A design method and system of a high-frequency transformer

Abstract

This invention discloses a design method and system for high-frequency transformers, belonging to the field of high-frequency transformer optimization design technology. The method includes: acquiring design parameter data of the high-frequency transformer to be optimized, a dataset of historical transformer design parameters, and a dataset of historical transformer actual performance indicators; acquiring a convolutional neural multilayer perceptron deep learning model based on the above data to obtain an initial particle population and predicted particle fitness; if there is a particle fitness greater than a preset historical particle fitness, updating the initial particle population, obtaining a first particle population, and then acquiring the optimization design parameter data of the high-frequency transformer to be optimized; acquiring the simulation experimental deviation value of the simulation data and experimental data based on it, and if it is less than or equal to a preset threshold, completing the design of the high-frequency transformer. The high-frequency transformer design method provided by this invention can achieve high-precision and high-efficiency optimization of high-frequency transformer design parameters.

Description

Technical Field

[0001] This invention relates to the field of high-frequency transformer optimization design technology, and in particular to a design method and system for high-frequency transformers. Background Technology

[0002] High-frequency transformers, as core components of power electronic transformers, possess functions such as voltage transformation, electrical isolation, power transmission and control, and bidirectional energy flow. They offer significant advantages in terms of size, capacity, and weight, and are widely used in medium-voltage distribution networks, lightweight offshore wind power, data centers, new energy charging stations, and aerospace and marine applications. However, high-frequency transformers generate winding, core, and dielectric losses during operation, which can easily lead to insulation risks such as thermal breakdown, flashover, and partial discharge. Therefore, insulation and temperature rise have become key factors limiting further improvements in the power density and reliability of high-frequency transformers. Optimizing the structure of high-frequency transformers to improve their insulation and temperature rise performance has become a research focus in related fields. Currently, transformer optimization design methods are mainly divided into two categories: one is to optimize based on loss-based analytical models. This method, based on a one-dimensional transformer loss model, decomposes copper losses into skin losses and proximity losses, and uses the ratio of copper thickness to skin depth to represent the single-layer winding loss. Subsequent research has expanded to include a two-dimensional model based on the frequency domain diffusion equation, achieving decoupling between skin and proximity losses. Simulation results demonstrate that the error between this model and the one-dimensional model is negligible. Furthermore, winding structure optimization has received attention; methods such as split cores and pyramid structures can reduce the number of PCB layers and effectively optimize losses, improving transformer design efficiency and performance. However, analytical models simplify and idealize the actual physical field, resulting in relatively low accuracy of performance parameters calculated using them. Another approach is performance calculation and optimization based on finite element simulation. Transformer performance parameters can be calculated by constructing two-dimensional or three-dimensional finite element models. Three-dimensional models can reflect the detailed structure of the transformer and obtain more accurate calculation results, but they suffer from a large number of model nodes and long computation time. Although dimensionality reduction methods can simplify the three-dimensional model to a two-dimensional model to reduce the number of mesh nodes and improve model efficiency, the computational cost of finite element models is still much higher than that of analytical models, which limits the efficiency of optimization design using finite element models.

[0003] Under the current technological background, existing optimization design methods for high-frequency transformers have several shortcomings. Firstly, there is a conflict between model accuracy and computational efficiency. Analytical models, in order to improve computational efficiency, typically simplify and idealize the actual physical processes, resulting in insufficient accuracy in the calculated performance parameters. While finite element simulation can provide higher accuracy, its large number of model nodes and complex calculation processes lead to low computational efficiency and long processing times, failing to meet the needs of large-scale design and rapid iteration. Secondly, computational costs are high. Running finite element models requires substantial computational resources and high-performance hardware support. Especially in the fine design scenario of high-frequency transformers, its computational cost is far higher than that of analytical models, significantly reducing the economic viability of optimization design. The problems include: firstly, the lack of engineering feasibility; secondly, the lack of multi-physics coupling optimization capabilities. Existing optimization methods are mostly focused on optimizing single physical fields (such as electromagnetic or thermal fields), making it difficult to effectively handle multi-physics coupling problems. In high-frequency transformer design, the interaction of multiple physical fields such as electromagnetics and thermodynamics is complex. Existing methods fail to fully consider these complex coupling effects, which can easily lead to incomplete or distorted optimization results; and thirdly, the poor adaptability and scalability of the methods. Current optimization methods often rely on specific models or assumptions, making it difficult to flexibly meet the design requirements of different types of transformers. Especially when facing complex factors such as new materials and new structures, their versatility and scalability are limited, making it difficult to meet rapidly changing industrial needs. Summary of the Invention

[0004] The present invention aims to provide a design method and system for high-frequency transformers to solve the above-mentioned technical problems, avoid the problems of low accuracy and poor calculation efficiency in the optimization design of high-frequency transformers, and achieve high-precision and high-efficiency optimization of high-frequency transformer design parameters.

