Method and system for designing frequency conversion range of CLLC direct-current transformer based on neural network

Abstract

The invention discloses a design method and system for a frequency conversion range of a CLLC direct-current transformer based on a neural network, and the method comprises the steps: carrying out the sampling in a preset variable range, and constructing a sampling database covering all working conditions; performing simulation verification on the sampled data to generate a physical constraint data set containing forward and reverse gains; inputting the processed simulation data into a neural network model for training, and establishing a precise mapping relation between parameter variables and electrical performance; and finally, optimizing model output by integrating an optimization algorithm (such as gradient descent and genetic algorithm), and realizing global optimal solution of resonant cavity parameters under the rigid constraints that the gain matching degree error is less than 2.1% and the frequency fluctuates. According to the method, a simulation verification-model training-parameter optimization closed-loop system is innovatively constructed, through a sampling mechanism and a cross-platform data interface, the three-day optimization period needed by traditional manual trial and error is compressed, and the design is more convenient and reasonable.

Description

Technical Field

[0001] The present invention belongs to the technical field of power electronic conversion of low-voltage DC distribution networks, and relates to a method and system for designing a frequency conversion range of a CLLC (isolated bidirectional resonant converter) DC transformer based on a neural network. Background Art

[0002] With the rapid development of distributed renewable energy, the application of low-voltage DC distribution networks has expanded from single-source power supply to diverse scenarios, including photovoltaic and energy storage integration, data center power supply, and building DC conversion. However, existing DC transformers face structural constraints: the fixed voltage gain characteristics of the traditional LLC (series resonant converter) topology make it difficult to adapt to a wide bidirectional input range, requiring customized designs for different scenarios and increasing hardware costs. Furthermore, minute-by-minute fluctuations in renewable energy (such as ±80% variations in photovoltaic irradiance) and random load disturbances (such as sudden changes in electric vehicle charging and discharging power) require equipment with real-time topology reconfiguration capabilities and redundancy.

[0003] Traditional CLLC design has technical constraints: First, the symmetrical resonant cavity designed by the fundamental wave approximation (FHA) method ( L r (Resonant Inductance) = C r (resonant capacitor) leads to imbalance in forward and reverse gain. In V2G (electric vehicles transmitting energy to the grid), the boost gain required in the forward mode and the reverse charging mode are quite different. Secondly, to cover the ±20% input fluctuation, 200-600 kHz Wideband regulation results in a large proportion of switching losses in the high-frequency band, and the core volume becomes larger, resulting in a decrease in power density.

[0004] Existing analysis and design methods are difficult to break through the barriers of coordinated optimization of topological parameters and dynamic operating conditions, and the cost of hardware modification increases, making it difficult to guide the design of CLLC DC transformers. Summary of the Invention

[0005] In response to the problems existing in the prior art, the present invention provides a design method and system for the frequency conversion range of a CLLC DC transformer based on a neural network, which realizes the optimized design of the CLLC DC transformer in application scenarios with different forward and reverse gains of bidirectional energy transmission.

[0006] The present invention is achieved through the following technical solutions: The design method of the frequency conversion range of the CLLC DC transformer based on neural network includes: After analyzing and constraining the parameters of the CLLC resonant cavity, the value range of each parameter of the CLLC resonant cavity is obtained; The Latin hypercube sampling method is used to perform sampling and restriction processing within the value range of each parameter of the CLLC resonant cavity to generate a sample set, thereby establishing a sampling database containing forward and reverse gain characteristics; Based on a sampling database containing forward and reverse gain characteristics, the forward and reverse gain data at different frequencies are obtained by simulating and verifying the sampling data, and the gain effect index G is calculated; The gain effect index G and the parameters of the CLLC resonant cavity are input into the neural network model for training; The optimization algorithm is used to perform multi-objective optimization on the trained neural network model to obtain the optimal resonant cavity parameter values ​​within the input variable range.

