Method, system and device for predicting internal moisture content of transformer

Abstract

The present application belongs to the technical field of transformer monitoring, in order to solve the poor real-time performance and inaccuracy of the existing transformer micro-water prediction, a transformer internal micro-water content prediction method, system and equipment are proposed, through simulation and simulation of the transformer multi-physical field coupling model, the correlation between the micro-water content of different weak points and the oil outlet under different operating parameters and / or external environmental parameters of the transformer is obtained, and then multiple training sample sets are constructed, and based on the differences in micro-water diffusion characteristics and the degree of influence of temperature / electric field of each weak point, a transformer micro-water content dynamic deduction model is constructed, avoiding the average processing of a single model for different parts characteristics, and more accurately capturing the specific correlation between each weak point and the micro-water content of the oil outlet, improving the accuracy of the deduction result.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of transformer monitoring, and particularly relates to a transformer internal micro-water content prediction method, system and device. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] Due to moisture intrusion and insulation aging, moisture widely exists in the insulating oil and insulating paperboard of the transformer, and is one of the most common insulation defects in oil-paper insulation. The increase of the moisture content of the oil-paper insulation not only accelerates the aging process of the oil-paper insulation, but also causes electric field distortion, induces and accelerates partial discharge. In engineering applications, offline detection means are often used for measurement, i.e. samples are taken from the oil outlet of the transformer during power-off maintenance or sampling analysis, and the water content in the solid insulation is estimated through Karl Fischer titration and oil-paper balance relationship curve. Offline detection can only take samples during power-off maintenance, and cannot realize real-time monitoring of the solid insulation of the transformer; offline detection means is indirect detection, which relies on the oil-paper balance curve under ideal conditions to estimate the water content in the solid insulation paperboard, while in actual operation, the temperature and moisture distribution are uneven, and it is difficult to achieve ideal balance state, resulting in poor calculation accuracy; only sampling from the oil outlet can only reflect the micro-water content of the insulating oil at the oil outlet, and cannot represent the water content distribution of the insulating oil at various places in the transformer, and it is more difficult to determine the position of the weak insulation inside the transformer.

[0004] In addition, online monitoring means only exists as auxiliary detection, i.e. installing sensors such as distributed optical fiber sensors, miniature humidity sensors, etc. However, the transformer is subjected to the coupling of multiple physical fields such as electric field, temperature field and fluid field during operation, and the distribution of each physical field is extremely uneven, the voltage grade is high, and the harmonic content is large, which puts higher requirements on the insulation level, high temperature resistance and corrosion resistance of the sensor, and further interferes with the sensor detection process and data accuracy. In addition, the internal space of the transformer is compact and the insulation structure is numerous, and it is difficult to install small-sized physical sensors at the weak insulation, and the actual problem of sensing the water content distribution at different positions of the transformer cannot be solved.

[0005] Therefore, how to realize accurate capture of the dynamic micro-water distribution of the transformer without relying on a large number of physical sensors susceptible to interference of multiple physical fields in a complex internal environment, while avoiding the indirect estimation error based on the oil-paper balance assumption and the sampling limitations in offline detection, is a problem that needs to be solved at present. SUMMARY

[0006] In order to overcome the above-mentioned deficiencies of the prior art, the present application provides a transformer internal micro-water content prediction method, system and device, a plurality of transformer micro-water content dynamic deduction models are constructed, the specificity correlation of the micro-water content of each weak point of the transformer and the oil outlet is accurately captured, and the accuracy of the results is improved.

[0007] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0008] In a first aspect, the present application provides a transformer internal micro-water content prediction method, comprising:

[0009] By simulating the transformer multi-physical field coupling model, the correlation of the micro-water content of different insulation weak points and the oil outlet under different operating parameters and / or external environmental parameters of the transformer is obtained, and a plurality of training data sets are constructed;

[0010] Based on the plurality of training sample sets, the regression decision tree model is trained to obtain the transformer micro-water content dynamic deduction model corresponding to different insulation weak points;

[0011] Based on the obtained micro-water content of the oil outlet of the transformer to be predicted, the transformer micro-water content dynamic deduction model corresponding to different insulation weak points is used to obtain the micro-water content of different insulation weak parts inside the transformer.

