Magnetic core parameter optimization method and system based on magnetic energy transmission and loss optimization

Abstract

The invention discloses a magnetic core parameter optimization method based on magnetic energy transmission and loss optimization, which relates to the technical field of high-frequency magnetic elements and comprises the following steps: extracting characteristic parameters of a magnetic core, establishing an excitation waveform classification model in combination with the frequency of a working thermometer, and classifying excitation waveforms; based on the excitation waveform classification result, a temperature correction coefficient is introduced into a SteinMazz equation, and a corrected SteinMazz equation is obtained and used for predicting magnetic core losses of different excitation waveforms; analyzing the independent or synergistic effect of the temperature, the excitation waveform and the magnetic core material on the magnetic core loss; constructing a neural network magnetic core loss prediction model by taking the frequency, the waveform, the magnetic flux density peak value, the temperature and the magnetic core material as characteristics, constructing a fitness function by using the maximum transmission magnetic energy and the minimum magnetic core loss, and optimizing the input characteristics by using a genetic algorithm to obtain optimized magnetic core parameters; and excellent fitting precision and generalization ability are obtained in the aspects of excitation waveform classification, stein-Mitz equation correction and magnetic core loss regression.

Description

Technical Field

[0001] The present invention relates to the technical field of high-frequency magnetic components, and in particular to a method and system for optimizing magnetic core parameters based on magnetic energy transmission and loss optimization. Background Art

[0002] As an important cornerstone of modern industrial and technological development, power electronics technology plays an irreplaceable role in improving energy conversion efficiency and realizing intelligent control of electrical systems. In particular, in the fields of new energy power generation, electric vehicles, smart grids and various information and communication technologies, the application of power electronics technology is becoming more and more extensive, playing a key role in promoting technological innovation and upgrading of related industries.

[0003] With the continuous advancement of power semiconductor technology, power electronic equipment is rapidly developing towards high frequency and high power density. This trend not only puts higher requirements on power conversion technology, but also brings new challenges and opportunities to the design and application of magnetic components. Magnetic components, such as transformers and inductors, play a vital role in power converters. They not only affect the efficiency of power conversion, but are also directly related to the volume and weight of the entire system.

[0004] However, under high-frequency conditions, the loss characteristics of magnetic components become particularly complex, mainly due to the complexity of the microstructure of high-frequency magnetic materials and their nonlinear behavior under high-frequency alternating magnetic fields. Traditional finite element simulation methods based on electromagnetic field theory are often incapable of accurately predicting the actual performance of magnetic components in such situations. Therefore, it is particularly important to study and develop a core loss model suitable for high-frequency conditions.

[0005] At present, the core loss models commonly used in the industry and academia include loss separation models and empirical calculation models. Typical models include the Preisach model, the Hodgdon model, the Jiles-Atherton model, and the Steinmetz equation (SE). Although these models can simulate and predict the loss characteristics of the core to a certain extent, the core loss is affected by multiple factors such as frequency, flux density, excitation waveform, temperature, and material properties. The existing models often have certain limitations, such as a narrow scope of application or insufficient accuracy in practical applications.

[0006] In view of the above challenges, in-depth research on the loss mechanism of magnetic components under high-frequency conditions and the exploration of more accurate and universal loss prediction models are of great significance for optimizing the design of magnetic components and improving the overall performance of power electronic equipment. This requires not only theoretical innovation, such as developing new mathematical models to more accurately describe and predict the behavior of high-frequency magnetic materials, but also the support of experimental data to ensure the effective connection between theoretical models and practical applications.

[0007] Considering the development trend of power electronics technology in the future, the study of the loss characteristics of high-frequency magnetic components also needs to focus on the application of new materials, the design of new magnetic components, and the corresponding manufacturing process innovation. Through multidisciplinary cross-collaboration and the comprehensive application of the latest research results in the fields of materials science, electromagnetism, circuit design, etc., it will help to break through the existing technical bottlenecks and promote the further development of power electronics technology. Therefore, it is urgent to develop a data-driven method to construct a high-precision and universal magnetic loss model to meet the needs of modern power electronic equipment for performance optimization and prediction. Summary of the invention

[0008] The purpose of the present invention is to provide a method for optimizing magnetic core parameters based on magnetic energy transmission and loss optimization, so as to solve the problems existing in the above-mentioned prior art and achieve technical effects of high applicability and high precision.

[0009] To achieve the above objectives, in a first aspect, the present invention provides the following solution: a method for optimizing magnetic core parameters based on magnetic energy transmission and loss optimization, comprising the following steps:

[0010] Step 1: Extract the characteristic parameters of the magnetic core, combine them with the working thermometer frequency, establish an excitation waveform classification model, and classify the excitation waveform;

[0011] Step 2: Based on the excitation waveform classification result, a temperature correction coefficient is introduced into the Steinmetz equation to obtain a modified Steinmetz equation for predicting the core loss of different excitation waveforms;

[0012] Step 3: Analyze the independent or synergistic effects of temperature, excitation waveform and core material on core loss;

[0013] Step 4: Use frequency, waveform, peak value of magnetic flux density, temperature and core material as features to construct a predictive neural network core loss prediction model, construct the maximum transmitted magnetic energy and the minimum core loss into a fitness function, use a genetic algorithm to optimize the input features, and obtain the optimized core parameters.

