Magnetic core loss prediction method

Abstract

The invention discloses a magnetic core loss prediction method in the technical field of electromagnetic material prediction, and the method comprises the steps: obtaining environment parameters of a to-be-measured magnetic core and a one-dimensional magnetic flux density time sequence signal, carrying out the prediction based on a pre-trained magnetic core loss prediction model, and obtaining a prediction value of the magnetic core loss; the data processing method of the magnetic core loss prediction model comprises the following steps: converting environmental parameters into a tensor form to obtain parameter feature vectors; converting the one-dimensional magnetic flux density time sequence signal into a two-dimensional time-frequency spectrogram, and performing feature extraction on the two-dimensional time-frequency spectrogram to obtain an image feature vector; splicing the parameter feature vector and the image feature vector to obtain comprehensive feature representation; performing full connection layer processing on the comprehensive characteristic representation to obtain a predicted value of the magnetic core loss; wherein the environment parameters comprise continuous variables and discrete variables. According to the method, the problem that the magnetic core loss is difficult to accurately predict under high-frequency or non-sine excitation in the prior art is solved.

Description

Technical Field

[0001] The invention relates to a magnetic core loss prediction method, belonging to the technical field of electromagnetic material prediction. Background Art

[0002] In the field of power electronic equipment, core loss, as a key parameter for evaluating equipment efficiency and performance, has always received widespread attention in the industry. Traditional core loss prediction technology is mainly based on experimental measurement and theoretical calculation. The experimental measurement method obtains the loss data of the core under different conditions through actual testing. Although the results are accurate, the measurement cycle is long, the cost is high, and it is not suitable for all application scenarios. On the other hand, the theoretical calculation method estimates the core loss by establishing a mathematical model based on the principle of electromagnetics. Although this method is low-cost, the calculation process is complicated, and it is often difficult to achieve high accuracy in the prediction of core loss under complex working conditions. In recent years, with the rapid development of big data and artificial intelligence technology, machine learning algorithms have been introduced into the field of core loss prediction, providing new ideas for solving this problem.

[0003] Although machine learning algorithms have shown great potential in core loss prediction, existing prediction methods still have some significant shortcomings. Specifically, existing prediction models often only consider data from a single mode, such as prediction based only on waveform data or only on material property data. This single-mode data processing method ignores the fact that core loss is affected by multiple factors, resulting in limited prediction accuracy. For example, relying solely on waveform data may not fully reflect the loss characteristics of core materials under different frequencies, temperatures, and other conditions; relying solely on material property data may ignore the impact of waveform changes on losses under actual working conditions. Therefore, under complex working conditions such as high frequency or non-sinusoidal excitation, existing prediction methods often find it difficult to accurately predict core loss, thereby limiting the performance optimization and efficiency improvement of power electronic equipment. To address this issue, the industry urgently needs a core loss prediction method that can comprehensively consider multiple factors and achieve high-precision prediction.

[0004] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgement or any form of suggestion that the information constitutes the prior art already known to ordinary technicians in this field. Summary of the invention

[0005] The technical problem to be solved by the present invention is that it is difficult to accurately predict the core loss under high frequency or non-sinusoidal excitation in the prior art.

[0006] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions.

[0007] The present invention provides a method for predicting magnetic core loss, comprising:

[0008] Obtain environmental parameters and one-dimensional magnetic flux density time series signal of the magnetic core to be tested;

[0009] According to the environmental parameters and the one-dimensional magnetic flux density time series signal, based on a pre-trained magnetic core loss prediction model, prediction is performed to obtain a predicted value of the magnetic core loss;

[0010] The data processing method of the magnetic core loss prediction model includes:

[0011] Convert the environmental parameters into tensor form to obtain parameter feature vectors;

[0012] Converting the one-dimensional magnetic flux density time series signal into a two-dimensional time-frequency spectrum diagram, performing feature extraction on the two-dimensional time-frequency spectrum diagram to obtain an image feature vector;

[0013] splicing the parameter feature vector and the image feature vector to obtain a comprehensive feature representation;

[0014] Processing the comprehensive feature representation through a fully connected layer to obtain a predicted value of the core loss;

[0015] The environmental parameters include continuous variables and discrete variables, the continuous variables include temperature and frequency, and the discrete variables include material properties and waveform types.