[0007] Preferably, the parameters of the CLLC resonant cavity are analyzed and constrained, specifically: Get the primary resonant inductance L 1. Primary resonant capacitor C 1. The ratio k of the primary excitation inductance to the primary resonant inductance, the ratio h of the secondary converted resonant parameter to the primary resonant parameter, and the parameters of the CLLC resonant cavity of the transformer ratio N; According to the hardware indicators and resonant frequency as constraints, the parameters of the CLLC resonant cavity are scalarized and the value ranges of the parameters of the CLLC resonant cavity are set.

[0008] Preferably, the Latin hypercube sampling method is used to perform sampling and restriction processing within the value range of each parameter of the CLLC resonant cavity, specifically: According to the value range of each parameter variable of the CLLC resonant cavity, data sampling is performed by the Latin lattice hypercube sampling method, and the value range of each CLLC resonant cavity parameter variable is divided into several intervals with equal probability. A sample value is selected from each parameter to generate a sample set, and the resonant frequency is combined with the resonant frequency. f r ∈[70kHz,250kHz] as the constraint condition to filter the valid samples in the sample set, and traverse the switching frequency f s ∈[0.5f r ,1.8f r ], and establish a sampling database containing forward and reverse gain characteristics at different frequencies.

[0009] Preferably, the sampling database containing the forward and reverse gain characteristics includes the primary resonant inductance L 1. Primary resonant capacitor C 1. The ratio of primary excitation inductance to primary resonant inductance k, the ratio of secondary converted resonant parameter to primary resonant parameter h, transformer ratio N, secondary resonant inductance L 2. Secondary side resonant capacitor C2 and primary excitation inductance L m ; The secondary side resonant inductor L 2. Secondary side resonant capacitor C 2 and primary excitation inductance L m , the specific calculation formula is: .

[0010] Preferably, based on a sampling database containing forward and reverse gain characteristics, the sampling data is simulated and verified through the PLECS simulation platform, and a soft start control strategy is adopted to detect the steady-state value of the output voltage during the simulation process. After the transformer reaches a steady state, the output voltage is sampled to obtain forward and reverse gain data at different frequencies, thereby obtaining the gain effect index G.

[0011] Preferably, the gain effect index G is obtained by: The gain data generated by each set of CLLC resonant cavity parameters in the sample database containing forward and reverse gain characteristics are fitted with a polynomial to obtain the minimum switching frequency that satisfies the gain in PFM control mode. f s_min and the maximum switching frequency f s_max , combined with f r , obtain the gain effect index G, the expression is:

[0012] Where, f r is the resonant frequency.

[0013] Preferably, the gain effect index G and the parameters of the CLLC resonant cavity are input into the neural network model for training, specifically: Based on the back propagation neural network model, the input layer is the CLLC resonant cavity parameter vector X=[L 1 ,C 1 ,k,h,N] , the output layer is the gain effect index G, the LeakyReLU activation function and the composite loss function are used to train the back propagation neural network model, and the trained back propagation neural network model is obtained through data normalization and regularization processing.

[0014] Preferably, an optimization algorithm is used to perform multi-objective optimization on the trained neural network model, specifically: An improved genetic algorithm is used to perform multi-objective optimization on the trained neural network model. The constraints are handled by the elite retention strategy and the quadratic penalty function to obtain the optimal resonant cavity parameter values ​​within the input variable range.

[0015] The design system of CLLC DC transformer frequency conversion range based on neural network includes: A parameter processing module is used to obtain the value range of each parameter of the CLLC resonant cavity by analyzing and constraining the CLLC resonant cavity parameters; The sampling module uses the Latin hypercube sampling method to perform sampling and restriction processing within the value range of each parameter of the CLLC resonant cavity to generate a sample set, thereby establishing a sampling database containing forward and reverse gain characteristics; The calculation module, based on a sampling database containing forward and reverse gain characteristics, obtains forward and reverse gain data at different frequencies by simulating and verifying the sampling data, and calculates the gain effect index G; The model training module inputs the gain effect index G and the parameters of the CLLC resonant cavity into the neural network model for training; The optimization solution module uses the optimization algorithm to perform multi-objective optimization on the trained neural network model to obtain the optimal resonant cavity parameter value within the input variable range.

