Absorbent electromagnetic parameter prediction method, system, equipment and medium

Abstract

The invention discloses an absorbent electromagnetic parameter prediction method, system and device and a medium, and relates to the technical field of material science. The method comprises the following steps: acquiring to-be-tested data; the to-be-detected data comprises frequency, particle size characteristics and physical inspiration characteristics; constructing an electromagnetic parameter prediction model based on a multi-branch neural network; the multi-branch neural network comprises an input layer, a shared feature extraction layer, a branch layer, a deep connection layer and an output layer which are connected in sequence; inputting the data to be measured into the electromagnetic parameter prediction model, and outputting an electromagnetic parameter prediction result of the absorbent; the electromagnetic parameter prediction result comprises a dielectric constant real part, a dielectric constant imaginary part, a magnetic conductivity real part and a magnetic conductivity imaginary part. The method can improve the prediction accuracy of the electromagnetic parameters of the absorbent.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of material science, in particular to an electromagnetic parameter prediction method, system, device and medium for an absorbent. BACKGROUND

[0002] For electromagnetic wave absorbers, the most core index is its electromagnetic parameter. Generally, the related requirements of the "ASTM D7449M −22a Standard Test Method for Measuring Relative Complex Permittivity and Relative Magnetic Permeability of Solid Materials at Microwave Frequencies Using Coaxial Air Line" are referred to, and a vector network analyzer is combined with an air transmission line to measure the dielectric constant and magnetic permeability of the absorbent in a certain frequency range, such as the 0.5-18GHz frequency band. However, this testing process is relatively complex, on the one hand, the absorbent needs to be fully mixed with paraffin in a molten state, and then a sample suitable for being placed in the air transmission line is prepared, on the other hand, the calibration of the vector network analyzer and the testing process are also relatively strict. Relatively speaking, it is simpler and faster to measure the particle size distribution of the absorbent. For example, according to the requirements of "GB / T 41949-2022", the particle size distribution parameters of the absorbent can be directly obtained by using a laser particle size analyzer, including D10 / D50 / D90, specific surface area, particle size distribution width, etc. If the electromagnetic parameters can be accurately predicted through the particle size distribution parameters, the testing process can be simplified, and the research and preparation of the absorbent can be accelerated.

[0003] However, the traditional electromagnetic parameter prediction relies on physical simulation calculation, which is time-consuming and requires accurate material models, and the existing machine learning method uses a single path network, which cannot effectively separate the physical mechanisms of dielectric / magnetic permeability, and the nonlinear relationship between the particle size distribution characteristics and the electromagnetic response is difficult to be represented by a simple model. SUMMARY

[0004] The purpose of the present application is to provide an electromagnetic parameter prediction method, system, device and medium for an absorbent, which aims to solve or improve at least one of the above technical problems.

[0005] To achieve the above purpose, the present application provides the following solutions: An electromagnetic parameter prediction method for an absorbent, comprising: obtaining test data; the test data includes frequency, particle size characteristics and physical heuristic characteristics; The electromagnetic parameter prediction model is constructed based on a multi-branch neural network; the multi-branch neural network comprises an input layer, a shared feature extraction layer, a branch layer, a deep connection layer and an output layer connected in sequence; The electromagnetic parameter prediction model is constructed based on a multi-branch neural network; the multi-branch neural network comprises an input layer, a shared feature extraction layer, a branch layer, a deep connection layer and an output layer connected in sequence;

[0006] Optionally, the construction process of the electromagnetic parameter prediction model comprises: Loading particle size characteristic data and integrating electromagnetic parameter data to obtain characteristic correlation data; the characteristic correlation data comprises electromagnetic parameters at all frequency points and corresponding particle size characteristics; Performing target separation on the characteristic correlation data and labeling original characteristics and corresponding target variables; the original characteristics are frequency and particle size characteristics; the target variables are real part of dielectric constant, imaginary part of dielectric constant, real part of magnetic permeability and imaginary part of magnetic permeability; Based on the characteristic correlation data, constructing a physical heuristic feature, and splicing the physical heuristic feature with the original characteristics to obtain an updated training data set; the updated training data set comprises target variables corresponding to the spliced characteristic data set; Inputting the updated training data set into the multi-branch neural network for network parameter optimization training, and determining the trained network as the electromagnetic parameter prediction model.

