Neural network-based magnetic core loss prediction method and related device

Abstract

The invention discloses a magnetic core loss prediction method based on a neural network and a related device, and the method comprises the steps: obtaining the state information of a to-be-measured magnetic core, the state information of the to-be-measured magnetic core comprises the magnetic flux density, amplitude, temperature and frequency; preprocessing the state information of the magnetic core to be tested to obtain the preprocessed state information of the magnetic core to be tested; the preprocessed state information of the magnetic core to be measured is input into the trained prediction model, the loss value of the magnetic core to be measured is obtained, the prediction model is constructed based on a CNN + Transformer + cross attention hybrid architecture, the method and the related device can predict the magnetic core loss power, and accuracy and applicability are high.

Description

Technical Field

[0001] This invention belongs to the field of power converter efficiency optimization technology, and relates to a method and related device for predicting magnetic core loss based on neural networks. Background Technology

[0002] In the pursuit of higher frequencies and higher power densities, the thermal management of magnetic components in power electronic equipment has become increasingly prominent. High-frequency inverters widely use small and lightweight high-frequency magnetic components, but high-frequency operation leads to a significant increase in core losses, causing core overheating and severely affecting the efficiency and reliability of the equipment.

[0003] Core loss prediction is crucial for thermal management; however, current mainstream prediction methods have significant limitations: Steinmetz's empirical formula and its modified model are only applicable to sinusoidal excitation, with parameters varying drastically with frequency, flux density, and temperature, resulting in complex calculations and large errors; models based on physical characteristics, such as Preisach and Jiles-Atherton, while highly accurate, are only suitable for static or low-frequency conditions and are difficult to extend to high-frequency non-sinusoidal excitation; methods based on experimental data fitting can adapt to non-sinusoidal waveforms, but rely on extensive experiments, have poor versatility, and cannot accurately reflect the impact of SPWM control parameters such as modulation and carrier ratio on hysteresis and eddy current losses. Furthermore, the DC bias effect is significant under high-frequency conditions, and residual losses and winding and core losses are severely coupled, further increasing the difficulty of modeling and prediction. Therefore, there is an urgent need to develop a universal core loss prediction model applicable to high-frequency non-sinusoidal excitation and considering the influence of modulation parameters to improve the thermal management and design optimization capabilities of power electronic equipment. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and device for predicting magnetic core loss based on neural networks. This method and device can predict the power of magnetic core loss with high accuracy and applicability.

[0005] To achieve the above objectives, this invention discloses a method for predicting magnetic core loss based on a neural network, comprising: The state information of the magnetic core under test is obtained, including magnetic flux density, amplitude, temperature and frequency. The state information of the magnetic core under test is preprocessed to obtain the preprocessed state information of the magnetic core under test. The preprocessed state information of the magnetic core under test is input into the trained prediction model to obtain the loss value of the magnetic core under test. The prediction model is constructed based on a hybrid architecture of CNN+Transformer+cross attention.

[0006] A further improvement of the neural network-based core loss prediction method described in this invention is as follows: Furthermore, the process of preprocessing the state information of the magnetic core under test to obtain the preprocessed state information of the magnetic core under test is as follows: Frequency and time domain analysis is performed on the waveform characteristics of magnetic flux density to extract duty cycle, slope, rise and fall time, and depth features from the waveform. The duty cycle, slope, rise and fall time, and depth features are then merged with the state information of the magnetic core under test. The merged result is then transformed using Log1p to obtain the preprocessed state information of the magnetic core under test.

[0007] Furthermore, before inputting the preprocessed state information of the magnetic core under test into the trained prediction model, the process also includes: Build the training dataset; A prediction model is built based on the aforementioned CNN+Transformer+cross-attention hybrid architecture; The prediction model is trained using the training dataset to obtain the trained prediction model.

[0008] Furthermore, during the training of the prediction model using the training dataset, the loss function corresponding to CNN is: total loss = λ1 × Huber loss + λ2 × physical consistency loss, where λ1 and λ2 are adjustable empirical values.

