Dynamic calibration method and system for multi-physical field data fusion of oil-immersed transformer

Abstract

The invention relates to the technical field of power systems, in particular to a dynamic calibration method and system for multi-physics field data fusion of an oil-immersed transformer, and the system comprises a data calibration unit which carries out the dynamic calibration of the multi-physics field data of the transformer through a cross-physics field coupling calibration model, and evaluates the calibration effect based on an objective function; wherein the multi-physical field data comprises temperature data, vibration data, UHF signal data and gas data; the cross-physics field coupling calibration model comprises establishment of a temperature-vibration compensation equation and a gas-UHF trigger mechanism; according to the method, data fusion is carried out on multiple sensors, dynamic calibration is carried out by adopting a cross-physics field coupling calibration model, an edge GAN generates an adversarial sample, updating of the cross-physics field coupling calibration model is triggered in combination with confidence, and comprehensive characteristic parameters are intelligently analyzed by a hierarchical fusion architecture based on a Shapley value. The health state of the transformer can be accurately judged, and the accuracy of fault diagnosis is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a dynamic calibration method and system for multi-physical field data fusion of an oil-immersed transformer. Background Art

[0002] Oil-immersed transformers, as crucial equipment in power systems, are widely used in power transmission and distribution networks, particularly in high-voltage and large-capacity power conversion scenarios. Their primary function is to dissipate heat, provide protection, and provide insulation through oil-immersed insulation. Therefore, monitoring and maintaining their operating status are crucial. However, due to their complex operating environment, long operating times, and susceptibility to external factors, oil-immersed transformers are susceptible to various faults, such as overheating, partial discharge, and mechanical damage. Failure to promptly detect and address these issues can lead to equipment failure or even major safety incidents.

[0003] Traditional transformer fault monitoring and diagnosis methods mainly rely on single detection methods, such as temperature monitoring and gas monitoring. Although they can monitor the operating status of equipment to a certain extent, these methods lack sufficient comprehensiveness and cannot accurately judge the health status of equipment in a timely manner. They are also easily affected by external interference, leading to misdiagnosis or missed diagnosis. Therefore, how to achieve intelligent monitoring, health assessment and fault warning of oil-immersed transformers through comprehensive analysis of multiple physical field data has become a major issue facing the power industry.

[0004] At present, with the rapid development of technologies such as smart grids, the Internet of Things, and artificial intelligence, the means of transformer health monitoring are gradually developing towards multi-dimensional data fusion and intelligent diagnosis. In order to improve the operational safety and maintenance efficiency of transformers, there is an urgent need for a transformer fault monitoring technology based on dynamic calibration and intelligent data fusion based on collaborative perception of multi-physical fields. Summary of the Invention

[0005] The present invention provides a dynamic calibration method and system for multi-physical field data fusion of oil-immersed transformers, which overcomes the shortcomings of the above-mentioned existing technologies and can effectively solve the problem that the health status of existing transformers cannot be judged in a timely and accurate manner due to the fact that they can only be diagnosed through a single diagnostic method such as temperature and gas monitoring.

[0006] To solve the above problems, one of the technical solutions described in the present invention is achieved in the following manner: a dynamic calibration method for multi-physical field data fusion of an oil-immersed transformer, comprising the following steps: The transformer's multi-physics data is dynamically calibrated using a cross-physics coupling calibration model, and the calibration effect is evaluated based on an objective function. The multi-physics data includes temperature data, vibration data, UHF signal data, and gas data. The cross-physics coupling calibration model includes establishing a temperature-vibration compensation equation and a gas-UHF trigger mechanism. Generate adversarial samples based on edge GAN using the transformer's multi-physics data, and trigger the update of the cross-physics coupling calibration model based on confidence. The dynamic features of each physical field data after dynamic calibration are extracted to generate comprehensive feature parameters. The comprehensive feature parameters are intelligently analyzed based on the hierarchical fusion architecture of Shapley values. The key spatiotemporal features are enhanced through the dual-channel attention mechanism, and the comprehensive diagnostic score is calculated to determine the fault type.

[0007] The above-mentioned temperature-vibration compensation equation includes the following calculation equations: , Where V is the original vibration characteristic value, unit: mm / s; V' is the vibration value after temperature compensation, unit: mm / s; T is the real-time monitoring temperature value, unit: °C; T0 is the reference temperature; α is the thermal expansion coefficient of the material, unit: β is the transient response factor, unit: mm / s·℃; t is the time variable of temperature change, unit: s; α and β are calibrated by finite element thermal coupling simulation. , .

[0008] The above-mentioned gas-UHF trigger mechanism includes, Set gas thresholds; Determine whether the gas data exceeds the threshold; In response to this, the UHF signal of the transformer is sampled at high frequency to capture the partial discharge pulse signal.

