Transformer magnetic field prediction method and system based on physical information neural network

Abstract

The invention provides a transformer magnetic field prediction method and system based on a physical information neural network, and the method comprises the steps: inputting a magnetic field distribution diagram sample of a transformer into a U-Net model, and obtaining a predicted magnetic field image, outputted by the U-Net model, of the transformer; determining a data driving loss function according to the pixel difference between the predicted magnetic field image of the transformer and the target magnetic field image, determining the magnetic field energy density of the transformer according to the predicted magnetic field image of the transformer, and determining a physical loss function according to the magnetic field energy density of the transformer; determining a comprehensive loss function according to the data-driven loss function and the physical loss function, and training the U-Net model by using the comprehensive loss function; and inputting the current magnetic field distribution diagram of the transformer into the trained U-Net model to obtain a current predicted magnetic field image of the transformer. According to the method, the deep learning model and the physical loss constraint are fused, and the precision, the physical consistency and the generalization ability of transformer magnetic field prediction are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a transformer magnetic field prediction method and system based on a physical information neural network. BACKGROUND

[0002] The magnetic field distribution of a transformer is an important representation of its operating state, which can reflect problems such as core saturation, local overheating, and electromagnetic interference. Therefore, accurate prediction of the magnetic field distribution is of great significance for equipment health management, fault diagnosis, and performance optimization. However, traditional methods such as sensor-based measurement and finite element analysis simulation are commonly used, but the former is limited by the number of measurement points, arrangement method, and environmental interference, making it difficult to obtain complete and continuous magnetic field distribution information, and the latter can provide high-precision prediction but has high computational cost and is difficult to apply in real time.

[0003] In recent years, deep learning technology has made breakthrough progress in the field of image processing, especially the U-Net architecture with its encoding-decoding structure, which has achieved efficient feature extraction and image reconstruction. However, in the field of transformer magnetic field distribution prediction, existing research mainly focuses on classification and segmentation tasks based on convolutional neural networks (CNN), while there are still few studies on the prediction of static magnetic field images, and there are challenges such as limited image quality, poor physical consistency of the model, insufficient generalization ability, and low computational efficiency. SUMMARY

[0004] The present application provides a transformer magnetic field prediction method and system based on a physical information neural network to solve the problem of low accuracy in existing transformer magnetic field prediction, improve the prediction accuracy of transformer magnetic field distribution images, and enhance their robustness in complex environments. The system combines physical information neural networks, deep learning model optimization, data preprocessing, and training strategy optimization to improve the accuracy and generalization ability of the prediction.

[0005] The present application provides a transformer magnetic field prediction method based on a physical information neural network, comprising:

[0006] Inputting the magnetic field distribution pattern sample of the transformer into the U-Net model to obtain the predicted magnetic field image of the transformer output by the U-Net model;

[0007] Determining a data-driven loss function based on the pixel difference between the predicted magnetic field image of the transformer and the target magnetic field image, determining the magnetic field energy density of the transformer based on the predicted magnetic field image of the transformer, and determining a physical loss function based on the magnetic field energy density of the transformer;

[0008] determine a comprehensive loss function according to the data-driven loss function and the physical loss function, and train the U-Net model using the comprehensive loss function;

[0009] input the current magnetic field distribution map of the transformer into the trained U-Net model to obtain a current predicted magnetic field image of the transformer.

[0010] According to the transformer magnetic field prediction method based on a physical information neural network provided by the application, before the magnetic field distribution map sample of the transformer is input into the U-Net model and the predicted magnetic field image of the transformer output by the U-Net model is obtained, the method further comprises the following steps:

[0011] Based on the spatial coordinates of the magnetic field measurement points of the transformer, the magnetic field distribution map of the transformer is calculated through a Gaussian distribution function to simulate the magnetic field intensity influence area of the transformer.

[0012] The diffusion range of the Gaussian distribution function is controlled through a standard deviation parameter.