[0007] Optionally, the physical heuristic feature comprises the reciprocal of frequency, the ratio of particle size parameter D50 to wavelength, the logarithm of specific surface area, and the ratio of particle size distribution width to normalized D50.

[0008] Optionally, the shared feature extraction layer comprises a first full connection layer, a first batch normalization layer, a first LeakyReLU activation layer and a Dropout layer connected in sequence; wherein the first full connection layer comprises 256 neurons, the first batch normalization layer adopts z-score for standardization processing, the first LeakyReLU activation layer adopts a negative slope of 0.2, and the Dropout layer adopts a dropout probability of 0.3.

[0009] Optionally, the branch layer comprises dielectric constant branches and magnetic permeability branches with the same structure; the dielectric constant branches and the magnetic permeability branches each comprise two operation modules connected in sequence, and each operation module comprises a full connection layer, a batch normalization layer and a LeakyReLU activation layer; Among them, the full connection layer in the first operation module of the two branches contains 128 neurons, the full connection layer in the second operation module contains 64 neurons, and the LeakyReLU activation layer in the two operation modules adopts a negative slope of 0.2.

[0010] The application further provides an absorbent electromagnetic parameter prediction system, comprising: A data acquisition unit is configured to acquire to-be-tested data, wherein the to-be-tested data comprises frequency, particle size characteristics and physical heuristic characteristics. A model construction unit is configured to construct an electromagnetic parameter prediction model based on a multi-branch neural network, wherein the multi-branch neural network comprises an input layer, a shared feature extraction layer, a deep connection layer and an output layer connected in sequence. A parameter prediction unit is configured to input the to-be-tested data into the electromagnetic parameter prediction model and output an electromagnetic parameter prediction result of the absorbent, wherein the electromagnetic parameter prediction result comprises a real part of a dielectric constant, an imaginary part of the dielectric constant, a real part of magnetic permeability and an imaginary part of the magnetic permeability.

[0011] The application further provides an electronic device comprising a memory and a processor, wherein the memory is configured to store a computer program, and the processor is configured to execute the computer program to enable the electronic device to perform the absorbent electromagnetic parameter prediction method according to the above.

[0012] The application further provides a computer-readable storage medium storing a computer program, wherein the computer program is configured to be executed by a processor to implement the absorbent electromagnetic parameter prediction method according to the above.

[0013] According to the embodiments of the application, the following technical effects are achieved: The application discloses an absorbent electromagnetic parameter prediction method, system, device and medium, which comprises the following steps: acquiring to-be-tested data; the to-be-tested data comprises frequency, particle size characteristics and physical heuristic characteristics; constructing an electromagnetic parameter prediction model based on a multi-branch neural network; the multi-branch neural network comprises an input layer, a shared feature extraction layer, a branch layer, a deep connection layer and an output layer connected in sequence; inputting the to-be-tested data into the electromagnetic parameter prediction model and outputting an electromagnetic parameter prediction result of the absorbent; the electromagnetic parameter prediction result comprises a real part of a dielectric constant, an imaginary part of the dielectric constant, a real part of magnetic permeability and an imaginary part of the magnetic permeability. The application can improve the prediction accuracy of the electromagnetic parameters of the absorbent. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0015] Figure 1 The flowchart of the method for predicting electromagnetic parameters of the absorbent in the embodiment is shown. Figure 2 The schematic diagram of the multi-branch neural network architecture in the embodiment is shown. Figure 3 The schematic diagram of the particle size distribution parameter results of the newly produced absorbent tested by the laser particle size analyzer in the embodiment is shown. Figure 4 The comparison diagram of the predicted value and the actual measured value in the embodiment is shown. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0017] The purpose of the present application is to provide a method, system, device and medium for predicting electromagnetic parameters of an absorbent, aiming to solve or improve at least one of the above technical problems.

[0018] In order to make the above purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0019] As shown in Figure 1 The present application provides a method for predicting electromagnetic parameters of an absorbent, comprising: Obtaining test data; the test data includes frequency, particle size characteristics and physical heuristic characteristics.