[0005] To address the aforementioned technical problems, this invention provides a design method for a high-frequency transformer, comprising: Obtain design parameter data of the high-frequency transformer to be optimized and several standard historical performance data, and obtain historical transformer design parameter dataset and historical transformer actual performance index dataset based on the standard historical performance data; Based on the historical transformer design parameter dataset, the historical transformer actual performance index dataset, and the preset initial deep learning model, a convolutional neural multilayer perceptual deep learning model is obtained. Based on the initial design parameter data of the high-frequency transformer to be optimized and the convolutional neural multilayer perception deep learning model, the initial particle population, the initial velocity of several particles and the performance index data of several particles are obtained, and the fitness of several predicted particles is obtained based on the particle performance index data. If the predicted particle fitness is greater than the preset historical particle fitness, the initial particle population is updated based on the particle swarm performance index data corresponding to the predicted particle fitness, the initial particle velocity corresponding to the predicted particle fitness, and the preset physical constraints until the first preset convergence condition is met, and the first particle population is obtained, so as to obtain the optimization design parameter data of the high-frequency transformer to be optimized based on the first particle population. Based on the optimization design parameter data of the high-frequency transformer to be optimized, simulation data and experimental data are obtained, and simulation experimental deviation values ​​are obtained based on the simulation experimental data and experimental data. If the simulation experimental deviation value is less than or equal to the preset simulation experimental deviation threshold, the design of the high-frequency transformer is completed.

[0006] In the above scheme, the preset initial deep learning model is trained using historical transformer design parameter datasets and historical transformer actual performance index datasets. This enables precise mapping between transformer design parameters and transformer performance index data, allowing the resulting convolutional neural multilayer perception deep learning model to balance prediction accuracy and computational efficiency. This provides highly efficient performance prediction capabilities, avoiding the problems of low accuracy and poor computational efficiency encountered when optimizing the design of high-frequency transformers. Then, using the initial design parameter data of the high-frequency transformer to be optimized and the convolutional neural multilayer perception deep learning model, an initial particle population, initial particle velocity, and particle performance index data are generated, and the predicted particle fitness is calculated. This provides a performance evaluation basis for optimizing the high-frequency transformer design parameters and improves computational efficiency in the early stages of optimization. Subsequently, by comparing the predicted particle fitness with the preset historical particle fitness, and combining the initial particle velocity and preset physical constraints, the particle population is iteratively updated until the first preset convergence condition is met. This efficiently filters out the high-frequency transformer design parameters that meet the performance requirements, obtaining the optimized design parameter data for the high-frequency transformer to be optimized. This ensures the optimization accuracy of the design parameters while avoiding the lengthy computation process of traditional simulations. Finally, by obtaining simulation and experimental data from the optimized design parameters of the high-frequency transformer to be optimized, calculating the simulation and experimental deviation values ​​and determining whether the simulation and experimental deviation values ​​are less than or equal to the preset simulation and experimental deviation threshold, the engineering feasibility of the optimized design parameters can be verified, ensuring that the optimized results have practical application value. At the same time, the design accuracy is further guaranteed by dual verification through simulation and experiment only in the final stage, achieving high-precision and high-efficiency optimization of the high-frequency transformer design parameters.

[0007] Furthermore, the process of acquiring the design parameter data of the high-frequency transformer to be optimized and several standard historical performance data, and acquiring a historical transformer design parameter dataset and a historical transformer actual performance index dataset based on the standard historical performance data, includes: Obtain the design parameter data of the high-frequency transformer to be optimized and some initial historical performance data; The initial historical performance data is cleaned to obtain clean historical performance data; The purified historical performance data is normalized to obtain standard historical performance data; Historical transformer design parameter datasets and historical transformer actual performance index datasets are obtained based on standard historical performance data.

