Proxy model training method and device

Abstract

The invention discloses a proxy model training method and device, and the method comprises the steps: carrying out the simulation calculation of a transformer, and obtaining training data; the training data is used for representing temperature field data of each component of the transformer; conducting wire current and environment temperature are sampled through a preset sampling method, and sampling data are obtained; training an agent model based on the sampling data and the training data to obtain model parameters of the agent model; calculating a model error of the agent model based on the model parameters and the test data of the agent model; and if the model error meets a preset error requirement, obtaining a trained proxy model. According to the method, the real-time calculation requirement of the transformer temperature field can be met, so that the calculation efficiency of the transformer temperature field is improved.

Description

Technical Field

[0001] The present application relates to the field of transformer technology, and in particular to a training method and device for an agent model. Background Art

[0002] With the rapid development of power systems, their intelligence level has also increased. As an important device in power systems, transformers are mainly responsible for voltage conversion and power transmission. Transformers face many challenges during operation, such as temperature field management and monitoring. The temperature distribution of transformers directly affects their performance and safe operation. Excessive temperature may cause equipment damage or even fire and other serious consequences. Therefore, real-time monitoring and analysis of transformer temperature fields become particularly important.

[0003] Currently, digital twin technology is used to carry out real-time monitoring of the transformer temperature field. Digital twin technology establishes a digital model of the transformer and presents various parameters and states in the physical system to operation and maintenance personnel in digital form, thereby realizing comprehensive monitoring and analysis of the internal temperature field of the transformer.

[0004] At present, the temperature field inside the transformer is relatively complex, and the existing finite element simulation technology is usually used to simulate the temperature field inside the transformer. However, the above method usually has the problem of long simulation calculation time and cannot meet the real-time calculation requirements of the transformer temperature field. Summary of the invention

[0005] In view of this, an embodiment of the present application provides a method and device for training a proxy model, which aims to meet the real-time calculation requirements of the transformer temperature field and improve the calculation efficiency of the transformer temperature field.

[0006] In a first aspect, an embodiment of the present application provides a method for training a proxy model, the method comprising:

[0007] Perform simulation calculation on the transformer to obtain training data; the training data is used to characterize the temperature field data of each component of the transformer;

[0008] The conductor current and the ambient temperature are sampled by a preset sampling method to obtain sampling data;

[0009] Based on the sampled data and the training data, the proxy model is trained to obtain model parameters of the proxy model;

[0010] Calculating a model error of the proxy model based on the model parameters of the proxy model and the test data;

[0011] If the model error meets the preset error requirement, a trained proxy model is obtained.

[0012] In a second aspect, an embodiment of the present application provides a training device for an agent model, the device comprising:

[0013] A simulation calculation module is used to perform simulation calculation on the transformer to obtain training data; the training data is used to characterize the temperature field data of each component of the transformer;

[0014] A sampling module is used to sample the conductor current and the ambient temperature by a preset sampling method to obtain sampling data;

[0015] A training module, used to train the proxy model based on the sampled data and the training data to obtain model parameters of the proxy model;

[0016] An error calculation module, used to calculate the model error of the proxy model based on the model parameters of the proxy model and the test data;

[0017] The judgment module is used to obtain the trained proxy model if the model error meets the preset error requirement.

[0018] In a third aspect, an embodiment of the present application provides a training device for an agent model, the device comprising:

[0019] Memory for storing computer programs;

[0020] The processor is used to execute the computer program so that the device performs the training method of the agent model described in the first aspect.

[0021] In a fourth aspect, an embodiment of the present application provides a computer storage medium, on which a computer program is stored. When the computer program is executed, the device running the computer program implements the training method of the agent model described in the first aspect above.

