A Method for Optimizing Resonant Cavity Parameters in LLC Circuits for Energy Storage Bidirectional DC-DC Converters

Abstract

This invention relates to a method for optimizing the resonant cavity parameters of an LLC circuit for a bidirectional DC-DC converter for energy storage. The method includes: selecting N resonant inductance values ​​and calculating the corresponding magnetizing inductance and resonant capacitance values; calculating the corresponding resonant inductance current, magnetizing inductance current, and resonant capacitor voltage; designing the resonant inductance, magnetizing inductance, and resonant capacitor separately, and calculating feasibility labels, costs, and losses; integrating the inductance and capacitor design data into an input dataset; using the comprehensive index calculated using cost and losses as the first output dataset; using the feasibility labels as the second output dataset; and calculating the maximum mutual information coefficient between each input data feature and each output dataset to form the final input dataset and the final output dataset. These datasets are then input into an SVM for training. Randomly generated input data is then input into the SVM for output. After eliminating infeasible solutions, the optimal design scheme is selected. This method is beneficial for improving the performance and efficiency of resonant cavity parameter optimization design.

Description

Technical Field

[0001] This invention relates to the field of power electronics technology, and more specifically to a method for optimizing the design of resonant cavity parameters for an LLC circuit of a bidirectional DC-DC converter for energy storage. Background Technology

[0002] Bidirectional LLC DC-DC resonant converters, as an important topology of resonant switching technology in DC / DC converters, have advantages such as high efficiency, low loss, zero-voltage switching, zero-current switching, and high integration, and are widely used. The parameters of the internal resonant circuit directly affect the converter's conversion efficiency and performance. However, the resonant circuit contains many magnetic components and elements, resulting in numerous design parameters. Currently, the hardware parameter design of resonant circuits mainly relies on engineering experience to calculate parameters and perform circuit model calculations. While the engineering experience method has low design costs, it is difficult to comprehensively consider cost, efficiency, and performance factors for systems with many parameters and extensive coupling relationships, leading to low design efficiency and difficulty in obtaining optimal parameter designs. Hardware parameter design can also be calculated using algorithmic models, which are highly efficient, but the design effect is limited by the accuracy of the model itself, and the model accuracy is largely determined by the training data. If the quality of the dataset used is low, the performance of the design scheme output by the trained model will not be good. Therefore, both design methods have their advantages and disadvantages, and using only one method for parameter design has significant drawbacks. There are already solutions that combine engineering experience with algorithm models for hardware parameter design. The specific method is to use engineering experience data to provide a dataset for training the algorithm model, and the subsequent parameter design is carried out by the trained algorithm model. However, this method needs to ensure the quality of the provided dataset. How to improve the quality of the dataset generated by engineering experience for training the algorithm model is an urgent problem to be solved.

[0003] Patent CN112069758A discloses a method for optimizing soft-start parameters of LLC converters based on a simplified numerical model, including sampling the input voltage of the LLC converter and setting the maximum overshoot current I during startup according to the current stress of the switching devices. r_peak Set the minimum soft-switching current I during startup based on soft-switching conditions. r_zvs Using this as the initial condition for parameter iteration, and based on the simplified circuit model under four modes, the duration of each mode at startup, the instantaneous resonant capacitor voltage, and the resonant current are obtained through iteration to determine the adaptive startup frequency. Next, the system stability is judged based on the conditions for LLC converter oscillation. If the oscillation condition is met, the startup frequency is obtained; if the oscillation condition is not met, I is gradually increased. r_peak At the same time, reduce I r_zvsThe iterative calculation is repeated until the oscillation condition is met, the starting frequency is obtained, and finally the initial starting frequency and initial duty cycle are output, and soft start is performed using these parameters. Patent CN115495936A discloses a parameter design method for LLC converters based on adaptive polynomial approximation, relating to the field of LLC resonant converter design technology. First, the parameter space of the resonant cavity operating only in the PO mode is determined by numerical analysis. Then, the final parameter space is determined by setting constraints. A surrogate model of the optimization objective and parameter space is established based on the adaptive polynomial approximation method, and a surrogate model of the parameter space is also established using the adaptive polynomial approximation method. The NSGA-II algorithm is used for multi-objective optimization to design the optimal parameters. This invention's method is effective and reasonable, with high efficiency, achieving an efficiency of 94.2% under worst-case conditions and a peak efficiency of 94.93%. While the above patents disclose the optimization design of LLC converter-related parameters through specific algorithm models, none address how to improve the quality of the dataset used for training the algorithm model, thereby further improving the performance and efficiency of resonant cavity parameter optimization design through algorithm models. Summary of the Invention

[0004] The purpose of this invention is to provide a method for optimizing the design of resonant cavity parameters for an LLC circuit of a bidirectional DC-DC converter for energy storage. This method is beneficial to improving the performance and efficiency of resonant cavity parameter optimization design.

[0005] To achieve the above objectives, the technical solution adopted by this invention is: a method for optimizing the resonant cavity parameters of an LLC circuit for energy storage bidirectional DC-DC converters, comprising:

[0006] N inductance values ​​are selected as resonant inductance values, and the corresponding magnetizing inductance and resonant capacitance values ​​are calculated based on the relationship between the resonant inductance, magnetizing inductance, and resonant capacitance, resulting in N sets of data consisting of the resonant inductance values ​​and the corresponding magnetizing inductance and resonant capacitance values; the corresponding resonant inductance current, magnetizing inductance current, and resonant capacitance voltage data are calculated sequentially using the N sets of data.

[0007] Based on the obtained resonant inductance value, magnetizing inductance value, simulated resonant inductance current and magnetizing inductance current data, and combined with the core data in the core database, the resonant inductance and magnetizing inductance are designed, and feasibility labels, costs and losses are calculated; based on the obtained resonant capacitance value and simulated resonant capacitance voltage, and combined with capacitance data from different manufacturers, the resonant capacitor is designed, and feasibility labels, costs and losses are calculated.

[0008] The design data related to inductors and capacitors are integrated into an initial input dataset. The comprehensive index data calculated using cost and loss data is used as the first output dataset, and the feasibility label is used as the second output dataset. The maximum mutual information coefficient between each feature of the input dataset and the first and second output datasets is calculated. The maximum mutual information coefficient is used to select the input data features that are highly correlated with the first and second output datasets, forming two new input datasets respectively.

[0009] The two new input datasets are integrated to obtain the final input dataset. The first output dataset and the second output dataset are integrated to obtain the final output dataset. The final input data and the final output data are then fed into an SVM for training. A large number of input data with the same format as the final input data are randomly generated and fed into the SVM for output. After eliminating infeasible solutions, the optimal design solution is selected based on comprehensive indicators.