[0008] In the above scheme, by acquiring the design parameter data of the high-frequency transformer to be optimized and several initial historical performance data, the basic parameters of the optimized high-frequency transformer are clarified, and raw data for subsequent model training is provided. Next, by cleaning the initial historical performance data, invalid data is removed, improving the purity and effectiveness of the data and preventing poor-quality data from interfering with subsequent model training and parameter optimization. Then, by normalizing the clean historical performance data, performance data of different dimensions and scales can be transformed into a unified scale, eliminating the influence of dimensional differences between data, adapting to the training requirements of subsequent convolutional neural multilayer perceptron deep learning models, and obtaining standard historical performance data. Finally, by extracting the standard historical performance data, a historical transformer design parameter dataset and a historical transformer actual performance index dataset are obtained, achieving accurate matching between the design parameters and corresponding performance indicators of the high-frequency transformer. This provides standardized and suitable input data for the subsequent training of the deep learning model, ensuring that the model can effectively learn the correlation between design parameters and performance indicators.

[0009] Furthermore, the step of obtaining a convolutional neural multilayer perceptron deep learning model based on historical transformer design parameter datasets, historical transformer actual performance index datasets, and a preset initial deep learning model includes: Based on a preset ratio, the historical transformer design parameter dataset is divided into a model building subset, a performance evaluation subset, and an independent verification subset. Based on the historical transformer actual performance index dataset, the model building subset, the performance evaluation subset, and the independent verification subset, the first performance index dataset corresponding to the model building subset, the second actual performance index dataset corresponding to the performance evaluation subset, and the third actual performance index dataset corresponding to the independent verification subset are obtained. Based on the model construction subset, performance evaluation subset, independent testing subset, first performance index dataset, second actual performance index dataset, third actual performance index dataset, and preset initial deep learning model, a convolutional neural multilayer perceptual deep learning model is obtained.

[0010] In the above scheme, the historical transformer design parameter dataset is divided into a model construction subset, a performance evaluation subset, and an independent verification subset according to a preset ratio. Corresponding first performance index dataset, second actual performance index dataset, and third actual performance index dataset are then matched to establish a hierarchical dataset system for the training, evaluation, and verification of the deep learning model, ensuring the independence and specificity of data at each stage of model training. Next, by training, evaluating, and verifying the model using the model construction subset, performance evaluation subset, independent verification subset, first performance index dataset, second actual performance index dataset, and third actual performance index dataset with the preset initial deep learning model, the parameters of the model are fitted, validated, and verified. Finally, a convolutional neural multilayer perceptron deep learning model with high-precision performance prediction capabilities is obtained, thus providing a reliable performance evaluation foundation for subsequent optimization of high-frequency transformer design parameters.

[0011] Furthermore, the step of obtaining a convolutional neural multilayer perceptron deep learning model based on a model construction subset, a performance evaluation subset, an independent testing subset, a first performance index dataset, a second actual performance index dataset, a third actual performance index dataset, and a preset initial deep learning model includes: Based on the model construction subset and the preset initial deep learning model, obtain the training performance index dataset corresponding to the model construction subset; An error function is constructed based on the first actual performance index dataset and the training performance index dataset. The preset initial deep learning model is updated by minimizing the error function until the second preset convergence condition is met, and the deep learning training model is obtained. Based on the performance evaluation subset and the deep learning training model, obtain the validation performance metric dataset corresponding to the performance evaluation subset; The model deviation fluctuation value is obtained based on the second actual performance index dataset and the validation performance index dataset. If the model deviation fluctuation value is less than the preset model deviation fluctuation threshold, the deep learning validation model is obtained. Based on the independent test subset and the deep learning verification model, obtain the test performance index dataset corresponding to the independent test subset; The indicator response curve is obtained based on the third actual performance indicator dataset and the test performance indicator dataset. If the indicator response curve meets the preset accuracy standard condition, the convolutional neural multilayer perceptual deep learning model is obtained.