[0022] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0023] The embodiment of the present application provides a method for training a proxy model, which provides a proxy model that can meet the real-time calculation requirements of the transformer temperature field. The method includes: performing simulation calculation on the transformer, obtaining training data that can characterize the temperature field data of each component in the transformer, thereby improving data support for the subsequent training process. Then, the wire current and the ambient temperature are sampled by a preset sampling method to obtain sampling data. After that, the proxy model is trained based on the sampling data and the training data to obtain the model parameters of the proxy model. After determining the model parameters of the proxy model, the model error of the proxy model can be calculated based on the model parameters of the proxy model and the test data to determine whether the model error meets the preset error requirements. If the model error of the proxy model calculated meets the preset error requirements, it is considered that the proxy model composed of the model parameters currently calculated can be used as the trained proxy model. Since the proxy model can reduce the amount of calculation and the calculation cycle, the proxy model is trained based on the sampling data and the training data, so that the trained proxy model can meet the real-time calculation requirements of the transformer temperature field, and then the trained proxy model is used to calculate the transformer temperature field, which can improve the calculation efficiency of the transformer temperature field. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in this embodiment or 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 application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 A flowchart of a method for training an agent model provided in an embodiment of the present application;

[0026] Figure 2 A schematic diagram of the structure of the training data provided in the embodiment of the present application;

[0027] Figure 3 A schematic diagram of a radial basis function interpolation space provided in an embodiment of the present application;

[0028] Figure 4 A schematic diagram of the structure of the training agent model provided in the embodiment of the present application;

[0029] Figure 5 A schematic diagram of the structure of a training device for an agent model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0030] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0031] Traditional temperature monitoring methods usually rely on limited sensor arrangements and can only provide local temperature information, making it difficult to fully grasp the complex temperature field distribution inside the transformer. In order to solve the above problems, digital twin technology is currently used to monitor the transformer temperature field in real time. Digital twin technology establishes a digital model of the transformer and presents various parameters and states in the physical system to the operation and maintenance personnel in a digital form, thereby achieving comprehensive monitoring and analysis of the transformer's internal temperature field.

[0032] Specifically, digital twin technology uses advanced numerical simulation methods combined with real-time monitoring data to accurately predict the temperature distribution of each component of the transformer. Through digital twin technology, operation and maintenance personnel can understand the working status of the transformer in real time, which helps operation and maintenance personnel to promptly discover potential anomalies and problems and take effective measures to adjust and optimize in time, thereby improving the reliability and operating efficiency of the transformer.

[0033] At present, the temperature field inside the transformer is relatively complex, and the existing finite element simulation technology is usually used to simulate the temperature field inside the transformer. However, the above method usually has the problem of long simulation calculation time and cannot meet the real-time calculation requirements of the transformer temperature field.

[0034] Based on this, in order to solve the above problems, in the embodiment of the present application, the transformer is simulated and calculated to obtain training data that can characterize the temperature field data of each component in the transformer, thereby improving data support for the subsequent training process. Then, the wire current and ambient temperature are sampled by a preset sampling method to obtain sampling data. After that, the proxy model is trained based on the sampling data and the training data to obtain the model parameters of the proxy model. After determining the model parameters of the proxy model, the model error of the proxy model can be calculated based on the model parameters of the proxy model and the test data to determine whether the model error meets the preset error requirements. If the model error of the proxy model calculated meets the preset error requirements, it is considered that the proxy model composed of the model parameters currently calculated can be used as the trained proxy model. Since the proxy model can reduce the amount of calculation and the calculation cycle, the proxy model is trained based on the sampling data and the training data, so that the trained proxy model can meet the real-time calculation requirements of the transformer temperature field, and then the trained proxy model is used to calculate the transformer temperature field, which can improve the calculation efficiency of the transformer temperature field.

[0035] The specific implementation of the training method and device of the agent model in the embodiment of the present application is described in detail below with reference to the accompanying drawings.

[0036] See also Figure 1 , which is a flow chart of a training method for an agent model provided in an embodiment of the present application, combined with Figure 1 As shown, it may specifically include:

[0037] S101: Perform simulation calculation on the transformer to obtain training data; the training data is used to characterize the temperature field data of each component of the transformer.