[0009] This invention discloses a magnetic core loss prediction system based on a neural network, comprising: The acquisition module is used to acquire the state information of the magnetic core under test, which includes magnetic flux density, amplitude, temperature and frequency. The preprocessing module is used to preprocess the state information of the magnetic core under test to obtain the preprocessed state information of the magnetic core under test. The prediction module is used to input the preprocessed state information of the magnetic core to be tested into the trained prediction model to obtain the loss value of the magnetic core to be tested. The prediction model is built based on a hybrid architecture of CNN+Transformer+cross attention.

[0010] A further improvement of the neural network-based core loss prediction method described in this invention is as follows: Furthermore, the process of preprocessing the state information of the magnetic core under test to obtain the preprocessed state information of the magnetic core under test is as follows: Frequency and time domain analysis is performed on the waveform characteristics of magnetic flux density to extract duty cycle, slope, rise and fall time, and depth features from the waveform. The duty cycle, slope, rise and fall time, and depth features are then merged with the state information of the magnetic core under test. The merged result is then transformed using Log1p to obtain the preprocessed state information of the magnetic core under test.

[0011] Furthermore, it also includes: The first building block is used to construct the training dataset; The second building module is used to build a prediction model based on the CNN+Transformer+cross-attention hybrid architecture; The training module is used to train the prediction model using the training dataset to obtain the trained prediction model.

[0012] Furthermore, during the training of the prediction model using the training dataset, the loss function corresponding to CNN is: total loss = λ1 × Huber loss + λ2 × physical consistency loss, where λ1 and λ2 are adjustable empirical values.

[0013] The present invention discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the neural network-based magnetic core loss prediction method.

[0014] The present invention discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the neural network-based magnetic core loss prediction method.

[0015] The present invention has the following beneficial effects: In specific operation, the magnetic core loss prediction method and related device based on neural networks of the present invention input the preprocessed state information of the magnetic core to be tested into the trained prediction model to obtain the loss value of the magnetic core to be tested. The prediction model is constructed based on a hybrid architecture of CNN+Transformer+cross-attention. The Transformer adopts a multi-head self-attention mechanism, a feedforward neural network, and a residual connection structure to realize parameter updates and long-range dependency extraction. At the same time, the present invention preprocesses the state information of the magnetic core to be tested, reducing the dependence on the amount of raw data, improving prediction accuracy and generalization ability, and enhancing the intelligence and physical consistency of the dataset. In addition, the present invention integrates multiple types of parameters for prediction to improve the accuracy of model prediction.

[0016] Furthermore, the Log1p transform is used to process the data to improve the model's ability to handle data of different orders of magnitude and optimize the model's dynamic range. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a graph showing how the average loss changes over the number of training epochs. Figure 2 For R 2 The graph shows how the value changes over training rounds. Figure 3 To verify the error distribution between the training results of the centralized model and the actual values; Figure 4 This is a flowchart of the method of the present invention; Figure 5 This is a structural diagram of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0021] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this invention generally indicates that the preceding and following objects have an "or" relationship.

[0023] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0024] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0026] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0027] Example 1 refer to Figure 4 The magnetic core loss prediction method based on neural networks described in this invention includes the following steps: 1) Obtain the state information of the magnetic core under test; the state information of the magnetic core under test includes magnetic flux density B, amplitude, temperature and frequency; 2) Preprocess the information of the magnetic core to be tested; The specific operation of step 2) is as follows: Frequency and time domain analysis is performed on the waveform characteristics of magnetic flux density to extract features such as duty cycle, slope, rise and fall time, and waveform symmetry index. These features are then spliced ​​with the information of the original magnetic core under test. This not only enhances the model's understanding of the essential physical laws of the waveform, but also reduces the dependence on the amount of original data, improves prediction accuracy and generalization ability, and enhances the intelligence and physical consistency of the dataset.

[0028] To address the significant differences in magnetic flux density data, this invention employs Log1p transformation to process the original data, thereby improving the model's ability to handle data of different orders of magnitude and optimizing the model's dynamic range.