[0010] Furthermore, N inductance values ​​are selected as resonant inductance values, and the corresponding magnetizing inductance and resonant capacitance values ​​are calculated based on the relationship between the resonant inductance, magnetizing inductance, and resonant capacitance, resulting in N sets of data consisting of the resonant inductance values ​​and their corresponding magnetizing inductance and resonant capacitance values. The corresponding resonant inductance current, magnetizing inductance current, and resonant capacitance voltage data are then calculated sequentially using these N sets of data. The specific method is as follows:

[0011] N inductance values ​​are selected from the inductance database as the resonant inductance values. The corresponding formulas for calculating the resonant capacitance and magnetizing inductance values ​​are as follows:

[0012]

[0013] Lm=k×Lr (2)

[0014] Where Cr is the resonant capacitance value corresponding to the resonant inductance value, fr is the resonant frequency, Lr is the resonant inductance value, Lm is the magnetizing inductance value corresponding to the resonant inductance value, and k is the proportionality coefficient between the magnetizing inductance and the resonant inductance.

[0015] After calculating the N resonant inductance values ​​and the corresponding N resonant capacitance values ​​and N magnetizing inductance values, these are combined into N sets of data consisting of the resonant inductance values ​​and their corresponding magnetizing inductance and resonant capacitance values. These N sets of data are then sequentially input into the simulation model for simulation. The maximum current I1max and the effective current I1 flowing through the resonant inductors and capacitors are calculated. The switching frequency f corresponding to the rated voltage output from the low-voltage DC side to the high-voltage DC side is then calculated. s The maximum voltage Ucmax that the resonant capacitor withstands, and the maximum and effective current I2max and current I2 flowing through the magnetizing inductor.

[0016] Furthermore, based on the obtained resonant inductance value, magnetizing inductance value, simulated resonant inductance current and magnetizing inductance current data, and combined with core data from the core database, the resonant inductor and magnetizing inductor are designed, and feasibility tags, costs, and losses are calculated. The specific method is as follows:

[0017] Select U magnetic core data from the magnetic core database and design either a resonant inductor or a magnetizing inductor to obtain different design schemes for the resonant inductor or the magnetizing inductor:

[0018] First, the winding conductor is designed. The formula for calculating the diameter of the enameled wire conductor of the inductor winding is as follows:

[0019]

[0020] Where, d w I is the diameter of the enameled wire conductor, I3 is the effective value of the inductor current, and J is the diameter of the enameled wire conductor. max h is the effective value of the maximum current density. ins This refers to the thickness of the insulation layer of the winding conductor;

[0021] After calculating the diameter of the enameled wire conductor, the maximum number of turns of the magnetic core is calculated by combining the magnetic core data in the magnetic core database. The calculation formula is as follows:

[0022]

[0023] Where, N max d is the maximum number of turns of the magnetic core. core h is the inner diameter of the magnetic core. ins d represents the thickness of the insulation layer of the winding conductor. w S is the diameter of the enameled wire. wire The spacing between the winding conductors;

[0024] After calculating the maximum number of turns in the magnetic core, the number of turns in the inductor winding, the cross-sectional area of ​​the winding, and the inductance window fill factor are calculated using the following formulas:

[0025]

[0026] Where, N Lr L3 is the number of turns in the inductor winding, I3max is the inductance value, ΔB is the flux swing, and A is the maximum current flowing through the inductor. e The effective magnetic flux cross-sectional area;

[0027]

[0028] Where A1 is the cross-sectional area of ​​the winding, I3 is the effective value of the current flowing through the inductor, and J max This represents the effective value of the maximum current density.

[0029]

[0030] Among them, K L A is the inductance window fill factor. w The area of ​​the magnetic core window;

[0031] After calculating the inductor's design parameters, the feasibility label is calculated based on these parameters. The calculation formula is as follows:

[0032]

[0033] Where Label is the feasibility label, and N Lr N represents the number of turns in the inductor winding. max K is the maximum number of turns of the magnetic core. L The fill factor for the inductance window;

[0034] After obtaining the feasibility label, the inductor cost and loss of different design schemes are calculated;

[0035] The cost of inductors is calculated by taking the average cost of producing the same type of iron core from different manufacturers. The formula is as follows:

[0036]

[0037] Among them, F L For the cost of the inductor, F1, F2, ..., F n The cost of producing inductors for different manufacturers, where n is the total number of different manufacturers;

[0038] The inductor loss is calculated using the average value of inductor losses obtained from calculators from different manufacturers. The calculation formula is as follows:

[0039]

[0040] Among them, P L For inductance losses, P1, P2, ..., P n The inductance loss is calculated using calculators from different manufacturers, where n is the total number of different manufacturers.

[0041] Furthermore, based on the obtained resonant capacitance value and the simulated resonant capacitance voltage, and combined with capacitance data from different manufacturers, a resonant capacitor is designed, and feasibility studies, cost, and loss calculations are performed. The specific method is as follows:

[0042] Select T capacitor data points from the capacitor database and design resonant capacitors for each, to obtain different design schemes for the resonant capacitors:

[0043] The formula for calculating the number of capacitors connected in series with different capacities is as follows:

[0044]

[0045] Where n2 is the number of capacitors that need to be connected in series, C i C represents the capacitance value corresponding to different capacitors. r Here is the capacitance value of the resonant capacitor, and round() is the rounding function;

[0046] After calculating the number of capacitors connected in series for different parameters, the feasibility label is calculated using the following formula:

[0047]

[0048] Where n is the number of capacitors that need to be connected in series, N cmax To maximize the number of stacked units, U ci U is the voltage value that a single capacitor can withstand. cmax The maximum voltage that the total capacitance can withstand; Label indicates feasibility.

[0049] After obtaining the feasibility label, the capacitor cost is calculated using the following formula:

[0050] F c =F ci ×n i (13)

[0051] Among them, F c F represents the total price required for different capacitors. ci The price required for individual capacitors of different capacitance values, n i The required quantity of capacitors with different capacitance values;

[0052] After calculating the cost of different capacitors, the equivalent series internal resistance and losses are calculated using the following formulas:

[0053]

[0054] Where ESR is the equivalent series resistance of the capacitor, tanδ is the loss angle of the capacitor, and f s The switching frequency corresponding to the rated voltage output from the low-voltage DC side to the high-voltage DC side; C is the capacitance value of the capacitor.

[0055] P C =I C 2 ×ESR (15)

[0056] Among them, I C P is the effective value of the current flowing through the capacitor. C This refers to the capacitor's losses.

[0057] Furthermore, the design data related to inductors and capacitors are integrated into an initial input dataset. The comprehensive index data calculated using cost and loss data is used as the first output dataset, and the feasibility label is used as the second output dataset. The maximum mutual information coefficient between each feature of the input dataset and the first and second output datasets is calculated. Using the maximum mutual information coefficient, input data features with high correlation to the first and second output datasets are selected, forming two new input datasets. The specific method is as follows:

[0058] Design data related to inductors and capacitors, including resonant inductance, resonant capacitance, magnetizing inductance, core inner diameter, maximum effective current density, effective magnetic flux cross-sectional area of ​​the core, core window area, and actual capacitance value, are selected and integrated into an initial input dataset.