[0012] In a second aspect, the present application provides a transformer internal micro-water content prediction system, comprising:

[0013] The construction module is configured to simulate the transformer multi-physical field coupling model, obtain the correlation of the micro-water content of different insulation weak points and the oil outlet under different operating parameters and / or external environmental parameters of the transformer, and construct a plurality of training data sets;

[0014] The training module is configured to train the regression decision tree model based on the plurality of training sample sets to obtain the transformer micro-water content dynamic deduction model corresponding to different insulation weak points;

[0015] The prediction module is configured to obtain the micro-water content of different insulation weak parts inside the transformer based on the obtained micro-water content of the oil outlet of the transformer to be predicted, and use the transformer micro-water content dynamic deduction model corresponding to different insulation weak points.

[0016] In a third aspect, the present application provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and running on the processor, when the computer instructions are run by the processor, the method of the first aspect is completed.

[0017] The above one or more technical solutions have the following beneficial effects:

[0018] In the present application, by means of a transformer multi-physical field coupling model, simulation is carried out to obtain the correlation between different weak points and the micro water content at the oil extraction port under different operating parameters and / or external environmental parameters of the transformer, and then a plurality of training sample sets are constructed, and based on the differences in the micro water diffusion characteristics and the degree of influence of temperature / electric field of each weak point, a transformer micro water content dynamic deduction model is respectively constructed, which avoids the averaging processing of a single model for different part characteristics, can more accurately capture the specific correlation between each weak point and the micro water content at the oil extraction port, and improves the accuracy of the deduction result.

[0019] In the present application, the dynamic deduction model based on the regression decision tree can combine the incremental updating mechanism, and when new operating data such as new micro water content data at the oil extraction port arrives, the node weight is adjusted to adapt to the data change. This feature enables the model to track the dynamic changes of the internal micro water content of the transformer for a long time, avoids the decrease in deduction accuracy caused by equipment aging and working condition evolution, and ensures the reliability in long-term operation.

[0020] In the present application, the tree complexity regularization term punishes the number of nodes of the regression decision tree, avoids generating too many branches to fit the training data, prevents overfitting, enables the model to more accurately deduce new data while maintaining good fitting to the training data, and improves the generalization ability to the actual scene; the time decay factor reduces the weight of old data, so that the regression decision tree model pays more attention to the change of the operating state of the transformer reflected by the new data. This feature is particularly suitable for long-term running transformers, the internal micro water distribution of which will dynamically evolve over time. The time decay factor can help the model adjust the dependence on historical data in time, ensure that the deduction result is consistent with the current state of the equipment, and improve the accuracy of long-term monitoring.

[0021] In the present application, the loss drop of the fitting piecewise linear model and the split point is calculated when the node is split, and the loss drop can quantify the effect of different split points, so that the node splitting has a clear optimization goal. At the same time, the fitting result of the piecewise linear model can intuitively reflect the trend difference of the data on both sides of the split point, enhance the interpretability of the decision tree node splitting, facilitate understanding of the processing logic of the regression decision tree model for the transformer micro water content and other variables, and the clear split logic also facilitates the combination of the time decay factor to more efficiently adjust the node weight during incremental updating, and improves the adaptability of the model to dynamic data.

[0022] The advantages of the additional aspects of the present application will be partially given in the following description, partially will become obvious from the following description, or will be learned by the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0023] The accompanying drawings, which form a part of this specification, are included to provide a further understanding of the application and are incorporated herein by reference. The illustrations are of exemplary embodiments of the application and explain the principles of the application, but do not limit the application.