[0038] As an example, the temperature field simulation calculation of each component in the transformer can be performed by finite element simulation software, and then, in the calculation process, the grid of the simulation model of each component of the transformer can be finely divided, that is, in the calculation process, the grid can be divided, and the temperature field data of each component in the transformer can be calculated in units of grids. After that, the calculated temperature field data can be exported in the format of unit coordinates, so that the temperature field distribution data of each component in the transformer under different working conditions (under different load rates) can be included in the derived temperature field data, and the derived temperature field data can be used as training data to provide data support for subsequent training processes. The training data is used to characterize the temperature field data of each component in the transformer. For example, the training data may include various types of data when calculating temperature data inside the transformer, such as load current value, ambient temperature, three-dimensional coordinates of each component, and temperature value.

[0039] As an example, the various components in the above transformer include: a coil part, an upper clamp part, a copper shield part, a core sheet part, a magnetic shunt part, an oil tank part, a pull plate part and a lower clamp part.

[0040] Please refer to Figure 2 , Figure 2 A schematic diagram of the structure of the training data provided in the embodiment of the present application, in the output data, the element is the basic unit, each element contains four nodes, each node contains coordinate data and temperature data. Taking the magnetic shunt as an example, there are 384,875 nodes, each node has three-dimensional coordinate data and temperature data.

[0041] S102: Sampling the conductor current and the ambient temperature by a preset sampling method to obtain sampling data.

[0042] In order to study the influence of conductor current and ambient temperature on the temperature of components in the transformer, the conductor current and ambient temperature can be uniformly sampled by a preset sampling method to obtain sampling data. As an example, the preset sampling method is the optimal Latin hypercube sampling method, which is a random sampling technique that further improves the efficiency and accuracy of sampling by optimizing the position of sampling points in space. The optimal Latin hypercube sampling method is used to sample the conductor current and ambient temperature, and the sampling strategy is uniform sampling.

[0043] For example, the optimal Latin hypercube sampling method is used to uniformly extract N sampling points in the design space of each component of the transformer. The range of I (conductor current) is I∈(0, a)A, and the range of Tc (ambient temperature) is Tc∈(b, c)℃. At each sampling point i∈{1,2,…,N}, the corresponding sampling data can be expressed as s i , the set of sampling data corresponding to all sampling points can refer to formula 1.

[0044] {s i =(I i ,Tc i )|I i ∈(0,a),Tc i ∈(b,c)},i=1,2,…N Formula 1

[0045] The sampling data includes the wire current and ambient temperature corresponding to each sampling point. The above sampling method can generate evenly distributed sampling points in the multi-dimensional parameter space, ensuring the representativeness and coverage of the sampling, reducing the number of simulations and improving the accuracy of the calculation results.

[0046] S103: Based on the sampled data and the training data, the proxy model is trained to obtain model parameters of the proxy model.

[0047] After the training data and the sampling data are determined, the temperature field data corresponding to the sampling data can be obtained through simulation calculation based on the sampling data and the training data. For example, when the sampling data is determined, the sampling data s corresponding to each sampling point can be calculated through simulation software. i If the component is expressed according to the horizontal and vertical coordinates (x, y), without loss of generality, if the transformer component has N2 coordinates, then it has a total of N2 temperature points. The temperature of each point is expressed according to the coordinates (x j ,y j ) and the sampling data s corresponding to the sampling point i The temperature field data corresponding to the sampled data is given in the form of, where i and j are s i The component temperature and coordinate (x j ,y j The total temperature set T of each coordinate point on the transformer component can refer to formula 2.

[0048] T={t(x j ,y j ,s i )|x j ∈(x min ,x max ),y j ∈(y min ,y max )},i=1,2,…N,j=1,2,…N2

[0049] Formula 2

[0050] Among them, the above t(x j ,y j ,s i ) represents the sample data s i Corresponding temperature field data.