[0014] Furthermore, in the step one, the characteristic parameters of the magnetic core include time domain characteristics, frequency domain characteristics, and geometric characteristics of the magnetic flux density data; based on the provided magnetic flux density data, characteristic variables are selected from its waveform characteristics, time domain characteristics, and frequency domain characteristics to distinguish different waveforms; a classification model is selected, and the selected characteristic variables, operating temperature, and frequency are input into the classification model, and the sine wave, triangular wave, and trapezoidal wave are classified and the classification accuracy of the three models is evaluated, and the importance of the features used is analyzed in combination with the classification effect of the model; the product of the frequency and the peak value of the magnetic flux density is used as the magnetic flux density data.

[0015] Furthermore, the characteristic variables include maximum value, minimum value, average slope, and number of zero crossing points, and the classification model adopts a random forest model, a support vector machine model, or a K-nearest neighbor algorithm model.

[0016] Furthermore, in the step 2, a linear temperature correction coefficient is added to the Steinmetz equation to fit the core loss under different materials, frequencies and flux densities, and the mean square error, absolute error and goodness of fit are used as equation accuracy evaluation indicators to perform polynomial temperature correction on the Steinmetz equation, and obtain the corrected core loss result by parameter fitting and waveform extraction of the data; the obtained polynomial temperature correction results are analyzed with the standard equation and the linear correction results respectively to obtain the temperature correction of the Steinmetz equation; the set temperature only affects the material coefficient, and does not affect the frequency correction coefficient and the maximum flux influence coefficient.

[0017] Furthermore, in the step three, the characteristic variables affecting the core loss and their response processes are subjected to variance analysis through single-factor analysis and two-factor interaction analysis methods, and the Bonferroni correction method and characteristic contribution calculation are used to minimize the core loss.

[0018] Furthermore, in step 4, characteristic variables with greater influence on core loss are extracted from the experimental data as inputs of subsequent models, and multiple machine learning models including BP neural network, GBDT model, decision tree regression, SVR and LSTM are used as prediction models to establish a mapping relationship between data set indicators and core loss, and to construct a core loss prediction model. Through model comparison and accuracy evaluation, MSE and R 2 As the model accuracy evaluation index, the prediction neural network core loss prediction model with the best prediction effect is selected to predict the actual core loss.

[0019] Furthermore, in step 4, the constructed prediction neural network core loss prediction model is used as the objective function, and a regression model for core loss is constructed according to temperature, frequency, waveform, magnetic flux density peak value and core material. On the basis of the regression model, a fitness function is constructed, and the minimum core loss and the operating point with the maximum transmission magnetic energy are obtained by genetic algorithm optimization calculation. The specific fitness function is:

[0020] Among them, E represents the transmitted magnetic energy, P represents the core loss, L represents the input-loss ratio, C represents the reward item, and W represents the penalty item.

[0021] In a second aspect, the present invention also provides a magnetic core parameter optimization system based on magnetic energy transmission and loss optimization, including a classification module, a correction module, an analysis module and an optimization module;

[0022] The classification module is used to extract the characteristic parameters of the magnetic core, combine the working thermometer frequency, establish the excitation waveform classification model, and classify the excitation waveform;

[0023] The correction module introduces a temperature correction coefficient into the Steinmetz equation based on the classification result of the excitation waveform to obtain a corrected Steinmetz equation for predicting the core loss of different excitation waveforms;

[0024] The analysis module is used to analyze the independent or synergistic effects of temperature, excitation waveform and core material on core loss;

[0025] The optimization module is used to construct a BP neural network using frequency, waveform, peak value of magnetic flux density, temperature and core material as features, construct the maximum transmitted magnetic energy and minimum core loss into a fitness function, and use a genetic algorithm to optimize the input features to obtain optimized core parameters.

[0026] In a third aspect, the present invention can also provide a computer device, including a processor and a memory, the memory being used to store a computer executable program, the processor reading the computer executable program from the memory and executing it, and when the processor executes the computer executable program, it can implement a magnetic core parameter optimization method based on magnetic energy transmission and loss optimization described in the present invention.

[0027] A computer-readable storage medium is also provided, in which a computer program is stored. When the computer program is executed by a processor, a magnetic core parameter optimization method based on magnetic energy transmission and loss optimization described in the present invention can be implemented.

[0028] Compared with the prior art, the advantages of the present invention are: when constructing the model, the present invention uses a multi-model comparison method, constructs different data-driven models for different problems, optimizes model parameters with data from the training set, and tests the generalization ability of the model with data from the verification set. Finally, relatively good fitting accuracy and generalization ability are achieved in the problems of excitation waveform classification, Steinmetz equation correction, and core loss regression.