[0024] Figure 1 The flow chart of the transformer internal micro water content prediction method in the embodiment of the application. DETAILED DESCRIPTION

[0025] It should be noted that the following detailed description is exemplary in nature and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0026] It should be noted that the terms used herein are only intended to describe specific embodiments and are not intended to limit the exemplary embodiments according to the application.

[0027] In the case of no conflict, the embodiments in the application and the features in the embodiments can be combined with each other.

[0028] Embodiment one

[0029] The embodiment discloses a transformer internal micro water content prediction method, comprising:

[0030] By simulating the transformer multi-physical field coupling model, the correlation between the micro water content at different insulation weak points and the oil outlet under different operating parameters and / or external environmental parameters of the transformer is obtained, and multiple training data sets are constructed;

[0031] Based on the multiple training sample sets, a regression decision tree model is trained to obtain a transformer micro water content dynamic deduction model corresponding to different insulation weak points;

[0032] Based on the obtained micro water content at the oil outlet of the transformer to be predicted, the transformer micro water content dynamic deduction model corresponding to different insulation weak points is used to obtain the micro water content at different insulation weak points inside the transformer.

[0033] In this embodiment, by simulating the transformer multi-physical field coupling model, the correlation between the micro water content at different insulation weak points and the oil outlet under different operating parameters and / or external environmental parameters of the transformer is obtained, and multiple training sample sets are constructed. Based on the differences in micro water diffusion characteristics and the degree of influence of temperature / electric field of each weak point, a transformer micro water content dynamic deduction model is constructed, which avoids the average processing of a single model for different parts characteristics, can more accurately capture the specific correlation between each weak point and the micro water content at the oil outlet, and improves the accuracy of the deduction result.

[0034] The transformer internal micro water content prediction method proposed in the embodiment is described in detail as follows:

[0035] Step 1: Obtain the correlation between the micro water content at different weak insulation points and the oil extraction port under different operating parameters and / or external environmental parameters of the transformer by simulating the transformer multi-physical field coupling model, and construct multiple training data sets.

[0036] In the embodiment, based on the constructed two-dimensional finite element model of the transformer, the electric field distribution and power loss are solved; the temperature field distribution is calculated based on the power loss obtained, and the material parameters are based on the temperature field distribution, so as to realize the coupling of the electric field and the temperature field; the correlation between the oil-paper insulation micro water diffusion and the temperature field is established based on the Fick's second diffusion law, the distribution of the micro water in the transformer oil and the insulation paper is simulated, the material parameters are changed based on the micro water distribution, and the electric field and the temperature field are reacted, so as to realize the coupling of the temperature field and the fluid field.

[0037] Firstly, the geometric size parameters of the transformer are obtained, and a transformer simulation grid model is constructed, specifically: through the cooperative calling of the grid modeling interface of MATLAB and COMSOL, the transformer geometric model is quickly and efficiently established by using programming language.

[0038] The MATLAB and COMSOL cooperative calling of the grid modeling interface includes: when modeling the transformer in COMSOL, the COMSOL Multiphysics with MATLAB coordination software interface is combined. The program command of the MATLAB control file is programmed by using JAVA language, so that the geometric model of the transformer can be controlled and drawn through MATLAB, and the transformer geometric model is quickly and accurately established. The method of manual partitioning instead of system automatic partitioning is adopted to complete the targeted partitioning of the key areas and ensure the convergence and efficiency of the simulation calculation.

[0039] The triangular network is selected for the initial partitioning of the geometric model network, and the key boundaries and areas are subjected to multiple super-fine partitioning, so as to complete the overall grid partitioning of the transformer model.

[0040] Obtain the experimental data of the related materials of the transformer at different temperatures, such as the variation law of the electrical conductivity, the relative dielectric constant and the like with temperature, and fit the nonlinear variation curve of the material parameters with temperature by the least square method.