[0051] Through the above steps, a series of temperature field data samples corresponding to the sampled data at different load rates can be obtained. These data samples not only reflect the thermal response characteristics of the transformer under different wire currents and ambient temperature conditions, but also provide data support for the subsequent training process of the proxy model, so that the proxy model can be trained based on the above data samples, which can further improve the simulation efficiency and the reliability of the calculation results.

[0052] Before training the proxy model, a proxy model can be constructed first to reduce the order and calculate the temperature field of the transformer based on the trained proxy model. Since the proxy model can provide an effective alternative to the mathematical expression of the relationship between variables and complex responses, the calculation of the proxy model has a lower computational cost, that is, the proxy model can reduce the amount of calculation and the calculation cycle. Among them, the proxy model refers to a technology that optimizes the calculation process by constructing a mathematical model with a small amount of calculation, a short calculation cycle, but a calculation result similar to the numerical analysis result without reducing the accuracy.

[0053] As an example, a radial basis function (RBF) can be used to construct a proxy model. In this regard, the present application does not specifically limit the method of constructing the proxy model. The proxy model constructed based on the radial basis function can refer to Formula 3. At the same time, Figure 3 , Figure 3 The schematic diagram of the radial basis function interpolation space provided in the embodiment of the present application can obtain a feasible calculation domain of the entire space by calculating the sampled data, so that the trained proxy model can achieve order reduction processing and fast calculation.

[0054]

[0055] In the above formula 3 represents the estimated value of the simulation result of the transformer temperature rise field, m is the order of the polynomial function, c j is the polynomial basis function p j The coefficient of (x), λ i Yes s indivual The weight coefficient, n s Refers to the number of sampling points (support points) used for interpolation or fitting, r(x,x i ) refers to x and x i The Euclidean distance between i is the supporting point, x is the point to be found, It refers to the radial basis function.

[0056] For example, the basis function can be a cubic function Gaussian function Multiquadric surface functions Inverse quadratic function

[0057] Due to the better stability of the proxy model, multi-quadratic surface functions can be selected as basis functions. In order to calculate the unknown coefficients in the proxy model, such as c and λ, without loss of generality, the orthogonality condition is used as a supplementary condition, which can be referred to Formula 4.

[0058]

[0059] Combining Formula 3 and Formula 4, the unknown coefficients in the proxy model can be calculated using Formula 5.

[0060]

[0061] Among them, the above Φ represents a matrix composed of the radial basis function and the temperature field data corresponding to its sampling data, P represents a matrix composed of the polynomial function and the temperature field data corresponding to its sampling data, and y is the estimated value of the simulation result, that is, the output value of the proxy model.

[0062] Please refer to Figure 4 , Figure 4 A schematic diagram of the structure of the training agent model provided in the embodiment of the present application, Figure 4 The left figure in the figure is a schematic diagram of the sampling points in the three-dimensional space. Red represents the temperature field data corresponding to the sampling data, and blue represents the temperature field data corresponding to the test sampling data. The right figure is a structural diagram of the proxy model. The Input Layer represents the input node of the data to be calculated, that is, during the training process, the temperature field data corresponding to the sampling data of the input node can be used. Layer represents the model parameters of the trained proxy model, and the Output Layer represents the calculation results output by the proxy model.

[0063] After the proxy model is determined, the proxy model can be trained by combining the N sampling data of the same coordinate point and the temperature field data corresponding to each of them, the constructed proxy model and the coefficients of the proxy model, to calculate the proxy model parameters. Taking the coordinate point (x1, y1) as an example, the set shown in Formula 6 is used as the output of the proxy model, and the model parameter p1 of the proxy model can be obtained through training. Among them, the purpose of training the proxy model is to make the output result of the proxy model as close as possible to the temperature field data corresponding to the sampling data determined based on the training data and the sampling data, that is, the purpose of training is to make the calculation accuracy of the proxy model as accurate as possible. Since the proxy model can reduce the amount of calculation and the calculation cycle, the trained proxy model can meet the real-time calculation requirements of the transformer temperature field while maintaining the calculation accuracy, so as to improve the calculation efficiency.