[0012] In the above scheme, by inputting a subset of the model construction into a preset initial deep learning model to obtain the training performance index dataset corresponding to the model construction subset, a prediction result reference can be provided for subsequent model parameter updates and error calculation. Next, an error function is constructed using the first actual performance index dataset and the training performance index dataset. The preset initial deep learning model is iteratively updated with the goal of minimizing the error function until a second preset convergence condition is met. This allows the model to initially learn the correlation between design parameters and performance indices, completing parameter fitting and optimization, and obtaining the deep learning training model. Then, by inputting a performance evaluation subset into the deep learning training model for performance prediction, a validation performance index dataset corresponding to the performance evaluation subset is obtained, providing data support for verifying the model's prediction stability. Subsequently, the model bias fluctuation value is calculated using the second actual performance index dataset and the validation performance index dataset, and it is determined whether the model bias fluctuation value is less than a preset model bias fluctuation threshold to avoid the problem of excessive local prediction bias, completing the evaluation of the deep learning training model. Finally, by inputting an independent verification subset into the deep learning validation model for performance verification, a test performance index dataset corresponding to the independent verification subset is obtained, providing experimental prediction data for verifying the deep learning validation model's generalization ability to unknown samples. Finally, the index response curves are generated using the third actual performance index dataset and the test performance index dataset, and it is verified whether the index response curves meet the preset accuracy standard conditions. This completes the verification of the model's generalization ability and overall prediction accuracy, and ultimately obtains a convolutional neural multilayer perception deep learning model that has both stable prediction ability and high-precision generalization ability, providing a reliable performance prediction basis for subsequent optimization of high-frequency transformer design parameters.

[0013] Furthermore, the process of obtaining initial particle population, initial particle velocities, and particle performance index data based on the initial design parameter data of the high-frequency transformer to be optimized and the convolutional neural multilayer perceptron deep learning model, and obtaining several predicted particle fitness values ​​based on the particle performance index data, includes: An initial particle population and several initial velocities of particles are generated based on the initial design parameter data of the high-frequency transformer to be optimized. Several particle performance index data are obtained based on the convolutional neural multilayer perceptual deep learning model and the initial particle population, and several predicted particle fitnesss are obtained based on the particle performance index data and the preset multi-objective weighted fitness function.

[0014] In the above scheme, an initial particle swarm and several initial velocities of particles are generated from the initial design parameter data of the high-frequency transformer to be optimized. This transforms the optimization problem of the high-frequency transformer design parameters into an optimization problem of the particle swarm optimization algorithm, thus establishing an initial parameter framework for the iterative optimization of the particle swarm optimization algorithm. Next, by inputting the initial particle swarm into a convolutional neural multilayer perceptron deep learning model, several particle performance index data are obtained. Combined with a preset multi-objective weighted fitness function, several predicted particle fitness values ​​are calculated. The convolutional neural multilayer perceptron deep learning model achieves rapid and accurate prediction of the design parameter performance, while the preset multi-objective weighted fitness function evaluates the particle performance, providing a basis for updating the initial particle swarm. This approach ensures the optimization accuracy of the design parameters while avoiding the lengthy calculation process of traditional simulations.

[0015] Further, the step of obtaining simulation data and experimental data based on the optimization design parameter data of the high-frequency transformer to be optimized, and obtaining the simulation experiment deviation value based on the simulation experiment data and experimental data, and completing the design of the high-frequency transformer if the simulation experiment deviation value is less than or equal to a preset simulation experiment deviation threshold, includes: The optimization design parameters of the high-frequency transformer to be optimized are simulated to obtain simulation data; Based on the optimized design parameter data of the high-frequency transformer to be optimized, a high-frequency transformer prototype was obtained, and experiments were conducted on the high-frequency transformer prototype to obtain experimental data. The simulation experiment deviation value is obtained based on the simulation experiment data and the experimental data. If the simulation experiment deviation value is less than or equal to the preset simulation experiment deviation threshold, the design of the high-frequency transformer is completed.

[0016] In the above scheme, by simulating the optimized design parameters of the high-frequency transformer to be optimized, simulation data is obtained, which can theoretically verify the transformer performance corresponding to the optimized design parameters, providing theoretical reference and comparison benchmark for subsequent experimental verification. Next, by fabricating a high-frequency transformer prototype based on the optimized design parameters and conducting experiments, experimental data is obtained, which can verify the feasibility and actual performance of the optimized design parameters from an engineering perspective, overcoming the theoretical limitations of pure simulation analysis. Then, by calculating the simulation-experiment deviation value using both simulation and experimental data, and determining whether the simulation-experiment deviation value is less than or equal to a preset simulation-experiment deviation threshold, a dual verification of the theoretical and practical aspects of the optimized design parameters is achieved, ensuring that the optimized design parameters meet the requirements of engineering practicality and performance accuracy.