[0016] Based on the aforementioned magnetic core loss prediction method, before converting the environmental parameters into a tensor form, it also includes standardizing the continuous variables using the quartile method, wherein the expression for standardizing the continuous variables using the quartile method is:

[0017] (1);

[0018] In the formula, represents the standardized environmental parameters, Represents the original environment parameters, represents the median of the original environmental parameters, represents the interquartile range, where , Q1 represents the first quartile of the original environmental parameters, and Q3 represents the third quartile of the original environmental parameters.

[0019] Based on the aforementioned magnetic core loss prediction method, before converting the environmental parameters into a tensor form, it also includes standardizing the discrete variables using One-Hot encoding.

[0020] Based on the aforementioned magnetic core loss prediction method, a method for standardizing the discrete variables using One-Hot coding includes:

[0021] determining all categories of material properties and waveform types;

[0022] For each material category and waveform category, create a material category binary column and a waveform category binary column respectively;

[0023] When the discrete variable belongs to the determined material category, a 1 is marked at the corresponding position of the material category binary column, otherwise a 0 is marked;

[0024] When the discrete variable belongs to the determined waveform category, a 1 is marked at the corresponding position of the waveform category binary column, otherwise a 0 is marked.

[0025] Based on the aforementioned core loss prediction method, the material category binary column is represented as:

[0026] M i × i = [ M 1 M 2 … M i ] = [ 1 0 … 0 0 1 … 0 0 0 … 0 0 0 … 1 ] (2);

[0027] In the formula, A binary column representing material category, The first category represents material properties, The second category represents material properties, The material properties categories, material category binary column Each row represents the One-Hot code corresponding to a material category;

[0028] The waveform category binary column representation is:

[0029] W j × j = [ W 1 W 2 … W j ] = [ 1 0 … 0 0 1 … 0 0 0 … 0 0 0 … 1 ] (3);

[0030] In the formula, represents the waveform class binary column, The first category represents the waveform type, The second category represents the waveform type. Indicates the waveform type. categories, waveform category binary columns Each row of represents the One-Hot encoding corresponding to a waveform category.

[0031] Based on the aforementioned magnetic core loss prediction method, the one-dimensional magnetic flux density time series signal is converted into a two-dimensional time-frequency spectrum diagram, including:

[0032] Using Morlet wavelet transform, the one-dimensional magnetic flux density time series signal is decomposed into a set of basis functions formed by wavelet contraction, flattening and translation;

[0033] Extracting and visualizing the coefficients of the basis function to generate a two-dimensional time-frequency spectrum diagram with a size of 1×24×24 pixels;

[0034] Among them, the principle expression of Morlet wavelet transform is:

[0035] (4);

[0036] In the formula, It represents a set of wavelet basis functions decomposed from the one-dimensional magnetic flux density time series signal f(t), Indicates the current time Wavelet The scaling factor of the expansion or contraction degree, The translation factor representing the degree of wavelet transform, Indicates the current moment The differential of represents positive infinity, Represents negative infinity.

[0037] Based on the aforementioned core loss prediction method, a multi-layer convolutional neural network (CNN) is used to extract features from the two-dimensional time-frequency spectrum to obtain an image feature vector, including:

[0038] The first convolution layer including 32 convolution kernels of size 3×3, stride 1 and padding technique is used to capture the basic texture and edge information in the two-dimensional time-frequency spectrum, generating a 32×24×24 high-order tensor;

[0039] The 32×24×24 high-order tensor is subjected to the first maximum pooling using a maximum pooling layer with a configuration window size of 2×2 and a stride of 2, and is compressed into a 32×12×12 high-order tensor;

[0040] The 32×12×12 high-order tensor is converted into a 64×12×12 high-order tensor using a second convolution layer with 64 convolution kernels of size 3×3 and a stride of 1;

[0041] Using a maximum pooling layer with a configuration window size of 2×2 and a stride of 2, the 64×12×12 high-order tensor is subjected to a second maximum pooling, and the dimension is reduced to a multi-dimensional tensor of 64×6×6;

[0042] Using a flattening operation, the 64×6×6 multidimensional tensor is converted into a one-dimensional vector of length 2304;

[0043] The one-dimensional vector with a length of 2304 is mapped to a new 512-dimensional space using a fully connected layer, and a ReLU activation function is introduced for processing to obtain an image feature vector.

