A method, device, equipment and medium for constructing a price reduction model for power transformers

Abstract

The invention discloses a method, device, equipment and medium for constructing a price reduction model of a power transformer, comprising obtaining a plurality of transformer operating values ​​and a plurality of transformer simulation current parameters of a power transformer; performing simulation parameterized calculations on the simulation current parameters of each transformer by multi-physical field coupling simulation software to generate a sample data set; constructing an initial Gaussian process regression price reduction model using the sample data set based on PLS decomposition technology and Bayesian optimization algorithm; inputting the operating values ​​of each transformer into the initial Gaussian process regression price reduction model for temperature prediction, generating a temperature field numerical distribution result, and comparing the result with the preset temperature field distribution result; if the comparison result meets the preset comparison condition, using the initial Gaussian process regression price reduction model as the target Gaussian process regression price reduction model; solving the technical problem that the traditional finite element model has a very large amount of calculation, resulting in a long simulation calculation time for the transformer physical field.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment modeling, and in particular to a method, device, equipment and medium for constructing a price reduction model of a power transformer. Background Art

[0002] The electric power industry is the lifeline of my country's economic development. In recent years, my country has vigorously developed ultra-high voltage direct current transmission projects, and the application of digital twins in electric power industry production mainly includes the following two aspects: one is power equipment management: reducing the risk of equipment operation and improving equipment operation reliability; the other is power system dispatching: using digital twin real-time data to reduce resource mobilization costs and improve energy efficiency.

[0003] Digital twins can serve the collaborative design and digital delivery of complex power equipment, and the predictive maintenance of high-value power equipment, and have broad research and application value. However, the realization of digital twins of power equipment also faces many challenges and urgently needs to be studied. For example, the model scale of power transformers is generally large, and the physical model is generally understood from the physical mechanism and operation process. It is difficult to incorporate historical data into the simulation process of the physical model, and the simulation cycle is long.

[0004] At present, most of the simulation calculations of the physical field of the transformer are carried out through traditional numerical simulation methods, such as the finite element method and the finite volume method. However, due to the high degree of freedom of the traditional finite element model and the need to discretize the space domain and the time domain at the same time, the amount of calculation is very large, resulting in a long simulation calculation time for the physical field of the transformer. Summary of the invention

[0005] The present invention provides a method, device, equipment and medium for constructing a price reduction model of a power transformer, which solves the technical problem that the traditional finite element model has a high degree of freedom and needs to discretize the space domain and the time domain at the same time, resulting in a very large amount of calculation and causing the simulation calculation time of the transformer physical field to be too long.

[0006] A first aspect of the present invention provides a method for constructing a price reduction model for a power transformer, comprising:

[0007] Acquire multiple transformer operation values ​​and multiple transformer simulation current parameters of the power transformer;

[0008] Performing simulation parameterization calculation on each of the transformer simulation current parameters by multi-physics field coupling simulation software to generate a sample data set;

[0009] Based on PLS decomposition technology and Bayesian optimization algorithm, the sample data set is used to construct an initial Gaussian process regression price reduction model;

[0010] Inputting each transformer operation value into the initial Gaussian process regression price reduction model to perform temperature prediction, generate a temperature field numerical distribution result, and compare it with a preset temperature field distribution result;

[0011] If the comparison result satisfies the preset comparison condition, the initial Gaussian process regression price reduction model is used as the target Gaussian process regression price reduction model.

[0012] Optionally, the step of performing simulation parameterized calculation on each of the transformer simulation current parameters by multi-physics field coupling simulation software to generate a sample data set includes:

[0013] Perform simulation calculation on each transformer simulation current parameter by multi-physics field coupling simulation software to generate field intensity distribution results corresponding to each transformer simulation current parameter;

[0014] The multi-physics field coupling simulation software is used to perform batch parameterized calculations on each of the transformer simulation current parameters and the field intensity distribution results corresponding to each of the transformer simulation current parameters to generate a sample data set.

[0015] Optionally, the sample data set includes multiple transformer current values ​​and multiple field strength value distributions; the step of constructing an initial Gaussian process regression price reduction model using the sample data set based on PLS decomposition technology and Bayesian optimization algorithm includes:

[0016] Centralizing each of the transformer current values ​​and each of the field strength value distributions to generate a plurality of centralized current values ​​and a plurality of centralized field strength value distributions;

[0017] The PLS decomposition technology is used to perform dimensionality reduction decomposition on the distribution of each of the centralized current values ​​and each of the centralized field strength values ​​to generate a plurality of initial linear combination values;

[0018] Selecting components of each of the initial linear combination values ​​by using a method explanation ratio method, and outputting a plurality of target linear combination values;

[0019] Based on the Bayesian optimization algorithm, an initial Gaussian process regression price reduction model is constructed using the target linear combination values ​​and the field strength value distributions.

[0020] Optionally, the step of inputting each transformer operation value into the initial Gaussian process regression price reduction model for temperature prediction to generate a temperature field numerical distribution result includes:

[0021] Centralizing each of the transformer operation values ​​to generate a plurality of transformer operation centralized values;

[0022] Probability estimation is performed on the central values ​​of each transformer operation through the initial Gaussian process regression price reduction model, and the posterior probability value distribution is output;

[0023] Based on the posterior probability numerical distribution, the temperature field numerical distribution result is determined.