[0020] Building an electromagnetic parameter prediction model based on a multi-branch neural network; the multi-branch neural network includes an input layer, a shared feature extraction layer, a branch layer, a deep connection layer and an output layer connected in turn.

[0021] Inputting the test data into the electromagnetic parameter prediction model and outputting the electromagnetic parameter prediction results of the absorbent; the electromagnetic parameter prediction results include the real part of the dielectric constant, the imaginary part of the dielectric constant, the real part of the magnetic permeability and the imaginary part of the magnetic permeability.

[0022] As a specific implementation, the electromagnetic parameter prediction model comprises an input layer, a shared feature extraction layer, a branch layer, a deep connection layer and an output layer connected in turn, wherein: The input layer receives feature data, the input size is the number of features; the input dimension is the original features and the physical heuristic features (about 15-20 features in total), including frequency, D10 / D50 / D90, specific surface area, particle size distribution width and other particle size parameters, and added physical heuristic features.

[0023] The shared feature extraction layer comprises a fully connected layer 1, a batch normalization layer 1, a LeakyReLU activation layer 1 and a Dropout layer connected in turn, wherein the fully connected layer 1 comprises 256 neurons, the batch normalization layer 1 adopts z-score standardization processing on data, the LeakyReLU activation layer 1 adopts a negative slope of 0.2, and the Dropout layer adopts a dropout probability of 0.3.

[0024] The branch layer comprises a permittivity branch and a permeability branch connected with the shared layer respectively, wherein the permittivity branch focuses on learning permittivity-related features by using particle size distribution characteristics (such as D50 / specific surface area), and focuses on dielectric mechanisms such as interfacial polarization and electronic displacement, and comprises a fully connected layer 2, a batch normalization layer 2, a LeakyReLU activation layer 2 and a fully connected layer 3, a batch normalization layer 3 and a LeakyReLU activation layer 3 connected in turn; the permeability branch focuses on distribution width indicators for learning permeability-related features, and focuses on magnetic properties such as magnetic moment precession and eddy current loss, and comprises a fully connected layer 4, a batch normalization layer 4, a LeakyReLU activation layer 4 and a fully connected layer 5, a batch normalization layer 5 and a LeakyReLU activation layer 5 connected in turn; the fully connected layer 2 and the fully connected layer 4 respectively comprise 128 neurons, the fully connected layer 3 and the fully connected layer 5 respectively comprise 64 neurons, and the LeakyReLU activation layer 2, the LeakyReLU activation layer 3, the LeakyReLU activation layer 4 and the LeakyReLU activation layer 5 adopt a negative slope of 0.2.

[0025] The deep connection layer comprises two inputs connected with the permittivity branch and the permeability branch respectively, retains branch-specific features while realizing information fusion, and outputs to an output fully connected layer of the output layer.

[0026] The output layer comprises an output fully connected layer and a regression layer, wherein the output fully connected layer comprises 4 neurons corresponding to 4 outputs, the output fully connected layer is connected to the regression layer, and the regression layer calculates the mean square error loss.

[0027] The input features include frequency, particle size characteristics and added physical heuristic features; the output includes four continuous values representing the real part of permittivity, the imaginary part of permittivity, the real part of permeability and the imaginary part of permeability respectively.

[0028] Particle size characteristics include D10 / D50 / D90, specific surface area, particle size distribution width, etc.; added physical heuristic characteristics include frequency inverse, particle size wavelength ratio, surface area to volume ratio, particle size distribution width index.

[0029] Particle size wavelength ratio is defined as: D50 / (300 / f), where f corresponds to the frequency value.

[0030] Surface area to volume ratio is defined as: log(SSA+1x10-6), where SSA corresponds to "specific surface area" in Table 1.

[0031] Particle size distribution width index is defined as: span / ((D50 - D50_mean) / D50_std), where span corresponds to "distribution width" in Table 1, and D50_mean and D50_std correspond to the mean and standard deviation of the particle size parameter D50.

[0032] The standardization of the data refers to subtracting the mean divided by the standard deviation.

[0033] Based on the above technical solution, the specific processing process is shown as follows.

[0034] Step 1: Load particle size characteristic data. Read the particle size characteristic data table and obtain the sample number.