[0029] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention in conjunction with the accompanying drawings. The specific implementation of the present invention is given in detail by the following embodiments and their accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0031] Figure 1 It is a flow chart of a method for optimizing magnetic core parameters based on optimization of magnetic energy transmission and loss according to an embodiment of the present invention;

[0032] Figure 2 A schematic diagram of a flow chart for establishing an excitation waveform classification model according to an embodiment of the present invention;

[0033] Figure 3 A schematic diagram of a temperature correction process for establishing the Steinmetz equation according to an embodiment of the present invention;

[0034] Figure 4 A schematic diagram of a process for establishing optimal conditions for core loss according to an embodiment of the present invention;

[0035] Figure 5 A schematic diagram of a flow chart for establishing a magnetic core loss prediction model according to an embodiment of the present invention;

[0036] Figure 6 A schematic diagram of a flow chart of establishing an optimized core loss prediction model according to an embodiment of the present invention;

[0037] Figure 7 Predict the results for the linear temperature correction equation;

[0038] Figure 8 The k value changes of the standard Steinmetz equation, linear temperature correction and polynomial temperature correction at 0-100℃;

[0039] Fig. 9 Prediction graph for polynomial temperature correction equation;

[0040] Fig.10 It is a scatter plot of predicted loss and actual loss;

[0041] Fig.11 It is a schematic diagram of single factor contribution rate;

[0042] Fig.12 It is the training set and test set in the prediction process of the same model;

[0043] Fig.13 The MSE change of the training set for the prediction of core loss using the BP neural network model;

[0044] Fig.14 is the prediction result of BP neural network model;

[0045] Fig.15 Schematic diagram of the average fitness curve of the genetic algorithm optimization process. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] like Figure 1 As shown, a method for optimizing magnetic core parameters based on magnetic energy transmission and loss optimization includes:

[0048] Step 1: Establish a feature classification system for different excitation waveforms, establish an excitation waveform classification model, and classify the excitation waveforms;

[0049] As an example that can be implemented, Figure 2 As shown, based on the provided flux density data, characteristic variables are selected from the waveform characteristics, time domain characteristics, and frequency domain characteristics to distinguish different waveforms. The characteristic variables include maximum value, minimum value, average slope, and number of zero crossing points. A suitable machine learning model is selected, and the selected characteristic variables as well as the operating temperature and frequency are input into three classification models: random forest model, support vector machine model, and K nearest neighbor algorithm model. Sine waves, triangle waves, and trapezoidal waves are classified and the classification accuracy of the three models is evaluated. The importance of the features used is analyzed in combination with the classification effect of the model.

[0050] Based on the extracted magnetic flux density related features and the working temperature and frequency of the magnetic core material, this paper conducts a correlation analysis. The Pearson correlation coefficient r is used to measure the strength of the association between each feature and between it and the waveform category; the three parameters with the highest correlation between the feature and the target are the median (-0.82), median (-0.47), and waveform area (0.41) of the magnetic flux intensity, indicating that the number of second-order derivative discontinuities of the waveform, the median of the magnetic flux intensity, and the waveform area have a strong influence on the waveform category; three classification models, random forest, support vector machine, or K nearest neighbor, are used to construct a waveform classification model using the above feature variables. The random forest model is completely correct in classifying sine waves and triangle waves, but there is only one error for trapezoidal waves, indicating that the model can well capture the mapping relationship between feature variables and target variables, and can well complete the waveform classification task.

[0051] As a preferred embodiment, the features are ranked by importance using the random forest feature optimization method, and the importance score of each feature is calculated using the random forest Gini index to rank the features by importance.

[0052] Step 2: Based on the excitation waveform classification result, a temperature correction coefficient is introduced into the Steinmetz equation to obtain a modified Steinmetz equation for predicting the core loss of different excitation waveforms;

[0053] As an example that can be implemented, Figure 3 As shown, a linear temperature correction coefficient is added to the Steinmetz equation to fit the core loss under different materials, frequencies and flux densities. The mean square error, absolute error and goodness of fit are used as equation accuracy evaluation indicators to perform polynomial temperature correction on the Steinmetz equation. The corrected core loss result is obtained by parameter fitting and waveform extraction of the data. The obtained polynomial temperature correction results are analyzed with the standard equation and the linear correction results, respectively, to obtain the temperature correction of the Steinmetz equation.

[0054] The Steinmetz equation is a classic model for estimating the energy loss of magnetic materials under sinusoidal current excitation. In practical applications, the characteristics of the magnetic core, such as the magnetization curve, magnetic permeability and coercive force, will change significantly due to temperature changes. When the temperature rises, the performance of the magnetic material will also change, resulting in an increase in core loss. In order to improve the applicability of the formula under different temperature conditions, a temperature variable T is introduced to realize the temperature correction of the Steinmetz equation. The invention adopts two correction methods: linear temperature and nonlinear temperature correction, in which the nonlinear temperature correction is a quadratic polynomial temperature correction. The two correction methods correct k as follows:

[0055] Linear temperature correction:

[0056] k(T)=k0·(1+a·T)

[0057] In the formula, k0 represents the constant at standard temperature; T represents temperature, unit: (℃); a represents the temperature coefficient of the first-order term.

[0058] Quadratic polynomial temperature correction:

[0059] k T =k0·(1+a·T+b·T 2 )

[0060] In the formula, k0 represents the constant at standard temperature; T represents temperature, unit: (℃); a represents the temperature coefficient of the linear term; b represents the temperature correction coefficient of the quadratic term.