[0059] After obtaining the initial input dataset, a comprehensive index is calculated using the total cost and total loss data, and this index is used as the first output dataset. The feasibility labels corresponding to each input data point are used as the second output dataset. The formulas for calculating the total cost, total loss, and comprehensive index are as follows:

[0060] F i总 =F iLr +F iLm +F iCr (16)

[0061] P i总 =P iLr +P iLm +P iCr (17)

[0062]

[0063] Among them, t i F is a comprehensive index of cost and loss for the i-th design scheme. i总 Let F be the total design cost of the i-th design scheme. max P represents the maximum design cost among all design options. i总 Let P be the total loss of the i-th design scheme. max F represents the maximum loss among all design schemes. iLr The total cost of the resonant inductor designed for the i-th design scheme, F iLm The total cost of the magnetizing inductor designed for the i-th design scheme, F iCr The total cost of the resonant capacitor designed for the i-th design scheme, P iLr P is the total loss of the resonant inductor. iLm For the total loss of the magnetizing inductor, P iCr This represents the total loss of the resonant capacitor.

[0064] After obtaining the first output dataset and the second output dataset, calculate the maximum mutual information coefficient between each feature of the input dataset and the first output dataset and the second output dataset;

[0065] After calculating the maximum mutual information coefficient between each feature and the output dataset, the features in the input dataset are sorted according to the maximum mutual information coefficient. The features with the largest maximum mutual information coefficient are selected as the features of the new input dataset according to the required data dimensions.

[0066] Furthermore, the calculation process for the maximum mutual information coefficient between features in the input dataset and a single output dataset is as follows:

[0067] (1) Select the data of one feature in the input dataset and the output dataset in turn to form a two-dimensional dataset, and draw it in the form of coordinate points in the two-dimensional coordinate plane;

[0068] (2) Calculate the dividing lines of the region based on the precision traversal. The calculation formulas for the dividing lines of the x-axis and y-axis are as follows:

[0069] x 分割线 =k×d x (19)

[0070] y 分割线 =k×d y (20)

[0071] Where, x 分割线 The x-coordinate and y-coordinate of the current x-axis dividing line are... 分割线 Let y be the y-coordinate corresponding to the current y-axis dividing line, k be the current traversal count, and d be the y-coordinate of the dividing line. x The step size d used to traverse the x-axis y The step size used for traversing the y-axis;

[0072] (3) After calculating the dividing line, the dividing line divides the area into four regions: "top left", "bottom left", "top right", and "bottom right". At this time, the coordinate points in each region are counted and the probability of the coordinate point in each region is calculated. The calculation formula is as follows:

[0073]

[0074] Where p is the probability of a coordinate point appearing in this region, and n i n represents the number of coordinate points within this region. 总 This represents the total number of coordinate points.

[0075] (4) After calculating the probability of the coordinate point in each region, the mutual information value between the feature and the output dataset under the current traversal is calculated according to the mutual information calculation formula, which is as follows:

[0076]

[0077] Where I(x, y) is the mutual information value of the two variables x and y, and p(X = x i Y = y j ) is in (x i y j The joint probability of the region ) p(x=x i ) is X at x i The probability of the region, p(Y=y j ) is Y at y j The probability of the region; when x i When the subscript i = 1, it represents the left side; when i = 2, it represents the right side; when y j When the subscript j=1, it represents the top side; when j=2, it represents the bottom side.

[0078] (5) After iterating through and changing the dividing line and calculating the mutual information value between each feature and the output dataset, stop the iteration when the dividing line meets the iteration condition and calculate the maximum mutual information coefficient. The calculation formula is as follows:

[0079]

[0080] Where mic(x, y) is the maximum mutual information coefficient between the feature and the output dataset, max() is the maximum value function, I(x, y) is the mutual information coefficient between the feature and the output dataset, a is the number of regions divided in the x direction, and b is the number of regions divided in the y direction.

[0081] Furthermore, the new input dataset has two rows: one is the dataset after feature filtering by the input dataset and the first output dataset, and the other is the dataset after feature filtering by the input dataset and the second output dataset. These two rows of data are combined in order of data number to obtain the final input dataset; the first output dataset and the second output dataset are combined in order of data number to obtain the final output dataset.

[0082] Furthermore, the final input data and final output data are input into the SVM for model training. After the model training is completed, M input data are randomly generated according to the data format of the final input data, where each input data represents a design scheme. These randomly generated input data are input into the SVM to output the corresponding comprehensive index and the feasibility label of each design scheme. Input data with a feasibility label of 0 are removed, i.e., the design schemes. The comprehensive indexes corresponding to the remaining input data are sorted, and the input data with the largest comprehensive index, i.e., the design scheme, is selected as the optimized optimal design scheme.

[0083] Compared with existing technologies, this invention has the following advantages: This method, based on combining engineering experience and algorithm model design, incorporates the maximum mutual information coefficient to perform feature filtering on input data obtained using engineering experience, thereby providing a higher-quality dataset for intelligent algorithm training. This allows the trained algorithm model to perform parameter optimization design more efficiently. The parameters optimized by this method can reduce the resonant cavity power consumption of the LLC circuit in the energy storage bidirectional DC-DC converter, improve the power conversion efficiency of the energy storage bidirectional DC-DC converter, and reduce hardware costs. Attached Figure Description

[0084] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0085] Figure 1 This is a flowchart illustrating the implementation of the resonant cavity parameter optimization design method for an energy storage bidirectional DC-DC converter LLC circuit provided in an embodiment of the present invention.

[0086] Figure 2 This is a circuit diagram of the bidirectional LLC resonant converter in an embodiment of the present invention;

[0087] Figure 3 This is a flowchart illustrating the process of obtaining test data in an embodiment of the present invention;

[0088] Figure 4 This is a flowchart illustrating the inductor design in an embodiment of the present invention;

[0089] Figure 5 This is a flowchart illustrating the capacitor design in an embodiment of the present invention;

[0090] Figure 6 This is a flowchart illustrating the process of filtering initial dataset features in an embodiment of the present invention;

[0091] Figure 7 This is a schematic diagram of the initial input dataset, the first output dataset, and the second output dataset in an embodiment of the present invention;

[0092] Figure 8 This is a schematic diagram of the region division for calculating the maximum mutual information coefficient in an embodiment of the present invention;

[0093] Figure 9 This is a flowchart illustrating the use of SVM to select the optimal design scheme in an embodiment of the present invention.

[0094] Figure 10This is a schematic diagram of the final input dataset and the final output dataset in an embodiment of the present invention. Detailed Implementation

[0095] The core of this invention is to provide a resonant cavity parameter optimization design method for LLC circuits in energy storage bidirectional DC-DC converters. This method generates an initial input dataset and two output datasets for the resonant cavity design using engineering experience. The maximum mutual information coefficient is used to filter features in the initial input dataset that are related to the two output datasets, forming two new input datasets. These two new input datasets are then integrated into a final input dataset, and the two output datasets are integrated into a final output dataset. The final input and output datasets are fed into an SVM for training, and the trained SVM is used to optimally select from a large number of randomly generated design schemes. This design method, combining engineering experience and intelligent algorithm design, introduces the maximum mutual information coefficient to filter features from the input data obtained using engineering experience, providing a higher-quality dataset for intelligent algorithm training. The trained algorithm model can perform parameter optimization design more efficiently. The optimized parameters obtained through this method can reduce the resonant cavity power consumption of the LLC circuit in the energy storage bidirectional DC-DC converter, improve the power conversion efficiency of the energy storage bidirectional DC-DC converter, and reduce hardware costs.