[0024] A second aspect of the present invention provides a device for constructing a price reduction model for a power transformer, comprising:

[0025] An acquisition module, used for acquiring transformer operation values ​​and multiple transformer simulation current parameters of the power transformer;

[0026] A calculation module, used for performing simulation parameterized calculation on each of the transformer simulation current parameters through multi-physics field coupling simulation software to generate a sample data set;

[0027] An adopting module is used to construct an initial Gaussian process regression price reduction model using the sample data set based on PLS decomposition technology and Bayesian optimization algorithm;

[0028] A prediction module, used for inputting the transformer operation value into the initial Gaussian process regression price reduction model to perform temperature prediction, generate a temperature field numerical distribution result, and compare it with a preset temperature field distribution result;

[0029] The comparison module is used to use the initial Gaussian process regression price reduction model as the target Gaussian process regression price reduction model if the comparison result meets the preset comparison condition.

[0030] Optionally, the calculation module includes:

[0031] A simulation calculation submodule is used to perform simulation calculation on each of the transformer simulation current parameters through multi-physics field coupling simulation software to generate a field intensity distribution result corresponding to each of the transformer simulation current parameters;

[0032] The parameterized calculation submodule is used to perform batch parameterized calculations on each of the transformer simulation current parameters and the field intensity distribution results corresponding to each of the transformer simulation current parameters using the multi-physics field coupling simulation software to generate a sample data set.

[0033] Optionally, the sample data set includes a plurality of transformer current values ​​and a plurality of field strength value distributions; and the adoption module includes:

[0034] A centralization processing submodule, used to centralize each of the transformer current values ​​and each of the field strength value distributions to generate a plurality of centralized current values ​​and a plurality of centralized field strength value distributions;

[0035] A dimensionality reduction decomposition submodule, used to perform dimensionality reduction decomposition on the distribution of each of the centralized current values ​​and each of the centralized field strength values ​​by using the PLS decomposition technology to generate a plurality of initial linear combination values;

[0036] A component selection submodule, used for performing component selection on each of the initial linear combination values ​​by using a method explanation ratio method, and outputting a plurality of target linear combination values;

[0037] A submodule is constructed, which is used to construct an initial Gaussian process regression price reduction model based on the Bayesian optimization algorithm by using each of the target linear combination values ​​and each of the field strength value distributions.

[0038] Optionally, the prediction module includes:

[0039] The first submodule is used to perform centralization processing on the transformer operation value to generate a transformer operation centralized value;

[0040] The second submodule is used to perform probability estimation on the transformer operation centralized value through the initial Gaussian process regression price reduction model, and output the posterior probability numerical distribution corresponding to the transformer operation centralized value;

[0041] The determination submodule is used to determine the temperature field numerical distribution result based on the posterior probability numerical distribution.

[0042] A third aspect of the present invention provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for constructing a price reduction model for a power transformer as described in any one of the above items.

[0043] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed, the method for constructing a price reduction model for a power transformer as described in any one of the above items is implemented.

[0044] It can be seen from the above technical solutions that the present invention has the following advantages:

[0045] The above technical scheme of the present invention provides a method for constructing a price reduction model of a power transformer, firstly obtaining a plurality of transformer operating values ​​and a plurality of transformer simulation current parameters of the power transformer; performing simulation parameterization calculation on each transformer simulation current parameter by multi-physics field coupling simulation software to generate a sample data set; then, based on PLS decomposition technology and Bayesian optimization algorithm, using the sample data set to construct an initial Gaussian process regression price reduction model; inputting the operating values ​​of each transformer into the initial Gaussian process regression price reduction model to perform temperature prediction, generating a temperature field numerical distribution result, and comparing it with a preset temperature field distribution result; finally, if the comparison result meets the preset If the comparison conditions are set, the initial Gaussian process regression price reduction model is used as the target Gaussian process regression price reduction model; the above scheme uses multi-physical field coupling simulation software to perform simulation parameterized calculations on the simulation current parameters of each transformer to generate a sample data set, and then based on the PLS decomposition technology and Bayesian optimization algorithm, the process of constructing the reduced-order model using the sample data set can effectively reduce the dimension of the sample data set, thereby obtaining a Gaussian process regression price reduction model with less computational complexity. Compared with the traditional finite element model with a calculation time of hours, the Gaussian process regression price reduction model can be calculated in seconds, reducing the simulation calculation time for the transformer physical field. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0047] Figure 1 A flowchart of a method for constructing a price reduction model for a power transformer provided by an embodiment of the present invention;

[0048] Figure 2 A black and white schematic diagram of a magnetic field cloud map of a transformer output by the Simdroid simulation software provided in an embodiment of the present invention;

[0049] Figure 3 A black and white schematic diagram of a magnetic field cloud map of a transformer output by a target Gaussian process regression price reduction model provided in an embodiment of the present invention;

[0050] Figure 4 A color schematic diagram of a magnetic field cloud map of a transformer output by the Simdroid simulation software provided in an embodiment of the present invention;

[0051] Figure 5 A color schematic diagram of a magnetic field cloud map of a transformer output by a target Gaussian process regression price reduction model provided in an embodiment of the present invention;

[0052] Figure 6 A flowchart of another method for constructing a price reduction model for a power transformer provided by an embodiment of the present invention;

[0053] Figure 7 A schematic diagram of a process for constructing a power transformer price reduction model provided by an embodiment of the present invention;

[0054] Figure 8 A structural block diagram of a device for constructing a price reduction model for a power transformer provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] The embodiments of the present invention provide a method, device, equipment and medium for constructing a price reduction model of a power transformer, which are used to solve the technical problem that the traditional finite element model has a high degree of freedom and needs to discretize the space domain and the time domain at the same time, resulting in a very large amount of calculation, resulting in a long simulation calculation time for the physical field of the transformer.