[0029] 3) Input the preprocessed information of the magnetic core to be tested into the trained prediction model to obtain the loss value of the magnetic core to be tested. The prediction model is constructed based on a hybrid architecture of CNN+Transformer+cross attention.

[0030] For CNN convolutional neural networks: the input magnetic flux density waveform sampling data is converted into a three-dimensional data structure to obtain three-dimensional magnetic flux data. The obtained three-dimensional magnetic flux data is then converted into a multi-channel format data structure through three convolutions of 1D CNN to obtain the local features at each time step.

[0031] Regarding the definition of the loss function, the traditional loss function is adjusted. Based on the physical control equation of core loss, the traditional loss function is optimized into a Huber-Steinmtz hybrid loss function. Total loss = λ1 × Huber loss (predicted value, actual value) + λ2 × physical consistency loss. The physical consistency loss is derived from the estimated value of the Steinmtz equation. When there is a significant deviation between the model's prediction and the result calculated by the Steinmtz formula, the physical consistency loss term will generate a gradient penalty, guiding the model to converge in a physically reasonable direction. This mechanism is beneficial for the model to make reasonable predictions under extreme conditions or in areas with insufficient training data coverage.

[0032] For the Transformer+ cross-attention part: the waveform feature data output by the CNN convolutional neural network is fed into a 3-layer Transformer encoder after adding position encoding to establish long-range dependencies.

[0033] Global pooling yields waveform features, and the three-dimensional parameters of amplitude, temperature, and frequency are processed by a fully connected network to increase the dimensionality to 64 dimensions.

[0034] Cross-attention is performed using parameter features as queries and waveform features as key-value pairs to achieve parameter focusing on key regions of the waveform.

[0035] Finally, the fused vector is processed through four fully connected layers to output the loss value.

[0036] Furthermore, the training process of the prediction model is as follows: Import the magnetic flux density waveform, magnetic flux amplitude, temperature, and frequency data from the database into the model. The input format is 1027 columns per row. Columns 1-1024 represent the 1024 sampling points of the magnetic flux density B in a single period, column 1025 represents the amplitude (maximum value) of B, and columns 1026 and 1027 represent the temperature (°C) and frequency (Hz).

[0037] For the magnetic flux density waveforms in columns 1-1024, their time-domain and frequency-domain features, along with higher-order features such as the waveform symmetry index, are extracted and combined into a 32-bit array, which is then concatenated after the waveform data.

[0038] To address the significant differences in magnetic flux density data, this invention employs Log1p transformation to process the original data, thereby improving the model's ability to handle data of different orders of magnitude and optimizing the model's dynamic range.

[0039] Feature extraction was performed on the processed magnetic flux density waveform using a convolutional neural network. The initial data was a B*1*1056 three-dimensional data structure, where B represents the magnetic flux density amplitude. The obtained three-dimensional magnetic flux data was transformed into a B*128*36 format data structure after three convolutions using a 1D CNN, which fully extracted features from the waveform at small, medium, and large time scales, resulting in 128-dimensional local features at each time step.

[0040] The loss function is: Total loss = λ1 × Huber loss + λ2 × Physical consistency loss, where λ1 and λ2 are adjustable empirical values, and the physical consistency loss is calculated by the Huber-Steinmtz equation for calculating core loss.

[0041] The next step is to import the local features of the magnetic flux waveform obtained after CNN processing into the Transformer model. In the Transformer model, a multi-head attention mechanism is used to extract different features, and finally, the features are concatenated and fused, while preserving both the conditional waveform features and the original parameter features.