[0096] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0097] Existing resonant converter cavity design methods mainly fall into two categories: engineering experience-based design and algorithmic design. Engineering experience-based design relies on engineers' expertise to calculate relevant parameters of the resonant cavity and select appropriate magnetic components and capacitors based on these calculations. However, this method is inefficient when dealing with numerous parameters and high complexity. Algorithmic design, on the other hand, uses algorithms and computers to calculate relevant data, enabling more efficient handling of complex optimization designs with many parameters. However, algorithmic design depends on the accuracy of the model; low model accuracy leads to poor resonant cavity performance. Currently, there are approaches that combine these two methods: using engineering experience to provide a training dataset for the model, and then using the trained model for subsequent design. However, the quality of training datasets generated using engineering experience is not always high, and the accuracy of the trained model cannot be guaranteed.

[0098] Therefore, this invention provides a method for optimizing the resonant cavity parameters of an LLC circuit for energy storage bidirectional DC-DC converters. By initially setting N resonant inductors, resonant capacitors, and magnetizing inductors and conducting simulations, relevant data for the subsequent design of resonant inductors, resonant capacitors, and magnetizing inductors are obtained. Using the previously obtained inductor test data and core datasheets, resonant inductors and magnetizing inductors are designed, and feasibility labels and cost / loss calculations are performed, providing a portion of the data for the initial input dataset, the first output dataset, and the second output dataset. Similarly, using the previously obtained capacitor test data and capacitor datasheets, resonant capacitors are designed, and feasibility labels and cost / loss calculations are performed, providing another portion of the data for the initial input dataset, the first output dataset, and the second output dataset. After the design of the resonant inductors, resonant capacitors, and magnetizing inductors is completed, a comprehensive index of cost and loss is calculated. Commonly used data in the design are selected as the initial dataset, the comprehensive index as the first output dataset, and the feasibility labels as the second output dataset. Based on the maximum mutual information coefficient between the initial dataset and the first and second output datasets, features with high correlation between the initial input dataset and the output datasets are selected, forming two new input datasets. Finally, the two new input datasets obtained in the previous step are integrated into the final input dataset, and the two output datasets are integrated into the final dataset. The SVM model is trained using the final input and output datasets. After training, a large number of input data (design schemes) with the same format as the final input data are randomly generated and fed into the SVM model for comprehensive index and feasibility label output. After eliminating infeasible design schemes, the optimal result of the feasible design scheme is selected based on the comprehensive index of cost loss. This method, combining engineering experience and intelligent algorithm design, incorporates the maximum mutual information coefficient to perform feature filtering on the input data obtained from engineering experience, thereby providing a higher quality dataset for intelligent algorithm training. This allows the trained algorithm model to perform parameter optimization design more efficiently. The parameters optimized by this method can reduce the resonant cavity power consumption of the LLC circuit in the energy storage bidirectional DC-DC converter, improve the power conversion efficiency of the energy storage bidirectional DC-DC converter, and reduce hardware costs.

[0099] Figure 1 This is a flowchart illustrating the method implementation in this embodiment; Figure 2 This is a schematic diagram of a bidirectional LLC resonant converter circuit. Figure 3 Flowchart for obtaining test data; Figure 4 Flowchart for inductor design; Figure 5 Flowchart for capacitor design; Figure 6 Flowchart for filtering features in the initial dataset; Figure 7 This is a schematic diagram of the initial input dataset, the first output dataset, and the second output dataset; Figure 8A schematic diagram of the region division process for calculating the maximum mutual information coefficient; Figure 9 Flowchart for using SVM to select the optimal design solution; Figure 10 This is a schematic diagram of the final input dataset and the final output dataset.

[0100] like Figure 1 As shown, this embodiment provides a method for optimizing the resonant cavity parameters of an LLC circuit for energy storage bidirectional DC-DC converters, including:

[0101] S1. Select N inductance values ​​as resonant inductance values, and calculate the corresponding magnetizing inductance and resonant capacitance values ​​based on the relationship between the resonant inductance, magnetizing inductance, and resonant capacitance, obtaining N sets of data consisting of the resonant inductance values ​​and their corresponding magnetizing inductance and resonant capacitance values; input the N sets of data into the simulation model sequentially to calculate the corresponding resonant inductance current, magnetizing inductance current, and resonant capacitance voltage data. The specific implementation method is as follows.

[0102] N inductance values ​​are selected from the inductance database as the resonant inductance values. The corresponding formulas for calculating the resonant capacitance and magnetizing inductance values ​​are as follows:

[0103]

[0104] Lm=k×Lr (2)

[0105] Where Cr is the resonant capacitance value corresponding to the resonant inductance value, fr is the resonant frequency (set according to requirements), Lr is the resonant inductance value, Lm is the excitation inductance value corresponding to the resonant inductance value, and k is the proportionality coefficient between the excitation inductance and the resonant inductance (generally 3-7, here 5 is selected as the proportionality coefficient).

[0106] After calculating the N resonant inductance values ​​and the corresponding N resonant capacitance values ​​and N magnetizing inductance values, these are combined into N sets of data consisting of the resonant inductance values ​​and their corresponding magnetizing inductance and resonant capacitance values. These N sets of data are then sequentially input into the simulation model for simulation. The maximum current I1max and the effective current I1 flowing through the resonant inductors and capacitors are calculated. The switching frequency f corresponding to the rated voltage output from the low-voltage DC side to the high-voltage DC side is then calculated. s The maximum voltage Ucmax that the resonant capacitor withstands, and the maximum and effective current I2max and current I2 flowing through the magnetizing inductor.

[0107] This step uses the selected N resonant inductor data to generate the values ​​of resonant capacitor and magnetizing inductor, and puts them into the simulation model to obtain data such as resonant inductor current, magnetizing inductor current, and resonant capacitor voltage. The purpose is to use the relevant data obtained in this step to design the resonant inductor, resonant capacitor and magnetizing inductor in the next two steps.

[0108] S2. Based on the obtained resonant inductance value, magnetizing inductance value, simulated resonant inductance current, and magnetizing inductance current data, and combined with core data from the core database, design the resonant inductor and magnetizing inductor, and perform feasibility labeling, cost, and loss calculations; based on the obtained resonant capacitance value and simulated resonant capacitance voltage, and combined with capacitance data from different manufacturers, design the resonant capacitor, and perform feasibility labeling, cost, and loss calculations. The specific implementation method is as follows.

