector into the attention mechanism layer, calculate the attention weights of each feature dimension of the integrated hidden layer state vector, and perform a weighted summation of the hidden layer state vector to obtain the context feature vector.

[0064] Step 1024: Input the context feature vector into the output layer to map the prediction residual vector of the mechanism model.

[0065] Specifically, the neural network model is used to accurately predict the residuals between measured and theoretical vibrations under real-time operating conditions.

[0066] In step 1021 above, current harmonics and oil temperature are the main variables causing nonlinear errors in the physical model. Using these as inputs, the neural network learns the systematic biases produced by the physical model under specific current configurations and temperature conditions. The full harmonic electrical excitation vector from step 100 is then used... (Formula 3) and environmental condition vector (Formula 5) Perform feature fusion to construct the network input feature vector at time t. : (11) The input features are then standardized to obtain the normalized input features of the network. : (12) in and These are the mean and standard deviation of the historical dataset, respectively.

[0067] In step 1022 above, features are extracted using a Bidirectional Long Short-Term Memory (Bi-LSTM) feature extraction layer. Bi-LSTM is used as the core hidden layer.

[0068] Mechanism of action: The vibration response of the converter transformer has significant thermal inertia and magnetic hysteresis characteristics. That is, the vibration residual at the current moment depends not only on the current input, but also on the thermal state and magnetization state of the equipment at the previous moment.

[0069] Processing logic: Bi-LSTM captures the input sequence simultaneously through two processing channels, forward and backward. The historical evolution information and future trend information of (Formula 12) are used to calculate the forward hidden state. and backward hidden state The two are then concatenated to obtain the comprehensive hidden state vector. : (13) ;(14) ; (15) In step 1023 above, attention weights are assigned by introducing an attention mechanism layer after the Bi-LSTM layer.

[0070] Mechanism of action: Excitation components of different frequencies contribute differently to the vibration residual. For example, even harmonics often induce more complex structural nonlinear responses than odd harmonics.

[0071] Processing logic: The network automatically calculates the attention scores for each feature dimension through fully connected layers. And then transformed into normalized attention weights through the Softmax function: (16) (17) in, , These represent the weight matrix, bias vector, and parameter vector of the attention layer, respectively. L is the hidden layer state vector. The feature dimension is j, which is the summation traversal index used for normalization calculation, and its value ranges from 1 to L.

[0072] Finally, the hidden states are weighted and summed to obtain the context feature vector. : (18) in, Representing vectors The i-th element in the equation. Specifically, this mechanism can strengthen the weight of key influencing factors (such as DC bias component) and suppress irrelevant noise, thereby extracting the deep features that have the most significant impact on the residual.

[0073] In step 1024 above, the weighted features are processed through a fully connected layer. Mapping back to the dimension of the vibration vector, outputting the predicted residual vector under the current operating condition. This is then superimposed as a compensation term onto the physical model (i.e., the electromagnetic-structural coupling mechanism model) from step 101 to construct the corrected theoretical vibration vector. : (19) The training method for this neural network model is as follows: a) Training data preparation: Select historical operating datasets of converter transformers under known health conditions as training samples.

[0074] 1) Calculate the theoretical value: Calculate the electrical excitation vector at historical moments. and environmental condition vector Input the mechanism model from step 101. At this point, the stiffness parameters to be identified in the model will be... The stiffness is fixed at the factory nominal value. The theoretical vibration vector under the nominal state is calculated. .

[0075] 2) Extracting residual labels: Define intrinsic residual vectors The difference between the measured vibration and the theoretical vibration: (20) in, (Formula 4) The mechanical response vector corresponding to the frequency is the measured mechanical response vector, and this residual vector This represents the nonlinear error portion that the physical model failed to explain.

[0076] b) Offline training and parameter fixing of the model: The training samples, i.e., the historical operating dataset of the converter transformer under known health conditions (input is...), will be used. , tag as Supervised learning of neural networks is performed using mean squared error (MSE) as the objective function to measure the prediction residual. The difference between the actual residual and the network weights is then analyzed. The Adam optimization algorithm is then used to iteratively update the network weights until the loss function converges, resulting in the fixed network model parameters.