[0044] Based on the aforementioned magnetic core loss prediction method, the parameter feature vector and the image feature vector are concatenated to obtain a comprehensive feature representation, including:

[0045] A fully connected layer is used to receive the parameter feature vector and convert it into a five-dimensional vector:

[0046] The five-dimensional vector is converted into a high-level parameter feature vector through a hidden layer containing 128 nodes;

[0047] The high-level parameter feature vector and the image feature vector are concatenated to obtain a comprehensive feature representation.

[0048] Based on the aforementioned core loss prediction method, the comprehensive feature representation is processed through a fully connected layer to obtain a predicted value of the core loss, including:

[0049] The comprehensive feature representation is processed by a fully connected layer and then predicted by another fully connected layer with a single output node to obtain a predicted value of the core loss.

[0050] Based on the aforementioned magnetic core loss prediction method, the mean square error function is used as the loss function, and the model parameters are adjusted based on the Adam optimization algorithm and the loss function to train the magnetic core loss prediction model, wherein the loss function is expressed as:

[0051] (5);

[0052] In the formula, represents the total number of samples, N represents the number of training samples, Indicates The actual core loss value of samples, Indicates The predicted core loss values ​​for samples.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. The present invention innovatively integrates environmental parameters and one-dimensional magnetic flux density time series signals, so that the core loss prediction model can comprehensively consider information from more dimensions, thereby improving the accuracy and comprehensiveness of the prediction. The present invention converts the one-dimensional magnetic flux density time series signal into a two-dimensional time-frequency spectrum, and obtains an image feature vector through feature extraction, which not only retains the time-frequency characteristics of the signal, but also further simplifies the data through feature extraction. The present invention also converts environmental parameters into parameter feature vectors and image feature vectors for fusion, thereby improving processing efficiency.

[0055] 2. The present invention constructs a core loss prediction model based on multimodal data fusion and multiple convolutional neural networks. The core loss prediction model uses a fully connected layer to process comprehensive feature representations to obtain the predicted value of the core loss, making full use of the complementarity of multimodal data to improve prediction accuracy. The present invention demonstrates high accuracy and strong generalization capabilities in core loss prediction through a core loss prediction method based on multimodal data and multiple convolutional neural networks. Compared with traditional single data source or empirical formula methods, the present invention can more accurately predict core loss under different materials, waveforms and working environments.

[0056] 3. The present invention provides a new solution for the performance prediction of magnetic components in power electronic equipment. By accurately predicting the core loss, the present invention not only solves the problem that the existing technology is difficult to accurately predict the core loss under high frequency or non-sinusoidal excitation, but also promotes technological innovation and development in related fields. By integrating multimodal data processing and multiple convolutional neural networks, new ideas and methods are provided for solving similar problems, which can optimize the design and operation of power electronic equipment, improve the efficiency and reliability of equipment, and reduce cost control. This is of great significance to the development and progress of the power electronics industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a technical route flow diagram of a magnetic core loss prediction method provided by an embodiment of the present invention;

[0058] Figure 2 It is a schematic diagram of a time-frequency domain signal processing process based on wavelet transform and multi-layer convolutional neural network provided by an embodiment of the present invention;

[0059] Figure 3 It is a schematic diagram of an environmental parameter processing process of a magnetic core loss prediction method provided by an embodiment of the present invention;

[0060] Figure 4 It is a schematic diagram of a characteristic melting process of a magnetic core loss prediction method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. The embodiments of the present invention and the technical features in the embodiments may be combined with each other unless there is a conflict.

[0062] The term "and / or" is only a description of the association relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " generally indicates that the related objects are in an "or" relationship.

[0063] Example 1

[0064] like Figure 1 As shown, this embodiment introduces a method for predicting magnetic core loss, including:

[0065] Step 1: Obtain the environmental parameters and one-dimensional magnetic flux density time series signal of the magnetic core to be tested.

[0066] The environmental parameters include continuous variables and discrete variables, the continuous variables include temperature and frequency, and the discrete variables include material properties and waveform types.

[0067] The environmental parameters of the magnetic core to be tested provide the external conditions when the magnetic core is working, such as temperature and frequency, as well as the material properties of the magnetic core itself and the waveform type of the excitation signal.

