A method for inductor design based on frequency modulation

Abstract

The invention discloses an inductor design method based on frequency modulation, and relates to the technical field of inductor. The method comprises: step S1: recording the frequency initial value and magnetic permeability value of the target inductor conductor in the current circuit, and calculating the reference resistivity value; step S2: setting the reference frequency range used as the frequency initial value modulation target; step S3: constructing a convolutional neural network, taking the frequency initial value mean as the input value, and setting all output values ​​as the frequency modulation value; step S4: calculating all the modulation resistivity values, and screening out all the frequency modulation values ​​whose modulation resistivity values ​​are lower than the reference resistivity value; step S5: setting the maximum frequency modulation value screened out to a new upper boundary value and a new lower boundary value, and setting the new upper boundary value and the new lower boundary value as the actual frequency range of the target inductor circuit. The invention can effectively avoid the circuit loss and inductance value deviation caused by the Skin effect, and has the advantages and beneficial effects of ensuring the stability and performance of the inductor.

Description

Technical Field

[0001] The present invention relates to the technical field of inductors, and in particular to an inductor design method based on frequency modulation. Background Art

[0002] Traditional planar inductors arrange coils and magnetic cores on a plane, while three-dimensional inductors allow coils and magnetic cores to be arranged in three-dimensional space. This design provides higher inductance density and better electromagnetic performance. Three-dimensional inductors are widely used in high-frequency and radio frequency circuits, such as communication equipment, radio frequency identification (RFID) systems, power converters, and power management. Under high-frequency conditions, when the current of a three-dimensional inductor is more concentrated on the surface of the conductor rather than inside the entire cross-section, the Skin effect is prone to occur. The Skin effect causes the equivalent resistance of the conductor to increase with increasing frequency, causing the resistance value of the inductor surface to change, making the inductor conductor prone to deviation and increasing circuit energy loss. Under high-frequency conditions, the Skin effect causes the current to be more concentrated on the surface of the wire rather than evenly distributed over the entire wire cross-section. This will increase the effective resistance, thereby increasing the resistance and loss of the wire. At present, the existing technology mainly avoids the Skin effect by changing the wire material or adjusting the frequency, but if the frequency of the inductor is too high, the Skin effect avoidance effect will be poor, and if it is too low, the inductor will function abnormally in the circuit, resulting in the lack of accuracy in the processing of the avoidance effect by the existing means. Summary of the invention

[0003] The present invention provides an inductor design method based on frequency modulation, which solves the problem of low accuracy in avoiding the skin effect during three-dimensional inductor frequency design in the prior art.

[0004] The present invention is achieved through the following technical solutions:

[0005] An inductor design method based on frequency modulation, the method comprising:

[0006] Step S1: recording the initial frequency value and the magnetic permeability value of the target inductor conductor in the current circuit, presetting the Skin scalar value, and calculating the reference resistivity value of the target inductor in the current circuit based on the initial frequency value, the Skin scalar value and the magnetic permeability value;

[0007] Step S2: setting a reference frequency range used as a frequency initial value modulation target, representing an upper limit value and a lower limit value in the reference frequency range as an upper boundary value and a lower boundary value, respectively, and dividing the modulation process into a plurality of working cycles of the same length;

[0008] Step S3: construct a convolutional neural network, input the average of the initial frequency values ​​of the inductance in each working cycle as an input value into the convolutional neural network, and set all output values ​​to frequency modulation values ​​representing the frequency value after modulation;

[0009] Step S4: all frequency modulation values ​​are calculated based on the skin scalar value and the magnetic permeability value to obtain all modulated resistivity values, and all frequency modulation values ​​whose modulated resistivity values ​​are lower than the reference resistivity value are screened out;

[0010] Step S5: Set the screened minimum frequency modulation value as the new lower boundary value, perform data processing based on the remaining frequency modulation values, determine the new upper boundary value according to the data processing results, and set the new upper boundary value and the new lower boundary value as the actual frequency range of the target inductor circuit.