[0009] The above multi-physics data of the transformer is used to generate adversarial samples based on edge GAN, and combined with confidence to trigger the update of the cross-physics coupling calibration model, including: Edge GAN generates adversarial samples based on multi-physics field data and outputs them to the cross-physics field coupling calibration model; Edge GAN receives new multi-physics field data and adversarial samples in real time, and calculates the confidence of the new multi-physics field data through the discriminator; The confidence calculation formula is as follows: , Where, is the confidence level of the new data, ranging from [0,1]; Predict output values ​​for the model; is the true labeled value; is the maximum value used to normalize the relative error; Determine whether the confidence level of the new multi-physics field data is less than 0.85; In response to this, the update of the cross-physical field coupling calibration model is triggered, that is, the cross-physical field coupling calibration model is compressed by adopting the knowledge distillation method to achieve the update of the cross-physical field coupling calibration model.

[0010] The above-mentioned layered fusion architecture based on Shapley value performs intelligent analysis on comprehensive feature parameters and strengthens key spatiotemporal features through a dual-channel attention mechanism, including: The Shapley value is used to quantify the contribution of each feature. The calculation method is as follows: , Where, represents the average marginal contribution value of the i-th feature in all feature combinations; F represents the full set of features, which here are temperature, vibration, gas, and UHF physical field features; |F| is the number of features in set F; S is a subset of F; |S| is the number of features in set S; Feature subset, which represents a combination of some features taken from F; is the predicted value or scoring function of the model when using the feature subset S; Representation characteristics i The predicted value of the model when subset S is not added; score item It is a weighting coefficient that indicates the probability of different feature combinations occurring; The dual-channel attention mechanism includes: Spatial attention module: calculates the sensor position weight ws=softmax(MLP(xyz)), Where xyz is the three-dimensional coordinate feature vector of the sensor; is a multi-layer perceptron that extracts spatial features through nonlinear mapping; ws is the attention weight of each sensor in the spatial dimension. The larger the value, the greater the influence of the position at the current moment; : Normalization function, used to map the output to the probability distribution interval [0,1]; Temporal Attention Module: Using LSTM to Extract Temporal Importance , Where ht is the hidden state of LSTM at time step t, which represents the feature representation at the current moment; ct is the cell state of LSTM at time step t, which represents the long-term memory feature; W is a trainable weight matrix used for the linear transformation of [ht;ct]; [ht;ct] is the concatenation vector of the hidden state and the cell state; wt is the attention weight in the time dimension. The larger the value, the greater the contribution of the time step to the final prediction of the model. Sigmoid activation function maps the value to the [0,1] interval.

[0011] The dynamic features of each physical field data after the above extraction and dynamic calibration include: Use sliding window techniques to calculate trend changes, volatility, and rate of change in temperature data; Perform frequency domain analysis and time domain statistics extraction on vibration data; Perform wavelet transform on UHF signal data to extract partial discharge features; Calculate the concentration change trend and key gas component ratio of gas data.

[0012] The above calculation of comprehensive diagnostic score and determination of fault type include: The composite diagnostic score is calculated as follows: Assume that the characteristics of each physical field are 、 、 、 , each single domain fault index function is 、 、 、 , the weight is 、 、 、 ;The bias term is b; The comprehensive diagnostic score formula is: , For fault judgment, the following logic function is used: , Where P(fault|X) is the probability of a fault occurring under input feature X; D is the comprehensive diagnosis score; is an exponential function, used to implement the S-shaped curve of logistic regression; When the temperature and gas weights in the comprehensive diagnosis score D are significantly higher, and When , it is determined to be an overheating fault; When the combined diagnostic score of the vibration and UHF parts exceeds the set threshold, it is determined to be a mechanical fault; When the score of the UHF part is significant and there is a combined effect of temperature and gas anomaly indicators, it is determined to be an insulation fault; When the comprehensive diagnosis score D is significantly higher, it corresponds to , and when all single domain indicators are abnormal at the same time, it is determined to be a multi-factor coupling fault.

[0013] The second technical solution of the present invention is achieved by the following method: a dynamic calibration system for oil-immersed transformer multi-physics field data fusion, using a dynamic calibration method for oil-immersed transformer multi-physics field data fusion, comprising: The data calibration unit dynamically calibrates the transformer's multi-physics data using a cross-physics coupling calibration model, and evaluates the calibration results based on an objective function. The multi-physics data includes temperature, vibration, UHF signal, and gas data. The cross-physics coupling calibration model includes establishing a temperature-vibration compensation equation and a gas-UHF trigger mechanism. The edge GAN unit generates adversarial samples based on the multi-physics field data of the transformer and triggers the update of the cross-physics coupling calibration model in combination with the confidence level; The analysis and diagnosis unit extracts the dynamic features of each physical field data after dynamic calibration, generates comprehensive feature parameters, performs intelligent analysis on the comprehensive feature parameters based on the hierarchical fusion architecture of Shapley values, strengthens key spatiotemporal features through the dual-channel attention mechanism, calculates the comprehensive diagnosis score and determines the fault type.