[0068] The one-dimensional magnetic flux density time series signal of the magnetic core to be tested reflects the change of the magnetic flux density inside the magnetic core over time and is an important input data for the prediction of magnetic core loss.

[0069] Environmental parameters and magnetic flux density signals can jointly determine the working state and loss of the magnetic core.

[0070] Step 2: According to the environmental parameters and the one-dimensional magnetic flux density time series signal, based on a pre-trained magnetic core loss prediction model, prediction is performed to obtain a predicted value of the magnetic core loss.

[0071] The pre-trained core loss prediction model has learned the complex relationship between environmental parameters, flux density signals and core loss. The pre-trained model, combined with the current environmental parameters and flux density signals, can quickly and accurately predict the core loss and provide a predicted value of the core loss.

[0072] Wherein, the data processing method of the magnetic core loss prediction model is as follows: Figure 3 As shown, including:

[0073] (1) Convert the environmental parameters into tensor form to obtain a parameter feature vector.

[0074] A tensor is a multidimensional array that can efficiently represent and store multidimensional data. By converting environmental parameters into tensor form, subsequent matrix operations and feature extraction are facilitated.

[0075] This embodiment converts environmental parameters including continuous variables and discrete variables into a tensor form suitable for core loss prediction processing, which facilitates subsequent feature extraction and model training.

[0076] (2) Converting the one-dimensional magnetic flux density time series signal into a two-dimensional time-frequency spectrum diagram, performing feature extraction on the two-dimensional time-frequency spectrum diagram, and obtaining an image feature vector.

[0077] The time-frequency spectrum can reflect the energy distribution of the signal at different times and frequencies, and is an important tool for time series signal analysis. Feature extraction technology can extract useful feature information from the image.

[0078] The present invention converts a one-dimensional time series signal into a two-dimensional time-frequency spectrum, which can intuitively display the changes of the one-dimensional magnetic flux density time series signal in the time and frequency domains, and can extract more useful feature information.

[0079] (3) Concatenate the parameter feature vector and the image feature vector to obtain a comprehensive feature representation, such as Figure 4 shown.

[0080] The concatenation operation can merge feature vectors from two different sources into a comprehensive feature representation, thereby retaining all the information of both.

[0081] The present invention splices the parameter feature vector of the environmental parameter and the image feature vector of the one-dimensional magnetic flux density time series signal to obtain a comprehensive feature representation, and can fully utilize the information of both to make a comprehensive and accurate prediction of the magnetic core to be tested.

[0082] (4) The comprehensive feature representation is processed by a fully connected layer to obtain a predicted value of the core loss.

[0083] The fully connected layer is a basic structure of a neural network, which can perform nonlinear transformation on the input feature vector to obtain the final prediction result.

[0084] The present invention utilizes a fully connected layer to transform the comprehensive feature representation through a weight matrix and an activation function, and learns the complex relationship between the comprehensive feature representation and the predicted value of the core loss. After multiple iterative training, the fully connected layer can continuously adjust the weight matrix and activation function parameters to make the output predicted value of the core loss more accurate.

[0085] The core loss prediction method provided by the present invention combines environmental parameters and one-dimensional magnetic flux density time series signals, and realizes rapid and accurate prediction of core loss through preprocessing, feature extraction and model prediction. The core loss prediction method provided by the present invention takes advantage of the deep learning model and can learn the complex relationship between input features and output prediction values, providing a new idea and method for the prediction of core loss.

[0086] Example 2

[0087] Based on the same inventive concept as Example 1, this example introduces a method for predicting core loss, which specifically includes the following steps:

[0088] Step 1: Obtain the environmental parameters and one-dimensional magnetic flux density time series signal of the magnetic core to be tested.

[0089] First, the experimental environment information such as the type data of the core material, current waveform and frequency data, and experimental temperature are collected, and the magnetic flux density time series signal under the corresponding experimental environment is recorded.

[0090] The collected data is cleaned to remove outliers and missing values ​​to ensure data quality and consistency, and the environmental parameters of the magnetic core to be tested and the one-dimensional magnetic flux density time series signal are obtained.

[0091] The environmental parameters include continuous variables and discrete variables, the continuous variables include temperature and frequency, and the discrete variables include material properties and waveform types.

[0092] Step 2: According to the environmental parameters and the one-dimensional magnetic flux density time series signal, based on a pre-trained magnetic core loss prediction model, prediction is performed to obtain a predicted value of the magnetic core loss.