[0056] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] See also Figure 1 , Figure 1 A flowchart of the steps of a method for constructing a price reduction model for a power transformer provided in an embodiment of the present invention.

[0058] The present invention provides a method for constructing a price reduction model for a power transformer, comprising:

[0059] Step 101: Acquire multiple transformer operation values ​​and multiple transformer simulation current parameters of the power transformer.

[0060] The transformer simulation current parameters are current data such as the transformer operating current and the short-circuit operating current of the power transformer.

[0061] The transformer operation values ​​include transformer temperature values, transformer concentrated electric field distribution area values ​​and other data. These data can be used to form an equipment operation database for power transformers.

[0062] In this embodiment, a plurality of transformer operation values ​​and a plurality of transformer simulation current parameters of the power transformer are obtained.

[0063] Step 102: Perform simulation parameterization calculations on simulation current parameters of each transformer using multi-physics field coupling simulation software to generate a sample data set.

[0064] The sample data set includes a plurality of transformer current values ​​and a plurality of field strength value distributions.

[0065] It should be noted that, firstly, the simulated current parameters of each transformer are simulated and calculated based on the numerical simulation model (Simdroid simulation model) in the multi-physics field coupling simulation software (Simdroid simulation software), and the field intensity distribution results corresponding to the simulated current parameters of each transformer are generated. Then, the multi-physics field coupling simulation software is continued to be used to perform batch parameterized calculations on the simulated current parameters of each transformer and the field intensity distribution results corresponding to the simulated current parameters of each transformer, so as to obtain a sample data set with electrical signal excitation as input, that is, the transformer current value as input, and the field intensity distribution of the entire solution domain as output (the field intensity value distribution is the output).

[0066] In this embodiment, simulation parameterization calculations are performed on simulation current parameters of each transformer using multi-physics field coupling simulation software to generate a sample data set.

[0067] Step 103: Based on the PLS decomposition technology and the Bayesian optimization algorithm, an initial Gaussian process regression price reduction model is constructed using the sample data set.

[0068] PLS (Partial Least Squares) decomposition technology is a dimensionality reduction technology used to process multivariate data. It is usually used to process data with multiple independent variables and dependent variables. At the same time, it is a dimensionality reduction technology that can maximize the covariance between the prediction matrix composed of each element in the dimensionality reduction space and the predicted matrix.

[0069] It should be noted that the transformer current values ​​and field strength value distributions in the sample data set are respectively centralized to generate multiple centralized current values ​​and multiple centralized field strength value distributions; this process aims to move the mean of the data to zero to eliminate the impact of the average value of the data on the model, which helps to ensure that the data has a consistent scale and distribution when building and training machine learning models. In the centralized data set, the mean of each feature is close to zero.

[0070] Furthermore, the PLS decomposition technology is used to reduce the dimensionality of the centralized current values ​​and the centralized field strength value distributions obtained after the centralization processing, and generate multiple initial linear combination values; through the PLS decomposition technology, the independent variables and dependent variables in the sample data set, that is, the centralized current values ​​and the centralized field strength value distributions, are decomposed into a series of latent variables (components), which are linear combinations, aiming to capture the variability in the original data to the greatest extent.

[0071] Further, an appropriate number of components are selected by the method explanation ratio method, that is, the method explanation ratio method is used to select components for each initial linear combination value, and multiple target linear combination values ​​are output; wherein, the use of the variance explanation ratio method for component selection is a very important step in data analysis and feature engineering, which helps to reduce the dimension, improve the interpretability and generalization ability of the model, and in PCA (principal component analysis) or other dimensionality reduction techniques, sufficient components can be selected to explain most of the variance in the data. Those skilled in the art can set the variance explanation ratio as needed, which is not specifically limited here, for example, 80% or 90% of the variance explanation ratio.

[0072] Furthermore, based on the Bayesian optimization algorithm, the initial Gaussian process regression price reduction model is constructed using the numerical values ​​of each target linear combination and the numerical distribution of each field strength. Based on the typical Gaussian process regression model, the numerical values ​​of each target linear combination are used as input and the numerical distribution of each field strength is used as output for modeling. At the same time, the Bayesian optimization algorithm is used to adjust the model parameters, and finally the initial Gaussian process regression price reduction model with the best hyperparameter combination is obtained. The model establishes a probabilistic relationship between the input and output, and can estimate the model parameters.

[0073] In this embodiment, based on the PLS decomposition technology and the Bayesian optimization algorithm, an initial Gaussian process regression price reduction model is constructed using a sample data set.

[0074] Step 104 , input the operating values ​​of each transformer into the initial Gaussian process regression price reduction model for temperature prediction, generate a temperature field numerical distribution result, and compare it with a preset temperature field distribution result.

[0075] The preset temperature field distribution results are the temperature field distribution results output by the Simdroid simulation software.