[0016] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for designing the frequency conversion range of a CLLC DC transformer based on a neural network are implemented.

[0017] Compared with the prior art, the present invention has the following beneficial technical effects: The present invention discloses a design method and system for the frequency conversion range of a CLLC DC transformer based on a neural network. First, intelligent sampling of high-dimensional parameter space is performed within a preset variable range to build a sampling database covering all operating conditions. The sampled data is then simulated and verified using the PLECS power electronics simulation platform to generate a physical constraint data set containing key performance indicators such as forward and reverse gain and switching frequency. The processed simulation data is input into a neural network model for training to establish a precise mapping relationship between parameter variables and electrical performance. Finally, an optimization algorithm (such as gradient descent and genetic algorithm) is integrated to optimize the model output. Under the hard constraints of gain matching error < 2.1% and frequency fluctuation, the resonant cavity parameters ( L 1. C 1, etc.) to find the global optimal solution. This method innovatively constructs a closed-loop system of "simulation verification-model training-parameter optimization." Through a sampling mechanism and a cross-platform data interface (MATLAB-PLECS), it compresses the three-day optimization cycle required for traditional manual trial and error, making the design more convenient and reasonable.

[0018] The neural network-based design method for the frequency conversion range of a CLLC DC transformer of the present invention obtains a prediction model based on the neural network and simulation data, and then optimizes the optimal design of the CLLC DC transformer under application scenarios with different forward and reverse gains of bidirectional energy transmission, obtaining a concise and effective model output that is easy to analyze and expand. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 CLLC resonant cavity diagram optimized in the present invention; Figure 2 Design the steps of the method for the present invention; Figure 3 is a topology diagram of a bidirectional CLLC resonant converter according to an embodiment. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0022] In the description of the present invention, it should be noted that, unless otherwise expressly specified and limited, the terms "include" and "comprise" indicate the existence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the existence or addition of one or more other features, wholes, steps, operations, elements, components and / or their collections.

[0023] It should also be understood that the terms used in the present specification are only for the purpose of describing particular embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0024] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0025] A design method for optimizing the frequency conversion range of asymmetric CLLC DC transformer based on neural network, such as Figure 2 As shown, the specific steps are: (1) After analyzing and constraining the parameters of the CLLC resonant cavity, the value range of each parameter of the CLLC resonant cavity is obtained.

[0026] like Figure 1 As shown, the resonant cavity has L 1 (primary resonant inductance), C 1 (primary resonant capacitor), L m Primary magnetizing inductance), L 2 (secondary resonant inductor), C 2 (secondary resonant capacitor) and N（ Transformer ratio ） By using the resonant frequency and scalar parameters to constrain the CLLC resonant cavity parameters, we can obtain the ratio k of the primary excitation inductance to the primary resonant inductance, and the ratio h of the secondary side converted resonant parameter (inductance or capacitance) to the primary resonant parameter. The specific expressions are: (1) in h =1 is the traditional symmetrical design, h ≠1 is an asymmetric design, so the parameters of the resonant cavity can be confirmed to be L 1, C 1, k , h and N . According to the hardware indicators and resonant frequency constraints, set the range of each parameter. like L 1:[1e-6,500e-6], etc. Here, for the convenience and rationality of transformer design, the transformer ratio adopts the fractional range, and the expression is: (2) (2) The Latin hypercube sampling method is used to perform sampling and restriction processing within the value range of each parameter of the CLLC resonant cavity to generate a sample set, thereby establishing a sampling database containing forward and reverse gain characteristics; According to the value range of each parameter variable of the CLLC resonant cavity, data sampling is performed by the Latin lattice hypercube sampling method, and the value range of each CLLC resonant cavity parameter variable is divided into several intervals with equal probability. A sample value is selected from each parameter to generate a sample set. The resonant frequency is used as a constraint condition to screen the valid samples in the sample set, and the switching frequency is traversed to obtain the valid samples. f s ∈[0.5f r ,1.8f r ], establish a sampling database containing forward and reverse gain characteristics at different frequencies, specifically: The Latin lattice hypercube sampling method is a special stratified sampling technique that divides the range of each variable into a number of equally probable intervals. Each variable is sampled from these intervals, ensuring a uniform distribution of samples across each dimension. Compared to traditional random sampling methods, Latin lattice hypercube sampling provides better coverage of the entire sample space and is particularly suitable for Monte Carlo simulations in high-dimensional parameter spaces.