[0077] The above method integrates multi-source information through a neural network architecture, uses a Bi-LSTM feature extraction layer to capture the temporal dependence and thermal inertia of winding vibration, and then enhances the influence of key harmonic components through an attention mechanism. Finally, it accurately predicts the system error of the electromagnetic-structure coupling mechanism model, providing a reliable data-driven correction benchmark for the high-precision inversion of the stiffness parameters to be identified, and significantly improving the accuracy and robustness of the state assessment.

[0078] Step 103: Based on the mechanical response vector and the theoretical vibration vector corrected by the predicted residual vector, construct an objective function, minimize the objective function, solve the objective function, and invert to obtain the optimal value of the stiffness parameter to be identified.

[0079] Specifically, this step aims to transform the problem of monitoring vibration signals into a problem of identifying structural parameters with clear physical meaning. Taking the measured vibration vector acquired in step 100 as the target and the physical model in step 101 and the residual prediction results in step 102 as the benchmark, the real-time equivalent stiffness coefficient of the converter transformer winding is retrieved online using the powerful global optimization capability of the Sparrow Search Algorithm (SSA).

[0080] In order to accurately identify the current stiffness parameters of the winding. (Also known as equivalent stiffness), construct an evaluation index to measure the degree of agreement between the "corrected theoretical model" and "real measurement data". The stiffness parameters to be identified... As optimization variables, the following objective function is defined. : (twenty one) in, Indicates the stiffness parameter to be identified; This represents the weighting coefficient; M is the upper limit of the harmonic order; This represents the measured vibration amplitude at the nth harmonic frequency; This represents the theoretical vibration amplitude at the nth harmonic frequency; This represents the prediction residual vector output by the neural network.

[0081] The physical meaning of this objective function is to find an optimal stiffness value. This allows the theoretical calculations of the physical model, after being compensated for the nonlinearity of the neural network predictions, to approximate the actual measured values ​​across the entire frequency range. This is due to the introduction of... Compensation is performed to effectively eliminate systematic errors caused by model simplification, significantly improving the accuracy of stiffness identification.

[0082] Optionally, the Sparrow Search algorithm is used to iteratively solve the objective function, and the solution that minimizes the objective function value is output as the optimal value of the stiffness parameter to be identified.

[0083] The specific method for global optimization using the SSA algorithm is as follows: (1) Initialization parameters and population: Set the population size and maximum number of iterations The population positions are randomly initialized within a predefined stiffness search space, with each individual's position representing a potential stiffness value. .

[0084] (2) Fitness calculation: Each individual represents a certain fitness level. Substitute the objective function formula into formula (21) to calculate its fitness value (i.e., objective function value J). The smaller the fitness, the closer the stiffness value is to the true state.

[0085] (3) Population Iterative Update Strategy: Based on the SSA algorithm mechanism, the population is dynamically divided into discoverers, joiners, and vigilants, each executing a different position update strategy. Among them, the discoverers undertake the global exploration task, responsible for searching stiffness values ​​in a wide area. If the objective function value corresponding to the current stiffness is better, the population is guided to move closer to that area to expand the search range. The joiners are responsible for local development, improving the identification accuracy by following the discoverers to conduct refined searches near the current optimal stiffness value. The vigilants rely on the anti-predation mechanism to randomly jump to other positions in the search space, thereby avoiding the algorithm from getting trapped in local optima (such as a pseudo solution at a harmonic resonance point) and enhancing the robustness of the algorithm.

[0086] (4) Output the optimal solution: when the number of iterations reaches or objective function value The iteration stops when the value falls below the preset convergence threshold. The output is the stiffness value corresponding to the globally optimal position at this point, which is the actual equivalent stiffness of the winding at the current moment. .

[0087] The above method constructs and minimizes the objective function, compares the measured vibration with the intelligently corrected theoretical benchmark across the entire frequency band, and inverts to obtain the optimal value of the equivalent stiffness parameter of the winding. This eliminates the influence of load current fluctuations, realizes the quantitative identification of the mechanical fastening state of the winding, and improves the accuracy of the assessment.