[0061] In non-extreme temperature environments, the frequency index and the magnetic flux density index fluctuate slightly. They can be assumed to be constants when building the model, and their values ​​are only revised under different correction conditions. The corrected Steinmetz equation is as follows:

[0062] The linear temperature coefficient is revised and adjusted model:

[0063]

[0064] Polynomial temperature revised adjusted model:

[0065]

[0066] The nonlinear least squares method is used to fit the objective function, and the entire data set is predicted through the fitted function. In order to ensure the generalization ability of the model and the stability of the prediction, K-fold cross validation is used to verify it to avoid overfitting of the model, aiming to provide a more robust model evaluation. At the same time, the mean square error (MSE), mean absolute error (MAE) and goodness of fit (R 2 ) is used as the model accuracy evaluation index, and the fitting parameters of the standard Steinmetz equation are k=1.53, α=1.43, and β=2.47. Among them, k represents the overall level of core loss, reflecting the core loss under specific materials and working conditions. The frequency influence index α describes the influence of the operating frequency on the core loss. α=1.43 indicates that the core loss shows a slightly higher than linear growth pattern with the increase of frequency. The influence index of the maximum value of the flux density is 2.47, indicating the significant influence of the flux density on the loss. The fitting formula is:

[0067]

[0068] The temperature correction coefficient a is added to the standard Steinmetz equation to construct a linear temperature correction model. The evaluation results of the model accuracy evaluation index show that the mean square error, mean absolute error, and goodness of fit are 282277074, 9809, and 0.9904, respectively. Compared with the standard Steinmetz equation, the accuracy of the mean square error and mean absolute error are optimized by 592% and 212%, respectively, and the goodness of fit performance is also significantly improved. The predicted result and the true value of the core loss reach a fit of 0.9904. At the same time, the fitting parameters of the linear temperature-corrected Steinmetz equation are k=1.87, α=1.44, a=-0.0054, and β=2.43, respectively. The material constant after correction is 1.87, which is higher than the original standard equation, indicating that the core loss level at the corrected standard temperature has increased. The value of a is -0.0054, indicating that temperature is inversely proportional to core loss. The α parameter of the equation before and after correction does not change much, which is 1.43 and 1.44 respectively, indicating that the correction of the formula has little to do with the effect of frequency on loss. The fitting formula is as follows:

[0069]

[0070] The actual core loss and the linear temperature correction equation prediction results are as follows Figure 7 As shown. In the comparison of core loss prediction, it can be seen that the error between the predicted value and the actual value is very small. The linear modified Steinmetz equation fitting results are smaller in the front and back data, while the prediction in the middle of the data is slightly larger than the true value. The prediction accuracy of the linear temperature correction equation for the entire data segment is much higher than that of the standard Steinmetz equation, and the model fitting optimization effect is better.

[0071] The fitting parameters of the polynomial temperature-corrected Steinmetz equation are k=1.71, α=1.47, a=-0.0124, β=2.452, and b=6.98e-05. The fitting formula of the polynomial temperature-corrected Steinmetz equation is as follows:

[0072]

[0073] The k values ​​of the standard Steinmetz equation, linear temperature correction, and polynomial temperature correction are varied from 0 to 100°C to observe the changes in the overall level of core loss at different temperatures. The results are as follows Figure 8 shown.

[0074] from Figure 8 It can be found that as the temperature increases, the overall level of core loss in both the linear temperature correction and polynomial temperature correction models shows a downward trend. The prediction results using polynomial temperature correction are shown in Figure 2. Fig. 9 shown.

[0075] from Fig. 9It can be seen that the core loss prediction value of this model is extremely close to the actual value, and there is no situation where the predicted data of a long section or a large section is higher or lower than the actual data. Compared with the standard formula and the linear temperature correction formula, this model can adjust faster when the predicted result value deviates, making the prediction result fit better. The predicted data of the whole section data is compared with the actual data, and the results are as follows Fig.10 As shown, the polynomial temperature correction predicts the core loss closer to the actual core loss.

[0076] Step 3: Establish an analytical model to evaluate the independent or synergistic effects of temperature, excitation waveform and core material on core loss;

[0077] As an example that can be implemented, Figure 4 As shown, through single-factor analysis and two-factor interaction analysis methods, the characteristic variables affecting the core loss and their response processes are subjected to variance analysis, and the Bonferroni correction method and characteristic contribution calculation are used to minimize the core loss.

[0078] ANOVA was selected to conduct a single factor analysis on the core loss. As a commonly used statistical method, ANOVA is suitable for testing whether different groups (such as temperature, excitation waveform, material) have a significant effect on the response variable (core loss). The basic formula is as follows:

[0079]

[0080] Where n i represents the number of samples in the i-th group; represents the mean of the i-th group; represents the overall mean; k represents the number of groups; n represents the total number of samples.

[0081] Then, in order to investigate the effect of the interaction between the two factors on the core loss (i.e., the interaction between temperature and excitation waveform, temperature and core material, and excitation waveform and core material), it is necessary to construct an interactive relationship between two or more characteristic variables during the statistical analysis process, and use the interaction term to analyze the effect of multiple factors on the response variable. The formula of the interaction model is as follows:

[0082] Lossβ=0+β1·T+β2·W+β3·M+β4·(T·W)+β5·(T·M)+β6·(W·M)+ε

[0083] Where β1, β2, and β3 represent single factor regression coefficients respectively; β4, β5, and β6 represent interaction coefficients of different factors respectively; T, W, and M represent three types of characteristic variables: temperature, waveform, and material respectively; β0 and ε represent the error terms of single factors and interaction terms respectively.