[0014] The above-mentioned edge GAN unit includes a sample generation module, a calculation module, a judgment module and a response module; In the sample generation module, the edge GAN generates adversarial samples based on multi-physics field data and outputs them to the cross-physics field coupling calibration model; In the computing module, the edge GAN receives new multi-physics field data and adversarial samples in real time, and calculates the confidence level of the new multi-physics field data through the discriminator; A judgment module determines whether the confidence level of new multi-physics field data is less than 0.85; The response module, when responding, triggers the update of the cross-physical field coupling calibration model, that is, the cross-physical field coupling calibration model is compressed by adopting the knowledge distillation method to realize the update of the cross-physical field coupling calibration model.

[0015] The present invention fuses data from multiple sensors, performs dynamic calibration using a cross-physical field coupling calibration model, generates adversarial samples using edge GAN, and combines confidence-triggered updates of the cross-physical field coupling calibration model with a hierarchical fusion architecture based on Shapley values ​​to intelligently analyze comprehensive characteristic parameters. This can accurately determine the health status of the transformer and improve the accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0017] Figure 1 This is a flow chart of the method of Example 1 of the present invention.

[0018] Figure 2 Flowchart of the gas-UHF trigger mechanism in Example 1 of the present invention.

[0019] Figure 3This is a flowchart of model updating based on edge GAN in Example 1 of the present invention.

[0020] Figure 4 This is a system block diagram of a dynamic calibration system for multi-physical field data fusion of an oil-immersed transformer in Example 2 of the present invention.

[0021] Figure 5 This is a block diagram of the edge GAN unit in Example 2 of the present invention. DETAILED DESCRIPTION

[0022] The present invention is not limited to the following embodiments, and specific implementation methods can be determined based on the technical solutions of the present invention and actual conditions.

[0023] Example 1: Figure 1 As shown, an embodiment of the present invention discloses a dynamic calibration method for multi-physical field data fusion of an oil-immersed transformer, comprising the following steps: S101, dynamically calibrating the transformer's multi-physics field data using a cross-physics field coupling calibration model, and evaluating the calibration effect based on an objective function; wherein the multi-physics field data includes temperature data, vibration data, UHF signal data, and gas data; the cross-physics field coupling calibration model includes establishing a temperature-vibration compensation equation and a gas-UHF trigger mechanism; S102, using the transformer's multi-physics data, generates adversarial samples based on edge GAN, and triggers the update of the cross-physics coupling calibration model in combination with confidence; S103 extracts the dynamic features of each physical field data after dynamic calibration, generates comprehensive feature parameters, performs intelligent analysis on the comprehensive feature parameters based on the hierarchical fusion architecture of Shapley values, strengthens key spatiotemporal features through a dual-channel attention mechanism, calculates the comprehensive diagnostic score and determines the fault type.

[0024] The above also includes the real-time collection of multi-physical field data of the oil-immersed transformer through multiple sensors, including temperature data, vibration data, UHF signal data and gas data; specifically, the temperature information of the transformer is obtained in real time by installing temperature sensors on the transformer windings, oil temperature and other parts, and vibration sensors are installed on the transformer casing and other parts to collect vibration signals during the operation of the transformer. The partial discharge signals that may be generated inside the transformer are captured using UHF signal sensors, and the composition and concentration of dissolved gases in the transformer oil, such as hydrogen, methane, ethylene, etc., are detected by gas sensors.

[0025] The dynamic calibration may further include: Linear regression (for processing linear data), support vector regression (for nonlinear data), random forest regression (an integrated model capable of processing complex relationships between features), gradient boosting (for solving high-dimensional, nonlinear problems), or neural networks (for highly complex data patterns) are used as candidate calibration algorithms. In actual applications, the appropriate algorithm can be selected for preliminary calibration based on the characteristics of the data and the calibration requirements. Evaluate calibration accuracy using mean square error, weighted mean square error, or adaptive error metrics; for example, calculate the mean square error of the data before and after calibration to evaluate the calibration effect; The mean square error (MSE) is used to measure the difference between the calibrated data and the true value. The goal is to minimize this error. The smaller the MSE, the better the calibration effect. The specific calibration method is as follows: , Where, Represents actual data, represents the calibrated data, N is the number of samples; The weighted mean square error (WMSE) assigns different weights to different sensor data to highlight the contribution of key sensors. The specific method is as follows: , Where, is the weight of sensor i, Represents actual data, represents the calibrated data, N is the number of samples; The Adaptive Error Metric (AEM) measures the error metric under specific sensor and environmental changes. It is adaptive and can be dynamically adjusted as follows: , Where, It is a dynamically adjusted function based on environment variables. Represents actual data, represents the calibrated data, N is the number of samples; Dynamically adjust the calibration model parameters by combining reinforcement learning (optimizing the calibration strategy through the interaction between the intelligent agent and the environment, maximizing the cumulative reward, and optimizing the model parameters), genetic algorithms (simulating natural selection, iteratively selecting the optimal solution, and optimizing the model hyperparameters), or particle swarm optimization (simulating the foraging process of bird flocks, searching for the optimal solution in the solution space, and optimizing the parameters of the calibration algorithm); taking reinforcement learning as an example, through continuous trial and error and feedback, the parameters of the calibration model are optimized to improve the calibration accuracy.