[0076] It should be noted that the operating values ​​of each transformer are centralized to generate multiple centralized operating values ​​of the transformer; the probability of each centralized operating value of the transformer is estimated through the initial Gaussian process regression price reduction model, and the posterior probability numerical distribution is output. The posterior probability distribution can indicate the uncertainty of the prediction, usually expressed in the form of mean and variance; finally, based on the posterior probability numerical distribution, the temperature field numerical distribution result is determined and compared with the preset temperature field distribution result.

[0077] In this embodiment, the operating values ​​of each transformer are input into the initial Gaussian process regression price reduction model for temperature prediction, and the temperature field numerical distribution results are generated and compared with the preset temperature field distribution results.

[0078] Step 105: If the comparison result meets the preset comparison condition, the initial Gaussian process regression price reduction model is used as the target Gaussian process regression price reduction model.

[0079] It should be noted that if the comparison result meets the preset comparison condition, the constructed initial Gaussian process regression price reduction model will be used as the final target Gaussian process regression price reduction model, wherein the preset comparison condition means that if the similarity between the temperature field numerical distribution result and the preset temperature field distribution result (comparison result) is greater than the preset similarity threshold, it can be indicated that the comparison result meets the condition; if the similarity between the temperature field numerical distribution result and the preset temperature field distribution result (comparison result) is less than the preset similarity threshold, it indicates that the comparison result does not meet the condition, wherein the preset similarity threshold can be set as needed and is not specifically limited here.

[0080] For further information, see Figure 2-Figure 5 In order to demonstrate the accuracy of the target Gaussian process regression price reduction model constructed in this application, by giving the same current excitation conditions, the Simdroid simulation software and the constructed target Gaussian process regression price reduction model are used for calculation, and the magnetic field cloud map of the transformer output by the Simdroid simulation software and the magnetic field cloud map of the transformer output by the target Gaussian process regression price reduction model are obtained. By comparing the cloud maps, it can be found that there is basically no obvious difference in the magnetic field intensity distribution calculated by the two methods (see Figure 4-Figure 5 ,in, Figure 4 This is the color magnetic field cloud map of the transformer output by Simdroid simulation software. Figure 5 The colored magnetic field cloud map of the transformer is output by the target Gaussian process regression price reduction model); then, by further post-processing the magnetic field intensity distribution, the loss distribution of the transformer, that is, the heat flux density on the bottom and the heat flux density on the top, etc., can be obtained in the magnetothermal coupling calculation, and the temperature field of the transformer can be quickly calculated and the real-time temperature change curve of the output focus point can be output, thus realizing the real-time simulation of the transformer temperature field.

[0081] In this embodiment, if the comparison result meets the preset comparison condition, the initial Gaussian process regression price reduction model is used as the target Gaussian process regression price reduction model.

[0082] In an embodiment of the present invention, the present invention provides a method for constructing a price reduction model for a power transformer, firstly obtaining a plurality of transformer operating values ​​and a plurality of transformer simulation current parameters of the power transformer; performing simulation parameterization calculations on the simulation current parameters of each transformer by means of multi-physical field coupling simulation software, and generating a sample data set; then, based on PLS decomposition technology and Bayesian optimization algorithm, using the sample data set to construct an initial Gaussian process regression price reduction model; inputting the operating values ​​of each transformer into the initial Gaussian process regression price reduction model for temperature prediction, generating a temperature field numerical distribution result, and comparing it with a preset temperature field distribution result; finally, if the comparison result satisfies Preset comparison conditions, then use the initial Gaussian process regression price reduction model as the target Gaussian process regression price reduction model; the above scheme, through multi-physical field coupling simulation software, performs simulation parameterized calculations on the simulation current parameters of each transformer, generates a sample data set, and then based on PLS decomposition technology and Bayesian optimization algorithm, uses the sample data set to construct the reduced-order model process, which can effectively reduce the dimension of the sample data set, thereby obtaining a Gaussian process regression price reduction model with less computational complexity. Compared with the traditional finite element model with a calculation time of hours, the Gaussian process regression price reduction model can be calculated in seconds, reducing the simulation calculation time of the transformer physical field.

[0083] See also Figure 6 , Figure 6 A flowchart of the steps of another method for constructing a power transformer price reduction model provided by an embodiment of the present invention.

[0084] The present invention provides a method for constructing a price reduction model for a power transformer, comprising:

[0085] Step 601: Acquire multiple transformer operation values ​​and multiple transformer simulation current parameters of the power transformer.

[0086] The transformer operation values ​​are data such as transformer temperature values ​​and transformer concentrated electric field distribution area values, which are independent variables matching the training data (sample data set) composed of the distribution of each transformer current value and each field strength value.

[0087] In this embodiment, a plurality of transformer operation values ​​and a plurality of transformer simulation current parameters of the power transformer are obtained.

[0088] Step 602: simulating and calculating each transformer simulation current parameter by multi-physics field coupling simulation software to generate a field intensity distribution result corresponding to each transformer simulation current parameter.

[0089] It should be noted that, firstly, the simulation current parameters of each transformer are simulated and calculated based on the numerical simulation model (Simdroid simulation model) in the multi-physics field coupling simulation software (Simdroid simulation software), and the field intensity distribution results corresponding to the simulation current parameters of each transformer are generated.

[0090] In this embodiment, the simulation current parameters of each transformer are simulated and calculated by multi-physics field coupling simulation software to generate field intensity distribution results corresponding to the simulation current parameters of each transformer.