[0093] The core loss prediction model includes:

[0094] The data preprocessing module is used to standardize the environmental parameters and convert them into tensor data;

[0095] Wavelet transform module, used to convert the one-dimensional magnetic flux density time series signal into a two-dimensional time-frequency spectrum;

[0096] An image processing branch is used to extract features from the two-dimensional time-frequency spectrum diagram according to a multi-layer convolutional neural network to obtain an image feature vector;

[0097] The environmental parameter processing branch is used to process the tensor form data of the environmental parameters according to the fully connected layer to obtain the parameter feature vector;

[0098] The feature fusion layer is used to concatenate the parameter feature vector and the image feature vector to obtain a comprehensive feature representation, and further process the representation through a series of fully connected layers to obtain a predicted value of the core loss.

[0099] During the training process, the data set is divided into training set and test set at a ratio of 4:1. The Adam optimization algorithm and MSE loss function are used to continuously adjust the model parameters to improve the prediction performance of the model.

[0100] Among them, the parameters of the Adam optimizer are preset:

[0101] The initial learning rate Set to 0.001;

[0102] Exponential decay rate to be used for first-order moment estimation The value of is set to 0.9;

[0103] Exponential decay rate to be used for second-order moment estimation The value of is set to 0.999. In addition, for numerical stability, a minimum positive value is introduced , the value is ;

[0104] The number of iterations of the training parameters is set to 100, the batch size is set to 32, and the learning rate scheduling is set to multiply the learning rate by 0.1 every 20 iteration cycles.

[0105] The mean square error function is used as the loss function, and the model parameters are adjusted based on the Adam optimization algorithm and the loss function to train the core loss prediction model, wherein the loss function is expressed as:

[0106] (1);

[0107] In the formula, represents the total number of samples, N represents the number of training samples, Indicates The actual core loss value of samples, Indicates The predicted core loss values ​​for samples.

[0108] (1) Convert the environmental parameters into a tensor form to obtain a parameter feature vector.

[0109] Before converting the environmental parameters into a tensor form, the method further includes standardizing the continuous variables using the quartile method, wherein the expression for standardizing the continuous variables using the quartile method is:

[0110] (2);

[0111] In the formula, represents the standardized environmental parameters, Represents the original environment parameters, represents the median of the original environmental parameters, represents the interquartile range, where , Q1 represents the first quartile of the original environmental parameters, and Q3 represents the third quartile of the original environmental parameters.

[0112] Before converting the environmental parameters into a tensor form, the method further includes standardizing the discrete variables using One-Hot encoding.

[0113] The method for standardizing the discrete variable using One-Hot coding includes:

[0114] determining all categories of material properties and waveform types;

[0115] For each material category and waveform category, create a material category binary column and a waveform category binary column respectively;

[0116] When the discrete variable belongs to the determined material category, a 1 is marked at the corresponding position of the material category binary column, otherwise a 0 is marked;

[0117] When the discrete variable belongs to the determined waveform category, a 1 is marked at the corresponding position of the waveform category binary column, otherwise a 0 is marked.

[0118] The material category binary column is represented as:

[0119] M i × i = [ M 1 M 2 … M i ] = [ 1 0 … 0 0 1 … 0 0 0 … 0 0 0 … 1 ] (3);

[0120] In the formula, A binary column representing material category, The first category represents material properties, The second category represents material properties, The material properties categories, material category binary column Each row represents the One-Hot code corresponding to a material category;

[0121] The waveform category binary column representation is:

[0122] W j × j = [ W 1 W 2 … W j ] = [ 1 0 … 0 0 1 … 0 0 0 … 0 0 0 … 1 ] (4);

[0123] In the formula, represents the waveform class binary column, The first category represents the waveform type, The second category represents the waveform type. Indicates the waveform type. categories, waveform category binary columns Each row of represents the One-Hot encoding corresponding to a waveform category.

[0124] (2) Converting the one-dimensional magnetic flux density time series signal into a two-dimensional time-frequency spectrum diagram, performing feature extraction on the two-dimensional time-frequency spectrum diagram, and obtaining an image feature vector.