[0026] The above-mentioned temperature-vibration compensation equation includes the following calculation equations: , Where V is the original vibration characteristic value, unit: mm / s; V' is the vibration value after temperature compensation, unit: mm / s; T is the real-time monitoring temperature value, unit: °C; T0 is the reference temperature; α is the thermal expansion coefficient of the material, unit: β is the transient response factor, unit: mm / s·℃; t is the time variable of temperature change, unit: s; α and β are calibrated by finite element thermal coupling simulation. , .

[0027] like Figure 2 As shown, the above-mentioned gas-UHF triggering mechanism includes, S201, setting gas threshold; S202, determining whether the gas data exceeds a threshold; S203: In response to yes, high-frequency sampling is performed on the UHF signal of the transformer to capture the partial discharge pulse signal.

[0028] Here, it is determined whether the gas data exceeds a threshold; in step S204 , if the response is no, high-frequency sampling of the UHF signal of the transformer is not performed.

[0029] Among them, UHF high-frequency sampling is activated to capture partial discharge pulse signals. Specifically, the system automatically starts the high-frequency data acquisition module to capture partial discharge pulse signals in the range of 100MHz~1GHz, and outputs the collected high-frequency signals for extraction of dynamic features in subsequent processes.

[0030] like Figure 3 As shown above, the multi-physics field data of the transformer is used to generate adversarial samples based on edge GAN, and combined with the confidence level to trigger the update of the cross-physics coupling calibration model, including, S301, edge GAN generates adversarial samples based on multi-physics field data and outputs them to the cross-physics field coupling calibration model; S302, the edge GAN receives new multi-physics field data and adversarial samples in real time, and calculates the confidence of the new multi-physics field data through the discriminator; Among them, the definition formulas of the generator and discriminator of edge GAN are as follows: , Where G(·) is the generator model; D(·) is the discriminator model; z is random noise or latent variable, input to the generator; x fake is the virtual sample output by the generator; x is the real multi-physics field data sample; θ g is the model parameter set of the generator; θ d is the set of model parameters of the discriminator.

[0031] The confidence calculation formula is as follows: , Where, is the confidence level of the new data, ranging from [0,1]; Predict output values ​​for the model; is the true labeled value; is the maximum value used to normalize the relative error; S303, determining whether the confidence level of the new multi-physics field data is less than 0.85; S304, in response to "yes", triggering the update of the cross-physical field coupling calibration model, that is, compressing the cross-physical field coupling calibration model by adopting the knowledge distillation method to achieve the update of the cross-physical field coupling calibration model.

[0032] Here, it is determined whether the confidence level of the new multi-physics field data is less than 0.85; S305, if the response is no, then the cross-physics field coupling calibration model update is not triggered.

[0033] Among them, the hierarchical fusion architecture based on Shapley value performs intelligent analysis on comprehensive feature parameters and strengthens key spatiotemporal features through the dual-channel attention mechanism, including: The Shapley value is used to quantify the contribution of each feature. The calculation method is as follows: , Where, represents the average marginal contribution value of the i-th feature in all feature combinations; F represents the full set of features, which here are temperature, vibration, gas, and UHF physical field features; |F| is the number of features in set F; S is a subset of F; |S| is the number of features in set S; Feature subset, which represents a combination of some features taken from F; is the predicted value or scoring function of the model when using the feature subset S; Representation characteristics i The predicted value of the model when subset S is not added; score item It is a weighting coefficient that indicates the probability of different feature combinations occurring; The dual-channel attention mechanism includes: Spatial attention module: calculates the sensor position weight ws=softmax(MLP(xyz)), Where xyz is the three-dimensional coordinate feature vector of the sensor; is a multi-layer perceptron that extracts spatial features through nonlinear mapping; ws is the attention weight of each sensor in the spatial dimension. The larger the value, the greater the influence of the position at the current moment; : Normalization function, used to map the output to the probability distribution interval [0,1]; Temporal Attention Module: Using LSTM to Extract Temporal Importance , Where ht is the hidden state of LSTM at time step t, which represents the feature representation at the current moment; ct is the cell state of LSTM at time step t, which represents the long-term memory feature; W is a trainable weight matrix used for the linear transformation of [ht;ct]; [ht;ct] is the concatenation vector of the hidden state and the cell state; wt is the attention weight in the time dimension. The larger the value, the greater the contribution of the time step to the final prediction of the model. Sigmoid activation function maps the value to the [0,1] interval.