[0035] Step 2: Load and integrate electromagnetic parameter data.

[0036] Check if the number of electromagnetic parameter files is consistent with the number of particle size characteristic samples; traverse each sample, find the corresponding electromagnetic parameter file according to the file name, and read it; verify that the electromagnetic parameter file contains necessary columns; merge the particle size characteristics of each sample with the electromagnetic parameter data by repeatedly adding the particle size characteristics of this sample for each row of electromagnetic parameter data; finally obtain a large table containing electromagnetic parameters and corresponding particle size characteristics at all frequency points.

[0037] Step 3: Data preprocessing.

[0038] Separate features (X) and targets (Y), where feature X includes frequency and all particle size characteristics (as shown in Table 1), and feature Y includes four target variables; standardize the data to obtain standardized features and targets, and save the mean and standard deviation.

[0039] Step 4: Construct physical heuristic features.

[0040] On the basis of the original features, four physical heuristic features are constructed: the inverse of frequency (to simulate the resonance effect), the ratio of particle size parameter D50 to wavelength, the logarithm of specific surface area (plus a very small constant to avoid log 0), and the ratio of particle size distribution width to normalized D50, where normalization means subtracting the mean and dividing by the standard deviation. Then these new features are concatenated with the standardized original features to obtain the feature `X_combined`.

[0041] Step 5: Build and train the neural network model.

[0042] Build a multi-branch neural network, as shown in Figure 2 , including: Input layer: feature input layer, input size is the number of columns of `X_combined`.

[0043] Shared layer: fully connected layer (256 nodes) -> batch normalization -> LeakyReLU activation -> Dropout (0.3).

[0044] Two branches: Branch 1 is the dielectric constant prediction path, including: fully connected (128 nodes) -> batch normalization -> LeakyReLU -> fully connected (64 nodes) -> batch normalization -> LeakyReLU; Branch 2 is the permeability prediction path, including: fully connected (128 nodes) -> batch normalization -> LeakyReLU -> fully connected (64 nodes) -> batch normalization -> LeakyReLU.

[0045] Deep connection layer: connect the outputs of the two branches in depth.

[0046] Output layer: fully connected layer (4 nodes corresponding to 4 targets) -> regression layer.

[0047] The training settings include: Manually divide the training set and validation set, with the validation set accounting for 15% of the weight; use the Adam optimizer, maximum 500 rounds, batch size 512; learning rate initially 0.001, multiplied by 0.5 every 100 rounds; use the manually divided validation set for validation; after training is complete, save the model and standardization parameters and particle size data; Step 6: Model validation and visualization.

[0048] Randomly select 3 samples, load the actual electromagnetic parameters for each sample, prepare the input data including the original features and physical heuristic features, and standardize them; use the trained model for prediction; draw a comparison chart of actual value and predicted value, and save the image.

[0049] Step 7: Evaluate the model performance.

[0050] The prediction is performed on the entire data set, the determination coefficient R2 and the root mean square error RMSE of each target variable are calculated, and the data required for analysis is saved. The determination coefficient R2 is a relatively comprehensive evaluation index, which considers the difference between the predicted value and the true value of the model, and its value range is [0, 1], the closer to 1 indicates that the prediction effect of the model is better; and the root mean square error RMSE directly represents the average difference between the predicted value and the true value of the model.

[0051] More than 100 groups of test data are used to train the model, a desktop computer with a CPU of 12th Gen Intel(R) Core(TM) i5-12400F (2.50 GHz) and 32G RAM is used, and the time consumption is about 1 minute. After completing the model training, first, a plurality of groups of training data are applied to carry out testing, and then for the newly produced absorbent product, the particle size distribution parameters obtained by the laser particle size instrument are used to predict the electromagnetic parameters of the absorbent, and the predicted value is compared with the actual electromagnetic parameter measurement value. The particle size distribution parameter results of the newly produced absorbent tested by the laser particle size instrument are shown in Table 2, and the comparison of the predicted value and the actual measurement value is shown in Table 3. Figure 3 Figure 4 The comprehensive R2=0.99 and RMSE=0.21 show that the predicted value is highly consistent with the actual measurement value.