[0042] Regarding learning rate settings, during pre-training, the learning rate is smoothly increased from the initial value to 10 along an exponential curve. The loss-learning rate curve is recorded, and the learning rate at the steepest gradient descent point is taken as the "optimal initial learning rate." During formal training, the optimizer restarts with this learning rate and uses the OneCycleLR scheduler: total steps = number of epochs × number of batches per epoch. The first 30% of steps are the ascending segment, where the learning rate linearly increases from "optimal initial learning rate / 25" to "optimal initial learning rate." Subsequently, the learning rate linearly decreases to "optimal initial learning rate / 10000" for the next 70% of steps, and then decreases slightly again for the last 10% of steps. The momentum changes synchronously in the opposite direction (momentum decreases during the ascending segment and increases during the descending segment), forming a single-cycle large-small-mini learning rate curve. Combined with gradient pruning, this enables the network to converge quickly and avoids oscillations.

[0043] In terms of data difficulty assessment, the initial difficulty classification is based on the harmonic components and frequencies of the data. Low-frequency, low-harmonic data are defined as low-difficulty data and prioritized for training. In subsequent training, the dataset difficulty is updated in real time based on the loss value after different data iterations, realizing a personalized learning path.

[0044] Loss values ​​and R values ​​at different training epochs 2 Values ​​such as Figure 1 and Figure 2 As shown, the performance of this invention in this embodiment is as follows: Figure 3 As shown, it exhibits good prediction accuracy.

[0045] The final training error of this invention is the original R. 2 =0.982568, weighted original R 2 =0.998752, 95% accuracy error, correct rate is 89.82%.

[0046] Example 2 refer to Figure 5 The magnetic core loss prediction system based on neural networks of the present invention includes: The acquisition module is used to acquire the state information of the magnetic core under test, which includes magnetic flux density, amplitude, temperature and frequency. The preprocessing module is used to preprocess the state information of the magnetic core under test to obtain the preprocessed state information of the magnetic core under test. The prediction module is used to input the preprocessed state information of the magnetic core to be tested into the trained prediction model to obtain the loss value of the magnetic core to be tested. The prediction model is built based on a hybrid architecture of CNN+Transformer+cross attention.

[0047] In this embodiment, the process of preprocessing the state information of the magnetic core under test to obtain the preprocessed state information of the magnetic core under test is as follows: Frequency and time domain analysis is performed on the waveform characteristics of magnetic flux density to extract duty cycle, slope, rise and fall time, and depth features from the waveform. The duty cycle, slope, rise and fall time, and depth features are then merged with the state information of the magnetic core under test. The merged result is then transformed using Log1p to obtain the preprocessed state information of the magnetic core under test.

[0048] This embodiment also includes: The first building block is used to construct the training dataset; The second building module is used to build a prediction model based on the CNN+Transformer+cross-attention hybrid architecture; The training module is used to train the prediction model using the training dataset to obtain the trained prediction model.

[0049] In this embodiment, during the training of the prediction model using the training dataset, the loss function corresponding to CNN is: total loss = λ1 × Huber loss + λ2 × physical consistency loss, where λ1 and λ2 are adjustable empirical values.

[0050] The module division in this embodiment is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in each embodiment of this application can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0051] Example 3 A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the neural network-based magnetic core loss prediction method. For example, the steps include: acquiring state information of a magnetic core under test, including magnetic flux density, amplitude, temperature, and frequency; preprocessing the state information of the magnetic core under test to obtain preprocessed state information; and inputting the preprocessed state information of the magnetic core under test into a trained prediction model to obtain the loss value of the magnetic core under test. The prediction model is constructed based on a CNN+Transformer+cross-attention hybrid architecture. The memory may include main memory, such as high-speed random access memory, or it may also include non-volatile memory, such as at least one disk storage device. The processor, network interface, and memory are interconnected via an internal bus, which may be an industry-standard architecture bus, a peripheral component interconnection standard bus, or an extended industry-standard architecture bus. The bus can be categorized as an address bus, data bus, and control bus. The memory stores the program, specifically, the program may include program code, which includes computer operation instructions. Memory can include main memory and non-volatile memory, and provides instructions and data to the processor.