[0034] Among them, the temporal attention model is used to extract the importance weight wt on the time series, fuse the hidden state ht and cell state ct output by the LSTM network, and calculate the importance of each time step to the overall task.

[0035] The spatial attention model is used to calculate the spatial weights ws of different sensor positions. It maps the three-dimensional coordinate information (x, y, z) of the sensor into a weight distribution through a multi-layer perceptron (MLP), which represents the importance of each sensor position in the current analysis task.

[0036] Therefore, the dual-channel attention mechanism determines which sensor position is more critical through the spatial attention module, and then identifies which time period is more important through the temporal attention module, thereby enhancing the contribution of key spatiotemporal features in the final diagnostic model. It is used to strengthen abnormal signals (such as temperature surges or partial discharge pulses) that appear in certain sensor nodes in specific time periods, thereby improving the accuracy of fault diagnosis.

[0037] Among them, if the confidence level of the new multi-physics data is less than 0.85, the cross-physics coupling calibration model is updated. That is, the cross-physics coupling calibration model is compressed by using the knowledge distillation method to achieve the cross-physics coupling calibration model update, including: The dynamic update of the cross-physics coupling calibration model is performed using the following formula: , Where, the KL divergence term KL(pteacher|pstudent) measures the difference in output distribution between the student model and the teacher model, and is used to ensure that the student model can imitate the feature judgment ability of the teacher model as much as possible; CrossEntropy: measures the difference between the student model's predictions and the true labeled values, ensuring that the student model can correctly classify or predict the actual multiphysics data state. : Weight parameter used to balance the influence of teacher model guidance (KL term) and true label supervision (CrossEntropy term).

[0038] The above uses student model and teacher model, specifically: Teacher Model: This is a high-precision, complex benchmark model used for initial training based on a large amount of historical data. It can accurately characterize the multi-physics field characteristics and failure modes of oil-immersed transformers.

[0039] In this paper, the prediction results of the teacher model pteacher are the guiding criteria for dynamic calibration and adversarial sample generation.

[0040] Student Model: A lightweight model designed to run quickly on real-time edge devices. It learns the behavior of the teacher model through knowledge distillation, maintaining high inference speed and a certain level of accuracy.

[0041] The output probability distribution pstudentp of the student model remains consistent with the teacher model through the Loss function.

[0042] The above-mentioned dynamic features of each physical field data after dynamic calibration are extracted, including: Use sliding window techniques to calculate trend changes, volatility, and rate of change in temperature data; Perform frequency domain analysis and time domain statistics extraction on vibration data; Perform wavelet transform on UHF signal data to extract partial discharge features; Calculate the concentration change trend and key gas component ratio of gas data.

[0043] The temperature data extraction process may include: calculating the trend (linear regression slope), first-order difference (rate of change), and volatility (standard deviation); The vibration data extraction process may include: combining time domain statistics (mean, variance) with frequency domain (FFT) and wavelet transform to extract local time-frequency features; The gas data extraction process may include: extracting concentration trends and fluctuations, combined with changes in gas composition ratios; The UHF signal data extraction process may include: extracting the time-frequency characteristics and abnormal mutation indicators of the partial discharge signal through wavelet transformation.

[0044] The above calculations calculate the comprehensive diagnostic score and determine the fault type, including: The composite diagnostic score is calculated as follows: Assume that the characteristics of each physical field are 、 、 、 , each single domain fault index function is 、 、 、 , the weight is 、 、 、 ;The bias term is b; The comprehensive diagnostic score formula is: , For fault judgment, the following logic function is used: , Where P(fault|X) is the probability of a fault occurring under input feature X; D is the comprehensive diagnosis score; is an exponential function, used to implement the S-shaped curve of logistic regression; When the temperature and gas weights in the comprehensive diagnosis score D are significantly higher, and When , it is determined to be an overheating fault; When the combined diagnostic score of the vibration and UHF parts exceeds the set threshold, it is determined to be a mechanical failure; When the score of the UHF part is significant and there is a combined effect of temperature and gas anomaly indicators, it is determined to be an insulation fault; When the comprehensive diagnosis score D is significantly higher, it corresponds to , and when all single domain indicators are abnormal at the same time, it is determined to be a multi-factor coupling fault.

[0045] Among them, the weight is 、 、 、 Used to measure the importance of each physical field feature; is the fault index function of temperature characteristics, which represents the Nonlinear mapping or scoring, is the fault index function of the vibration characteristic, which represents the Nonlinear mapping or scoring, is the fault index function of gas characteristics, which represents the Nonlinear mapping or scoring, is the fault indicator function of UHF characteristics, which represents the The nonlinear mapping or scoring of ; the bias term b is used to adjust the benchmark of the comprehensive diagnostic score.

[0046] Among them, the characteristic conditions of overheating fault include: The temperature trend is greater than the set threshold (e.g. slope > 0.2); Excessive temperature fluctuation (standard deviation outside the normal range); At the same time, the concentration of combustible gases (such as hydrogen and acetylene) in the gas increases.