[0084] Modeling uses analysis of variance (ANOVA) to quantify the impact of interactions. In the calculation of the F value of the interaction, the sum of squares (SS) of the independent variable and the interaction term is split. The formula is:

[0085]

[0086] In the formula, y ij is the observed value in group i and level j; is the interaction term predicted by the model; is the mean of a single factor; is the mean of another factor; is the population mean.

[0087] The significance of the interaction on the response variable is explored through the interaction sum of squares and degrees of freedom, as well as the error sum of squares and degrees of freedom. The calculation formula of the interaction F value is as follows:

[0088]

[0089] Where: SS interaction is the sum of squares of the interaction terms; SS residual is the error sum of squares; df interaction is the degree of freedom of the interaction term, which is equal to the product of the two factor levels minus 1; df residual is the error degrees of freedom.

[0090] Based on the regression coefficient of each factor, the standardized regression coefficient is further used to quantify the contribution of each variable, and the relative contribution rate is used to compare the contribution of different features to the response variable. The calculation formula is as follows:

[0091]

[0092] Compared with a single factor, the regression coefficient β of the interaction term can better characterize the impact of the core loss under the synergistic effect of multiple factors. The modeling uses the regression coefficient β of the interaction term to calculate the contribution of the interaction term. The calculation formula is as follows:

[0093]

[0094] Based on the Python platform and Statsmodels tool, ANOVA one-way analysis of variance and two-way cross analysis of variance were performed. The analysis results are shown in Table 1:

[0095] Table 1 Single factor analysis results

[0096]

[0097]

[0098] It can be found from the above table that the F value and P value of the temperature effect on the core loss are 17.76 and 1.70E-11 respectively, which shows that the temperature has less influence on the core loss than the other two factors. By comparing the P value and F value under different factors, it can be found that the minimum P value of the excitation waveform on the core loss is 1.49E-118 and the maximum F value is 277.33, which shows that the influence of the excitation waveform on the core loss is extremely significant. The F value and P value of the core material on the loss are 99.12 and 2.12E-63 respectively, which shows that the core material has a certain influence on the core loss. In order to more accurately explore the influence of the three on the core loss, this paper uses the regression coefficient obtained by fitting to quantify the contribution rate of each factor to the loss change, such as Fig.11 As shown in Table 2, the contribution rates of core material, temperature and excitation waveform to core loss are 40.1%, 30.6% and 29.4% respectively. Therefore, under the premise of considering only the independent effect of a single factor, core material is the factor with the greatest impact on core loss. This shows that investing more energy in core material may be more beneficial to reducing core loss than the other two factors. In summary, temperature, excitation waveform and core material are all key factors affecting core loss. The effect of the interaction of the two factors on core loss is analyzed. As shown in Table 2, among the interaction of factors, the interaction effect of excitation waveform and core material is the most significant, with a P value of 2.72E-17. By comparing the P value and F value under different interactions, it can be found that the interaction between temperature and core material is not significant, with an F value and a P value of 0.68 and 0.732 respectively. From the analysis results of the interaction between temperature and excitation waveform, it can be seen that the interaction between temperature and excitation waveform has a certain effect on core loss.

[0099] Table 2 Results of two-factor analysis

[0100]

[0101] In order to better reflect the contribution rate of the two-factor interaction to the loss change, the variance contribution rate ω is used to quantify its contribution rate. The calculation formula is as follows. The results are shown in Table 3, which shows that the contribution rate of waveform and material to the loss change is the largest, which is 80.44. However, the interaction between temperature and material has the least significant effect on the loss change, which is 5.73.

[0102]

[0103] Table 3 Contribution rate of two-factor interaction to loss change

[0104]

[0105] As the temperature increases, the loss values ​​under various excitation waveforms show an overall downward trend, and the fluctuation range gradually decreases. The loss values ​​of the sine wave are most concentrated at all temperatures, showing a smaller fluctuation range. However, the loss distribution of the triangular wave and the trapezoidal wave under low temperature conditions is significantly expanded and the outliers increase, which indicates that these two waveforms will cause greater loss fluctuations at low temperatures. There are obvious differences in the loss performance of different core materials under different temperature conditions. As the temperature increases, the core loss of different materials has different trends, but generally shows a downward trend. In contrast, the loss range of material 3 is wider, especially under high temperature conditions, the loss increases significantly, showing poor high temperature performance. The loss of material 4 remains at a low level at all temperatures, and the fluctuation is small, showing good high temperature adaptability. Under high temperature (i.e. 90℃), the core loss is smaller than that under other temperature conditions. The core loss of different materials under the sine wave is the smallest, and its distribution is significantly reduced. The sine wave has the smallest core loss under different temperatures and core materials, which shows that the waveform with the smallest loss is the sine wave. The core loss of material 4 is the smallest at different temperatures and excitation waveforms, which means that the core material with the smallest loss is material 4. In summary, the core loss is the smallest when the temperature is 90°C, the material is material 4, and the excitation waveform is a sine wave.