[0064] {t(x1,y1,s i )},i=1,2,…N Formula 6

[0065] Afterwards, the model parameters of the corresponding proxy model can be calculated based on the N sampling data of different coordinate points on the transformer components. The above-mentioned calculated model parameters constitute a model parameter set P which can be expressed by Formula 7.

[0066] P = {p i},i=1,2,…N Formula 7

[0067] Among them, the above p i Represents the model parameters of the proxy model. When the model parameters of the proxy model are determined, the model parameters can be substituted into the proxy model, that is, based on the proxy model containing the model parameters, for any sampled data s(I, Tc), its value in (x i ,y i ) The temperature field data is t(x i ,y i , s).

[0068] S104: Calculate a model error of the proxy model based on the model parameters of the proxy model and the test data.

[0069] As an example, after determining the model parameters of the proxy model, the calculation results of the proxy model can be verified. For example, N3 sample points can be uniformly selected as test sampling points, combined with the test data, and the temperature field data corresponding to the test sampling data can be obtained through simulation calculation. It should be noted that the steps of determining the sampling data corresponding to the test sampling points and determining the temperature field data corresponding to the test sampling data are the same as the steps in S102 and S103 above, and are not repeated here. After determining the model parameters of the proxy model, the model parameters can be substituted into the proxy model, and the temperature field data corresponding to the above test sampling points can be calculated based on the proxy model containing the model parameters. The calculation results of the proxy model are compared with the temperature field data corresponding to the test sampling data to calculate the model error of the proxy model to verify the accuracy of the proxy model. Among them, the calculation formula of the model error can refer to Formula 8.

[0070]

[0071] Among them, the above erro represents the model error, treal i It represents the simulated calculated temperature at the i-th coordinate output by the proxy model, that is, the temperature field data corresponding to the test sampling point output by the proxy model.

[0072] S105: If the model error meets the preset error requirement, a trained proxy model is obtained.

[0073] When the model error is determined, if the model error meets the preset error requirement, it is considered that the proxy model containing the current model parameters has completed training and can be directly used as the trained proxy model. If the model error does not meet the preset error requirement, it is considered that the proxy model containing the current model parameters still needs to be trained. For example, the sampling data can be adjusted, such as increasing the sampling rate, so that the model error calculated by it meets the preset error requirement. Alternatively, the proxy model can be adjusted, such as selecting other methods to construct the proxy model, or adjusting the basis function, so that the model error calculated by it meets the preset error requirement. For example, the preset error requirement can be erro>ε, where ε represents the upper limit of the error. If the model error is greater than ε, it means that the training is completed.

[0074] In this way, the calculation accuracy of the proxy model can be guaranteed, and at the same time, it can be ensured that the trained proxy model can meet the real-time calculation requirements of the transformer temperature field.

[0075] Finally, after determining the trained proxy model, the temperature field of the transformer can be quickly calculated based on the trained proxy model, that is, after the specific operating parameters of the transformer are obtained, the trained proxy model can quickly calculate the corresponding temperature field data. In addition, in the process of calculating the temperature field data based on the trained proxy model, a parallel computing method can be used, that is, the calculation task of calculating the temperature field data based on the trained proxy model is distributed to multiple processors or computing nodes for simultaneous processing to shorten the data calculation time. Through the above parallel computing method, not only the data processing speed can be improved, but also the computing efficiency and resource utilization can be improved.