[0047] Among them, the characteristic conditions of mechanical failure include: Abnormal peaks (abnormal frequency components) appear in the spectrum of the vibration signal; The volatility and rate of change of vibration data deviate significantly from the historical mean; When partial discharge occurs in combination with UHF signals, it is easier to determine that it is a mechanical failure or poor contact.

[0048] Among them, the characteristic conditions of insulation fault include: The obvious partial discharge characteristics of UHF signals were extracted by wavelet transform; It may also be accompanied by a slight rise in temperature and changes in the concentration of abnormal gases in the gas (such as acetylene); Vibration data may show weak anomalies, but they are mainly UHF.

[0049] Among them, the characteristic conditions of multi-factor coupling failure include: Abnormal indicators exist in multiple physical fields, such as continuous temperature rise, abnormal vibration, gas anomalies, and partial discharge. The fusion diagnosis engine gives a high comprehensive score.

[0050] Specifically: When the transformer winding temperature continues to rise and the concentration of gases such as carbon monoxide and carbon dioxide in the dissolved gas in the oil increases, it is judged as an overheating fault; when the vibration spectrum is abnormal and accompanied by a sudden change in the UHF signal, it is judged as a mechanical fault; if an abnormal peak of a specific frequency appears in the vibration signal and the UHF signal suddenly increases, it is judged as a mechanical fault; when the UHF signal is significantly abnormal and the gas composition changes, it is judged as a partial discharge fault; if the UHF signal intensity increases significantly and the concentration of gases such as hydrogen and acetylene in the dissolved gas in the oil increases, it is judged as a partial discharge fault; when multiple physical field data are abnormal at the same time, it is judged as a multi-factor coupling fault.

[0051] In summary, the present invention fuses data from multiple sensors, performs dynamic calibration by adopting a cross-physical field coupling calibration model, generates adversarial samples using edge GAN, and combines confidence-triggered cross-physical field coupling calibration model updates with a hierarchical fusion architecture based on Shapley values ​​to perform intelligent analysis of comprehensive characteristic parameters. This can accurately determine the health status of the transformer and improve the accuracy of fault diagnosis.

[0052] Example 2: Figure 4 As shown, an embodiment of the present invention discloses a dynamic calibration system for multi-physical field data fusion of an oil-immersed transformer, comprising: The data calibration unit dynamically calibrates the transformer's multi-physics data using a cross-physics coupling calibration model, and evaluates the calibration results based on an objective function. The multi-physics data includes temperature, vibration, UHF signal, and gas data. The cross-physics coupling calibration model includes establishing a temperature-vibration compensation equation and a gas-UHF trigger mechanism. The edge GAN unit generates adversarial samples based on the multi-physics field data of the transformer and triggers the update of the cross-physics coupling calibration model in combination with the confidence level; The analysis and diagnosis unit extracts the dynamic features of each physical field data after dynamic calibration, generates comprehensive feature parameters, performs intelligent analysis on the comprehensive feature parameters based on the hierarchical fusion architecture of Shapley values, strengthens key spatiotemporal features through the dual-channel attention mechanism, calculates the comprehensive diagnosis score, determines the fault type, and generates a health assessment result.

[0053] like Figure 5 As shown, the above-mentioned edge GAN unit includes a sample generation module, a calculation module, a judgment module and a response module; In the sample generation module, the edge GAN generates adversarial samples based on multi-physics field data and outputs them to the cross-physics field coupling calibration model; In the computing module, the edge GAN receives new multi-physics field data and adversarial samples in real time, and calculates the confidence level of the new multi-physics field data through the discriminator; A judgment module determines whether the confidence level of new multi-physics field data is less than 0.85; The response module, when responding, triggers the update of the cross-physical field coupling calibration model, that is, the cross-physical field coupling calibration model is compressed by adopting the knowledge distillation method to realize the update of the cross-physical field coupling calibration model.

[0054] The above-mentioned dynamic calibration system for multi-physical field data fusion of oil-immersed transformers also includes an operation and maintenance management platform, which may include: a data visualization interface, which displays multi-physical field data and health status in real time, and intuitively displays the operating data and health status of the transformer in the form of charts, curves, etc.; an early warning push module, which sends fault warning information to operation and maintenance personnel. When the transformer has an abnormality, the early warning information is sent to the operation and maintenance personnel in a timely manner through text messages, emails, etc.; an intelligent decision support module, which provides maintenance suggestions and optimizes operation and maintenance strategies, and provides corresponding maintenance suggestions and operation and maintenance strategies according to the health status and fault type of the transformer to improve operation and maintenance efficiency.

[0055] Through the collection, dynamic calibration, feature extraction, and fusion analysis of multi-physics field data, intelligent fault diagnosis and health status assessment of oil-immersed transformers are achieved. At the same time, the operation and maintenance management platform provides intuitive data display and decision support for operation and maintenance personnel, improving the operational reliability and efficiency of the transformer.