[0017] Furthermore, the step of obtaining simulation data and experimental data based on the optimization design parameter data of the high-frequency transformer to be optimized, and obtaining the simulation experiment deviation value based on the simulation experiment data and experimental data, and completing the design of the high-frequency transformer if the simulation experiment deviation value is less than or equal to a preset simulation experiment deviation threshold, further includes: Based on the optimization design parameter data of the high-frequency transformer to be optimized, simulation data and experimental data are obtained, and simulation experimental deviation values ​​are obtained based on the simulation experimental data and experimental data. If the simulation experimental deviation value is greater than the preset simulation experimental deviation threshold, the convolutional neural multilayer perceptual deep learning model is updated based on the experimental data until the simulation experimental deviation value is less than or equal to the preset simulation experimental deviation threshold.

[0018] In the above scheme, when the simulation experiment deviation exceeds a preset simulation experiment deviation threshold, the parameters of the convolutional neural multilayer perception deep learning model are updated using experimental data. The model is then iteratively optimized and its accuracy calibrated again to compensate for the deficiencies in the initial training, making the newly obtained convolutional neural multilayer perception deep learning model more closely aligned with actual engineering scenarios. Then, the convolutional neural multilayer perception deep learning model is continuously iterated and updated until the simulation experiment deviation meets the preset simulation experiment deviation threshold, ensuring that the final optimized design parameters for the high-frequency transformer meet both the requirements of engineering practicality and performance accuracy.

[0019] This invention provides a design system for high-frequency transformers, including a data acquisition module, a deep learning model training module, a particle fitness calculation module, a design parameter data acquisition module, and a simulation experiment verification module, specifically: The data acquisition module is used to acquire the design parameter data of the high-frequency transformer to be optimized and several standard historical performance data, and to acquire the historical transformer design parameter dataset and the historical transformer actual performance index dataset based on the standard historical performance data. The deep learning model training module is used to obtain a convolutional neural multilayer perceptual deep learning model based on the historical transformer design parameter dataset, the historical transformer actual performance index dataset, and a preset initial deep learning model. The particle fitness calculation module is used to obtain the initial particle population, several initial particle velocities, and several particle performance index data based on the initial design parameter data of the high-frequency transformer to be optimized and the convolutional neural multilayer perception deep learning model, and to obtain several predicted particle fitness based on the particle performance index data. The design parameter data acquisition module is used to update the initial particle population based on the particle swarm performance index data corresponding to the predicted particle fitness, the initial particle velocity corresponding to the predicted particle fitness, and the preset physical constraints if the predicted particle fitness is greater than the preset historical particle fitness, until the first preset convergence condition is met, and to acquire the first particle population, so as to acquire the optimization design parameter data of the high-frequency transformer to be optimized based on the first particle population. The simulation experiment verification module is used to obtain simulation data and experimental data based on the optimization design parameter data of the high-frequency transformer to be optimized, and to obtain the simulation experiment deviation value based on the simulation experiment data and experimental data. If the simulation experiment deviation value is less than or equal to the preset simulation experiment deviation threshold, the design of the high-frequency transformer is completed.

[0020] This invention provides a high-frequency transformer design system. In practical applications, it only requires a deep learning model training module. By training a pre-set initial deep learning model using historical transformer design parameter datasets and historical transformer actual performance index datasets, it can accurately map transformer design parameters and transformer performance index data. This allows the resulting convolutional neural multilayer perceptron deep learning model to balance prediction accuracy and computational efficiency, possessing highly efficient performance prediction capabilities and avoiding the problems of low accuracy and poor computational efficiency encountered when optimizing the design of high-frequency transformers. Then, a particle fitness calculation module is used to generate initial particle populations, initial particle velocities, and particle performance index data using the initial design parameter data of the high-frequency transformer to be optimized and the convolutional neural multilayer perceptron deep learning model, and calculates and predicts particle fitness. This provides a performance evaluation basis for optimizing the high-frequency transformer design parameters and improves computational efficiency in the early stages of optimization. Subsequently, a design parameter data acquisition module is employed. By comparing the predicted particle fitness with the preset historical particle fitness, and combining the initial particle velocity and preset physical constraints, the particle population is iteratively updated until the first preset convergence condition is met. This efficiently filters out the high-frequency transformer design parameters that meet the performance requirements, obtaining the optimized design parameter data for the high-frequency transformer to be optimized. This ensures the optimization accuracy of the design parameters while avoiding the lengthy calculation process of traditional simulations. Finally, a simulation experiment verification module is used. By acquiring simulation data and experimental data from the optimized design parameter data of the high-frequency transformer to be optimized, the simulation experiment deviation value is calculated, and it is determined whether the simulation experiment deviation value is less than or equal to the preset simulation experiment deviation threshold. This verifies the engineering feasibility of the optimized design parameters, ensuring that the optimized results have practical application value. Furthermore, the dual verification through simulation and experiment only in the final stage further guarantees the design accuracy, achieving high-precision and high-efficiency optimization of the high-frequency transformer design parameters.