[0106] Step 4: Establish a core loss prediction model suitable for different core materials and complex working conditions;

[0107] As an example that can be implemented, Figure 5 As shown in the figure, the characteristic variables that have a greater impact on the core loss are extracted from the experimental data as the input of the subsequent model, and multiple machine learning models are selected as candidate models for the prediction model. As an example, the machine learning models are BP neural network, GBDT model, decision tree regression, SVR and LSTM; based on this, the mapping relationship between the data set indicators and the core loss is established, and the core loss prediction model is constructed. MSE and R are selected 2 The model with the best prediction effect is selected as the model accuracy evaluation index to predict the actual core loss.

[0108] The present invention preferably adopts a BP neural network model for predicting the core loss.

[0109] Using python as the development platform and Scikit-Learn-1.5.0 as the development tool, we built the gradient boosting tree (GBDT), support vector machine regression (SVR) and decision tree regression models. Using Tensorflow-2.7.0 as the development tool, we built the BP neural network and long short-term memory network (LSTM). The core loss was predicted by gradient boosting tree, support vector machine regression, decision tree regression, BP neural network and long short-term memory network. The training set and test set in the prediction process of different models are shown in Figure 2. Fig.12 As shown. Also through MSE and R 2 The prediction accuracy of different models is evaluated, as shown in Table 4.

[0110] Table 4 Prediction accuracy evaluation of different models

[0111]

[0112] Depend on Fig.12 It can be clearly seen that when using SVR to predict core loss, the performance of the five models is the worst, whether from the test set or the training set. 2 It can be found that when the SVR model is used for prediction, overfitting or underfitting may occur. Therefore, the SVR regression model is not suitable for the prediction of core loss. When the LSTM model is used to predict the core loss, the predicted values ​​in the training set are lower than the true values ​​in all time periods. On the other hand, in the test set, some time periods are higher than the true set. MSE and R of the LSTM model 2 They are 6.0×10 10 and 0.56, which shows from another perspective that the LSTM has a large error in predicting the core loss.

[0113] When using the decision tree regression model to predict the core loss, it is not difficult to see that the training set is close to the true value both in terms of trend and value. From the accuracy evaluation results of the decision tree regression model in Table 4, it can be found that its R 2 The MSE of the model is 0.95, and it is smaller than that of the LSTM model. This shows that the applicability of the decision tree regression model to predict the core loss is higher than that of the SVR and LSTM. When the GBDT model is used to predict the core loss, its performance in both the training set and the test set is not as good as that of the decision tree regression model. From the perspective of the prediction accuracy evaluation parameters, its R 2 The MSE of the GBDT model and the decision tree regression model is slightly smaller than that of the decision tree regression model. This shows that the applicability of the GBDT model and the decision tree regression model for predicting core loss is not much different. When the BP neural network model is used to predict core loss, its performance is the best in both the training set and the test set. In addition, its R 2 It is as high as 0.99 and its MSE is much smaller than other models.

[0114] In summary, the BP neural network model is the most suitable for the prediction of core loss compared with the other four models. Fig.13As shown in the figure, with the increase of the number of training times, the MSE of the training set first shows a rapid decline trend and then tends to be flat. The decline rate reaches a sudden change at the 12th iteration and begins to gradually slow down. And it is almost close to 0 after 103 iterations. Between 12-103 iterations, although the decline rate fluctuates, it generally shows a decreasing trend.

[0115] Step 5: Further optimize the core loss prediction model.

[0116] As an example that can be implemented, Figure 6 As shown in the figure, the BP neural network core loss prediction model constructed in step 4 is used as the objective function, and a regression model for core loss is constructed according to temperature, frequency, waveform, magnetic flux density peak and core material. Then, based on the regression model, a fitness function is constructed, and the minimum core loss and the operating point with the maximum transmitted magnetic energy are obtained through genetic algorithm optimization calculation.

[0117] Firstly, BP neural network is used to build a regression model of core loss related to temperature, frequency, waveform, peak value of magnetic flux density and core material, and then genetic algorithm is used to optimize it.

[0118] When constructing the model, the present invention uses the K-fold verification and test set division multi-model comparison method, constructs different data-driven models for different problems, optimizes model parameters with the data of the training set, and tests the generalization ability of the model with the data of the verification set. Finally, relatively good fitting accuracy and generalization ability are achieved in the problems of excitation waveform classification, Steinmetz equation correction, and core loss regression; the importance of some parameters is analyzed, the contribution of each variable to the model performance is analyzed, and the decision process of the decision tree is visualized, which is particularly important for the evolution from data laws to physical laws; when constructing the classification model and the regression model, multiple models are compared to improve the accuracy and correctness of the model prediction results.

[0119] The core of using genetic algorithm to optimize it lies in the construction of fitness function. The following is a step-by-step introduction to the construction process of fitness function. For this problem, the ratio L of transmitted magnetic energy E to core loss P (hereinafter referred to as transmission loss ratio) is used as the measurement parameter, and its calculation method is as follows:

[0120] L=E / P

[0121] Where, E represents the transmitted magnetic energy; P represents the core loss, unit: (w / m 3 ); L represents the loss ratio (the ratio of transmitted magnetic energy to core loss).