[0091] Step 603: Use multi-physics field coupling simulation software to perform batch parameterized calculations on each transformer simulation current parameter and the field intensity distribution results corresponding to each transformer simulation current parameter to generate a sample data set.

[0092] It should be noted that the multi-physics field coupling simulation software is continued to be used to perform batch parameterized calculations on the simulated current parameters of each transformer and the field intensity distribution results corresponding to the simulated current parameters of each transformer, so as to obtain a sample data set with electrical signal excitation as input, that is, the transformer current value as input, and the field intensity distribution of the entire solution domain as output (field intensity value distribution as output).

[0093] Exemplarily, the constructed current values ​​of each transformer are used as input data, and the constructed field strength value distributions are used as output data. The input data and output data are the same as those in the simulation. The above data are used to construct and output a reduced-order model (initial Gaussian process regression reduced-order model) to achieve the characterization of the physical phenomena and mechanisms of the operation of power equipment using a data model, thereby completing the twinning of the equipment.

[0094] In this embodiment, multi-physics field coupling simulation software is used to perform batch parameterized calculations on each transformer simulation current parameter and the field intensity distribution results corresponding to each transformer simulation current parameter to generate a sample data set.

[0095] Step 604: Based on the PLS decomposition technology and the Bayesian optimization algorithm, an initial Gaussian process regression price reduction model is constructed using the sample data set.

[0096] The sample data set includes a plurality of transformer current values ​​and a plurality of field strength value distributions.

[0097] Further, step 604 may include the following sub-steps:

[0098] S41, performing centralization processing on each of the transformer current values ​​and each of the field strength value distributions to generate a plurality of centralized current values ​​and a plurality of centralized field strength value distributions;

[0099] S42, using the PLS decomposition technology to perform dimensionality reduction decomposition on the distribution of each of the centralized current values ​​and each of the centralized field strength values ​​to generate a plurality of initial linear combination values;

[0100] S43, performing component selection on each of the initial linear combination values ​​by using a method explanation ratio method, and outputting a plurality of target linear combination values;

[0101] S44. Based on the Bayesian optimization algorithm, an initial Gaussian process regression price reduction model is constructed using the target linear combination values ​​and the field strength value distributions.

[0102] It should be noted that PLS is a dimensionality reduction technique that maximizes the covariance between the prediction matrix and the predicted matrix composed of each element in the dimensionality reduction space. Simply put, the prediction matrix and the predicted matrix are first preprocessed to make them averaged and centered, and then the prediction matrix is ​​decomposed, which is specifically:

[0103] X=TP T +E;

[0104] Where X∈R n×m is the prediction matrix; T∈R n×a is the score matrix; P∈R m×a is the load matrix, a is the number of nodes; E∈R m×n is the residual matrix, n is the number of samples, and m is the number of features; the matrix product TP T It can be expressed as the score vector t j (the jth column of T) and the load vector p j (the jth column of P), then

[0105]

[0106] Similarly, the prediction matrix is ​​decomposed into

[0107] Y=UQ T +F;

[0108] Where Y∈R n×m is the predicted matrix, which is U∈R n×a is the score matrix; Q∈R m×a is the load matrix; F∈R m×n is the residual matrix; UQ T It can be expressed as the score vector u j (j-th column of U) and the load vector q j (the jth column of Q), so it can be expressed as

[0109]

[0110]

[0111] Where b j is the regression coefficient, then U=TB. B∈R a×a is the regression matrix, so Y = TBQ T +F.

[0112] Based on the above foundation, PLS dimensionality reduction technology fully considers the impact of the combined effects of individual factors on the predicted value. In terms of regression rate, PLS is faster than the general multivariate regression method and has looser requirements on samples. The Gaussian process algorithm is a machine learning method based on Bayesian theory. Its purpose is to find the posterior probability distribution of the predicted output value. According to its mean and variance, the predicted output result with a confidence interval can be obtained. Its properties are completely determined by the mean function and covariance function. It can achieve the advantages of adaptive acquisition of hyperparameters and output with probabilistic meaning. It has been widely used in the fields of time series analysis, automatic control, etc. to deal with predictive regression problems with high dimensions, small samples, nonlinearity, etc.

[0113] For further information, see Figure 7 , through the numerical simulation model (Simdroid simulation model) calculation as the way to obtain empirical data, after completing the construction of the sample data set, the PLS decomposition technology and Gaussian process algorithm (Bayesian optimization algorithm) in the machine learning algorithm are used to reduce the model to generate a data reduction model (initial Gaussian process regression reduction model) that can realize the mapping from input parameters to simulation result data; wherein, the data set in the sample space is divided into a training set and a test set according to a certain ratio during the process of machine learning training. The training set is used as a known quantity to construct and train the initial Gaussian process regression reduction model; the test set is used as an unknown quantity to test and evaluate the accuracy and generalization ability of the initial Gaussian process regression reduction model, that is, to adjust the hyperparameters of the initial Gaussian process regression reduction model, so as to obtain the initial Gaussian process regression reduction model with the best hyperparameter combination. Through calculation accuracy evaluation and model optimization, the initial Gaussian process regression reduction model based on the PLS algorithm and the Bayesian optimization algorithm is finally obtained. The model evaluation accuracy adopts the accuracy convergence evaluation, for example, the accuracy is set to 1%, and the iterative optimization is stopped when the convergence accuracy reaches the set value.