[0027] The sample database containing the forward and reverse gain characteristics includes the primary side resonant inductance L 1. Primary resonant capacitor C 1. The ratio of primary excitation inductance to primary resonant inductance k, the ratio of secondary converted resonant parameter to primary resonant parameter h, transformer ratio N, secondary resonant inductance L 2. Secondary side resonant capacitor C 2 and primary excitation inductance L m ; The secondary side resonant inductor L 2. Secondary side resonant capacitor C 2 and primary excitation inductance L m , the specific calculation formula is: (3) According to the resonant frequency f r Limit the parameters after sampling and remove invalid data, such as f r ∈[70kHz,250kHz], and then after obtaining the sample set, use PFM modulation mode (Pulse Frequency Modulation, PFM) to control, which requires changing the switching frequency f s To obtain the gain at different frequencies in the forward and reverse directions, it is necessary to take points and traverse the switching frequency based on the parameter database. f s ∈[0.5 f r ,1.8 fr ]. The final sampling database containing forward and reverse gain characteristics at different frequencies is obtained.

[0028] (3) Based on the sampling database containing forward and reverse gain characteristics, the forward and reverse gain data at different frequencies are obtained by simulation verification of the sampling data, and the gain effect index G is calculated.

[0029] Based on a sampling database containing forward and reverse gain characteristics, the sampled data was simulated and verified using the PLECS simulation platform. A soft start control strategy was adopted. During the simulation process, the output voltage steady-state value was detected. After the transformer reached steady-state, the output voltage was sampled to obtain forward and reverse gain data at different frequencies, thereby obtaining the gain effect index G, which is specifically: After obtaining the sampling database containing the forward and reverse gain characteristics at different frequencies, the design program sets the parameters of the database ( L 1, C 1, L m , L 2, C 2， N,f s ) is substituted into the simulation model and run. To save simulation time, parallel simulation is used in the program and a soft-start control strategy is configured. After the transformer reaches steady state, the output voltage is sampled and the forward and reverse gain data at different frequencies are obtained according to the formula.

[0030] This patent uses forward gain as an example for explanation, and reverse gain is vice versa, specifically: like Figure 3 As shown in the topology diagram of the bidirectional CLLC resonant converter, V1 is the primary side DC bus voltage, V2 is the secondary side DC bus voltage, V ab is the primary bridge arm midpoint voltage, V cd This article defines the forward and reverse directions of power transfer as power transfer from V1 to V2, and vice versa. When the converter is operating in the forward direction, drive signals are applied to switches Q1-Q4 to achieve inversion on the primary side. Switches Q5-Q8 are not driven, and their anti-parallel diodes achieve rectification on the secondary side. When the converter is operating in the reverse direction, the operating states of the primary and secondary switches are opposite.

[0031] When working in the forward direction, the forward gain is the product of the output voltage and the transformer ratio divided by the input voltage, and the reverse gain is the product of the output voltage and the transformer ratio divided by the input voltage. (4) Where N is the transformer ratio, Gain forward is the forward gain, Gain backward is the reverse gain, V 1 is the primary side DC bus voltage, V 2 is the secondary side DC bus voltage.