[0041] The nonlinear curve of the material parameters changing with temperature includes: the values of the relative dielectric constant and the electrical conductivity of the transformer insulating oil and insulating paper at different temperatures are obtained, and the functional relationship of the relative dielectric constant γ and the electrical conductivity σ changing with temperature T is obtained by the least square method, wherein c represents the micro water content in the transformer oil-paper insulation, as shown in Table 1.

[0042] Table 1:

[0043]

[0044] The electromagnetic field dynamic parameters of the transformer two-dimensional finite element model are set, the power loss and electric field distribution are solved by using the electromagnetic analysis module, and then the temperature field distribution is obtained, which determines the material parameters at each position.

[0045] The Fick second diffusion law is used to establish the influence model of oil-paper insulation micro-water diffusion and temperature field, the fluid field adopts the dilute substance transfer module, and finally the micro-water content at each position in the fluid field directly affects the material parameters, realizing the coupling of electric field, temperature field and fluid field.

[0046] Among them, the power loss includes core loss and winding loss. The core loss includes hysteresis loss, eddy current loss and abnormal loss, and the winding loss includes ohmic loss and eddy current loss.

[0047] The core loss expression is shown in formula (1):

[0048] (1)

[0049] Among them, is the sum of hysteresis loss and abnormal loss, is the eddy current loss. B is a function of magnetic flux density, s is the thickness ratio of silicon steel sheet, f represents the frequency. Coefficients , , α and β are obtained by fitting the input loss curve data.

[0050] When containing high-order harmonics, the eddy current loss under distorted magnetic flux is derived as shown in formula (2):

[0051] (2)

[0052] Among them, is the conductivity of silicon steel sheet, is the thickness of single silicon steel sheet, NH is the highest number of harmonics, is the amplitude of the th harmonic.

[0053] The ohmic loss calculation formula of the winding under harmonic condition is shown in formula (3):

[0054] (3)

[0055] Among them, is the ratio of the th harmonic current to the fundamental current effective value . The ohmic loss of the winding under the fundamental condition.

[0056] The formula for calculating the eddy current loss of the winding under the harmonic condition is shown as formula (4):

[0057] (4)

[0058] Wherein, is the eddy current loss of the winding under the fundamental frequency, is the ratio of the harmonic current frequency to the fundamental current frequency .

[0059] The mathematical expression of Fick's second diffusion law is:

[0060] (5)

[0061] In the formula, is the micro-water concentration (%) in the insulation paper, is the diffusion coefficient of micro-water in the insulation paper (m 2 / s), T represents the temperature, represents the gradient.

[0062] The empirical formula for calculating the diffusion coefficient D is shown as formula (6):

[0063] (6)

[0064] In the formula, T0 is the reference temperature; Ea is the activation energy (K) in the diffusion process; D0 is a pre-exponential factor, with the unit of m 2 / s; k is a dimensionless parameter.

[0065] The oil-paper insulation micro-water diffusion calculation model in Fick's second diffusion law and the empirical formula for calculating the micro-water diffusion coefficient D are input into the physical field of rare substance transfer, to build the heat-mass coupling interface, so as to realize the docking of the relevant interfaces of the electric-thermal coupling model to the heat-mass coupling model.

[0066] By establishing the influence relationship between the temperature field and the micro-water diffusion, the internal heat-mass transfer coupling model of the transformer is established, and the electric-thermal-flow multi-physical field coupling model is built. Finally, the COMSOL free triangular grid is used for grid partitioning, the volume heat source is applied by using the Joule heat calculated in the foregoing, the boundary of the solid heat transfer module is set as the convective heat flux 28 , the porosity of the oil-paper insulation boundary of the porous rare substance transfer module is set to 60%, and the transient field is used for simulation.

[0067] ​In this embodiment, according to the constructed transformer electric-thermal-water multi-physical field coupling model, the internal micro-water content distribution of the converter transformer under normal operating state is obtained, and according to the actual engineering experience, a plurality of micro-water content concentration places are selected as typical insulation weak positions.