[0013] According to the transformer magnetic field prediction method based on a physical information neural network provided by the application, the data-driven loss function is determined according to the pixel difference between the predicted magnetic field image and the target magnetic field image of the transformer through the following formula:

[0014]

[0015] wherein L1 loss is the data-driven loss function, N is the total number of pixel points in the magnetic field distribution map sample of the transformer, P i and T i respectively are the pixel value of the i th pixel in the predicted magnetic field image of the transformer and the pixel value of the i th pixel in the target magnetic field image of the transformer.

[0016] According to the transformer magnetic field prediction method based on a physical information neural network provided by the application, the magnetic field energy density of the transformer is determined according to the predicted magnetic field image of the transformer through the following formula:

[0017]

[0018] wherein ω m is the magnetic field energy density of the transformer in a unit volume, B is the magnetic induction intensity of the transformer, which is determined through a normalized mapping relationship between the pixel value in the predicted magnetic field image and the magnetic induction intensity, and μ is the magnetic permeability.

[0019] According to the transformer magnetic field prediction method based on the physical information neural network provided by the application, a physical loss function is determined according to the magnetic field energy density of the transformer through the following formula:

[0020]

[0021] Physics loss is a physical loss function, N is the total number of pixel points in the magnetic field distribution pattern sample of the transformer, B i is the magnetic induction intensity mapped by the i th pixel point in the predicted magnetic field image of the transformer.

[0022] According to the transformer magnetic field prediction method based on the physical information neural network provided by the application, a comprehensive loss function is determined according to the data-driven loss function and the physical loss function through the following formula:

[0023] L total =L1 loss +λPhysics loss

[0024] Wherein, L total is a comprehensive loss function, L1 loss is a data-driven loss function, Physics loss is a physical loss function, and lambda is an adjustment coefficient for controlling the influence degree of the physical loss function.

[0025] According to the transformer magnetic field prediction method based on the physical information neural network provided by the application, the method further comprises:

[0026] The U-Net model is trained by using the Adam optimizer.

[0027] According to the transformer magnetic field prediction method based on the physical information neural network provided by the application, the U-Net model comprises a multi-scale feature extraction module and an attention mechanism.

[0028] The application also provides a transformer magnetic field prediction system based on a physical information neural network, comprising:

[0029] The first prediction module is used for inputting the magnetic field distribution pattern sample of the transformer into the U-Net model to obtain the predicted magnetic field image of the transformer output by the U-Net model;

[0030] The loss calculation module is used for determining a data-driven loss function according to the pixel difference between the predicted magnetic field image and the target magnetic field image of the transformer, determining the magnetic field energy density of the transformer according to the predicted magnetic field image of the transformer, and determining a physical loss function according to the magnetic field energy density of the transformer.

[0031] a model training module configured to determine a comprehensive loss function according to the data-driven loss function and the physical loss function, and train the U-Net model using the comprehensive loss function;

[0032] a second prediction module configured to input the current magnetic field distribution of the transformer into the trained U-Net model to obtain a current predicted magnetic field image of the transformer.

[0033] The application further provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the transformer magnetic field prediction method based on a physical information neural network according to any one of the above when executing the program.

[0034] The application further provides a non-transitory computer-readable storage medium having a computer program stored thereon, and the computer program is executable on a processor to implement the transformer magnetic field prediction method based on a physical information neural network according to any one of the above.

[0035] The application further provides a computer program product, including a computer program, and the computer program is executable on a processor to implement the transformer magnetic field prediction method based on a physical information neural network according to any one of the above.

[0036] The transformer magnetic field prediction method and system based on a physical information neural network provided by the application introduce the physical law in the magnetic field distribution as a physical loss term in the training process of the U-Net model on the basis of retaining the traditional image reconstruction capability of the U-Net model, so as to improve the physical rationality of the predicted image in the non-measurement area; in the model training process, in addition to minimizing the reconstruction error between the predicted magnetic field image and the target magnetic field image, the residual error of the predicted result under the physical law is also minimized, so that the data-driven modeling and the electromagnetic physical law are fused, and the accuracy, the physical consistency and the generalization ability of the transformer magnetic field prediction are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0038] Figure 1 is one of the flowcharts of the transformer magnetic field prediction method based on a physical information neural network provided by the application;