[0125] The one-dimensional magnetic flux density time series signal is converted into a two-dimensional time-frequency spectrum diagram, including:

[0126] Using Morlet wavelet transform, the one-dimensional magnetic flux density time series signal is decomposed into a set of basis functions formed by wavelet contraction, flattening and translation;

[0127] Extracting and visualizing the coefficients of the basis function to generate a two-dimensional time-frequency spectrum diagram with a size of 1×24×24 pixels;

[0128] Among them, the principle expression of Morlet wavelet transform is:

[0129] (5);

[0130] In the formula, It represents a set of wavelet basis functions decomposed from the one-dimensional magnetic flux density time series signal f(t), Indicates the current time Wavelet The scaling factor of the expansion or contraction degree, The translation factor representing the degree of wavelet transform, Indicates the current moment The differential of represents positive infinity, Represents negative infinity. Among them, by scanning the translation factor of the wavelet transform degree, the time series information can be extracted.

[0131] The scale factor a corresponds to 1024 sampling points in one cycle, that is, the specific value is an integer sequence from 1 to 1023. The translation factor b is automatically obtained by the one-dimensional convolution process and depends on the input one-dimensional magnetic flux density time series signal and the length of the wavelet. The wavelet transform can retain the information in the time domain and frequency domain at the same time, which is convenient for subsequent feature extraction.

[0132] like Figure 2 As shown, a multi-layer convolutional neural network CNN is used to extract features from the two-dimensional time-frequency spectrum to obtain an image feature vector, including:

[0133] The first convolution layer, which includes 32 convolution kernels with a size of 3×3, a stride of 1 and a padding technique, is used to capture the basic texture and edge information in the two-dimensional time-frequency spectrum, generating a 32×24×24 high-order tensor.

[0134] The purpose of using the padding technology in this embodiment is to keep the size of the two-dimensional time-frequency spectrum unchanged. Each convolution kernel in the first layer of convolution is responsible for detecting different local features and generating a 32×24×24 high-order tensor. This multi-channel output enhances the core loss prediction model's understanding of image details.

[0135] The 32×24×24 high-order tensor is subjected to the first maximum pooling using a maximum pooling layer with a configured window size of 2×2 and a stride of 2, and is compressed into a 32×12×12 high-order tensor.

[0136] The main purpose of performing the first maximum pooling in this embodiment is to reduce the spatial dimension of the two-dimensional time-frequency spectrum while retaining the most representative features; during the first maximum pooling process, the amount of calculation is reduced and the robustness of the core loss prediction model to small displacement changes in the input data is increased.

[0137] The 32×12×12 high-order tensor is converted into a 64×12×12 high-order tensor using a second layer of convolution with 64 convolution kernels of size 3×3 and a stride of 1.

[0138] The level of abstraction is further improved. As the number of convolution layers increases, the learned visual patterns become more complex. Through the second layer of convolution, the high-order tensor data is converted from 32×12×12 to 64×12×12. The newly increased number of channels means that more feature combinations are taken into account, which helps to improve the expressiveness of the core loss prediction model.

[0139] The 64×12×12 high-order tensor is subjected to a second maximum pooling using a maximum pooling layer with a configured window size of 2×2 and a stride of 2, and is reduced to a multi-dimensional tensor of 64×6×6.

[0140] The purpose of performing the second maximum pooling in this embodiment is to further reduce the spatial resolution of the feature map, enhance the recognition ability of the core loss prediction model for key features, and reduce the risk of overfitting.

[0141] The 64×6×6 multidimensional tensor is converted into a one-dimensional vector with a length of 2304 by using a flattening operation.

[0142] By using the flattening operation, the 64×6×6 multi-dimensional tensor is converted into a one-dimensional vector of length 2304, which is conducive to the subsequent fully connected layer to directly process the 64×6×6 multi-dimensional tensor, making the entire network structure more compact and efficient.

[0143] The one-dimensional vector with a length of 2304 is mapped to a new 512-dimensional space using a fully connected layer, and a ReLU activation function is introduced for processing to obtain an image feature vector.

[0144] The purpose of introducing the ReLU activation function in this embodiment is to help the core loss prediction model better learn the complex mapping relationship between input data and core loss.

[0145] (3) Concatenating the parameter feature vector and the image feature vector to obtain a comprehensive feature representation, including:

[0146] A fully connected layer is used to receive the parameter feature vector and convert it into a five-dimensional vector:

[0147] The five-dimensional vector is converted into a high-level parameter feature vector through a hidden layer containing 128 nodes;

[0148] The high-level parameter feature vector and the image feature vector are concatenated to obtain a comprehensive feature representation.