[0011] Under high-frequency conditions, when the current of a three-dimensional inductor is more concentrated on the surface of the conductor rather than inside the entire cross-section, the Skin effect is prone to occur. The Skin effect causes the equivalent resistance of the conductor to increase as the frequency increases, causing the resistance value of the inductor surface to change, making the inductor conductor prone to deviations and increasing circuit energy loss. Under high-frequency conditions, the Skin effect causes the current to be more concentrated on the surface of the wire rather than evenly distributed over the entire wire cross-section. This will cause the effective resistance to increase, thereby increasing the resistance and loss of the wire. Currently, the prior art mainly avoids the Skin effect by changing the wire material or adjusting the frequency, but if the frequency of the inductor is too high, the Skin effect avoidance effect will be poor, and if it is too low, the inductor's function in the circuit will be abnormal, resulting in a lack of accuracy in the processing of the effect avoidance by the existing means. Based on this, the present invention provides an inductor design method based on frequency modulation to solve the problem of low accuracy in avoiding the Skin effect in the prior art when designing the frequency of a three-dimensional inductor.

[0012] Furthermore, the construction of the convolutional neural network includes: normalizing the input frequency value preprocessing, collecting the skin scalar value and the magnetic permeability value of the target inductor under several different current values, and setting each group of skin scalar value and magnetic permeability value as an inductance data set in a set form after weighted addition; constructing a first distance matrix and a second distance matrix for the mean of the frequency initial value and the inductance data set based on the Manhattan distance, taking all the first distance matrices and the second distance matrices as input values ​​of the convolutional neural network, and outputting the frequency modulation value after the frequency value is modulated in the form of a single neuron.

[0013] Furthermore, the input and output process of the convolutional neural network includes: merging the first distance matrix and the second distance matrix into a two-dimensional input tensor, and using convolution layers and pooling layers to sequentially extract input data features of the two-dimensional input tensor; adding a fully connected layer to the input data features to map the input data features to the output layer; and setting a linear activation function for the output layer to map a single neuron output.

[0014] Furthermore, the construction process of the first distance matrix includes: assuming that the frequency value is represented by f, and the mean of the initial frequency value is represented by , let the first distance matrix be expressed as D, then the first distance matrix D is expressed as: , where i and j represent the ordinal numbers of the means of the initial values ​​of the frequencies.

[0015] Furthermore, a multi-dimensional feature space is constructed for computational expansion when the Manhattan distance constructs the second distance matrix.

[0016] Furthermore, the PCA principal component analysis method is used to perform feature dimensionality reduction on the input data features of the two-dimensional input tensor, and the input data features after feature dimensionality reduction are decomposed into eigenvalues ​​through covariance matrix calculation, and a feature threshold is set to filter out the eigenvalues ​​whose number of principal components is lower than the feature threshold, and the filtered input data features are input into the convolution layer and pooling layer of the convolutional neural network.

[0017] Furthermore, the PCA principal component analysis method uses an incremental PCA analysis method to gradually update the covariance matrix.

[0018] Further, the process of data processing the remaining frequency modulation values ​​includes: the new upper boundary value after the data aggregation includes a first reference value, a second reference value and an implementation cutoff value, the first reference value is obtained by taking the mean of the remaining frequency modulation values, the second reference value is obtained by taking the standard deviation of the remaining frequency modulation values, and the implementation cutoff value includes the first modulation value and the second modulation value extracted by arranging the remaining frequency modulation values ​​from small to large;

[0019] When the first reference value is greater than the second reference value, selecting the first modulation value as a new upper boundary value;

[0020] When the first reference value is less than the second reference value, the second modulation value is selected as a new upper boundary value.

[0021] Furthermore, the implementation cutoff value also includes a third modulation value extracted from the remaining frequency modulation value, and the size of the third modulation value is between the first modulation value and the second modulation value; when the integer value of the first reference value is equal to the integer value of the second reference value, the third modulation value is selected as the new upper boundary value.

[0022] Further, the first modulation value, the second modulation value and the third modulation value are respectively set to represent the 25th percentile, the median and the 75th percentile of the remaining frequency modulation values ​​arranged from small to large.