[0039] Figure 2A structure schematic diagram of a U-Net model in a transformer magnetic field prediction method based on a physical information neural network provided by the application is shown in the figure;

[0040] Figure 3 A second flow schematic diagram of the transformer magnetic field prediction method based on the physical information neural network provided by the application is shown in the figure;

[0041] Figure 4 A comparison schematic diagram of a transformer magnetic field prediction result output by the model and a real target in the transformer magnetic field prediction method based on the physical information neural network provided by the application is shown in the figure;

[0042] Figure 5 A structure schematic diagram of a transformer magnetic field prediction system based on the physical information neural network provided by the application is shown in the figure;

[0043] Figure 6 A structure schematic diagram of an electronic device provided by the application is shown in the figure. DETAILED DESCRIPTION

[0044] To make the objectives, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described below in connection with the drawings in the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0045] The present application will be described below in connection with Figure 1 A transformer magnetic field prediction method based on a physical information neural network is described in the present application, which comprises:

[0046] In step 101, a magnetic field distribution pattern sample of a transformer is input into a U-Net model to obtain a predicted magnetic field image of the transformer output by the U-Net model.

[0047] In this embodiment, the magnetic field distribution pattern sample of the transformer is taken as an input image I (such as an image with a resolution of 256x256), and the input image I is not directly sampled from the image.

[0048] The input image I is subjected to standardization processing to reduce the influence of data distribution difference on training. At the same time, the measurement error is simulated by combining Gaussian noise to improve the adaptability of the model to real data. In data division, a reasonable training set and validation set division strategy is adopted to ensure the stability of the training process and the generalization ability of the model.

[0049] For the feature extraction and reconstruction task of transformer magnetic field distribution image, the U-Net model is selected as the basis because it performs well in image segmentation and reconstruction tasks. Optionally, the input layer and output layer of the U-Net model are single channels, and the hidden layer contains multiple 64-channel convolution modules for layer-by-layer extraction of image features and spatial structure.

[0050] The embodiment adopts an end-to-end training method, and the training set and the validation set are obtained by image index division. During the training process, the input image I is input into the U-Net model, and the predicted magnetic field image P is output. The structure of the U-Net model based on the convolutional neural network is as shown in Figure 2 .

[0051] Step 102, determining a data-driven loss function according to the pixel difference between the predicted magnetic field image of the transformer and the target magnetic field image, determining the magnetic field energy density of the transformer according to the predicted magnetic field image of the transformer, and determining a physical loss function according to the magnetic field energy density of the transformer;

[0052] The target magnetic field image T can be the measured magnetic field image of the transformer, which is processed by gray scale conversion for model training.

[0053] The pixel difference between the predicted magnetic field image P and the target magnetic field image T (measured magnetic field gray image) can be measured using the L1 loss function as the main supervision signal.

[0054] To improve the accuracy and physical consistency of transformer magnetic field distribution prediction, a physics-informed neural network (PINN) is introduced based on the U-Net model. This method embeds the physical laws in the electromagnetic field into the model training process, and by introducing the physical loss term into the loss function, the U-Net model learns the data features while following the actual physical mechanism. Compared with the pure data-driven model, the introduction of the physics-informed neural network not only improves the prediction accuracy, but also significantly enhances the generalization ability of the model under complex working conditions and the robustness to input changes.

[0055] Further introduce the magnetic field energy density loss term, which is a key component of PINN, to constrain the U-Net model output to satisfy the energy conservation characteristics under Maxwell's equations.

[0056] Step 103, determining a comprehensive loss function according to the data-driven loss function and the physical loss function, and training the U-Net model using the comprehensive loss function;

[0057] To ensure that the prediction result has both data-driven accuracy and satisfies the physical law constraint, a physics-informed neural network is introduced to construct a comprehensive loss function for joint optimization.

[0058] Using L total A comprehensive loss function is used to evaluate the model's prediction error, and the model's convergence trend is determined by plotting loss curves during training and validation. Furthermore, the model is deployed on a GPU (Graphics Processing Unit) computing platform to accelerate the training process, ensuring both training efficiency and real-time prediction.