[0114] Exemplarily, model parameter tuning is to find the best hyperparameter configuration to obtain the best performance. In the reduced-order model, parameter tuning is required to ensure the performance and stability of the model. First, select key hyperparameters: determine the key hyperparameters to be adjusted: these hyperparameters are derived from the input key parameter database of the reduced-order model, based on the previous transformer input parameters, such as input current, heat dissipation coefficient, convection coefficient, etc.; second, initialize parameters: initialize the parameters of the model according to prior knowledge; third, hyperparameter search: use the validation set to search for the best hyperparameter combination; fourth, train the constructed reduced-order model: use the best hyperparameter combination to train the reduced-order model. During the training process, monitor the performance of the model and perform early stopping when necessary to prevent overfitting; fifth, evaluate the model: use the test set to evaluate the performance of the finally trained reduced-order model. You can use the accuracy evaluation index to select the appropriate index according to the task; sixth, adjust: according to the evaluation results, if the model performance is not satisfactory, you can return to the hyperparameter search step and further adjust the model; model parameter tuning is an iterative process, and you need to constantly try different hyperparameter combinations to find the optimal model configuration. At the same time, be careful when using the test set to evaluate model performance to avoid leaking information. The ultimate goal is to create a model that performs well on new data.

[0115] Furthermore, the above process includes data preparation, PLS decomposition, and subsequent use of the constructed initial Gaussian process regression reduced-order model to estimate the posterior probability distribution. Through these steps, the present application establishes a reduced-order model to predict the values ​​of the predicted matrix and provide uncertainty information about these predictions.

[0116] In this embodiment, based on the PLS decomposition technology and the Bayesian optimization algorithm, an initial Gaussian process regression price reduction model is constructed using a sample data set.

[0117] Step 605: Input the operating values ​​of each transformer into the initial Gaussian process regression price reduction model for temperature prediction, generate a temperature field numerical distribution result, and compare it with the preset temperature field distribution result.

[0118] Further, step 405 may include the following sub-steps:

[0119] S51, performing centralization processing on each of the transformer operation values ​​to generate a plurality of transformer operation centralized values;

[0120] S52, performing probability estimation on each of the transformer operation centralized values ​​through the initial Gaussian process regression price reduction model, and outputting the posterior probability numerical distribution corresponding to the transformer operation centralized value;

[0121] S53. Determine the temperature field numerical distribution result based on the posterior probability numerical distribution.

[0122] It should be noted that the operating values ​​of each transformer are centralized to generate multiple centralized operating values ​​of the transformer; the probability of each centralized operating value of the transformer is estimated through the initial Gaussian process regression price reduction model, and the posterior probability numerical distribution is output. The posterior probability distribution can indicate the uncertainty of the prediction, usually expressed in the form of mean and variance; finally, based on the posterior probability numerical distribution, the temperature field numerical distribution result is determined and compared with the preset temperature field distribution result.

[0123] In this embodiment, the operating values ​​of each transformer are input into the initial Gaussian process regression price reduction model for temperature prediction, and the temperature field numerical distribution results are generated and compared with the preset temperature field distribution results.

[0124] Step 606: If the comparison result meets the preset comparison condition, the initial Gaussian process regression price reduction model is used as the target Gaussian process regression price reduction model.

[0125] It should be noted that if the comparison result meets the preset comparison condition, the constructed initial Gaussian process regression price reduction model will be used as the final target Gaussian process regression price reduction model, wherein the preset comparison condition means that if the similarity between the temperature field numerical distribution result and the preset temperature field distribution result (comparison result) is greater than the preset similarity threshold, it can be indicated that the comparison result meets the condition; if the similarity between the temperature field numerical distribution result and the preset temperature field distribution result (comparison result) is less than the preset similarity threshold, it indicates that the comparison result does not meet the condition, wherein the preset similarity threshold can be set as needed and is not specifically limited here.

[0126] In this embodiment, if the comparison result meets the preset comparison condition, the initial Gaussian process regression price reduction model is used as the target Gaussian process regression price reduction model.

[0127] Exemplarily, the monitoring data based on power transformers mainly include: electrical quantity parameters, acoustic vibration data, photoelectric sensing data, heating data, etc. Taking the converter transformer as an example, its monitoring quantities include: top oil temperature, winding temperature, oil level, oil pressure, core and clamp grounding current, test data, etc. The target Gaussian process regression price reduction model obtained by construction can have an overall architecture including: user layer, fault diagnosis layer, case library construction layer, digital twin feedback layer, and includes information such as ledger data, operation data, fault data, and historical cases under the operation state of the transformer as system input, and then uses the twin model to evaluate the operation state of the transformer and diagnose potential faults, and obtain fault diagnosis results and solutions, which are fed back to the user as output; wherein, the user layer includes an input fault phenomenon module, a fault phenomenon preprocessing module, a fault processing module, and a result feedback module. After describing and preprocessing the fault phenomenon, the status information is input into the fault diagnosis layer; the fault diagnosis layer accepts the input from the user layer, completes the fault diagnosis and feeds back the results to the user layer. It is the core of the diagnosis system, including the fault diagnosis model and the fault type and solution output module; the case library construction layer is the underlying knowledge base of the system, including the phenomenon case mapping module and the case library update module. Based on historical fault diagnosis experience, the case mapping module collects cases, including the mapping relationship between fault phenomena, causes, types and solutions verified by practice; after receiving the data, the digital twin evaluates the operating status of the transformer and diagnoses potential faults, obtains fault diagnosis results and solutions, and feeds them back to the user as output.