[0023] Compared with the prior art, the present invention calculates the reference resistivity value and uses a convolutional neural network to process frequency modulation. The upper and lower limits of the frequency range can be adjusted according to the target inductance data to dynamically adjust the designed inductance circuit frequency range. This can effectively avoid the problem of poor frequency accuracy caused by the Skin effect, and has the advantages and beneficial effects of ensuring the stability and performance of the inductor within the entire designed frequency range. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0025] Figure 1 It is a flowchart of the present invention. DETAILED DESCRIPTION

[0026] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention. Example

[0027] like Figure 1 As shown, this embodiment is an inductor design method based on frequency modulation, and the method includes:

[0028] Step S1: recording the initial frequency value and the magnetic permeability value of the target inductor conductor in the current circuit, presetting the Skin scalar value, and calculating the reference resistivity value of the target inductor in the current circuit based on the initial frequency value, the Skin scalar value and the magnetic permeability value;

[0029] Step S2: setting a reference frequency range used as a frequency initial value modulation target, representing an upper limit value and a lower limit value in the reference frequency range as an upper boundary value and a lower boundary value, respectively, and dividing the modulation process into a plurality of working cycles of the same length;

[0030] Step S3: construct a convolutional neural network, input the average of the initial frequency values ​​of the inductance in each working cycle as an input value into the convolutional neural network, and set all output values ​​to frequency modulation values ​​representing the frequency value after modulation;

[0031] Step S4: all frequency modulation values ​​are calculated based on the skin scalar value and the magnetic permeability value to obtain all modulated resistivity values, and all frequency modulation values ​​whose modulated resistivity values ​​are lower than the reference resistivity value are screened out;

[0032] Step S5: Set the screened minimum frequency modulation value as the new lower boundary value, perform data processing based on the remaining frequency modulation values, determine the new upper boundary value according to the data processing results, and set the new upper boundary value and the new lower boundary value as the actual frequency range of the target inductor circuit.

[0033] The Skin effect refers to the fact that in high-frequency circuits, current tends to flow on the surface of the conductor rather than being evenly distributed throughout the entire conductor cross-section. This causes the resistance of the conductor surface to increase, affecting the high-frequency performance of the circuit. Therefore, this embodiment needs to limit or reduce the occurrence of the Skin effect by improving the precise control of the target inductance frequency setting. Prepare measuring equipment, such as a frequency meter and an electromagnetic conductivity meter, to record the current frequency value and magnetic permeability value of the target inductor conductor, and mark the frequency value before modulation as the initial frequency value. As a feasible implementation method, in specific applications, assume that the calculation relationship between the reference resistivity value and the frequency value, the Skin scalar value, and the magnetic permeability value is calculated by the calculation formula Calculated, where δ represents the Skin scalar value, f represents the frequency value, μ represents the magnetic permeability value, and ρ represents the resistivity value, which is also the reference resistivity value to be calculated in this embodiment. The frequency value f is used to calculate the angular frequency value ω in the form of ω=2πf, that is, the original calculation formula of the Skin scalar value is . The Skin scalar value is used to represent the Skin reference value that the target inductor needs to limit in the current circuit, that is, when the Skin value is lower than the Skin scalar value, it is determined that the current inductor is less affected by the Skin. Since the Skin scalar value is positively correlated with the reference resistivity value, it can be determined that the size of the reference resistivity value in the current inductor circuit obtained by calculation can be used as the adjustment limit value of the resistivity value. Since the frequency value of the modulation process may fluctuate with different amplitudes, multiple working cycles are evenly divided to take the average of the frequency value for the optimization calculation of the convolutional neural network. At the same time, the number of working cycles is the same as the average of the initial frequency value, the frequency modulation value, and the number of modulation resistivity values. The reference frequency range is a frequency range value used for reference that can reduce the influence of the Skin effect on the inductor circuit. The frequency range commonly used in the field for optimizing the Skin effect can be used. The purpose of using the convolutional neural network is to further optimize and modulate the mean frequency value in each working cycle based on the reference frequency range. The convolutional neural network is used to analyze the mean value of the initial frequency value of the inductor in each working cycle. By learning historical data and patterns, the convolutional neural network can predict the frequency modulation value in the frequency modulation process, that is, calculate a frequency modulation value representing the tendency prediction according to the current frequency value and the change trend in the working cycle. The frequency modulation value is the frequency value after the modulation process.