[0032] The gain data generated by each set of CLLC resonant cavity parameters in the sampling database containing forward and reverse gain characteristics at different frequencies are fitted with a polynomial to obtain the minimum switching frequency that satisfies the gain in PFM control mode. f s_min and the maximum switching frequency f s_max , combined with the resonant frequency f r , the gain effect index G (resonant frequency / maximum switching frequency - minimum switching frequency) is obtained to express the gain effect: (5) Under the same variable frequency range, a larger resonant frequency can reduce the volume of magnetic components and reduce losses, while under the same resonant frequency, a smaller frequency range is better.

[0033] (4) Input the gain effect index G and the parameters of the CLLC resonant cavity into the neural network model for training; Based on the back propagation neural network model, the input layer is the CLLC resonant cavity parameter vector X=[L 1 ,C 1 ,k,h,N] , the output layer is the gain effect index G, the LeakyReLU activation function and the composite loss function are used to train the back propagation neural network model, and the trained back propagation neural network model is obtained through data normalization and regularization. Specifically: A BP neural network (BPNN) with two hidden layers is constructed. The network topology is defined as 5-16-8-1, where the input layer corresponds to the parameter vectors X=[ L 1, C 1, k , h , N ], the output layer is the gain effect index G. The network activation function uses LeakyReLU (LeakyRectified Linear Unit (ReLU) activation function) ( α = 0.1) gradient vanishing problem, the output layer uses linear activation.

[0034] Based on the simulation data set (including group samples) generated by the PLECS power electronics simulation platform, the following training optimization process is implemented for the network: Data preprocessing: Min-Max normalization is performed on the input parameters, and the output parameters are standardized using Z-Score (standard score): ; in, Xnorm is the normalized value, Xmax is the maximum value of the input parameter, Xmin is the maximum value of the input parameter; Gstd is the normalized value, μG For output parameters G The mean of σG For output parameters G The standard deviation of Loss function design: Define a composite loss function L and suppress overfitting by introducing L2 regularization.

[0035] ; In the formula, the mean square error (MSE) term is used to measure the prediction gain Gpred With real gain Gture The degree of deviation directly optimizes the model prediction accuracy and weight coefficient λ1 Control the contribution ratio of MSE item in the total loss, if λ1 If it is too large, the model may overfit the noisy data; if it is too small, the prediction accuracy will decrease; L2 regularization term|| W || is the sum of squares of the penalty model weights (W), which suppresses the risk of overfitting, and the weight coefficient λ2 Control the regularization strength. Too large a regularization strength will lead to underfitting of the model. wj is the model parameter vector w The j weight parameters.

[0036] Training and Validation: The training / validation / test sets were split into a 7:2:1 ratio, and early stopping monitoring (patience = 10 epochs, a training process in which all samples in the training dataset are trained once (and only once)) was implemented. After 300 epochs of training, the RMSE (root mean square error) of the validation set converged to 0.023, establishing a preliminary mapping relationship between "circuit parameters → gain effect".

[0037] (5) Use the optimization algorithm to perform multi-objective optimization on the trained neural network model to obtain the optimal resonant cavity parameter values ​​within the input variable range.

[0038] An improved genetic algorithm is used to perform multi-objective optimization on the trained neural network model. By processing the constraints with an elite retention strategy and a quadratic penalty function, the optimal resonant cavity parameter values ​​within the input variable range are obtained, specifically: After obtaining the "parameters → output results" model, an improved genetic algorithm with an elite retention strategy is used for global optimization, and the optimization objective function is defined as:

[0039] Design constraints, impose a quadratic penalty function on individuals that go beyond the bounds, and impose penalties on individuals that violate the constraints, forcing the optimization direction to converge to the feasible region.

[0040]

[0041] ; In the formula gi(X) is the offset of the parameter beyond the constraint range, λ is the penalty coefficient, which is allocated according to the constraint priority, 103 for the lower limit of the resonant frequency, and 102 for the upper limit of the resonant frequency.