[0068] Different transformer operating parameters and external environment parameters are set to obtain the correlation between the micro-water content at different insulation weak positions and the oil outlet. Among them, the operating parameters of the converter transformer are set, such as normal state, damp state and typical defects. A plurality of different typical defects are set at the insulation weak positions, such as burr defect, recess defect, metal discontinuity defect and the like. The micro-water diffusion characteristics are affected by the environmental temperature, environmental humidity, voltage and current on the network side and the valve side, and by changing these external parameters, the micro-water content distribution of the converter transformer under different external parameters is obtained.

[0069] In this embodiment, the correlation between the micro-water content at different insulation weak positions and the oil outlet, that is, the micro-water content at different insulation weak positions and the oil outlet of the transformer under different transformer operating parameters and external environment parameters is constructed into a data set, and a fitting curve of the micro-water content at different insulation weak positions and the oil outlet is established.

[0070] Step 2: Based on a plurality of training sample sets, a regression decision tree model is trained to obtain a transformer micro-water content dynamic deduction model corresponding to different insulation weak positions; based on the obtained micro-water content at the oil outlet of the transformer to be predicted, the transformer micro-water content dynamic deduction model corresponding to different insulation weak positions is used to obtain the micro-water content at different insulation weak positions inside the transformer.

[0071] The training data set established in step 1 is imported into the improved regression decision tree algorithm for training, and then a transformer micro-water content dynamic deduction model is constructed. Based on the transformer micro-water content dynamic deduction model, the micro-water content at different insulation weak positions inside the converter transformer can be deduced according to the micro-water content at the oil outlet.

[0072] The regression decision tree is selected in this embodiment because the correlation between the micro-water content at the oil outlet and the insulation weak position presents a linear relationship, and the regression decision tree is suitable for generating a decision tree model.

[0073] The principle of the regression decision tree is that: X and Y are input and output variables, respectively, and Y is a continuous variable, and in this embodiment, the mass concentration of micro-water substances in the transformer oil at the oil outlet is taken as the input variable, and the mass concentration of micro-water substances in the transformer oil paper at the insulation weak position is taken as the output variable.

[0074] The given training set is , and the feature parameter set is , is the number of feature parameters, is the sample size. The heuristic method is used to divide the feature space, and the optimal feature parameter is determined as the split point according to the principle of minimizing the square error in each process.

[0075] To overcome the shortcomings of overfitting of traditional algorithms, the tree complexity regularization term and the time decay factor are added to the loss function in the embodiment:

[0076] (7)

[0077] wherein, is the number of nodes, which controls the complexity of the regression decision tree model; is the predicted value of the node; and is the time decay coefficient; is the regularization strength parameter, represents the loss function, represents the true value of the i-th sample, i.e. the moisture content of the weak insulation; represents the predicted value of the i-th sample; t represents the time variable.

[0078] Based on the information loss caused by greedy splitting, the embodiment introduces piecewise linear regression and spline function to improve the processing accuracy of continuous variables. When splitting the node, a piecewise linear model is fitted, the loss reduction of the split point s is calculated, a linear function is fitted for the data set on both sides of the split point, and the optimal split point is selected by comparing the loss change before and after the split. The loss reduction of the split point s is calculated as:

[0079] (8)

[0080] wherein, and are the left and right fitting values of the piecewise linear function respectively; s is the split point, a node, and a split point is selected as S when calculating the loss function, which aims to improve the processing accuracy of continuous variables; L refers to the loss function of the previous step.

[0081] For continuous variables such as temperature, T cubic spline interpolation is used:

[0082] …(9)

[0083] wherein, is the spline node, which is optimized by minimizing the residual sum of squares; , , , , are coefficients of the corresponding expanded expression.