[0122] The transmission of magnetic energy usually involves the interaction between the magnetic field and the magnetic material. The concept of this is relatively complicated to explain. Therefore, this article only uses the frequency f and the peak value of the magnetic flux density Bmax The product of is used to measure the size of the transmitted magnetic energy, and the formula is as follows:

[0123] Transmitted magnetic energy (E) = f × B max

[0124] Furthermore, the construction of the fitness function of the genetic algorithm requires finding the optimal operating point. First, when E>E max , and P<P min When , it indicates that the current working condition obtains a huge amount of transmitted magnetic energy with minimal loss. At this time, a larger reward term is added to the fitness function, that is:

[0125] F(P,E)=C+L........(E>E max )and(L<L min )

[0126] In the formula, E max represents the maximum value of the transmitted magnetic energy in the data set; L min represents the minimum value of the loss-input ratio in the data set; C represents the reward item, which takes a value of 10000.

[0127] When the core loss P is greater than the upper quartile of the core loss (P q3 ) and its loss ratio L is greater than its maximum value L max When , it is considered that the target operating condition is achieved, while taking into account both higher magnetic energy transmission and lower core loss, a bonus item is added, and the calculation method is as follows:

[0128] F(L,P)=Q+L........(P>P max )and(L>L max )

[0129] Where Q represents the reward item, and its value is 1000.

[0130] When the transmitted magnetic energy E is less than E min It is considered that the magnetic energy transmitted in this condition is small and has no practical application value, so a negative correction term is added.

[0131] F(L,P)=W

[0132] Where W represents the penalty term, and its value is -1000.

[0133] In other cases, L is used as the value of the fitness function, that is, F(L,P)=L

[0134] The final fitness function of this model is expressed as follows:

[0135]

[0136] Based on python3.8.2 as the platform, combined with tensorflow2.7.0 and deap 1.4.1, the BP neural network structure is built and the genetic algorithm is implemented. Only temperature, frequency, waveform, magnetic flux density peak and core material are required as independent variables. In order to detect the effect of the BP neural network model in this prediction process, the BP neural network model is first used to predict it. The results are as follows Fig.14 As shown. 2 The prediction accuracy was evaluated by using MSE, and the results are shown in Table 5.

[0137] Table 5 Prediction accuracy evaluation results

[0138]

[0139] From the evaluation results, it can be found that the BP neural network model has a high prediction accuracy, especially R 2 . It shows that the BP neural network model is suitable for this prediction and has good generalization ability. Because it is necessary to explore the temperature, frequency, waveform, magnetic flux density peak and core material under what conditions the core loss is minimum and the transmission magnetic energy is maximum. When applying the genetic algorithm, 50 populations are set, 100 iterations are performed, the population crossover rate is set to 0.5, the mutation rate is set to 0.2, and 30% of the inferior offspring are eliminated each time. The optimization process is as follows Fig.15 As shown in Figure 6, as the optimization process progresses, the average fitness of the offspring shows a trend of first rising rapidly and then stabilizing. The final prediction results of the genetic algorithm optimized BP neural network model are shown in Table 6.

[0140] Table 6 Final optimization results

[0141]

[0142]

[0143] According to historical data, the highest value of the transmission loss ratio is 3.63. The initial goal of optimization is to seek a working condition that can exceed this value. The final result after optimization is calculated by L=E / P, and the transmission loss ratio is 3.6328. This result shows that the optimization process has achieved the expected effect and further verifies the effectiveness of the optimization method.

[0144] Embodiment 2, the present invention also provides a magnetic core parameter optimization system based on magnetic energy transmission and loss optimization, including a classification module, a correction module, an analysis module and an optimization module;

[0145] The classification module is used to extract the characteristic parameters of the magnetic core, combine the working thermometer frequency, establish the excitation waveform classification model, and classify the excitation waveform;

[0146] The correction module introduces a temperature correction coefficient into the Steinmetz equation based on the classification result of the excitation waveform to obtain a modified Steinmetz equation for predicting the core loss of different excitation waveforms;

[0147] The analysis module is used to analyze the independent or synergistic effects of temperature, excitation waveform and core material on core loss;

[0148] The optimization module is used to construct a BP neural network using frequency, waveform, peak value of magnetic flux density, temperature and core material as features, construct the maximum transmitted magnetic energy and minimum core loss into a fitness function, and use a genetic algorithm to optimize the input features to obtain optimized core parameters.

[0149] On the other hand, the present invention provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, it can implement a magnetic core parameter optimization method based on magnetic energy transmission and loss optimization described in the present invention.

[0150] The present invention can also provide a computer device, including a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, it can implement a magnetic core parameter optimization method based on magnetic energy transmission and loss optimization described in the present invention.

[0151] The computer device may be a laptop computer, a desktop computer or a workstation.

[0152] The processor may be a central processing unit (CPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), or an off-the-shelf field programmable gate array (FPGA).

[0153] The memory of the present invention may be an internal storage unit of a laptop computer, a desktop computer or a workstation, such as a memory or a hard disk; or an external storage unit, such as a mobile hard disk or a flash memory card.

[0154] Computer-readable storage media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer-readable storage media may include: read-only memory (ROM), random access memory (RAM), solid-state drive (SSD) or optical disk, etc. Among them, random access memory may include resistance random access memory (ReRAM) and dynamic random access memory (DRAM).