[0076] Correspondingly, the parallel computing method can quickly calculate the temperature field data corresponding to each component in the transformer. Since the temperature field calculation process of each component can be carried out simultaneously, the calculation speed of the entire simulation process is accelerated, thereby providing a data basis for the real-time display of the transformer thermal field.

[0077] Finally, the calculated temperature field data can be converted into a three-dimensional image of the transformer thermal field through real-time rendering technology, which can not only improve the calculation efficiency, but also ensure the accuracy and real-time performance of the thermal field distribution. The temperature distribution of the transformer can then be monitored and evaluated in real time under a variety of operating conditions, providing important data support and decision-making basis for optimizing design and operation and maintenance.

[0078] The above is a training method for a proxy model provided in an embodiment of the present application, and the method provides a proxy model that can meet the real-time calculation requirements of the transformer temperature field. The method includes: performing simulation calculation on the transformer, obtaining training data that can characterize the temperature field data of each component in the transformer, thereby improving data support for the subsequent training process. Then, the wire current and the ambient temperature are sampled by a preset sampling method to obtain sampling data. After that, the proxy model is trained based on the sampling data and the training data to obtain the model parameters of the proxy model. After determining the model parameters of the proxy model, the model error of the proxy model can be calculated based on the model parameters of the proxy model and the test data to determine whether the model error meets the preset error requirements. If the model error of the proxy model calculated meets the preset error requirements, it is considered that the proxy model composed of the model parameters currently calculated can be used as the trained proxy model. Since the proxy model can reduce the amount of calculation and the calculation cycle, the proxy model is trained based on the sampling data and the training data, so that the trained proxy model can meet the real-time calculation requirements of the transformer temperature field, and then the trained proxy model is used to calculate the transformer temperature field, which can improve the calculation efficiency of the transformer temperature field.

[0079] The above are some specific implementations of the training method of the proxy model provided in the embodiment of the present application. Based on this, the present application also provides a corresponding device. The device provided in the embodiment of the present application will be introduced from the perspective of functional modularization.

[0080] See also Figure 5 , which is a schematic diagram of the structure of a training device 500 for an agent model provided in an embodiment of the present application, and the device 500 may include:

[0081] The simulation calculation module 501 is used to perform simulation calculation on the transformer to obtain training data; the training data is used to characterize the temperature field data of each component of the transformer;

[0082] The sampling module 502 is used to sample the conductor current and the ambient temperature by a preset sampling method to obtain sampling data;

[0083] A training module 503 is used to train the proxy model based on the sampled data and the training data to obtain model parameters of the proxy model;

[0084] An error calculation module 504, used to calculate a model error of the proxy model based on the model parameters of the proxy model and the test data;

[0085] The judgment module 505 is used to obtain the trained proxy model if the model error meets the preset error requirement.

[0086] Optionally, the preset sampling method is an optimal Latin hypercube sampling method, and the sampling module 502 includes:

[0087] The sampling unit is used to sample the conductor current and the ambient temperature respectively by using the optimal Latin hypercube sampling method to obtain sampling data.

[0088] Optionally, the training module 503 includes:

[0089] A simulation calculation unit, used for obtaining temperature field data corresponding to the sampled data based on the sampled data and the training data through simulation calculation;

[0090] The training unit is used to train the proxy model based on the temperature field data corresponding to the sampled data to obtain model parameters of the proxy model.

[0091] Optionally, the device further comprises:

[0092] A construction unit for building proxy models based on radial basis functions.

[0093] Optionally, the simulation calculation module 501 includes:

[0094] The export unit is used to simulate and calculate the temperature field of each component in the transformer, export the calculated temperature field data in the format of grid and unit coordinates, and use the exported data as training data.

[0095] Optionally, the device further comprises:

[0096] A first adjustment unit, configured to adjust the sampled data if the model error does not meet the preset error requirement, until the model error meets the preset error requirement, thereby obtaining a trained proxy model;

[0097] Or, a second adjustment unit is used to adjust the proxy model if the model error does not meet the preset error requirement, until the model error meets the preset error requirement, thereby obtaining a trained proxy model.