[0109] Select U magnetic core data points from the magnetic core database and design either a resonant inductor or a magnetizing inductor to obtain different design schemes for either. It should be noted that both resonant and magnetizing inductors can be designed using the following methods.

[0110] First, the winding conductor is designed using the data obtained above. The formula for calculating the diameter of the enameled wire conductor of the inductor winding is as follows:

[0111]

[0112] Where, d w I is the diameter of the enameled wire conductor, I3 is the effective value of the inductor current, and J is the diameter of the enameled wire conductor. max h is the maximum effective current density (set according to requirements). ins This refers to the thickness of the insulation layer of the winding conductor.

[0113] After calculating the diameter of the enameled wire conductor, the maximum number of turns of the magnetic core is calculated by combining the magnetic core data in the magnetic core database. The calculation formula is as follows:

[0114]

[0115] Where, N max d is the maximum number of turns of the magnetic core. core h is the inner diameter of the magnetic core. ins d represents the thickness of the insulation layer of the winding conductor. w S is the diameter of the enameled wire. wire The spacing between the winding conductors (the conductor spacing is set according to requirements).

[0116] After calculating the maximum number of turns in the magnetic core, the number of turns in the inductor winding, the cross-sectional area of ​​the winding, and the inductance window fill factor are calculated using the following formulas:

[0117]

[0118] Where, N Lr L3 is the number of turns in the inductor winding, I3max is the inductance value, ΔB is the flux swing, and A is the maximum current flowing through the inductor. e It is the effective magnetic flux cross-sectional area.

[0119]

[0120] Where A1 is the cross-sectional area of ​​the winding, I3 is the effective value of the current flowing through the inductor, and J max This represents the effective value of the maximum current density.

[0121]

[0122] Among them, K L A is the inductance window fill factor. w Let be the area of ​​the magnetic core window.

[0123] After calculating the inductor's design parameters, the feasibility label is calculated based on these parameters. The calculation formula is as follows:

[0124]

[0125] Where Label is the feasibility label, and N Lr N represents the number of turns in the inductor winding. max K is the maximum number of turns of the magnetic core. L is the fill factor for the inductance window.

[0126] The judgment logic of the above formula is as follows: when the required number of inductor winding turns exceeds the maximum number of winding turns that the magnetic core can withstand, the design scheme cannot be realized; the window fill factor is generally empirically located in the range of 0.3-0.4. If the window fill factor exceeds the range of 0.3-0.4, the design scheme cannot be realized; for design schemes that cannot be realized, a label with a value of 0 is added, and for design schemes that can be realized, a label with a value of 1 is added to distinguish them.

[0127] After obtaining the feasibility label, the inductor cost and loss of different design schemes are calculated;

[0128] Since the same model of iron core produced by different manufacturers varies slightly, the average cost of producing the same model of iron core from different manufacturers is used as the cost of the inductor. The calculation formula is as follows:

[0129]

[0130] Among them, F L For the cost of the inductor, F1, F2, ..., F n The cost of producing inductors for different manufacturers, where n is the total number of different manufacturers;

[0131] After calculating the cost, we begin calculating the inductor loss. For a resonant converter, the total inductor loss includes the coil DC loss, coil AC loss, and core loss. Among these, the coil AC loss and core loss are relatively difficult to calculate. Different manufacturers provide their own core inductance loss calculators for ease of design; simply input the inductance value and other data, and the calculator will output an estimated inductance loss. Therefore, we use the average of the inductance losses calculated by the calculators from different manufacturers as the inductance loss. The calculation formula is as follows:

[0132]

[0133] Among them, P L For inductance losses, P1, P2, ..., P n The inductance loss is calculated using calculators from different manufacturers, where n is the total number of different manufacturers.

[0134] Based on the obtained resonant capacitance value and the simulated resonant capacitance voltage, a resonant capacitor is designed using capacitance data from different manufacturers. Feasibility studies, cost calculations, and loss calculations are then performed. The specific method is as follows:

[0135] Select T capacitor data points from the capacitor database and design resonant capacitors for each, to obtain different design schemes for the resonant capacitors.

[0136] The formula for calculating the number of capacitors connected in series with different capacities is as follows:

[0137]

[0138] Where n2 is the number of capacitors that need to be connected in series, C i C represents the capacitance value corresponding to different capacitors. r Here is the capacitance value of the resonant capacitor, and round() is the rounding function.

[0139] After calculating the number of capacitors connected in series for different parameters, the feasibility label is calculated using the following formula:

[0140]

[0141] Where n is the number of capacitors that need to be connected in series, N cmax U is the maximum number of stacks (usually set to 5). ci U is the voltage value that a single capacitor can withstand. cmax The maximum voltage that the total capacitance can withstand; Label indicates feasibility.

[0142] After obtaining the feasibility label, the capacitor cost is calculated using the following formula:

[0143] F c =Fci ×n i (13)

[0144] Among them, F c Fc represents the total price required for different capacitors. i The price required for individual capacitors of different capacitance values, n i The required quantity of capacitors with different capacitance values.

[0145] After calculating the cost of different capacitors, the equivalent series internal resistance and losses are calculated using the following formulas:

[0146]

[0147] Where ESR is the equivalent series resistance of the capacitor, tanδ is the loss angle of the capacitor, and f s The switching frequency corresponding to the rated voltage output from the low-voltage DC side to the high-voltage DC side is given by C, where C is the capacitance value of the capacitor.

[0148] P C =I C 2 ×ESR (15)

[0149] Among them, I C P is the effective value of the current flowing through the capacitor. C This refers to the capacitor's losses.

[0150] Building upon the previous step, this step aims to design the resonant inductor and the magnetizing inductor using the inductance test data obtained earlier, and then perform feasibility labeling and cost and power consumption calculations to provide a portion of the dataset for the initial input dataset, the first output dataset, and the second output dataset.

[0151] This step also aims to use the previously obtained capacitance test data to design the resonant capacitor, and then perform feasibility labeling and cost and power consumption calculations to provide another part of the dataset for the initial input dataset, the first output dataset, and the second output dataset.

[0152] S3. Integrate the design data related to inductors and capacitors into an initial input dataset. Use the comprehensive index data calculated using cost and loss data as the first output dataset, and the feasibility label as the second output dataset. Calculate the maximum mutual information coefficient between each feature of the input dataset and the first and second output datasets. Use the maximum mutual information coefficient to filter out the input data features that are highly correlated with the first and second output datasets, forming two new input datasets. The specific implementation method is as follows.

[0153] Select relevant design data for inductors and capacitors, including: resonant inductance value, resonant capacitance value, magnetizing inductance value, core inner diameter, maximum effective current density, effective magnetic flux cross-sectional area of ​​the core, core window area, actual capacitance value, etc., and integrate them into the initial input dataset.