[0059] Step 104: Input the current magnetic field distribution map of the transformer into the trained U-Net model to obtain the current predicted magnetic field image of the transformer.

[0060] This embodiment retains the traditional image reconstruction capabilities of the U-Net model while introducing the physical laws of magnetic field distribution as a physical loss term. This term is embedded in the U-Net model training process to improve the physical rationality of the predicted image in non-measurement areas. During the model training process, in addition to minimizing the reconstruction error between the predicted magnetic field image and the target magnetic field image, the residual of the prediction result under the physical laws is also minimized. This integrates data-driven modeling with the laws of electromagnetic physics, significantly improving the accuracy, physical consistency, and generalization capability of transformer magnetic field prediction.

[0061] Based on the above embodiment, this embodiment further includes: before inputting the magnetic field distribution map sample of the transformer into the U-Net model to obtain the predicted magnetic field image of the transformer output by the U-Net model:

[0062] Based on the spatial coordinates of the magnetic field measurement points of the transformer, a magnetic field distribution diagram of the transformer is calculated using a Gaussian distribution function to simulate an area affected by the magnetic field strength of the transformer;

[0063] The diffusion range of the Gaussian distribution function is controlled by the standard deviation parameter std, which is used to normalize the spatial impact.

[0064] Based on the above embodiment, in this embodiment, the data-driven loss function is determined according to the pixel difference between the predicted magnetic field image and the target magnetic field image of the transformer by the following formula:

[0065]

[0066] Among them, L1 loss is the data-driven loss function, N is the total number of pixels in the magnetic field distribution sample of the transformer, and P i and T i are respectively the pixel value of the i-th pixel in the predicted magnetic field image of the transformer and the pixel value of the i-th pixel in the target magnetic field image of the transformer.

[0067] On the basis of the above-mentioned embodiments, in this embodiment, the magnetic field energy density of the transformer is determined according to the predicted magnetic field image of the transformer by the following formula:

[0068]

[0069] Wherein, ω m is the magnetic field energy density of the transformer in unit volume, B is the magnetic induction intensity of the transformer, which is determined by the normalized mapping relationship between the pixel value in the predicted magnetic field image and the magnetic induction intensity, and μ is the magnetic permeability.

[0070] Each pixel value in the predicted magnetic field image P reflects the local magnetic induction intensity B through normalization. B is the magnetic induction intensity in the predicted magnetic field image P, which is determined by the normalized mapping relationship between the pixel value in the predicted magnetic field image P and the magnetic induction intensity, and is used to calculate the magnetic field energy density in the physical loss function. μ is the magnetic permeability, which is given by an external constant parameter.

[0071] On the basis of the above-mentioned embodiments, in this embodiment, the physical loss function is determined according to the magnetic field energy density of the transformer by the following formula:

[0072]

[0073] Wherein, Physics loss is the physical loss function, N is the total number of pixel points in the magnetic field distribution pattern sample of the transformer, B i is the magnetic induction intensity mapped by the i-th pixel point in the predicted magnetic field image of the transformer.

[0074] The physical loss function Physics loss is the average of the square sum of the magnetic field energy density of all pixel points in the predicted magnetic field image of the transformer. This loss term measures the rationality of the predicted magnetic field in energy distribution, punishes the output results that are not consistent, and encourages the network to learn the distribution pattern that conforms to the electromagnetic energy conservation.

[0075] On the basis of the above-mentioned embodiments, in this embodiment, the comprehensive loss function is determined according to the data-driven loss function and the physical loss function by the following formula:

[0076] L total =L1 loss +λPhysics loss

[0077] Wherein, L total is the comprehensive loss function, L1 loss is the data-driven loss function, and Physics lossis a physical loss function, and λ is an adjustment coefficient for controlling the degree of influence of the physical loss function.

[0078] The joint optimization target is constructed by merging the data-driven loss and the physical loss by weighting. The λ controls the influence of the physical constraint in the training, which can be adjusted through experiments to balance the prediction accuracy and physical consistency.