[0149] (4) The comprehensive feature representation is processed through a fully connected layer to obtain a predicted value of the core loss, including:

[0150] The comprehensive feature representation is processed by a fully connected layer and then predicted by another fully connected layer with a single output node to obtain a predicted value of the core loss.

[0151] Example 3

[0152] A computer-readable storage medium having computer instructions stored thereon, characterized in that when the computer instructions are executed by a processor, the steps of the method in the above-mentioned embodiment 1 or 2 are implemented.

[0153] Example 4

[0154] The present invention also provides a computer program product, including computer instructions, characterized in that when the computer instructions are executed by a processor, the steps of the method in the above-mentioned embodiment 1 or 2 are implemented.

[0155] In summary, the present invention innovatively integrates environmental parameters and one-dimensional magnetic flux density time series signals, so that the core loss prediction model can comprehensively consider information in more dimensions, thereby improving the accuracy and comprehensiveness of the prediction. The present invention converts the one-dimensional magnetic flux density time series signal into a two-dimensional time-frequency spectrum, and obtains an image feature vector through feature extraction, which not only retains the time-frequency characteristics of the signal, but also further simplifies the data through feature extraction. The present invention also converts environmental parameters into parameter feature vectors and image feature vectors for fusion, thereby improving processing efficiency.

[0156] The present invention constructs a core loss prediction model based on multimodal data fusion and multiple convolutional neural networks. The core loss prediction model uses a fully connected layer to process comprehensive feature representations to obtain the predicted value of the core loss, making full use of the complementarity of multimodal data to improve prediction accuracy. The present invention demonstrates high accuracy and strong generalization capabilities in core loss prediction through a core loss prediction method based on multimodal data and multiple convolutional neural networks. Compared with traditional single data source or empirical formula methods, the present invention can more accurately predict core loss under different materials, waveforms and working environments.

[0157] The present invention provides a new solution for the performance prediction of magnetic components in power electronic equipment. By accurately predicting the core loss, the present invention not only solves the problem that the prior art is difficult to accurately predict the core loss under high frequency or non-sinusoidal excitation, but also promotes technological innovation and development in related fields. By integrating multimodal data processing with multiple convolutional neural networks, new ideas and methods are provided for solving similar problems, which can optimize the design and operation of power electronic equipment, improve the efficiency and reliability of equipment, and reduce cost control. This is of great significance to the development and progress of the power electronics industry. It should be understood by those skilled in the art that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention can take the form of a computer program product implemented 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.

[0158] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0159] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0160] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0161] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A method for predicting core loss, characterized in that: include: Obtain environmental parameters and one-dimensional magnetic flux density time series signal of the magnetic core to be tested; According to the environmental parameters and the one-dimensional magnetic flux density time series signal, based on a pre-trained magnetic core loss prediction model, prediction is performed to obtain a predicted value of the magnetic core loss; The data processing method of the magnetic core loss prediction model includes: Convert the environmental parameters into tensor form to obtain parameter feature vectors; Converting the one-dimensional magnetic flux density time series signal into a two-dimensional time-frequency spectrum diagram, performing feature extraction on the two-dimensional time-frequency spectrum diagram to obtain an image feature vector; splicing the parameter feature vector and the image feature vector to obtain a comprehensive feature representation; Processing the comprehensive feature representation through a fully connected layer to obtain a predicted value of the core loss; The environmental parameters include continuous variables and discrete variables, the continuous variables include temperature and frequency, and the discrete variables include material properties and waveform types.

2. The method for predicting magnetic core loss according to claim 1, characterized in that: Before converting the environmental parameters into a tensor form, the method further includes standardizing the continuous variables using the quartile method, wherein the expression for standardizing the continuous variables using the quartile method is: （1）； In the formula, represents the standardized environmental parameters, Represents the original environment parameters, represents the median of the original environmental parameters, represents the interquartile range, where , Q1 represents the first quartile of the original environmental parameters, and Q3 represents the third quartile of the original environmental parameters.

3. The method for predicting core loss according to claim 1, characterized in that: Before converting the environmental parameters into a tensor form, the method further includes standardizing the discrete variables using One-Hot encoding.