[0154] After obtaining the initial input dataset, a comprehensive index is calculated using the total cost and total loss data, and this index is used as the first output dataset. The feasibility labels corresponding to each input data point are used as the second output dataset. The formulas for calculating the total cost, total loss, and comprehensive index are as follows:

[0155] F i总 =F iLr +F iLm +F iCr (16)

[0156] P i总 =P iLr +P iLm +P iCr (17)

[0157]

[0158] Among them, t i F is a comprehensive index of cost and loss for the i-th design scheme. i总 Let F be the total design cost of the i-th design scheme. max P represents the maximum design cost among all design options. i总 Let P be the total loss of the i-th design scheme. max F represents the maximum loss among all design schemes. iLr The total cost of the resonant inductor designed for the i-th design scheme, F iLm The total cost of the magnetizing inductor designed for the i-th design scheme, F iCr The total cost of the resonant capacitor designed for the i-th design scheme, P iLr P is the total loss of the resonant inductor. iLm For the total loss of the magnetizing inductor, P iCr This represents the total loss of the resonant capacitor.

[0159] After obtaining the first output dataset and the second output dataset, calculate the maximum mutual information coefficient between each feature of the input dataset and the first output dataset and the second output dataset.

[0160] The calculation process for the maximum mutual information coefficient between features in the input dataset and a single output dataset is as follows:

[0161] (1) Select the data of one feature in the input dataset and the output dataset in turn to form a two-dimensional dataset, and draw it in the form of coordinate points in the two-dimensional coordinate plane.

[0162] (2) Calculate the dividing lines of the region based on the precision traversal. The calculation formulas for the dividing lines of the x-axis and y-axis are as follows:

[0163] x 分割线 =k×d x (19)

[0164] y 分割线 =k×d y (20)

[0165] Where, x 分割线 The x-coordinate and y-coordinate of the current x-axis dividing line are... 分割线 Let y be the y-coordinate corresponding to the current y-axis dividing line, k be the current traversal count, and d be the y-coordinate of the dividing line. x The step size d used to traverse the x-axis y The step size used for traversing the y-axis.

[0166] (3) After calculating the dividing line, the dividing line divides the area into four regions: "top left", "bottom left", "top right", and "bottom right". At this time, X has two values: "left" and "right", and Y has two values: "top" and "bottom". Then, the coordinate points in each region are counted and the probability of the coordinate point in each region is calculated. The calculation formula is as follows:

[0167]

[0168] Where p is the probability of a coordinate point appearing in this region, and n i n represents the number of coordinate points within this region. 总 This represents the total number of coordinate points.

[0169] (4) After calculating the probability of the coordinate point in each region, the mutual information value between the feature and the output dataset under the current traversal is calculated according to the mutual information calculation formula, which is as follows:

[0170]

[0171] Where I(x, y) is the mutual information value of the two variables x and y, and p(X = x i Y = y j ) is in (x i y j The joint probability of the region ) p(x=x i ) is X at x i The probability of the region, p(Y=y j ) is Y at y j The probability of the region; when x i When the subscript i = 1, it represents the left side; when i = 2, it represents the right side; when y jWhen the subscript j=1, it represents the top side; when j=2, it represents the bottom side.

[0172] (5) After iterating through and changing the dividing line and calculating the mutual information value between each feature and the output dataset, stop the iteration when the dividing line meets the iteration condition and calculate the maximum mutual information coefficient. The calculation formula is as follows:

[0173]

[0174] Where mic(x, y) is the maximum mutual information coefficient between the feature and the output dataset, max() is the maximum value function, I(x, y) is the mutual information coefficient between the feature and the output dataset, a is the number of regions divided in the x direction, and b is the number of regions divided in the y direction.

[0175] After calculating the maximum mutual information coefficient between each feature and the output dataset, the features in the input dataset are sorted according to the maximum mutual information coefficient. The features with the largest maximum mutual information coefficient are selected as the features of the new input dataset according to the required data dimensions.

[0176] This step primarily uses the maximum mutual information coefficients between each feature of the initial input dataset and the first and second output datasets to filter out features in the initial input dataset that are highly correlated with the first and second output datasets. The aim is to provide a higher quality dataset for the next step of SVM training and improve the accuracy of the trained model.

[0177] S4. The two new input datasets are integrated to obtain the final input dataset. The first output dataset and the second output dataset are integrated to obtain the final output dataset. The final input data and the final output data are then fed into an SVM for training. Then, a large number of input data with the same format as the final input data are randomly generated and fed into the SVM for output. After eliminating infeasible solutions, the optimal design solution is selected based on comprehensive indicators. The specific implementation method is as follows.

[0178] The resulting new input dataset has two rows: one is the dataset after feature filtering by the input dataset and the first output dataset, and the other is the dataset after feature filtering by the input dataset and the second output dataset. These two rows are combined in order of data number to obtain the final input dataset. The first output dataset and the second output dataset are combined in order of data number to obtain the final output dataset.

[0179] The final input data and final output data are input into the SVM for model training. After the model training is completed, M input data are randomly generated according to the data format of the final input data, where each input data represents a design scheme. These randomly generated input data are input into the SVM to make it output the corresponding comprehensive index and the feasibility label of each design scheme. Input data with a feasibility label of 0 are removed, i.e., the design schemes. The comprehensive indexes of the remaining input data are sorted, and the input data with the largest comprehensive index is selected as the optimal design scheme after optimization.

[0180] This step primarily integrates the two new input datasets and two output datasets from the previous step to obtain the final input and output datasets. The final input and output datasets are then fed into an SVM for training. After training, input data (design schemes) in the same format as the final input data are randomly generated and fed into the SVM for comprehensive index and feasibility label output. Infeasibility schemes are filtered out using the feasibility labels, and the optimal input data from the remaining data is selected based on the comprehensive index—the optimized design scheme. The aim is to optimize the resonant cavity design through an algorithmic model, thereby reducing the design cost and power consumption.

[0181] In summary, this embodiment initially selects N resonant inductor, resonant capacitor, and magnetizing inductor values ​​and runs simulations to obtain relevant data for subsequent inductor and capacitor design. Using the previously obtained inductor test data and core datasheet, it designs the resonant inductor and magnetizing inductor, performing feasibility labeling and cost loss calculations, providing a portion of the dataset for the initial input dataset, the first output dataset, and the second output dataset. Similarly, using the previously obtained capacitor test data and capacitor datasheet, it designs the resonant capacitor, performing feasibility labeling and cost loss calculations, providing another portion of the dataset for the initial input dataset, the first output dataset, and the second output dataset. A comprehensive index is calculated, integrating the inductor and capacitor design data from the first two steps into the initial input dataset. The comprehensive index serves as the first output dataset, and the feasibility label serves as the second output dataset. Based on the maximum mutual information coefficient between the initial dataset and the first and second output datasets, features with high correlation between the initial input dataset and the output datasets are selected, forming two new input datasets to prepare for the final input dataset in the next step. Finally, the two new input datasets obtained in the previous step are integrated into the final input dataset, and the two output datasets are integrated into the final dataset. The SVM model is trained using the final input and output datasets. After training, a large number of input data (design schemes) with the same format as the final input data are randomly generated and fed into the SVM model for comprehensive index and feasibility label output. After eliminating infeasible design schemes, the optimal result of the feasible design scheme is selected based on the comprehensive index of cost loss. This design method combines engineering experience design methods and algorithm design methods, and uses the maximum mutual information coefficient to perform feature filtering on the input data obtained from engineering experience, thereby providing a higher quality dataset for intelligent algorithm training. This allows the trained algorithm model to perform parameter optimization design more efficiently. The parameters optimized by this method can reduce the resonant cavity power consumption of the LLC circuit of the energy storage bidirectional DC-DC converter, improve the power conversion efficiency of the energy storage bidirectional DC-DC converter, and reduce hardware costs.