[0034] Similarly, the modulated resistivity value is also the resistivity value after the modulation process. In this embodiment, as a preference, the process of calculating the modulated resistivity value may be to express the modulated resistivity value as ρ, substitute the frequency modulation value into Calculate and obtain the modulated resistivity value ρ. Since the resistivity value is positively correlated with the Skin scalar value, a value lower than the Skin scalar value can meet the requirements of the inductor circuit, so a modulated resistivity value lower than the reference resistivity value is a resistivity value that meets the requirements of the inductor circuit. The frequency modulation value whose modulated resistivity value meets the value lower than the reference resistivity value is screened out as the set value of the actual frequency range. Since the frequency value is inversely correlated with the Skin scalar value, the lower boundary value of the actual frequency range must be defined first, that is, the frequency value must be greater than the minimum frequency modulation value screened out, and the remaining frequency modulation values ​​are aggregated to set the upper boundary value, that is, each frequency modulation value is used to ensure the stability of the inductor circuit parameters.

[0035] Furthermore, as a feasible implementation method, the construction of the convolutional neural network includes: performing normalization preprocessing on the input frequency value, collecting the skin scalar value and the magnetic permeability value of the target inductor under several different current values, and setting each group of skin scalar value and magnetic permeability value as an inductance data set in a set form after weighted addition; constructing a first distance matrix and a second distance matrix for the mean of the initial frequency value and the inductance data set based on the Manhattan distance, taking all the first distance matrices and the second distance matrices as input values ​​of the convolutional neural network, and outputting the frequency modulation value after the frequency value is modulated in the form of a single neuron.

[0036] The skin scalar value and permeability value under different current values ​​are collected to deeply analyze the electromagnetic characteristics of the inductor under different working conditions, including the evaluation of the frequency response, loss characteristics and magnetic coupling effect in the actual application of the inductor circuit. The weighted addition of the skin effect and the permeability constitutes the inductor data set, which is used to expand the dimension of each set of data, and helps to provide more information about the inductor characteristics, so that the convolutional neural network can better learn and understand the response of the inductor under different working conditions. The role of constructing the first and second distance matrices as the input of the neural network is to provide rich feature information and complex distance relationships, so that the neural network can better understand and regulate the behavior of the inductor at different frequencies, so as to achieve the purpose of accurately controlling and optimizing frequency modulation. The use of Manhattan distance can effectively measure the difference between two vectors. By calculating the distance between the mean of the initial frequency value and each sample in the inductor data set, these distances can be used as input features. That is, the method of constructing a distance matrix based on Manhattan distance and using it as the input of the convolutional neural network can improve the accuracy and adaptability of the inductor frequency modulation method. In the frequency modulation process, using a single neuron to complete the output helps to reduce the complexity of the model and the computational cost of training, while reducing the risk of overfitting. At the same time, a single neuron can be optimized during the training process to reduce the error between the prediction and the true value. The optimization process can usually be performed through a standard regression loss function (such as mean square error).

[0037] Furthermore, as a feasible implementation method, the input and output process of the convolutional neural network includes: merging the first distance matrix and the second distance matrix into a two-dimensional input tensor, and using convolution layers and pooling layers to sequentially extract input data features of the two-dimensional input tensor; adding a fully connected layer to the input data features to map the input data features to the output layer; setting a linear activation function for the output layer to map a single neuron output; and constructing a multi-dimensional feature space for calculation expansion when constructing the second distance matrix using Manhattan distance.