[0079] On the basis of the above embodiments, the present embodiment further comprises:

[0080] The U-Net model is trained using the Adam optimizer.

[0081] The optimizer uses Adam, and the initial learning rate is optionally set to 0.0002, and a linear decay strategy is used to adjust the learning rate after the 40th round. During the training process, the training and validation losses are continuously recorded, and when the validation loss reaches the optimal value, the current model weight is saved to net_G.pth.

[0082] The present embodiment uses the Adam optimization algorithm for gradient update, and combines a dynamic learning rate adjustment strategy to adaptively adjust the learning rate to speed up convergence and avoid falling into local optima. At the same time, an early stopping mechanism is introduced to dynamically determine the training stopping time according to the validation set loss, avoiding overfitting.

[0083] On the basis of the above embodiments, the U-Net model in the present embodiment comprises a multi-scale feature extraction module and an attention mechanism.

[0084] The improved U-Net model is used in the present embodiment, which adds a multi-scale feature extraction module and an attention mechanism to enhance the perception ability of key areas and different scale information based on the preservation of the encoding-decoding structure. At the same time, depth separable convolution is introduced to reduce the parameter quantity and computational complexity, and improve the efficiency and generalization ability of the model.

[0085] The present embodiment enhances the model's ability to express spatial details and complex distributions by introducing a multi-scale feature extraction module and a noise suppression mechanism.

[0086] Load the trained optimal model to predict and infer the new magnetic field distribution map. The test image is uniformly processed to 256x256 size and input into the U-Net model for inference, and the predicted magnetic field image P is output. The predicted magnetic field image P and the original target magnetic field image T are jointly visualized by matplotlib. The final displayed images include: input image (input_xxx), target image (target_xxx) and predicted image (out_xxx), and are saved as corresponding files. The complete flowchart of transformer magnetic field prediction is shown in Figure 3 .

[0087] The transformer magnetic field prediction result output by the model is compared with the real target as shown in the following table. Figure 4 As shown in the figure, by comparing the training effects of the model under different n_epochs and lr, and observing the changes of the training loss and the validation loss, it can be concluded that when n_epochs = 50 and lr = 0.0001, the training loss = 0.0078 and the validation loss = 0.0116, the training and validation losses are both low, and the training effect is optimal.

[0088] The embodiment adopts the PyTorch deep learning framework to realize the U-Net image prediction model based on the physical information neural network. The model introduces electromagnetic physical laws as part of the loss function on the basis of retaining the traditional image reconstruction capability, effectively improving the physical consistency and accuracy of the magnetic field distribution prediction. The embodiment can provide more accurate and reliable data support for the prediction, state monitoring and fault diagnosis of the transformer magnetic field distribution.

[0089] The transformer magnetic field prediction system based on the physical information neural network provided by the application is described below, and the transformer magnetic field prediction system based on the physical information neural network described below can be mutually corresponding and referred to the transformer magnetic field prediction method based on the physical information neural network described above.

[0090] As shown in the figure, the system comprises a first prediction module 501, a loss calculation module 502, a model training module 503 and a second prediction module 504, wherein: Figure 5

[0091] The first prediction module 501 is used for inputting the magnetic field distribution pattern sample of the transformer into the U-Net model to obtain the predicted magnetic field image of the transformer output by the U-Net model;

[0092] The loss calculation module 502 is used for determining a data-driven loss function according to the pixel difference between the predicted magnetic field image of the transformer and the target magnetic field image, determining the magnetic field energy density of the transformer according to the predicted magnetic field image of the transformer, and determining a physical loss function according to the magnetic field energy density of the transformer;

[0093] The model training module 503 is used for determining a comprehensive loss function according to the data-driven loss function and the physical loss function, and training the U-Net model using the comprehensive loss function;

[0094] The second prediction module 504 is used for inputting the current magnetic field distribution pattern of the transformer into the trained U-Net model to obtain the current predicted magnetic field image of the transformer.