4. The method for predicting core loss according to claim 3, characterized in that: The method of standardizing the discrete variable by using One-Hot coding includes: determining all categories of material properties and waveform types; For each material category and waveform category, create a material category binary column and a waveform category binary column respectively; When the discrete variable belongs to the determined material category, a 1 is marked at the corresponding position of the material category binary column, otherwise a 0 is marked; When the discrete variable belongs to the determined waveform category, a 1 is marked at the corresponding position of the waveform category binary column, otherwise a 0 is marked.

5. The method for predicting magnetic core loss according to claim 4, characterized in that: The material category binary column is represented as: （2）； In the formula, A binary column representing material category, The first category represents material properties, The second category represents material properties, The material properties categories, material category binary column Each row represents the One-Hot code corresponding to a material category; The waveform category binary column representation is: （3）； In the formula, represents the waveform class binary column, The first category represents the waveform type, The second category represents the waveform type. Indicates the waveform type. categories, waveform category binary columns Each row of represents the One-Hot encoding corresponding to a waveform category.

6. The method for predicting magnetic core loss according to claim 1, characterized in that: Converting the one-dimensional magnetic flux density time series signal into a two-dimensional time-frequency spectrum diagram includes: Using Morlet wavelet transform, the one-dimensional magnetic flux density time series signal is decomposed into a set of basis functions formed by wavelet contraction, flattening and translation; Extracting and visualizing the coefficients of the basis function to generate a two-dimensional time-frequency spectrum diagram with a size of 1×24×24 pixels; Among them, the principle expression of Morlet wavelet transform is: （4）； In the formula, It represents a set of wavelet basis functions decomposed from the one-dimensional magnetic flux density time series signal f(t), Indicates the current time Wavelet The scaling factor of the expansion or contraction degree, The translation factor representing the degree of wavelet transform, Indicates the current moment The differential of represents positive infinity, Represents negative infinity.

7. The method for predicting magnetic core loss according to claim 6, characterized in that: Using a multi-layer convolutional neural network (CNN), feature extraction is performed on the two-dimensional time-frequency spectrum to obtain an image feature vector, including: The first convolution layer including 32 convolution kernels of size 3×3, stride 1 and padding technique is used to capture the basic texture and edge information in the two-dimensional time-frequency spectrum, generating a 32×24×24 high-order tensor; The 32×24×24 high-order tensor is subjected to the first maximum pooling using a maximum pooling layer with a configuration window size of 2×2 and a stride of 2, and is compressed into a 32×12×12 high-order tensor; The 32×12×12 high-order tensor is converted into a 64×12×12 high-order tensor using a second convolution layer with 64 convolution kernels of size 3×3 and a stride of 1; Using a maximum pooling layer with a configuration window size of 2×2 and a stride of 2, the 64×12×12 high-order tensor is subjected to a second maximum pooling, and the dimension is reduced to a multi-dimensional tensor of 64×6×6; Using a flattening operation, the 64×6×6 multidimensional tensor is converted into a one-dimensional vector of length 2304; The one-dimensional vector with a length of 2304 is mapped to a new 512-dimensional space using a fully connected layer, and a ReLU activation function is introduced for processing to obtain an image feature vector.

8. The method for predicting magnetic core loss according to claim 7, characterized in that: The parameter feature vector and the image feature vector are concatenated to obtain a comprehensive feature representation, including: A fully connected layer is used to receive the parameter feature vector and convert it into a five-dimensional vector: The five-dimensional vector is converted into a high-level parameter feature vector through a hidden layer containing 128 nodes; The high-level parameter feature vector and the image feature vector are concatenated to obtain a comprehensive feature representation.

9. The method for predicting magnetic core loss according to claim 8, characterized in that: The comprehensive feature representation is processed through a fully connected layer to obtain a predicted value of the core loss, including: The comprehensive feature representation is processed by a fully connected layer and then predicted by another fully connected layer with a single output node to obtain a predicted value of the core loss.

10. The method for predicting magnetic core loss according to claim 1, characterized in that: The mean square error function is used as the loss function, and the model parameters are adjusted based on the Adam optimization algorithm and the loss function to train the core loss prediction model, wherein the loss function is expressed as: （5）； In the formula, represents the total number of samples, N represents the number of training samples, Indicates The actual core loss value of samples, Indicates The predicted core loss values ​​for samples.

Images