[0038] The two-dimensional input tensor refers to a tensor with two dimensions, which is usually used to pass input data to a convolutional neural network. A convolutional layer refers to a neural network layer that applies a convolution operation. Convolution operation is a common operation in image processing, which can effectively extract local features of an image. In a neural network, a convolutional layer performs a convolution operation on the input data by using a set of learnable filters (or convolution kernels). A pooling layer is a technique used to reduce the number of parameters and computation in a convolutional neural network. It reduces the spatial dimension size of each feature map by downsampling certain dimensions of the input data. The first distance matrix and the second distance matrix are combined into a two-dimensional input tensor. This means that the distance values ​​in each distance matrix are organized into a two-dimensional matrix form. The distance values ​​can be normalized distances or other pre-processed values. Convolutional layers and pooling layers are sequentially applied to the two-dimensional input tensor to extract features of the input data. Convolutional layers help detect spatial patterns in the input data, while pooling layers help reduce the spatial dimension of the data while retaining important feature information. After feature extraction, the fully connected layer maps the extracted features to the output layer, which contains a single neuron and usually uses a linear activation function to map to the final frequency modulation value. The linear activation function is suitable for regression tasks and can output continuous numerical results. The multi-dimensional feature space means that more feature dimensions can be considered, such as the skin scalar value and magnetic permeability value of the inductor, and may even include other related features, which can more comprehensively capture the characteristics and behaviors of the inductor under different working conditions and improve the accuracy of the second distance matrix.

[0039] Further, as a feasible implementation method, the construction process of the first distance matrix includes: assuming that the frequency value is represented by f, and the mean of the initial frequency value is represented by , let the first distance matrix be expressed as D, then the first distance matrix D is expressed as: , where i and j represent the ordinal numbers of the means of the initial values ​​of the frequencies.

[0040] The first distance matrix D can be used to represent the degree of difference in the initial values ​​of the inductor frequency in different working cycles. It is based on the absolute difference calculation and is used to quantify the change in the initial frequency value between different working cycles. In the inductor frequency modulation method, the first distance matrix D may be used as one of the inputs of the convolutional neural network, or as one of the key features of the data in the model training and optimization process. By comparing and analyzing the differences in the initial frequency values ​​in different cycles, it can help optimize the frequency modulation algorithm and strategy. In practical applications, the average initial frequency value and The selection and calculation method of needs to be determined according to the specific working characteristics of the inductor and the frequency of data collection. At the same time, considering the calculation efficiency and model accuracy, it may be combined with other characteristics and data for comprehensive analysis and processing.

[0041] Furthermore, as a feasible implementation method, the PCA principal component analysis method is used to perform feature dimensionality reduction on the input data features of the two-dimensional input tensor, and the input data features after feature dimensionality reduction are decomposed into eigenvalues ​​through covariance matrix calculation, and a feature threshold is set to filter out the eigenvalues ​​whose number of principal components is lower than the feature threshold, and the filtered input data features are input into the convolution layer and pooling layer of the convolutional neural network.

[0042] PCA transforms the original high-dimensional feature space into a low-dimensional feature space through linear transformation while retaining as much data variance as possible. In this case, PCA applied to a two-dimensional input tensor may reduce redundant information between features and retain the main data variation. This feature dimension reduction can help reduce the dimensionality of the data, thereby reducing computational complexity and improving the training efficiency of the model. PCA finds the principal components, i.e., the main feature directions, by calculating the covariance matrix of the input data and then performing eigenvalue decomposition on the covariance matrix. The eigenvalues ​​represent the variance of the data in each principal component direction, while the eigenvectors represent the principal component directions corresponding to these variances. Setting a feature threshold can help determine how many principal components to retain for dimensionality reduction. Typically, the number of principal components to retain is selected based on the percentage of the total variance retained or a specific threshold is set. Based on the set feature threshold, PCA can screen out principal components that contribute less to the overall variance. These principal components usually contain relatively unimportant information in the data, and retaining important principal components can more effectively characterize the variability of the data set. The features after PCA dimensionality reduction are input into the convolutional layers and pooling layers of the convolutional neural network, which will use the reduced features for feature extraction and further reduce the spatial dimension, thereby reducing the computational burden and number of parameters of subsequent layers while maintaining efficient use of important features of the data.