[0056] Embodiment 3: A storage medium stores a computer program that can be read by a computer, wherein the computer program is configured to execute a dynamic calibration method for multi-physical field data fusion of an oil-immersed transformer when the computer program is run.

[0057] The above storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory, a mobile hard disk, a magnetic disk, or an optical disk.

[0058] Example 4: An embodiment of the present invention discloses an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the computer program is loaded and executed by the processor to implement a dynamic calibration method for multi-physical field data fusion of an oil-immersed transformer.

[0059] The electronic device further includes a transmission device and an input / output device, wherein the transmission device and the input / output device are both connected to the processor.

[0060] The processor may be a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), an ASIC, an FPGA, or other programmable logic device, a transistor logic device, a hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. It may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and so on. Memory may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memories, removable hard drives, magnetic disks, or optical disks.

[0061] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiment of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal translation scripting language JavaScript, etc.

[0062] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0063] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0064] The present application also provides a computer program product comprising a non-transitory computer-readable storage medium storing a computer program, wherein the computer program is operable to cause a computer to perform some or all of the steps of any of the methods described in the above method embodiments. The computer program product may be a software installation package, and the computer may comprise an electronic device.

[0065] Example 5: An embodiment of the present invention discloses a terminal, including a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the programs include instructions for executing steps in a dynamic calibration method for multi-physical field data fusion of an oil-immersed transformer.

[0066] The above technical features constitute the embodiments of the present invention, which have strong adaptability and optimal implementation effects. Non-essential technical features can be added or removed according to actual needs to meet the requirements of different situations.

Claims

1. A dynamic calibration method for multi-physics field data fusion of an oil-immersed transformer, characterized in that: The following steps are involved: The transformer's multi-physics data is dynamically calibrated using a cross-physics coupling calibration model, and the calibration effect is evaluated based on an objective function. The multi-physics data includes temperature data, vibration data, UHF signal data, and gas data. The cross-physics coupling calibration model includes establishing a temperature-vibration compensation equation and a gas-UHF trigger mechanism. Generate adversarial samples based on edge GAN using the transformer's multi-physics data, and trigger the update of the cross-physics coupling calibration model based on confidence. The dynamic features of each physical field data after dynamic calibration are extracted to generate comprehensive feature parameters. The comprehensive feature parameters are intelligently analyzed based on the hierarchical fusion architecture of Shapley values. The key spatiotemporal features are enhanced through the dual-channel attention mechanism, and the comprehensive diagnostic score is calculated to determine the fault type.

2. The dynamic calibration method for multi-physics field data fusion of an oil-immersed transformer according to claim 1 is characterized in that: The temperature-vibration compensation equation is established, including , the calculation equation is as follows: , Where, V is the original vibration characteristic value, unit: mm / s; V' is the vibration value after temperature compensation, unit: mm / s; T is the real-time monitored temperature value, unit: °C; T0 is the reference temperature; α is the thermal expansion coefficient of the material, unit: ; β is the transient response factor, unit: mm / s·℃; t is the time variable of temperature change, unit: s; α and β are calibrated by finite element thermal coupling simulation. , .

3. The dynamic calibration method for multi-physics field data fusion of an oil-immersed transformer according to claim 1 is characterized in that: The gas-UHF trigger mechanism includes, Set gas thresholds; Determine whether the gas data exceeds the threshold; In response to this, the UHF signal of the transformer is sampled at high frequency to capture the partial discharge pulse signal.

4. The dynamic calibration method for multi-physics field data fusion of an oil-immersed transformer according to claim 1 is characterized in that: The multi-physics data of the transformer is used to generate adversarial samples based on edge GAN, and combined with confidence to trigger the update of the cross-physics coupling calibration model, including: Edge GAN generates adversarial samples based on multi-physics field data and outputs them to the cross-physics field coupling calibration model; Edge GAN receives new multi-physics field data and adversarial samples in real time, and calculates the confidence of the new multi-physics field data through the discriminator; The confidence calculation formula is as follows: , Where, is the confidence level of the new data, ranging from [0,1]; Predict output values ​​for the model; is the true labeled value; is the maximum value used to normalize the relative error; Determine whether the confidence level of the new multi-physics field data is less than 0.85; In response to this, the update of the cross-physical field coupling calibration model is triggered, that is, the cross-physical field coupling calibration model is compressed by adopting the knowledge distillation method to achieve the update of the cross-physical field coupling calibration model.