[0155] The present invention uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for optimizing magnetic core parameters based on magnetic energy transmission and loss optimization, characterized in that: The following steps are involved: Step 1: Extract the characteristic parameters of the magnetic core, combine them with the working thermometer frequency, establish an excitation waveform classification model, and classify the excitation waveform; Step 2: Based on the excitation waveform classification result, a temperature correction coefficient is introduced into the Steinmetz equation to obtain a modified Steinmetz equation for predicting the core loss of different excitation waveforms; Step 3: Analyze the independent or synergistic effects of temperature, excitation waveform and core material on core loss; Step 4: Use frequency, waveform, peak value of magnetic flux density, temperature and core material as features to construct a predictive neural network core loss prediction model, construct the maximum transmitted magnetic energy and the minimum core loss into a fitness function, use a genetic algorithm to optimize the input features, and obtain the optimized core parameters.

2. A magnetic core parameter optimization method based on magnetic energy transmission and loss optimization according to claim 1, characterized in that: In the step 1, the characteristic parameters of the magnetic core include time domain characteristics, frequency domain characteristics, and geometric characteristics of the magnetic flux density data; Based on the provided magnetic flux density data, characteristic variables are selected from the waveform characteristics, time domain characteristics, and frequency domain characteristics to distinguish different waveforms; a classification model is selected, and the selected characteristic variables, operating temperature, and frequency are input into the classification model. The sine wave, triangular wave, and trapezoidal wave are classified and the classification accuracy of the three models is evaluated. The importance analysis of the features used is performed based on the classification effect of the model; the product of the frequency and the peak value of the magnetic flux density is used as the magnetic flux density data.

3. A method for optimizing magnetic core parameters based on magnetic energy transmission and loss optimization according to claim 2, characterized in that: The characteristic variables include maximum value, minimum value, average slope, and number of zero crossing points. The classification model adopts random forest model, support vector machine model or K nearest neighbor algorithm model.

4. The method for optimizing magnetic core parameters based on magnetic energy transmission and loss optimization according to claim 1, characterized in that: In the step 2, a linear temperature correction coefficient is added to the Steinmetz equation to fit the core loss under different materials, frequencies and magnetic flux densities, and the mean square error, absolute error and goodness of fit are used as equation accuracy evaluation indicators to perform polynomial temperature correction on the Steinmetz equation, and the corrected core loss result is obtained by parameter fitting and waveform extraction of the data; The obtained polynomial temperature correction results are analyzed with the standard equation and the linear correction results to obtain the temperature correction of the Steinmetz equation. The set temperature only affects the material coefficient, and does not affect the frequency correction coefficient and the maximum magnetic flux influence coefficient.

5. The method for optimizing magnetic core parameters based on magnetic energy transmission and loss optimization according to claim 1, characterized in that: In the step three, the characteristic variables affecting the core loss and their response processes are subjected to variance analysis through single-factor analysis and two-factor interaction analysis methods, and the Bonferroni correction method and characteristic contribution calculation are used to minimize the core loss.

6. A method for optimizing magnetic core parameters based on magnetic energy transmission and loss optimization according to claim 1, characterized in that: In step 4, characteristic variables that have a greater impact on core loss are extracted from the experimental data as inputs to the subsequent model. Multiple machine learning models including BP neural network, GBDT model, decision tree regression, SVR and LSTM are used as prediction models to establish a mapping relationship between data set indicators and core loss, and to construct a core loss prediction model. Through model comparison and accuracy evaluation, MSE and R are taken. 2 As the model accuracy evaluation index, the prediction neural network core loss prediction model with the best prediction effect is selected to predict the actual core loss.

7. A method for optimizing magnetic core parameters based on magnetic energy transmission and loss optimization according to claim 5, characterized in that: In the step 4, the constructed prediction neural network core loss prediction model is used as the objective function, and a regression model for core loss is constructed according to temperature, frequency, waveform, magnetic flux density peak value and core material. On the basis of the regression model, a fitness function is constructed, and the minimum core loss and the operating point with the maximum transmission magnetic energy are obtained by genetic algorithm optimization calculation. The specific fitness function is: Among them, E represents the transmitted magnetic energy, P represents the core loss, L represents the input-loss ratio, C represents the reward item, and W represents the penalty item.

8. A magnetic core parameter optimization system based on magnetic energy transmission and loss optimization, characterized in that: It includes classification module, correction module, analysis module and optimization module; The classification module is used to extract the characteristic parameters of the magnetic core, combine the working thermometer frequency, establish the excitation waveform classification model, and classify the excitation waveform; The correction module introduces a temperature correction coefficient into the Steinmetz equation based on the classification result of the excitation waveform to obtain a corrected Steinmetz equation for predicting the core loss of different excitation waveforms; The analysis module is used to analyze the independent or synergistic effects of temperature, excitation waveform and core material on core loss; The optimization module is used to construct a BP neural network using frequency, waveform, peak value of magnetic flux density, temperature and core material as features, construct the maximum transmitted magnetic energy and minimum core loss into a fitness function, and use a genetic algorithm to optimize the input features to obtain optimized core parameters.

9. A computer device, characterized in that: It includes a processor and a memory, the memory is used to store a computer executable program, the processor reads part or all of the computer executable program from the memory and executes it, and when the processor executes part or all of the computer executable program, it can implement a magnetic core parameter optimization method based on magnetic energy transmission and loss optimization as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: A computer program is stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement a magnetic core parameter optimization method based on magnetic energy transmission and loss optimization as described in any one of claims 1 to 7.

Images