[0043] Among them, as a preferred embodiment, the PCA principal component analysis method uses an incremental PCA analysis method to gradually update the covariance matrix. Incremental PCA is a special PCA method that allows new data blocks to be processed step by step and the principal component analysis results to be updated without knowing the entire data set. This method is suitable for data stream scenarios or when the data set is too large to be processed at one time. Since the core of PCA is to calculate the covariance matrix of the data set and perform eigenvalue decomposition on the matrix to obtain the principal components. Incremental PCA is implemented by updating the estimated value of the covariance matrix each time a new data block is processed. Specifically, for each new data block, incremental PCA updates the current mean estimate and the estimate of the covariance matrix. In convolutional neural networks, incremental PCA can be used in the preprocessing step, that is, before the input data enters the convolution layer. Processing input data features by gradually updated incremental PCA can enable the network to handle dynamically changing data streams or large-scale data sets while reducing the demand for memory and computing resources.

[0044] Further, as a feasible implementation manner, the process of performing data processing on the remaining frequency modulation values ​​includes: the new upper boundary value after the data aggregation includes a first reference value, a second reference value and an implementation cutoff value, the first reference value is obtained by taking the average of the remaining frequency modulation values, the second reference value is obtained by taking the standard deviation of the remaining frequency modulation values, and the implementation cutoff value includes the first modulation value and the second modulation value extracted by arranging the remaining frequency modulation values ​​from small to large;

[0045] When the first reference value is greater than the second reference value, selecting the first modulation value as a new upper boundary value;

[0046] When the first reference value is less than the second reference value, selecting the second modulation value as a new upper boundary value;

[0047] Among them, the implementation cutoff value also includes a third modulation value extracted from the remaining frequency modulation value, and the size of the third modulation value is between the first modulation value and the second modulation value; when the integer value of the first reference value is equal to the integer value of the second reference value, the third modulation value is selected as the new upper boundary value.

[0048] The new upper boundary value is determined based on the comparison between the first reference value and the second reference value. If the first reference value (mean) is greater than the second reference value (standard deviation), it means that the data is more discrete. In order to maintain frequency stability, a relatively small first modulation value is selected as the new upper boundary value to control the frequency range. If the first reference value is less than the second reference value, it means that the central tendency of the data is more obvious. In this case, the second modulation value is selected as the new upper boundary value to cover as many feasible frequency modulation values ​​as possible. When the first reference value is equal to the second reference value after rounding, it means that most of the values ​​in the data set are distributed around the mean, and the data points are relatively concentrated. In this case, the numerical distribution of the data set is relatively stable, without obvious bias or outliers. At this time, a value in the center of the frequency modulation value can be directly selected as the third modulation value as the new upper boundary value to set the actual frequency range.

[0049] As a specific application, in a specific implementation, preferably, the first modulation value, the second modulation value and the third modulation value are respectively set to represent the 25th percentile, the median and the 75th percentile of the remaining frequency modulation values ​​arranged from small to large. Selecting percentiles as modulation values ​​can more accurately reflect the distribution characteristics of the data. These percentiles can better describe the position and distribution of data in the data set, and more comprehensively reflect the characteristics of the data set than simple averages or center positions. At the same time, percentiles have less impact on outliers because they are based on the sorting position of the data rather than the specific numerical value, and therefore can provide more robust statistical indicators.