[0088] Step 104: Determine the state level of the winding based on the optimal value of the stiffness parameter to be identified.

[0089] Specifically, after obtaining the optimal values ​​of the stiffness parameters to be identified, the winding condition level can be determined by combining the nominal stiffness of the transformer at the manufacturer's factory. This value can then be compared with preset grading evaluation standards to comprehensively determine the loosening level of the converter transformer windings and the fault status of the sensors. For example, this can be achieved by defining a dimensionless stiffness degradation index. : (twenty two) in, Indicates nominal stiffness; This represents the optimal value (actual equivalent stiffness) of the stiffness parameter to be identified obtained through inversion.

[0090] Based on stiffness degradation index Based on the numerical range and combined with engineering experience and the mechanical properties of insulation materials, a graded evaluation standard for the mechanical condition of converter transformer windings is established: State I (Healthy and Stable State): When When the winding axial clamping force is within the design allowable range and the mechanical structure is tight, it is recommended to maintain the regular inspection cycle; State II (early loosening warning): when When the insulation pads show slight aging creep or springback, and the overall winding stiffness shows a decreasing trend, although it does not immediately endanger operational safety, it is recommended to shorten the sampling interval of online monitoring data and closely monitor the deterioration trend; State III (Severe loosening alarm): When When the winding is determined to be significantly loose or potentially deformed, the equipment's ability to withstand short-circuit current surges is greatly weakened, posing an extremely high operational risk. It is recommended to conduct a cover inspection or re-tighten the winding in conjunction with the most recent power outage opportunity. State IV (Evaluation Result Verification State): When the residual value after the objective function converges... If the residual value exceeds the preset residual abnormality threshold, but no obvious signal quality abnormalities are observed in the current, vibration, and temperature acquisition channels, the model fitting consistency is deemed insufficient, and a monitoring system fault cannot be confirmed. It is recommended to increase the sampling frequency and verify the model parameters and on-site conditions. State V (Monitoring System Abnormal State): When the residual value after the objective function converges... If the residual difference exceeds the preset abnormal threshold and there are issues such as disconnection of the acquisition channel, saturation, long-term constant value, sudden change, asynchronous sampling, or multiple consecutive monitoring windows failing to converge to a reasonable stiffness range, the reliability of the monitoring data is deemed insufficient. It is recommended to prioritize checking the sensor installation, signal transmission link, and acquisition module, and to suspend the output of winding looseness levels from state I to state III.

[0091] The above method establishes a graded evaluation standard based on physical parameters rather than vibration manifestations, which enables accurate and reliable graded early warning of the loosening state of converter transformer windings, providing a direct and clear quantitative basis for equipment maintenance decisions.

[0092] To facilitate understanding of the embodiments of the present invention, a specific example of a converter transformer valve-side winding condition assessment process will be used for illustrative explanation below.

[0093] like Figure 4 As shown, this evaluation method mainly includes five core steps: First, in step S1, the load current signal of the converter transformer valve-side winding, the vibration signal of the tank surface, and environmental operating parameters are collected, followed by time-domain synchronization and noise reduction preprocessing, and extraction of full harmonic frequency domain features. Second, in step S2, the full harmonic electromagnetic excitation force is calculated based on the load current signal, and a temperature correction coefficient is introduced, finally outputting a theoretical vibration vector containing the stiffness parameter to be identified. Next, in step S3, using electrical excitation and environmental parameters as input, a deep neural network (such as Bi-LSTM) is used to predict the inherent residual between the mechanism model and the actual equipment, and the predicted residual vector is output. Subsequently, in step S4, an objective function is constructed to minimize the difference between the measured vibration and the theoretical value superimposed residual, and the Sparrow Search Algorithm (SSA) is used for global optimization to find the actual equivalent stiffness coefficient of the winding at the current moment. Finally, in step S5, the stiffness degradation index is calculated, compared with the preset grading evaluation standard, and the loosening level of the converter transformer winding and the fault status of the sensor are comprehensively determined.

[0094] The following describes a converter transformer winding condition assessment device based on electromagnetic vibration and stiffness identification provided by the present invention. The converter transformer winding condition assessment device based on electromagnetic vibration and stiffness identification described below and the converter transformer winding condition assessment method based on electromagnetic vibration and stiffness identification described above can be referred to in correspondence.