[0182] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for optimizing the resonant cavity parameters of an LLC circuit for energy storage bidirectional DC-DC converters, characterized in that, include: N inductance values ​​are selected as resonant inductance values, and the corresponding magnetizing inductance and resonant capacitance values ​​are calculated based on the relationship between the resonant inductance, magnetizing inductance, and resonant capacitance, resulting in N sets of data consisting of the resonant inductance values ​​and the corresponding magnetizing inductance and resonant capacitance values; the corresponding resonant inductance current, magnetizing inductance current, and resonant capacitance voltage data are calculated sequentially using the N sets of data. Based on the obtained resonant inductance value, magnetizing inductance value, simulated resonant inductance current and magnetizing inductance current data, and combined with the core data in the core database, the resonant inductance and magnetizing inductance are designed, and feasibility labels, costs and losses are calculated; based on the obtained resonant capacitance value and simulated resonant capacitance voltage, and combined with capacitance data from different manufacturers, the resonant capacitor is designed, and feasibility labels, costs and losses are calculated. The design data related to inductors and capacitors are integrated into an initial input dataset. The comprehensive index data calculated using cost and loss data is used as the first output dataset, and the feasibility label is used as the second output dataset. The maximum mutual information coefficient between each feature of the input dataset and the first and second output datasets is calculated. The maximum mutual information coefficient is used to select the input data features that are highly correlated with the first and second output datasets, forming two new input datasets respectively. The two new input datasets are integrated to obtain the final input dataset. The first output dataset and the second output dataset are integrated to obtain the final output dataset. The final input data and the final output data are then fed into the SVM for training. A large number of input data with the same format as the final input data are randomly generated and fed into the SVM for output. After eliminating infeasible solutions, the optimal design solution is selected based on the comprehensive index. Based on the obtained resonant inductance value, magnetizing inductance value, simulated resonant inductance current and magnetizing inductance current data, and combined with core data from the core database, the resonant inductance and magnetizing inductance are designed, and feasibility tags, costs, and losses are calculated. The specific method is as follows: Select U magnetic core data from the magnetic core database and design either a resonant inductor or a magnetizing inductor to obtain different design schemes for the resonant inductor or the magnetizing inductor: First, the winding conductor is designed. The formula for calculating the diameter of the enameled wire conductor of the inductor winding is as follows: Where, d w I is the diameter of the enameled wire conductor, I3 is the effective value of the inductor current, and J is the diameter of the enameled wire conductor. max h is the effective value of the maximum current density. ins This refers to the thickness of the insulation layer of the winding conductor; After calculating the diameter of the enameled wire conductor, the maximum number of turns of the magnetic core is calculated by combining the magnetic core data in the magnetic core database. The calculation formula is as follows: Where, N max d is the maximum number of turns of the magnetic core. core h is the inner diameter of the magnetic core. ins d represents the thickness of the insulation layer of the winding conductor. w S is the diameter of the enameled wire. wire This refers to the spacing between the winding conductors; After calculating the maximum number of turns in the magnetic core, the number of turns in the inductor winding, the cross-sectional area of ​​the winding, and the inductance window fill factor are calculated using the following formulas: Where, N Lr L3 is the number of turns in the inductor winding, I3max is the inductance value, ΔB is the flux swing, and A is the maximum current flowing through the inductor. e The effective magnetic flux cross-sectional area; Where A1 is the cross-sectional area of ​​the winding, I3 is the effective value of the current flowing through the inductor, and J max This represents the effective value of the maximum current density. Among them, K L A is the fill factor of the inductance window. w The area of ​​the magnetic core window; After calculating the inductor's design parameters, the feasibility label is calculated based on these parameters. The calculation formula is as follows: Where Label is the feasibility label, and N Lr N represents the number of turns in the inductor winding. max K is the maximum number of turns of the magnetic core. L The fill factor for the inductance window; After obtaining the feasibility label, the inductor cost and loss of different design schemes are calculated; The cost of inductors is calculated by taking the average cost of producing the same type of iron core from different manufacturers. The formula is as follows: Among them, F L For the cost of the inductor, F1, F2, ..., F n The cost of producing inductors for different manufacturers, where n is the total number of different manufacturers; The inductor loss is calculated using the average value of inductor losses obtained from calculators from different manufacturers. The calculation formula is as follows: Among them, P L For inductance losses, P1, P2, ..., P n The inductance loss is calculated using calculators from different manufacturers, where n is the total number of different manufacturers. Based on the obtained resonant capacitance value and the simulated resonant capacitance voltage, a resonant capacitor is designed using capacitance data from different manufacturers. Feasibility studies, cost calculations, and loss calculations are then performed. The specific method is as follows: Select T capacitor data points from the capacitor database and design resonant capacitors for each, to obtain different design schemes for the resonant capacitors: The formula for calculating the number of capacitors connected in series with different capacities is as follows: Where n2 is the number of capacitors that need to be connected in series, C i C represents the capacitance value corresponding to different capacitors. r Here is the capacitance value of the resonant capacitor, and round() is the rounding function; After calculating the number of capacitors connected in series for different parameters, the feasibility label is calculated using the following formula: Where n is the number of capacitors that need to be connected in series, N cmax To maximize the number of stacked units, U ci U is the voltage value that a single capacitor can withstand. cmax The maximum voltage that the total capacitance can withstand; Label indicates feasibility. After obtaining the feasibility label, the capacitor cost is calculated using the following formula: F c =F ci ×n i (13) Among them, F c F represents the total price required for different capacitors. ci The price required for individual capacitors of different capacitance values, n i The required quantity of capacitors with different capacitance values; After calculating the cost of different capacitors, the equivalent series internal resistance and losses are calculated using the following formulas: Where ESR is the equivalent series resistance of the capacitor, tanδ is the loss angle of the capacitor, and f s The switching frequency corresponding to the rated voltage output from the low-voltage DC side to the high-voltage DC side; C is the capacitance value of the capacitor. P C =I C 2 ×ESR (15) Among them, I C P is the effective value of the current flowing through the capacitor. C This refers to the capacitor's losses.