[0128] Furthermore, in data-driven modeling methods, artificial neural networks can fit highly complex and nonlinear input-output relationships well with large amounts of data. Support vector machines can achieve a good compromise between model complexity and generalization with limited sample capacity. The advantage of data-driven modeling is that as more and more data or experience is input, the model will continue to improve and perfect. The training part of data-driven modeling may encounter problems related to instability, but once training is completed, these models are stable when used.

[0129] In an embodiment of the present invention, the present invention provides a method for constructing a price reduction model for a power transformer, firstly obtaining a plurality of transformer operating values ​​and a plurality of transformer simulation current parameters of the power transformer; performing simulation parameterization calculations on the simulation current parameters of each transformer by means of multi-physical field coupling simulation software, and generating a sample data set; then, based on PLS decomposition technology and Bayesian optimization algorithm, using the sample data set to construct an initial Gaussian process regression price reduction model; inputting the operating values ​​of each transformer into the initial Gaussian process regression price reduction model for temperature prediction, generating a temperature field numerical distribution result, and comparing it with a preset temperature field distribution result; finally, if the comparison result satisfies Preset comparison conditions, then use the initial Gaussian process regression price reduction model as the target Gaussian process regression price reduction model; the above scheme, through multi-physical field coupling simulation software, performs simulation parameterized calculations on the simulation current parameters of each transformer, generates a sample data set, and then based on PLS decomposition technology and Bayesian optimization algorithm, uses the sample data set to construct the reduced-order model process, which can effectively reduce the dimension of the sample data set, thereby obtaining a Gaussian process regression price reduction model with less computational complexity. Compared with the traditional finite element model with a calculation time of hours, the Gaussian process regression price reduction model can be calculated in seconds, reducing the simulation calculation time of the transformer physical field.

[0130] See also Figure 8 , Figure 8 A structural block diagram of a device for constructing a price reduction model for a power transformer provided in an embodiment of the present invention.

[0131] The present invention provides a device for constructing a price reduction model for a power transformer, comprising:

[0132] An acquisition module 801 is used to acquire a transformer operation value and a plurality of transformer simulation current parameters of a power transformer;

[0133] A calculation module 802 is used to perform simulation parameterization calculation on each of the transformer simulation current parameters through multi-physics field coupling simulation software to generate a sample data set;

[0134] Adopting module 803, for constructing an initial Gaussian process regression price reduction model using the sample data set based on PLS decomposition technology and Bayesian optimization algorithm;

[0135] Prediction module 604, used for inputting the transformer operation value into the initial Gaussian process regression price reduction model to perform temperature prediction, generate temperature field value distribution results, and compare with preset temperature field distribution results;

[0136] The comparison module 805 is used to use the initial Gaussian process regression price reduction model as the target Gaussian process regression price reduction model if the comparison result meets the preset comparison condition.

[0137] Furthermore, the calculation module 802 includes:

[0138] A simulation calculation submodule is used to perform simulation calculation on each of the transformer simulation current parameters through multi-physics field coupling simulation software to generate a field intensity distribution result corresponding to each of the transformer simulation current parameters;

[0139] The parameterized calculation submodule is used to perform batch parameterized calculations on each of the transformer simulation current parameters and the field intensity distribution results corresponding to each of the transformer simulation current parameters using the multi-physics field coupling simulation software to generate a sample data set.

[0140] Further, module 803 is adopted, including:

[0141] A centralization processing submodule, used to centralize each of the transformer current values ​​and each of the field strength value distributions to generate a plurality of centralized current values ​​and a plurality of centralized field strength value distributions;

[0142] A dimensionality reduction decomposition submodule, used to perform dimensionality reduction decomposition on the distribution of each of the centralized current values ​​and each of the centralized field strength values ​​by using the PLS decomposition technology to generate a plurality of initial linear combination values;

[0143] A component selection submodule, used for performing component selection on each of the initial linear combination values ​​by using a method explanation ratio method, and outputting a plurality of target linear combination values;

[0144] A submodule is constructed, which is used to construct an initial Gaussian process regression price reduction model based on the Bayesian optimization algorithm by using each of the target linear combination values ​​and each of the field strength value distributions.

[0145] Furthermore, the prediction module 804 includes:

[0146] The first submodule is used to perform centralization processing on the transformer operation value to generate a transformer operation centralized value;

[0147] The second submodule is used to perform probability estimation on the transformer operation centralized value through the initial Gaussian process regression price reduction model, and output the posterior probability numerical distribution corresponding to the transformer operation centralized value;

[0148] The determination submodule is used to determine the temperature field numerical distribution result based on the posterior probability numerical distribution.

[0149] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, modules and sub-modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0150] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory; when the computer program is executed by the processor, the processor executes the steps of the method for constructing a price reduction model for a power transformer as described in any of the above embodiments.

[0151] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed, a method for constructing a price reduction model for a power transformer as described in any of the above embodiments is implemented.