[0084] In this embodiment, the online learning mechanism is combined to enhance dynamic adaptability to overcome the batch learning of traditional decision trees, which cannot adapt to dynamic data flow. Specifically, an incremental updating rule is introduced to adjust the node weight. When new data arrives, only the node weight of the relevant path is updated:

[0085] (10)

[0086] wherein, is the learning rate, represents new data, represents the predicted value of the new data based on the old weight, the node weight before updating.

[0087] If the new data causes the error of a certain node to continuously increase, local re-splitting is triggered, and only the node and its subtree are adjusted to adjust the local structure.

[0088] In addition, an adaptive learning rate strategy is introduced to dynamically adjust the learning rate according to the error fluctuation amplitude of the new data and the historical data. When the error fluctuation is large, the learning rate is reduced to avoid overshoot, and when the fluctuation is small, the learning rate is increased to accelerate convergence and improve the tracking ability of the model to the latest state.

[0089] In this embodiment, the training data set is imported into the regression decision tree model for training, and the verification data group is set to 30% of the data set for deducing the model output value verification to improve the training and verification accuracy. The fitting degree of the deducing model is judged by the root mean square error of the deducing model.

[0090] Embodiment Two

[0091] The purpose of this embodiment is to provide a transformer internal micro-water content prediction system, comprising:

[0092] A construction module is configured to: obtain the correlation between the micro-water content at different insulation weak points and oil outlets under different operating parameters and / or external environmental parameters of the transformer by simulating the transformer multi-physical field coupling model, and construct multiple training data sets;

[0093] A training module is configured to: train a regression decision tree model based on multiple training sample sets to obtain a transformer micro-water content dynamic deduction model corresponding to different insulation weak points;

[0094] A prediction module is configured to: based on the obtained micro-water content at the oil outlet of the transformer to be predicted, use the transformer micro-water content dynamic deduction model corresponding to different insulation weak points to obtain the micro-water content of different insulation weak parts inside the transformer.

[0095] In more embodiments, there are also provided:

[0096] An electronic device comprising a memory and a processor and computer instructions stored on the memory and running on the processor, when the computer instructions are run by the processor, the method described in embodiment one is completed. For brevity, it will not be described here.

[0097] It should be understood that in the embodiments, the processor can be a central processing unit CPU, and the processor can also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, ready programmable gate arrays FPGA or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0098] The memory can include read-only memory and random access memory, and provide instructions and data to the processor, and a portion of the memory can also include non-volatile random access memory. For example, the memory can also store device type information.

[0099] A computer readable storage medium for storing computer instructions, when the computer instructions are executed by the processor, the method described in embodiment one is completed.

[0100] The method in embodiment one can be directly embodied as a hardware processor to complete, or be completed by a combination of hardware and software modules in the processor. The software module can be located in a storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, or electrically erasable programmable memory, register, etc. The storage medium is located in the memory, and the processor reads the information in the memory, and combines the hardware to complete the steps of the above method. To avoid repetition, it will not be described in detail here.

[0101] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0102] The above describes the specific embodiments of the application in combination with the drawings, but is not a limitation on the protection scope of the application. Those skilled in the art should understand that various modifications or variations made by those skilled in the art on the basis of the technical solutions of the application without inventive labor are still within the protection scope of the application.