5. The dynamic calibration method for multi-physics field data fusion of an oil-immersed transformer according to claim 1 is characterized in that: The hierarchical fusion architecture based on Shapley values ​​performs intelligent analysis on comprehensive feature parameters and strengthens key spatiotemporal features through a dual-channel attention mechanism, including: The Shapley value is used to quantify the contribution of each feature. The calculation method is as follows: ， Where, represents the average marginal contribution value of the i-th feature in all feature combinations; F represents the full set of features, which here are temperature, vibration, gas, and UHF physical field features; |F| is the number of features in set F; S is a subset of F; |S| is the number of features in set S; Feature subset, which represents a combination of some features taken from F; is the predicted value or scoring function of the model when using the feature subset S; Representation characteristics i The predicted value of the model when subset S is not added; score item It is a weighting coefficient that indicates the probability of different feature combinations occurring; The dual-channel attention mechanism includes: Spatial attention module: calculates the sensor position weight ws=softmax(MLP(xyz)), Where xyz is the three-dimensional coordinate feature vector of the sensor; is a multi-layer perceptron that extracts spatial features through nonlinear mapping; ws is the attention weight of each sensor in the spatial dimension. The larger the value, the greater the influence of the position at the current moment; : Normalization function, used to map the output to the probability distribution interval [0,1]; Temporal Attention Module: Using LSTM to Extract Temporal Importance , Where ht is the hidden state of LSTM at time step t, which represents the feature representation at the current moment; ct is the cell state of LSTM at time step t, which represents the long-term memory feature; W is a trainable weight matrix used for the linear transformation of [ht;ct]; [ht;ct] is the concatenation vector of the hidden state and the cell state; wt is the attention weight in the time dimension. The larger the value, the greater the contribution of the time step to the final prediction of the model. Sigmoid activation function maps the value to the [0,1] interval.

6. The dynamic calibration method for multi-physics field data fusion of an oil-immersed transformer according to claim 1 is characterized in that: The extracting of dynamic features of each physical field data after dynamic calibration includes: Use sliding window techniques to calculate trend changes, volatility, and rate of change in temperature data; Perform frequency domain analysis and time domain statistics extraction on vibration data; Perform wavelet transform on UHF signal data to extract partial discharge features; Calculate the concentration change trend and key gas component ratio of gas data.

7. The dynamic calibration method for multi-physics field data fusion of an oil-immersed transformer according to claim 1 is characterized in that: The calculation of the comprehensive diagnosis score and determination of the fault type include: The composite diagnostic score is calculated as follows: Assume that the characteristics of each physical field are 、 、 、 , each single domain fault index function is 、 、 、 , the weight is 、 、 、 ;The bias term is b; The comprehensive diagnostic score formula is: ， For fault judgment, the following logic function is used: ， Where P(fault|X) is the probability of a fault occurring under input feature X; D is the comprehensive diagnosis score; is an exponential function, used to implement the S-shaped curve of logistic regression; When the temperature and gas weights in the comprehensive diagnosis score D are significantly higher, and When , it is determined to be an overheating fault; When the combined diagnostic score of the vibration and UHF parts exceeds the set threshold, it is determined to be a mechanical failure; When the score of the UHF part is significant and there is a combined effect of temperature and gas anomaly indicators, it is determined to be an insulation fault; When the comprehensive diagnosis score D is significantly higher, it corresponds to , and when all single domain indicators are abnormal at the same time, it is determined to be a multi-factor coupling fault.

8. A dynamic calibration system for multi-physics field data fusion of oil-immersed transformers, characterized by: A dynamic calibration method for multi-physical field data fusion of an oil-immersed transformer according to any one of claims 1 to 7, comprising: The data calibration unit dynamically calibrates the transformer's multi-physics data using a cross-physics coupling calibration model, and evaluates the calibration results based on an objective function. The multi-physics data includes temperature, vibration, UHF signal, and gas data. The cross-physics coupling calibration model includes establishing a temperature-vibration compensation equation and a gas-UHF trigger mechanism. The edge GAN unit generates adversarial samples based on the multi-physics field data of the transformer and triggers the update of the cross-physics coupling calibration model in combination with the confidence level; The analysis and diagnosis unit extracts the dynamic features of each physical field data after dynamic calibration, generates comprehensive feature parameters, performs intelligent analysis on the comprehensive feature parameters based on the hierarchical fusion architecture of Shapley values, strengthens key spatiotemporal features through the dual-channel attention mechanism, calculates the comprehensive diagnosis score and determines the fault type.

9. The dynamic calibration system for multi-physics field data fusion of an oil-immersed transformer according to claim 8, characterized in that: The edge GAN unit includes a sample generation module, a calculation module, a judgment module, and a response module; In the sample generation module, the edge GAN generates adversarial samples based on multi-physics field data and outputs them to the cross-physics field coupling calibration model; In the computing module, the edge GAN receives new multi-physics field data and adversarial samples in real time, and calculates the confidence level of the new multi-physics field data through the discriminator; A judgment module determines whether the confidence level of new multi-physics field data is less than 0.85; The response module, when responding, triggers the update of the cross-physical field coupling calibration model, that is, the cross-physical field coupling calibration model is compressed by adopting the knowledge distillation method to realize the update of the cross-physical field coupling calibration model.

Images