[0152] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0153] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0154] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0155] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0156] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a price reduction model for power transformers, characterized in that: include: Acquire multiple transformer operation values ​​and multiple transformer simulation current parameters of the power transformer; Performing simulation parameterization calculation on each of the transformer simulation current parameters by multi-physics field coupling simulation software to generate a sample data set; Based on PLS decomposition technology and Bayesian optimization algorithm, the sample data set is used to construct an initial Gaussian process regression price reduction model; Inputting each transformer operation value into the initial Gaussian process regression price reduction model to perform temperature prediction, generate a temperature field numerical distribution result, and compare it with a preset temperature field distribution result; If the comparison result satisfies the preset comparison condition, the initial Gaussian process regression price reduction model is used as the target Gaussian process regression price reduction model; The sample data set includes a plurality of transformer current values ​​and a plurality of field strength value distributions; The step of constructing an initial Gaussian process regression price reduction model using the sample data set based on the PLS decomposition technology and the Bayesian optimization algorithm includes: Centralizing each of the transformer current values ​​and each of the field strength value distributions to generate a plurality of centralized current values ​​and a plurality of centralized field strength value distributions; The PLS decomposition technology is used to perform dimensionality reduction decomposition on the distribution of each of the centralized current values ​​and each of the centralized field strength values ​​to generate a plurality of initial linear combination values; Selecting components of each of the initial linear combination values ​​by using a method explanation ratio method, and outputting a plurality of target linear combination values; Based on the Bayesian optimization algorithm, an initial Gaussian process regression price reduction model is constructed using the target linear combination values ​​and the field strength value distributions.

2. The method for constructing a price reduction model for a power transformer according to claim 1, characterized in that: The step of performing simulation parameterized calculation on each of the transformer simulation current parameters by multi-physics field coupling simulation software to generate a sample data set includes: Perform simulation calculation on each transformer simulation current parameter by multi-physics field coupling simulation software to generate field intensity distribution results corresponding to each transformer simulation current parameter; The multi-physics field coupling simulation software is used to perform batch parameterized calculations on each of the transformer simulation current parameters and the field intensity distribution results corresponding to each of the transformer simulation current parameters to generate a sample data set.

3. The method for constructing a price reduction model for a power transformer according to claim 1, characterized in that: The step of inputting each transformer operation value into the initial Gaussian process regression price reduction model for temperature prediction to generate a temperature field numerical distribution result includes: Centralizing each of the transformer operation values ​​to generate a plurality of transformer operation centralized values; Probability estimation is performed on the central values ​​of each transformer operation through the initial Gaussian process regression price reduction model, and the posterior probability value distribution is output; Based on the posterior probability numerical distribution, the temperature field numerical distribution result is determined.

4. A device for constructing a price reduction model for a power transformer, characterized in that: include: An acquisition module, used for acquiring transformer operation values ​​and multiple transformer simulation current parameters of the power transformer; A calculation module, used for performing simulation parameterized calculation on each of the transformer simulation current parameters through multi-physics field coupling simulation software to generate a sample data set; An adopting module is used to construct an initial Gaussian process regression price reduction model using the sample data set based on PLS decomposition technology and Bayesian optimization algorithm; A prediction module, used for inputting the transformer operation value into the initial Gaussian process regression price reduction model to perform temperature prediction, generate a temperature field numerical distribution result, and compare it with a preset temperature field distribution result; A comparison module, configured to use the initial Gaussian process regression price reduction model as a target Gaussian process regression price reduction model if the comparison result satisfies a preset comparison condition; The sample data set includes a plurality of transformer current values ​​and a plurality of field strength value distributions; The adoption module comprises: A centralization processing submodule, used for centralizing each transformer current value and each field strength value distribution, to generate a plurality of centralized current values ​​and a plurality of centralized field strength value distributions; A dimensionality reduction decomposition submodule, used to perform dimensionality reduction decomposition on the distribution of each of the centralized current values ​​and each of the centralized field strength values ​​by using the PLS decomposition technology to generate a plurality of initial linear combination values; A component selection submodule, used for performing component selection on each of the initial linear combination values ​​by using a method explanation ratio method, and outputting a plurality of target linear combination values; A submodule is constructed, which is used to construct an initial Gaussian process regression price reduction model based on the Bayesian optimization algorithm by using each of the target linear combination values ​​and each of the field strength value distributions.

5. The power transformer price reduction model building device according to claim 4 is characterized in that: The computing module comprises: A simulation calculation submodule is used to perform simulation calculation on each of the transformer simulation current parameters through multi-physics field coupling simulation software to generate a field intensity distribution result corresponding to each of the transformer simulation current parameters; The parameterized calculation submodule is used to perform batch parameterized calculations on each of the transformer simulation current parameters and the field intensity distribution results corresponding to each of the transformer simulation current parameters using the multi-physics field coupling simulation software to generate a sample data set.

6. The power transformer price reduction model building device according to claim 4, characterized in that: The prediction module comprises: The first submodule is used to perform centralization processing on the transformer operation value to generate a transformer operation centralized value; The second submodule is used to perform probability estimation on the transformer operation centralized value through the initial Gaussian process regression price reduction model, and output the posterior probability numerical distribution corresponding to the transformer operation centralized value; The determination submodule is used to determine the temperature field numerical distribution result based on the posterior probability numerical distribution.

7. An electronic device, characterized in that: It comprises a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the method for constructing a price reduction model of a power transformer as described in any one of claims 1 to 3.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the method for constructing a price reduction model for a power transformer as described in any one of claims 1 to 3 is implemented.

Images