Claims

1. A method for predicting the internal moisture content of a transformer, characterized in that, include: By simulating a multiphysics coupling model of a transformer, the correlation between the moisture content at different insulation weak points and the oil tap under different operating parameters and / or external environmental parameters of the transformer was obtained. Multiple training datasets were constructed, specifically: Based on the multi-physics coupling model of transformer, several locations with concentrated micro-water content were selected as typical weak insulation parts. By setting different operating parameters of the transformer and / or external environmental parameters, the correlation between different weak insulation parts and micro-water content at the oil intake was obtained, and multiple training datasets were constructed. The construction of the transformer multiphysics coupling model is specifically as follows: based on the constructed two-dimensional finite element model of the transformer, the electric field distribution and power loss are solved; The temperature field distribution is calculated using the power loss obtained from the solution as the heat source. Based on the reaction of the temperature field distribution to the material parameters, the coupling of the electric field and the temperature field is realized. Based on Fick's second diffusion law, the relationship between micro-water diffusion and temperature field in oil-paper insulation is established. The distribution of micro-water in transformer oil and insulating paperboard is simulated. Based on the micro-water distribution, the material parameters are changed, which in turn affect the electric field and temperature field, thus realizing the coupling of temperature field and fluid field. The construction of the transformer multiphysics coupling model also includes: obtaining the relative permittivity and conductivity values ​​of transformer insulating oil and insulating paperboard at different temperatures, and fitting the functional relationship between the relative permittivity and conductivity and temperature using the least squares method. The regression decision tree model was trained based on multiple training sample sets to obtain a dynamic inference model of transformer micro-water content corresponding to different insulation weak points. Based on the obtained micro-water content at the transformer oil inlet to be predicted, the micro-water content of different insulation weak points inside the transformer is obtained by using a dynamic extrapolation model of the transformer micro-water content corresponding to different insulation weak points.

2. The method for predicting the internal moisture content of a transformer as described in claim 1, characterized in that, In the training of the regression decision tree model, a segmented linear model is introduced when a node splits. By calculating the loss reduction at the split point, linear functions are fitted to the datasets on both sides of the split point. The optimal split point is selected by comparing the loss changes before and after the split.

3. The method for predicting the internal moisture content of a transformer as described in claim 1, characterized in that, The different operating parameters of the transformer are specifically: normal transformer state, damp transformer state, and typical transformer defects; the typical transformer defects include burr defects, dent defects, and metal discontinuity defects.

4. The method for predicting the internal moisture content of a transformer as described in claim 1, characterized in that, The loss function of the regression decision tree model incorporates a tree complexity regularization term and a time decay factor.

5. The method for predicting the internal moisture content of a transformer as described in claim 1, characterized in that, Also includes: An incremental update mechanism is introduced to adjust the node weights of the regression decision tree model, so that the node weights are dynamically corrected according to the error of new data; an adaptive learning rate strategy is introduced to dynamically adjust the learning rate according to the error fluctuation of new data and historical data.

6. A transformer internal moisture content prediction system, characterized in that, include: The module is configured to: simulate the transformer using a multi-physics coupling model to obtain the correlation between the moisture content at different insulation weak points and the oil tap under different operating parameters and / or external environmental parameters of the transformer, and construct multiple training datasets, specifically: Based on the multi-physics coupling model of transformer, several locations with concentrated micro-water content were selected as typical weak insulation parts. By setting different operating parameters of the transformer and / or external environmental parameters, the correlation between different weak insulation parts and micro-water content at the oil intake was obtained, and multiple training datasets were constructed. The temperature field distribution is calculated using the power loss obtained from the solution as the heat source. Based on the reaction of the temperature field distribution to the material parameters, the coupling of the electric field and the temperature field is realized. Based on Fick's second diffusion law, the relationship between micro-water diffusion and temperature field in oil-paper insulation is established. The distribution of micro-water in transformer oil and insulating paperboard is simulated. Based on the micro-water distribution, the material parameters are changed, which in turn affect the electric field and temperature field, thus realizing the coupling of temperature field and fluid field. The construction of the transformer multiphysics coupling model also includes: obtaining the relative permittivity and conductivity values ​​of transformer insulating oil and insulating paperboard at different temperatures, and fitting the functional relationship between the relative permittivity and conductivity and temperature using the least squares method. The training module is configured to train the regression decision tree model based on multiple training sample sets to obtain a dynamic inference model of transformer micro-water content corresponding to different insulation weak points. The prediction module is configured to: based on the obtained trace moisture content at the oil port of the transformer to be predicted, use a dynamic extrapolation model of the trace moisture content of the transformer corresponding to different weak insulation points to obtain the trace moisture content of different weak insulation points inside the transformer.

7. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-5.

Images