[0042] The Pareto front solution set is obtained through 200 generations of iteration, and a set of resonant cavity parameter values ​​with the largest G value is obtained.

[0043] The design system of CLLC DC transformer frequency conversion range based on neural network includes: A parameter processing module is used to obtain the value range of each parameter of the CLLC resonant cavity by analyzing and constraining the CLLC resonant cavity parameters; The sampling module uses the Latin hypercube sampling method to perform sampling and restriction processing within the value range of each parameter of the CLLC resonant cavity to generate a sample set, thereby establishing a sampling database containing forward and reverse gain characteristics; The calculation module, based on a sampling database containing forward and reverse gain characteristics, obtains forward and reverse gain data at different frequencies by simulating and verifying the sampling data, and calculates the gain effect index G; The model training module inputs the gain effect index G and the parameters of the CLLC resonant cavity into the neural network model for training; The optimization solution module uses the optimization algorithm to perform multi-objective optimization on the trained neural network model to obtain the optimal resonant cavity parameter value within the input variable range.

[0044] One embodiment of the present invention provides a terminal device comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the processor implements the steps of the aforementioned method for deblurring dynamic scene images; alternatively, when the processor executes the computer program, the processor implements the functions of the modules of the aforementioned system for deblurring dynamic scene images.

[0045] The computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to accomplish the present invention.

[0046] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0047] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0048] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0049] If the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0050] Based on this understanding, the present invention can implement all or part of the processes in the above-mentioned embodiment method by means of a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the above-mentioned dynamic scene image deblurring method. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form.

[0051] The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media does not include electric carrier signals and telecommunication signals. The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any ordinary technician in this industry can smoothly implement the present invention as shown in the drawings and described above. However, any equivalent changes, modifications and evolutions made by technicians familiar with this profession without departing from the scope of the technical solution of the present invention using the technical content disclosed above are all equivalent embodiments of the present invention. At the same time, any equivalent changes, modifications and evolutions made to the above embodiments based on the essential technology of the present invention are still within the scope of protection of the technical solution of the present invention.

Claims

1. A design method for the frequency conversion range of a CLLC DC transformer based on a neural network, characterized in that: include, After analyzing and constraining the parameters of the CLLC resonant cavity, the value range of each parameter of the CLLC resonant cavity is obtained; The Latin hypercube sampling method is used to perform sampling and restriction processing within the value range of each parameter of the CLLC resonant cavity to generate a sample set, thereby establishing a sampling database containing forward and reverse gain characteristics; Based on a sampling database containing forward and reverse gain characteristics, the forward and reverse gain data at different frequencies are obtained by simulating and verifying the sampling data, and the gain effect index G is calculated; The gain effect index G and the parameters of the CLLC resonant cavity are input into the neural network model for training; The optimization algorithm is used to perform multi-objective optimization on the trained neural network model to obtain the optimal resonant cavity parameter values ​​within the input variable range.

2. The method for designing the frequency conversion range of a CLLC DC transformer based on a neural network according to claim 1, wherein: By analyzing and constraining the parameters of the CLLC resonant cavity, specifically: Get the primary resonant inductance L 1. Primary resonant capacitor C 1. The ratio k of the primary excitation inductance to the primary resonant inductance, the ratio h of the secondary converted resonant parameter to the primary resonant parameter, and the parameters of the CLLC resonant cavity of the transformer ratio N; According to the hardware indicators and resonant frequency as constraints, the parameters of the CLLC resonant cavity are scalarized and the value ranges of the parameters of the CLLC resonant cavity are set.

3. The method for designing the frequency conversion range of a CLLC DC transformer based on a neural network according to claim 1, wherein: The Latin hypercube sampling method is used to perform sampling and restriction processing within the value range of each parameter of the CLLC resonant cavity, specifically: According to the value range of each parameter variable of the CLLC resonant cavity, data sampling is performed by the Latin lattice hypercube sampling method, and the value range of each CLLC resonant cavity parameter variable is divided into several intervals with equal probability. A sample value is selected from each parameter to generate a sample set, and the resonant frequency is combined with the resonant frequency. f r ∈[70kHz,250kHz] as the constraint condition to filter the valid samples in the sample set, and traverse the switching frequency f s ∈[0.5f r ,1.8f r ], and establish a sampling database containing forward and reverse gain characteristics at different frequencies.