[0098] The embodiments of the present application also provide corresponding devices and computer storage media for implementing the solutions provided by the embodiments of the present application.

[0099] The device includes a memory and a processor, the memory is used to store a computer program, and the processor is used to execute the computer program, so that the device executes the training method of the agent model described in any embodiment of the present application.

[0100] The computer storage medium stores a computer program. When the code is executed, the device executing the computer program implements the training method of the agent model described in any embodiment of the present application.

[0101] The "first" and "second" in the names such as "first" and "second" (if any) mentioned in the embodiments of the present application are only used as name identifiers and do not represent the first or second in order.

[0102] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0103] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is merely schematic, in which the unit described as a separate component may or may not be physically separated, and the component prompted as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0104] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for training a proxy model, characterized in that: The method comprises: Perform simulation calculation on the transformer to obtain training data; the training data is used to characterize the temperature field data of each component of the transformer; The conductor current and the ambient temperature are sampled by a preset sampling method to obtain sampling data; Based on the sampled data and the training data, the proxy model is trained to obtain model parameters of the proxy model; Calculating a model error of the proxy model based on the model parameters of the proxy model and the test data; If the model error meets the preset error requirement, a trained proxy model is obtained.

2. The method according to claim 1, characterized in that The preset sampling method is an optimal Latin hypercube sampling method. The wire current and ambient temperature are sampled by the preset sampling method to obtain sampling data, including: The conductor current and the ambient temperature are sampled respectively by the optimal Latin hypercube sampling method to obtain sampling data.

3. The method according to claim 1, characterized in that The step of training the proxy model based on the sampled data and the training data to obtain model parameters of the proxy model includes: Obtaining temperature field data corresponding to the sampled data based on the sampled data and the training data through simulation calculation; Based on the temperature field data corresponding to the sampled data, the proxy model is trained to obtain model parameters of the proxy model.

4. The method according to claim 1, characterized in that: The method further includes: constructing a proxy model based on a radial basis function.

5. The method according to claim 1, characterized in that The step of performing simulation calculation on the transformer to obtain training data includes: The temperature field of each component in the transformer is simulated and calculated, and the calculated temperature field data is exported in the format of grid and unit coordinates, and the exported data is used as training data.

6. The method according to claim 1, characterized in that The method further comprises: If the model error does not meet the preset error requirement, the sampled data is adjusted until the model error meets the preset error requirement, thereby obtaining a trained proxy model; Alternatively, if the model error does not meet the preset error requirement, the proxy model is adjusted until the model error meets the preset error requirement, thereby obtaining a trained proxy model.

7. A training device for an agent model, characterized in that: The device comprises: A simulation calculation module is used to perform simulation calculation on the transformer to obtain training data; the training data is used to characterize the temperature field data of each component of the transformer; A sampling module is used to sample the conductor current and the ambient temperature by a preset sampling method to obtain sampling data; A training module, used to train the proxy model based on the sampled data and the training data to obtain model parameters of the proxy model; An error calculation module, used to calculate the model error of the proxy model based on the model parameters of the proxy model and the test data; The judgment module is used to obtain the trained proxy model if the model error meets the preset error requirement.

8. The device according to claim 7, characterized in that The preset sampling method is an optimal Latin hypercube sampling method, and the sampling module includes: The sampling unit is used to sample the conductor current and the ambient temperature respectively by using the optimal Latin hypercube sampling method to obtain sampling data.

9. A training device for an agent model, characterized in that: The device comprises: Memory for storing computer programs; A processor, configured to execute the computer program so that the device performs the training method of the agent model as described in any one of claims 1 to 6.

10. A computer storage medium, characterized in that: The computer storage medium stores a computer program, and when the computer program is executed by a processor, the training method of the agent model as described in any one of claims 1 to 6 is implemented.

Images