[0095] like Figure 5 As shown, this invention provides a converter transformer winding condition assessment device based on electromagnetic vibration and stiffness identification, comprising the following modules: The acquisition module 500 is used to acquire comprehensive monitoring data of the valve-side winding of the converter transformer to be evaluated. The data includes: multi-band electrical excitation vector extracted based on load current signal, multi-band mechanical response vector extracted based on vibration signal, and environmental condition vector. The construction module 510 is used to construct an electromagnetic structure coupling mechanism model containing the stiffness parameters to be identified based on the electrical excitation vector and the environmental condition vector, so as to obtain the theoretical vibration vector. The prediction module 520 is used to input the electrical excitation vector and the environmental condition vector into the trained neural network model to obtain the prediction residual vector of the mechanism model. The inversion module 530 is used to construct an objective function based on the mechanical response vector and the theoretical vibration vector corrected by the predicted residual vector, and to solve the objective function with the goal of minimizing the objective function, thereby inverting to obtain the optimal value of the stiffness parameter to be identified. The determination module 540 is used to determine the state level of the winding based on the optimal value of the stiffness parameter to be identified.

[0096] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the converter transformer winding state assessment method based on electromagnetic vibration and stiffness identification provided by the above methods.

[0097] In another aspect, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the converter transformer winding state assessment method based on electromagnetic vibration and stiffness identification provided by the above methods.

[0098] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0099] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing the condition of converter transformer windings based on electromagnetic vibration and stiffness identification, characterized in that, include: Acquire comprehensive monitoring data of the valve-side winding of the converter transformer to be evaluated. The data includes: multi-band electrical excitation vector extracted based on load current signal, multi-band mechanical response vector extracted based on vibration signal, and environmental condition vector. Based on the electrical excitation vector and the environmental condition vector, an electromagnetic structure coupling mechanism model containing the stiffness parameters to be identified is constructed to obtain the theoretical vibration vector. The electrical excitation vector and the environmental condition vector are input into the trained neural network model to obtain the prediction residual vector of the mechanism model. Based on the mechanical response vector and the theoretical vibration vector corrected by the predicted residual vector, an objective function is constructed. The objective function is minimized, and the optimal value of the stiffness parameter to be identified is obtained by inversion. Based on the optimal value of the stiffness parameter to be identified, the state level of the winding is determined.

2. The method for evaluating the winding condition of a converter transformer based on electromagnetic vibration and stiffness identification according to claim 1, characterized in that, Based on the electrical excitation vector and the environmental condition vector, an electromagnetic structure coupling mechanism model containing the stiffness parameters to be identified is constructed to obtain the theoretical vibration vector, specifically including: Based on the electrical excitation vector, the electromagnetic excitation force at each frequency point is calculated by frequency domain discrete convolution operation; Based on the temperature parameters in the environmental condition vector, the stiffness parameters and damping coefficients to be identified are corrected for temperature. Based on the temperature-corrected stiffness parameters to be identified, the damping coefficient, the electromagnetic excitation force, and the equivalent mass of the winding, the frequency domain response transfer function of the mechanism model is constructed, and the theoretical vibration amplitude at each frequency point is calculated to form the theoretical vibration vector.

3. The method for evaluating the winding condition of a converter transformer based on electromagnetic vibration and stiffness identification according to claim 1, characterized in that, The temperature correction of the stiffness parameter and damping coefficient to be identified based on the temperature parameter in the environmental condition vector specifically includes: The corrected stiffness parameter to be identified is obtained using the following formula. : ； The corrected damping coefficient is obtained using the following formula. : ； Where T represents the temperature value; Indicates reference temperature The following stiffness parameters need to be identified; It represents the thermal softening coefficient of insulating materials, characterizing the rate at which stiffness decreases with increasing temperature; Indicates the average oil temperature; Indicates the foundation damping coefficient. This indicates the reference temperature corresponding to the base damping. This represents the viscosity-temperature constant of the oil.