4. The method for designing the frequency conversion range of a CLLC DC transformer based on a neural network according to claim 3, wherein: The sample database containing the forward and reverse gain characteristics includes the primary side resonant inductance L 1. Primary resonant capacitor C 1. The ratio of primary excitation inductance to primary resonant inductance k, the ratio of secondary converted resonant parameter to primary resonant parameter h, transformer ratio N, secondary resonant inductance L 2. Secondary side resonant capacitor C 2 and primary excitation inductance L m ; The secondary side resonant inductor L 2. Secondary side resonant capacitor C 2 and primary excitation inductance L m , the specific calculation formula is: 。 5. The method for designing the frequency conversion range of a CLLC DC transformer based on a neural network according to claim 1, wherein: Based on a sampling database containing forward and reverse gain characteristics, the sampling data was simulated and verified through the PLECS simulation platform. A soft start control strategy was adopted. The steady-state value of the output voltage was detected during the simulation process. The output voltage was sampled after the transformer reached the steady state, and the forward and reverse gain data at different frequencies were obtained, thereby obtaining the gain effect index G.

6. The method for designing the frequency conversion range of a CLLC DC transformer based on a neural network according to claim 5, characterized in that: The specific acquisition process of the gain effect index G is as follows: The gain data generated by each set of CLLC resonant cavity parameters in the sample database containing forward and reverse gain characteristics are fitted with a polynomial to obtain the minimum switching frequency that satisfies the gain in PFM control mode. f s_min and the maximum switching frequency f s_max , combined with f r , obtain the gain effect index G, the expression is: Where, f r is the resonant frequency.

7. The method for designing the frequency conversion range of a CLLC DC transformer based on a neural network according to claim 1, wherein: The gain effect index G and the parameters of the CLLC resonant cavity are input into the neural network model for training, specifically: Based on the back propagation neural network model, the input layer is the CLLC resonant cavity parameter vector X=[L 1 ,C 1 ,k,h,N] , the output layer is the gain effect index G, the LeakyReLU activation function and the composite loss function are used to train the back propagation neural network model, and the trained back propagation neural network model is obtained through data normalization and regularization processing.

8. The method for designing the frequency conversion range of a CLLC DC transformer based on a neural network according to claim 1, wherein: Use the optimization algorithm to perform multi-objective optimization on the trained neural network model, specifically: An improved genetic algorithm is used to perform multi-objective optimization on the trained neural network model. The constraints are processed by the elite retention strategy and the quadratic penalty function to obtain the optimal resonant cavity parameter values ​​within the input variable range.

9. A neural network-based CLLC DC transformer frequency conversion range design system, characterized in that: include, A parameter processing module is used to obtain the value range of each parameter of the CLLC resonant cavity by analyzing and constraining the CLLC resonant cavity parameters; The sampling module uses the Latin hypercube sampling method to perform sampling and restriction processing within the value range of each parameter of the CLLC resonant cavity to generate a sample set, thereby establishing a sampling database containing forward and reverse gain characteristics; The calculation module, based on a sampling database containing forward and reverse gain characteristics, obtains forward and reverse gain data at different frequencies by simulating and verifying the sampling data, and calculates the gain effect index G; The model training module inputs the gain effect index G and the parameters of the CLLC resonant cavity into the neural network model for training; The optimization solution module uses the optimization algorithm to perform multi-objective optimization on the trained neural network model to obtain the optimal resonant cavity parameter value within the input variable range.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method for designing the frequency conversion range of a CLLC DC transformer based on a neural network as described in any one of claims 1 to 8 are implemented.

Images