[0095] ​The embodiment introduces the physical law in the magnetic field distribution as a physical loss term in the training process of the U-Net model on the basis of reserving the traditional image reconstruction capability of the U-Net model, so as to improve the physical rationality of the predicted image in the non-measurement area; in the model training process, in addition to minimizing the reconstruction error between the predicted magnetic field image and the target magnetic field image, the residual of the predicted result under the physical law is also minimized, the data-driven modeling and electromagnetic physical law are fused, and the accuracy, physical consistency and generalization ability of the transformer magnetic field prediction are significantly improved.

[0096] Figure 6 An example of a schematic diagram of a physical structure of an electronic device is shown as Figure 6 The electronic device can include a processor 610, a communications interface 620, a memory 630, and a communications bus 640, wherein the processor 610, the communications interface 620, and the memory 630 communicate with each other through the communications bus 640. The processor 610 can invoke the logical instructions in the memory 630 to execute the transformer magnetic field prediction method based on the physical information neural network, which includes: inputting a magnetic field distribution pattern sample of a transformer into a U-Net model to obtain a predicted magnetic field image of the transformer output by the U-Net model; determining a data-driven loss function according to the pixel difference between the predicted magnetic field image of the transformer and a target magnetic field image, determining the magnetic field energy density of the transformer according to the predicted magnetic field image of the transformer, and determining a physical loss function according to the magnetic field energy density of the transformer; determining a comprehensive loss function according to the data-driven loss function and the physical loss function, and training the U-Net model using the comprehensive loss function; inputting a current magnetic field distribution pattern of the transformer into the trained U-Net model to obtain a current predicted magnetic field image of the transformer.

[0097] In addition, the logical instructions in the memory 630 described above can be implemented in the form of a software function unit and sold or used as an independent product, which can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0098] In another aspect, the present application also provides a computer program product comprising a computer program, which can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to perform the transformer magnetic field prediction method based on a physical information neural network provided by the above method, which comprises: inputting a magnetic field distribution pattern sample of a transformer into a U-Net model to obtain a predicted magnetic field image of the transformer output by the U-Net model; determining a data-driven loss function according to the pixel difference between the predicted magnetic field image of the transformer and a target magnetic field image, determining the magnetic field energy density of the transformer according to the predicted magnetic field image of the transformer, and determining a physical loss function according to the magnetic field energy density of the transformer; determining a comprehensive loss function according to the data-driven loss function and the physical loss function, and training the U-Net model using the comprehensive loss function; and inputting a current magnetic field distribution pattern of the transformer into the trained U-Net model to obtain a current predicted magnetic field image of the transformer.

[0099] In yet another aspect, the present application also provides a non-transitory computer readable storage medium having a computer program stored thereon, which can be executed by a processor to implement the transformer magnetic field prediction method based on a physical information neural network provided by the above method, which comprises: inputting a magnetic field distribution pattern sample of a transformer into a U-Net model to obtain a predicted magnetic field image of the transformer output by the U-Net model; determining a data-driven loss function according to the pixel difference between the predicted magnetic field image of the transformer and a target magnetic field image, determining the magnetic field energy density of the transformer according to the predicted magnetic field image of the transformer, and determining a physical loss function according to the magnetic field energy density of the transformer; determining a comprehensive loss function according to the data-driven loss function and the physical loss function, and training the U-Net model using the comprehensive loss function; and inputting a current magnetic field distribution pattern of the transformer into the trained U-Net model to obtain a current predicted magnetic field image of the transformer.

[0100] The system embodiments described above are only illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the present embodiment scheme according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0101] Those skilled in the art can clearly understand the technical solutions of the various embodiments from the above description of the embodiments, and the various embodiments can be implemented by means of software with the necessary general hardware platforms, and of course, can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part of the prior art that makes a contribution, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, and the like, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0102] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A transformer magnetic field prediction method based on physical information neural network, characterized in that: include: Inputting a sample of the magnetic field distribution map of the transformer into a U-Net model to obtain a predicted magnetic field image of the transformer output by the U-Net model; determining a data-driven loss function based on a pixel difference between a predicted magnetic field image and a target magnetic field image of the transformer, determining a magnetic field energy density of the transformer based on the predicted magnetic field image of the transformer, and determining a physical loss function based on the magnetic field energy density of the transformer; Determine a comprehensive loss function based on the data-driven loss function and the physical loss function, and train the U-Net model using the comprehensive loss function; The current magnetic field distribution map of the transformer is input into the trained U-Net model to obtain the current predicted magnetic field image of the transformer.