[0052] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0053] The principles and implementation modes of the present application are described by applying specific examples in this paper, and the above embodiment description is only used to help understand the core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In view of the above, the content of the specification should not be understood as a limitation of the present application.​

Claims

1. A method of predicting the electromagnetic parameters of an absorber, characterized in that, The method comprises the steps of: acquiring test data; the test data comprises frequency, particle size characteristics and physical heuristic characteristics; constructing an electromagnetic parameter prediction model based on a multi-branch neural network; the multi-branch neural network comprises an input layer, a shared feature extraction layer, a branch layer, a deep connection layer and an output layer connected in sequence; inputting the test data into the electromagnetic parameter prediction model, and outputting electromagnetic parameter prediction results of the absorbent; the electromagnetic parameter prediction results comprise a real part of a dielectric constant, an imaginary part of the dielectric constant, a real part of magnetic permeability and an imaginary part of the magnetic permeability.

2. The method of claim 1, wherein, The construction process of the electromagnetic parameter prediction model comprises the steps of: loading particle size characteristic data, and integrating electromagnetic parameter data to obtain characteristic correlation data; the characteristic correlation data comprises electromagnetic parameters at all frequency points and corresponding particle size characteristics; performing target separation on the characteristic correlation data, and labeling original characteristics and corresponding target variables; the original characteristics are frequency and particle size characteristics; the target variables are a real part of a dielectric constant, an imaginary part of the dielectric constant, a real part of magnetic permeability and an imaginary part of the magnetic permeability; constructing physical heuristic characteristics based on the characteristic correlation data, and splicing the physical heuristic characteristics with the original characteristics to obtain an updated training data set; the updated training data set comprises target variables corresponding to the spliced characteristic data set; inputting the updated training data set into the multi-branch neural network for network parameter optimization training, and determining a trained network as the electromagnetic parameter prediction model.

3. The method of claim 2, wherein, The physical heuristic characteristics comprise the reciprocal of frequency, the ratio of particle size parameter D50 to wavelength, the logarithm of specific surface area, and the ratio of particle size distribution width to normalized D50.

4. The method of claim 1, wherein, The shared feature extraction layer comprises a fully connected layer 1, a batch normalization layer 1, a LeakyReLU activation layer 1 and a Dropout layer connected in sequence; the fully connected layer 1 comprises 256 neurons; the batch normalization layer 1 adopts z-score standardization processing on data; the LeakyReLU activation layer 1 adopts a negative slope of 0.2; and the Dropout layer adopts a dropout probability of 0.

3.

5. The method of claim 1, wherein, The branch layer comprises dielectric constant branches and magnetic permeability branches which are of the same structure; the dielectric constant branches and the magnetic permeability branches each comprise two operation modules connected in sequence; each operation module comprises a fully connected layer, a batch normalization layer and a LeakyReLU activation layer; The fully connected layer in the first operation module of the two branches comprises 128 neurons, the fully connected layer in the second operation module comprises 64 neurons, and the LeakyReLU activation layer in the two operation modules each adopts a negative slope of 0.

2.

6. An absorbent electromagnetic parameter prediction system characterized by, The method comprises the steps of: a data acquisition unit configured to acquire test data; the test data comprises frequency, particle size characteristics and physical heuristic characteristics; a model construction unit configured to construct an electromagnetic parameter prediction model based on a multi-branch neural network; the multi-branch neural network comprises an input layer, a shared feature extraction layer, a deep connection layer and an output layer connected in sequence; A parameter prediction unit is configured to input the to-be-tested data into the electromagnetic parameter prediction model and output an electromagnetic parameter prediction result of the absorbent, wherein the electromagnetic parameter prediction result comprises a real part of a dielectric constant, an imaginary part of the dielectric constant, a real part of a magnetic permeability, and an imaginary part of the magnetic permeability.

7. An electronic device, comprising: The electronic device comprises a memory and a processor, wherein the memory is configured to store a computer program, and the processor is configured to execute the computer program to enable the electronic device to perform the absorbent electromagnetic parameter prediction method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer program is stored in the memory and is executed by the processor to implement the absorbent electromagnetic parameter prediction method according to any one of claims 1-5.

Images