4. The method for evaluating the winding condition of a converter transformer based on electromagnetic vibration and stiffness identification according to claim 3, characterized in that, The frequency domain response transfer function of the aforementioned mechanism model is expressed by the following formula: ； in, This represents the theoretical vibration amplitude at the nth harmonic frequency; This represents the angular frequency of the nth harmonic. Represents the equivalent mass constant of the winding; electromagnetic excitation force ; Indicates the electromechanical conversion factor; The loop variable for summing the convolutions; and These are the complex spectra of the current components at the corresponding frequencies.

5. The method for evaluating the winding condition of a converter transformer based on electromagnetic vibration and stiffness identification according to claim 1, characterized in that, The neural network model includes an input fusion layer, a feature extraction layer, an attention mechanism layer, and an output layer; Correspondingly, the step of inputting the electrical excitation vector and the environmental condition vector into the trained neural network model to obtain the prediction residual vector of the mechanism model specifically includes: The electrical excitation vector and the environmental condition vector are input into the input fusion layer to obtain normalized input features; The normalized features are input into the feature extraction layer, and its forward hidden state and backward hidden state are calculated respectively. The two are then concatenated to obtain the comprehensive hidden layer state vector. The integrated hidden layer state vector is input into the attention mechanism layer, the attention weights of each feature dimension of the integrated hidden layer state vector are calculated, and the hidden layer state vector is weighted and summed to obtain the context feature vector. The context feature vector is input into the output layer, and the prediction residual vector of the mechanism model is obtained by mapping.

6. The method for evaluating the condition of converter transformer windings based on electromagnetic vibration and stiffness identification according to claim 5, characterized in that, The objective function is as follows: ； in, Indicates the stiffness parameter to be identified; This represents the weighting coefficient; M is the upper limit of the harmonic order; This represents the measured vibration amplitude at the nth harmonic frequency; This represents the theoretical vibration amplitude at the nth harmonic frequency; This represents the prediction residual vector output by the neural network.

7. The method for evaluating the condition of converter transformer windings based on electromagnetic vibration and stiffness identification according to claim 5, characterized in that, The process of minimizing the objective function and solving for it to obtain the optimal value of the stiffness parameter to be identified specifically includes: The objective function is solved iteratively using the sparrow search algorithm, and the solution that minimizes the objective function value is output as the optimal value of the stiffness parameter to be identified.

8. The method for evaluating the winding condition of a converter transformer based on electromagnetic vibration and stiffness identification according to claim 7, characterized in that, Determining the state level of the winding based on the optimal value of the stiffness parameter to be identified specifically includes: Based on the optimal value of the stiffness parameter to be identified and the nominal stiffness of the winding, a stiffness degradation index is obtained. Based on the stiffness degradation index, the state level of the winding is determined by comparing it with the preset grading evaluation standard.

9. The method for evaluating the winding condition of a converter transformer based on electromagnetic vibration and stiffness identification according to claim 7, characterized in that, The stiffness degradation index It can be obtained through the following formula: ； in, Indicates nominal stiffness; This represents the optimal value of the stiffness parameter to be identified obtained through inversion.

10. A converter transformer winding condition assessment device based on electromagnetic vibration and stiffness identification, characterized in that, include: The acquisition module is used to acquire comprehensive monitoring data of the valve-side winding of the converter transformer to be evaluated. The data includes: multi-band electrical excitation vector extracted based on load current signal, multi-band mechanical response vector extracted based on vibration signal, and environmental condition vector. The construction module is used to construct an electromagnetic structure coupling mechanism model containing the stiffness parameters to be identified based on the electrical excitation vector and the environmental condition vector, so as to obtain the theoretical vibration vector; The prediction module is used to input the electrical excitation vector and the environmental condition vector into the trained neural network model to obtain the prediction residual vector of the mechanism model. The inversion module is used to construct an objective function based on the mechanical response vector and the theoretical vibration vector corrected by the predicted residual vector, and to solve the objective function with the goal of minimizing the objective function, thereby obtaining the optimal value of the stiffness parameter to be identified. The determination module is used to determine the state level of the winding based on the optimal value of the stiffness parameter to be identified.

Images