2. The transformer magnetic field prediction method based on physical information neural network according to claim 1 is characterized in that: Before inputting the magnetic field distribution map sample of the transformer into the U-Net model to obtain the predicted magnetic field image of the transformer output by the U-Net model, the method further includes: Based on the spatial coordinates of the magnetic field measurement points of the transformer, a magnetic field distribution diagram of the transformer is calculated using a Gaussian distribution function to simulate an area affected by the magnetic field strength of the transformer; The diffusion range of the Gaussian distribution function is controlled by a standard deviation parameter.

3. The transformer magnetic field prediction method based on physical information neural network according to claim 1 is characterized in that: The data-driven loss function is determined based on the pixel difference between the predicted magnetic field image and the target magnetic field image of the transformer by the following formula: Among them, L1 loss is the data-driven loss function, N is the total number of pixels in the magnetic field distribution sample of the transformer, and P i and T i are respectively the pixel value of the i-th pixel in the predicted magnetic field image of the transformer and the pixel value of the i-th pixel in the target magnetic field image of the transformer.

4. The transformer magnetic field prediction method based on physical information neural network according to claim 1 is characterized in that: The magnetic field energy density of the transformer is determined based on the predicted magnetic field image of the transformer using the following formula: Among them, ω m is the magnetic field energy density of the transformer per unit volume, B is the magnetic induction intensity of the transformer, which is determined by the normalized mapping relationship between the pixel value in the predicted magnetic field image and the magnetic induction intensity, and μ is the magnetic permeability.

5. The transformer magnetic field prediction method based on physical information neural network according to claim 4 is characterized in that: The physical loss function is determined based on the magnetic field energy density of the transformer using the following formula: Among them, Physics loss is the physical loss function, N is the total number of pixels in the magnetic field distribution sample of the transformer, B i is the magnetic induction intensity mapped by the i-th pixel point in the predicted magnetic field image of the transformer.

6. The transformer magnetic field prediction method based on physical information neural network according to claim 1 is characterized in that: The comprehensive loss function is determined based on the data-driven loss function and the physical loss function using the following formula: L total =L1 loss +λPhysics loss Among them, L total is the comprehensive loss function, L1 loss For data-driven loss functions, Physics loss is the physical loss function, and λ is the adjustment coefficient, which is used to control the influence of the physical loss function.

7. The transformer magnetic field prediction method based on physical information neural network according to any one of claims 1 to 6, characterized in that: Also includes: The U-Net model is trained using the Adam optimizer.

8. The transformer magnetic field prediction method based on physical information neural network according to any one of claims 1 to 6, characterized in that: The U-Net model includes a multi-scale feature extraction module and an attention mechanism.

9. A transformer magnetic field prediction system based on physical information neural network, characterized in that: include: A first prediction module is used to input a sample of the magnetic field distribution map of the transformer into a U-Net model to obtain a predicted magnetic field image of the transformer output by the U-Net model; a loss calculation module, configured to determine a data-driven loss function based on a pixel difference between a predicted magnetic field image and a target magnetic field image of the transformer, determine a magnetic field energy density of the transformer based on the predicted magnetic field image of the transformer, and determine a physical loss function based on the magnetic field energy density of the transformer; A model training module, configured to determine a comprehensive loss function based on the data-driven loss function and the physical loss function, and train the U-Net model using the comprehensive loss function; The second prediction module is used to input the current magnetic field distribution map of the transformer into the trained U-Net model to obtain the current predicted magnetic field image of the transformer.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the transformer magnetic field prediction method based on physical information neural network as described in any one of claims 1 to 8 is implemented.

Images