[0052] Example 4 A computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the neural network-based magnetic core loss prediction method. For example, the method includes: acquiring state information of a magnetic core under test, including magnetic flux density, amplitude, temperature, and frequency; preprocessing the state information to obtain preprocessed state information; and inputting the preprocessed state information into a trained prediction model to obtain the loss value of the magnetic core. The prediction model is constructed based on a CNN+Transformer+cross-attention hybrid architecture. Specifically, the computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. The volatile memory may include random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include read-only memory (ROM), hard disk, flash memory, optical disk, magnetic disk, etc.

[0053] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0054] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0055] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0056] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0057] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and disclosure of the invention. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the following claims.

[0058] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

[0059] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any simple modifications, alterations, or equivalent structural changes made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for predicting magnetic core loss based on neural networks, characterized in that, include: The state information of the magnetic core under test is obtained, including magnetic flux density, amplitude, temperature and frequency. The state information of the magnetic core under test is preprocessed to obtain the preprocessed state information of the magnetic core under test. The preprocessed state information of the magnetic core under test is input into the trained prediction model to obtain the loss value of the magnetic core under test. The prediction model is constructed based on a hybrid architecture of CNN+Transformer+cross attention.

2. The method for predicting magnetic core loss based on neural networks according to claim 1, characterized in that, The process of preprocessing the state information of the magnetic core under test to obtain the preprocessed state information of the magnetic core under test is as follows: Frequency and time domain analysis is performed on the waveform characteristics of magnetic flux density to extract duty cycle, slope, rise and fall time, and depth features from the waveform. The duty cycle, slope, rise and fall time, and depth features are then merged with the state information of the magnetic core under test. The merged result is then transformed using Log1p to obtain the preprocessed state information of the magnetic core under test.

3. The method for predicting magnetic core loss based on neural networks according to claim 1, characterized in that, Before inputting the preprocessed state information of the magnetic core under test into the trained prediction model, the following steps are also included: Build the training dataset; A prediction model is built based on the aforementioned CNN+Transformer+cross-attention hybrid architecture; The prediction model is trained using the training dataset to obtain the trained prediction model.

4. The method for predicting magnetic core loss based on neural networks according to claim 3, characterized in that, During the training of the prediction model using the training dataset, the loss function corresponding to CNN is: total loss = λ1 × Huber loss + λ2 × physical consistency loss, where λ1 and λ2 are adjustable empirical values.

5. A magnetic core loss prediction system based on neural networks, characterized in that, include: The acquisition module is used to acquire the state information of the magnetic core under test, which includes magnetic flux density, amplitude, temperature and frequency. The preprocessing module is used to preprocess the state information of the magnetic core under test to obtain the preprocessed state information of the magnetic core under test. The prediction module is used to input the preprocessed state information of the magnetic core to be tested into the trained prediction model to obtain the loss value of the magnetic core to be tested. The prediction model is built based on a hybrid architecture of CNN+Transformer+cross attention.

6. The core loss prediction system based on neural networks according to claim 5, characterized in that, The process of preprocessing the state information of the magnetic core under test to obtain the preprocessed state information of the magnetic core under test is as follows: Frequency and time domain analysis is performed on the waveform characteristics of magnetic flux density to extract duty cycle, slope, rise and fall time, and depth features from the waveform. The duty cycle, slope, rise and fall time, and depth features are then merged with the state information of the magnetic core under test. The merged result is then transformed using Log1p to obtain the preprocessed state information of the magnetic core under test.

7. The core loss prediction system based on neural networks according to claim 5, characterized in that, Also includes: The first building block is used to construct the training dataset; The second building module is used to build a prediction model based on the CNN+Transformer+cross-attention hybrid architecture; The training module is used to train the prediction model using the training dataset to obtain the trained prediction model.

8. The core loss prediction system based on neural networks according to claim 7, characterized in that, During the training of the prediction model using the training dataset, the loss function corresponding to CNN is: total loss = λ1 × Huber loss + λ2 × physical consistency loss, where λ1 and λ2 are adjustable empirical values.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the neural network-based magnetic core loss prediction method as described in any one of claims 1-4.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the neural network-based core loss prediction method as described in any one of claims 1-4.

Images