2. The method for optimizing resonant cavity parameters for an LLC circuit of a bidirectional DC-DC converter for energy storage according to claim 1, characterized in that, N inductance values ​​are selected as resonant inductance values. Based on the relationship between the resonant inductance, magnetizing inductance, and resonant capacitance, the corresponding magnetizing inductance and resonant capacitance values ​​are calculated, resulting in N sets of data consisting of the resonant inductance values ​​and their corresponding magnetizing inductance and resonant capacitance values. The corresponding resonant inductance current, magnetizing inductance current, and resonant capacitance voltage data are then calculated sequentially using these N sets of data. The specific method is as follows: N inductance values ​​are selected from the inductance database as the resonant inductance values. The corresponding formulas for calculating the resonant capacitance and magnetizing inductance values ​​are as follows: Lm=k×Lr (2) Where Cr is the resonant capacitance value corresponding to the resonant inductance value, fr is the resonant frequency, Lr is the resonant inductance value, Lm is the magnetizing inductance value corresponding to the resonant inductance value, and k is the proportionality coefficient between the magnetizing inductance and the resonant inductance. After calculating the N resonant inductance values ​​and the corresponding N resonant capacitance values ​​and N magnetizing inductance values, these are combined into N sets of data consisting of the resonant inductance values ​​and their corresponding magnetizing inductance and resonant capacitance values. These N sets of data are then sequentially input into the simulation model for simulation. The maximum current I1max and the effective current I1 flowing through the resonant inductors and capacitors are calculated. The switching frequency f corresponding to the rated voltage output from the low-voltage DC side to the high-voltage DC side is then calculated. s The maximum voltage Ucmax that the resonant capacitor withstands, and the maximum and effective current I2max and current I2 flowing through the magnetizing inductor.

3. The method for optimizing resonant cavity parameters for an LLC circuit of a bidirectional DC-DC converter for energy storage according to claim 1, characterized in that, The design data related to inductors and capacitors are integrated into an initial input dataset. The comprehensive index data calculated using cost and loss data is used as the first output dataset, and the feasibility label is used as the second output dataset. The maximum mutual information coefficient between each feature of the input dataset and the first and second output datasets is calculated. Using the maximum mutual information coefficient, input data features with high correlation to the first and second output datasets are selected, forming two new input datasets. The specific method is as follows: Design data related to inductors and capacitors, including resonant inductance, resonant capacitance, magnetizing inductance, core inner diameter, maximum effective current density, effective magnetic flux cross-sectional area of ​​the core, core window area, and actual capacitance value, are selected and integrated into an initial input dataset. After obtaining the initial input dataset, a comprehensive index is calculated using the total cost and total loss data, and this index is used as the first output dataset. The feasibility labels corresponding to each input data point are used as the second output dataset. The formulas for calculating the total cost, total loss, and comprehensive index are as follows: F i总 =F iLr +F iLm +F iCr (16) P i总 =P iLr +P iLm +P iCr (17) Among them, t i F is a comprehensive index of cost and loss for the i-th design scheme. i总 Let F be the total design cost of the i-th design scheme. max P represents the maximum design cost among all design options. i总 Let P be the total loss of the i-th design scheme. max F represents the maximum loss among all design schemes. iLr The total cost of the resonant inductor designed for the i-th design scheme, F iLm The total cost of the magnetizing inductor designed for the i-th design scheme, F iCr The total cost of the resonant capacitor designed for the i-th design scheme, P iLr P is the total loss of the resonant inductor. iLm For the total loss of the magnetizing inductor, P iCr This represents the total loss of the resonant capacitor. After obtaining the first output dataset and the second output dataset, calculate the maximum mutual information coefficient between each feature of the input dataset and the first output dataset and the second output dataset; After calculating the maximum mutual information coefficient between each feature and the output dataset, the features in the input dataset are sorted according to the maximum mutual information coefficient. The features with the largest maximum mutual information coefficient are selected as the features of the new input dataset according to the required data dimensions.

4. The method for optimizing the resonant cavity parameters of an LLC circuit for energy storage bidirectional DC-DC converter according to claim 3, characterized in that, The calculation process for the maximum mutual information coefficient between features in the input dataset and a single output dataset is as follows: (1) Select the data of one feature in the input dataset and the output dataset in turn to form a two-dimensional dataset, and draw it in the form of coordinate points in the two-dimensional coordinate plane; (2) Calculate the dividing lines of the region based on the precision traversal. The calculation formulas for the dividing lines of the x-axis and y-axis are as follows: x 分割线 =k×d x (19) and 分割线 =k×d y (20) Where, x 分割线 The x-coordinate and y-coordinate of the current x-axis dividing line are... 分割线 Let y be the y-coordinate corresponding to the current y-axis dividing line, k be the current traversal count, and d be the y-coordinate of the dividing line. x The step size d used to traverse the x-axis y The step size used for traversing the y-axis; (3) After calculating the dividing line, the dividing line divides the area into four regions: "top left", "bottom left", "top right", and "bottom right". At this time, the coordinate points in each region are counted and the probability of the coordinate point in each region is calculated. The calculation formula is as follows: Where p is the probability of a coordinate point appearing in this region, and n i n represents the number of coordinate points within this region. 总 This represents the total number of coordinate points. (4) After calculating the probability of the coordinate point in each region, the mutual information value between the feature and the output dataset under the current traversal is calculated according to the mutual information calculation formula, which is as follows: Where I(x,y) is the mutual information value of the two variables x and y, and p(X=x i Y = y j ) is in (x i ,y j The joint probability of the region ) p(x=x i ) is X at x i The probability of the region, p(Y=y j ) is Y at y j The probability of the region; when x i When the subscript i = 1, it represents the left side; when i = 2, it represents the right side; when y j When the subscript j=1, it represents the top side; when j=2, it represents the bottom side. (5) After iterating through and changing the dividing line and calculating the mutual information value between each feature and the output dataset, stop the iteration when the dividing line meets the iteration condition and calculate the maximum mutual information coefficient. The calculation formula is as follows: Where mic(x,y) is the maximum mutual information coefficient between the feature and the output dataset, max() is the maximum value function, I(x,y) is the mutual information coefficient between the feature and the output dataset, a is the number of regions divided in the x-direction, and b is the number of regions divided in the y-direction.

5. The method for optimizing resonant cavity parameters for an LLC circuit of a bidirectional DC-DC converter for energy storage according to claim 1, characterized in that, The new input dataset has two rows: one is the dataset after feature filtering by the input dataset and the first output dataset, and the other is the dataset after feature filtering by the input dataset and the second output dataset. These two rows are combined according to their data numbers to obtain the final input dataset. The first output dataset and the second output dataset are combined according to their data numbers to obtain the final output dataset.

6. The method for optimizing resonant cavity parameters for an LLC circuit of a bidirectional DC-DC converter for energy storage according to claim 1, characterized in that, The final input data and final output data are fed into the SVM for model training. After the model training is completed, M input data are randomly generated according to the data format of the final input data, where each input data represents a design scheme. These randomly generated input data are fed into the SVM to make it output the corresponding comprehensive index and the feasibility label of each design scheme. Remove the input data (design schemes) with a feasibility label of 0, sort the comprehensive indicators of the remaining input data, and select the input data (design schemes) with the largest comprehensive indicator as the optimized design scheme.

Images