[0050] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for designing an inductor based on frequency modulation, characterized in that: The method includes: Step S1: recording the initial frequency value and the magnetic permeability value of the target inductor conductor in the current circuit, presetting the Skin scalar value, and calculating the reference resistivity value of the target inductor in the current circuit based on the initial frequency value, the Skin scalar value and the magnetic permeability value; Step S2: setting a reference frequency range used as a frequency initial value modulation target, representing an upper limit value and a lower limit value in the reference frequency range as an upper boundary value and a lower boundary value, respectively, and dividing the modulation process into a plurality of working cycles of the same length; Step S3: construct a convolutional neural network, input the average of the initial frequency values ​​of the inductance in each working cycle as an input value into the convolutional neural network, and set all output values ​​to frequency modulation values ​​representing the frequency value after modulation; Step S4: all frequency modulation values ​​are calculated based on the skin scalar value and the magnetic permeability value to obtain all modulated resistivity values, and all frequency modulation values ​​whose modulated resistivity values ​​are lower than the reference resistivity value are screened out; Step S5: setting the screened minimum frequency modulation value as a new lower boundary value, performing data processing based on the remaining frequency modulation values, determining a new upper boundary value according to the data processing result, and setting the new upper boundary value and the new lower boundary value as the actual frequency range of the target inductor circuit; The process of processing the remaining frequency modulation values ​​includes: the new upper boundary value after the data aggregation includes a first reference value, a second reference value and an implementation cutoff value, the first reference value is obtained by taking the mean of the remaining frequency modulation values, the second reference value is obtained by taking the standard deviation of the remaining frequency modulation values, and the implementation cutoff value includes the first modulation value and the second modulation value extracted by arranging the remaining frequency modulation values ​​from small to large; When the first reference value is greater than the second reference value, selecting the first modulation value as a new upper boundary value; When the first reference value is less than the second reference value, selecting the second modulation value as a new upper boundary value; The implementation cutoff value also includes a third modulation value extracted from the remaining frequency modulation value, the third modulation value being between the first modulation value and the second modulation value; when the integer value of the first reference value is equal to the integer value of the second reference value, the third modulation value is selected as the new upper boundary value; The first modulation value, the second modulation value and the third modulation value are respectively set to represent the 25th percentile, the median and the 75th percentile of the remaining frequency modulation values ​​arranged from small to large.

2. The inductor design method based on frequency modulation according to claim 1, characterized in that: The construction of the convolutional neural network includes: normalizing the input frequency value, collecting the skin scalar value and the magnetic permeability value of the target inductor under several different current values, and setting each group of skin scalar value and magnetic permeability value as an inductance data set in a set form after weighted addition; constructing a first distance matrix and a second distance matrix for the mean of the initial frequency value and the inductance data set based on the Manhattan distance, taking all the first distance matrices and the second distance matrices as input values ​​of the convolutional neural network, and outputting a frequency modulation value after the frequency value is modulated in the form of a single neuron.

3. The inductor design method based on frequency modulation according to claim 2, characterized in that: The input and output process of the convolutional neural network includes: merging the first distance matrix and the second distance matrix into a two-dimensional input tensor, using a convolution layer and a pooling layer to sequentially extract input data features of the two-dimensional input tensor; adding a fully connected layer to the input data features to map the input data features to an output layer; and setting a linear activation function for the output layer to map a single neuron output.

4. The inductor design method based on frequency modulation according to claim 2, characterized in that: The construction process of the first distance matrix includes: assuming that the frequency value is represented by f, and the mean of the initial frequency value is represented by , let the first distance matrix be expressed as D, then the first distance matrix D is expressed as: , where i and j represent the ordinal numbers of the means of the initial values ​​of the frequencies.

5. The inductor design method based on frequency modulation according to claim 2, characterized in that: Construct a multi-dimensional feature space for Manhattan distance to perform computational expansion when constructing the second distance matrix.

6. The inductor design method based on frequency modulation according to claim 3, characterized in that: The PCA principal component analysis method is used to reduce the dimension of the input data features of the two-dimensional input tensor. The input data features after feature dimension reduction are decomposed into eigenvalues ​​through covariance matrix calculation, and a feature threshold is set to filter the principal components of the eigenvalues ​​whose number is lower than the feature threshold. The filtered input data features are input into the convolution layer and pooling layer of the convolutional neural network.

7. The inductor design method based on frequency modulation according to claim 6, characterized in that: The PCA principal component analysis method uses an incremental PCA analysis method to gradually update the covariance matrix.

Images