[0021] Furthermore, the data acquisition module is used to acquire design parameter data of the high-frequency transformer to be optimized and several standard historical performance data, and to acquire a historical transformer design parameter dataset and a historical transformer actual performance index dataset based on the standard historical performance data, including: Obtain the design parameter data of the high-frequency transformer to be optimized and some initial historical performance data; The initial historical performance data is cleaned to obtain clean historical performance data; The purified historical performance data is normalized to obtain standard historical performance data; Historical transformer design parameter datasets and historical transformer actual performance index datasets are obtained based on standard historical performance data.

[0022] In the above scheme, by acquiring the design parameter data of the high-frequency transformer to be optimized and several initial historical performance data, the basic parameters of the optimized high-frequency transformer are clarified, and raw data for subsequent model training is provided. Next, by cleaning the initial historical performance data, invalid data is removed, improving the purity and effectiveness of the data and preventing poor-quality data from interfering with subsequent model training and parameter optimization. Then, by normalizing the clean historical performance data, performance data of different dimensions and scales can be transformed into a unified scale, eliminating the influence of dimensional differences between data, adapting to the training requirements of subsequent convolutional neural multilayer perceptron deep learning models, and obtaining standard historical performance data. Finally, by extracting the standard historical performance data, a historical transformer design parameter dataset and a historical transformer actual performance index dataset are obtained, achieving accurate matching between the design parameters and corresponding performance indicators of the high-frequency transformer. This provides standardized and suitable input data for the subsequent training of the deep learning model, ensuring that the model can effectively learn the correlation between design parameters and performance indicators.

[0023] Furthermore, the deep learning model training module is used to obtain a convolutional neural multilayer perceptron deep learning model based on a historical transformer design parameter dataset, a historical transformer actual performance index dataset, and a preset initial deep learning model, including: Based on a preset ratio, the historical transformer design parameter dataset is divided into a model building subset, a performance evaluation subset, and an independent verification subset. Based on the historical transformer actual performance index dataset, the model building subset, the performance evaluation subset, and the independent verification subset, the first performance index dataset corresponding to the model building subset, the second actual performance index dataset corresponding to the performance evaluation subset, and the third actual performance index dataset corresponding to the independent verification subset are obtained. Based on the model construction subset, performance evaluation subset, independent testing subset, first performance index dataset, second actual performance index dataset, third actual performance index dataset, and preset initial deep learning model, a convolutional neural multilayer perceptual deep learning model is obtained.

[0024] In the above scheme, the historical transformer design parameter dataset is divided into a model construction subset, a performance evaluation subset, and an independent verification subset according to a preset ratio. Corresponding first performance index dataset, second actual performance index dataset, and third actual performance index dataset are then matched to establish a hierarchical dataset system for the training, evaluation, and verification of the deep learning model, ensuring the independence and specificity of data at each stage of model training. Next, by training, evaluating, and verifying the model using the model construction subset, performance evaluation subset, independent verification subset, first performance index dataset, second actual performance index dataset, and third actual performance index dataset with the preset initial deep learning model, the parameters of the model are fitted, validated, and verified. Finally, a convolutional neural multilayer perceptron deep learning model with high-precision performance prediction capabilities is obtained, thus providing a reliable performance evaluation foundation for subsequent optimization of high-frequency transformer design parameters. Attached Figure Description

[0025] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 A flowchart illustrating a design method for a high-frequency transformer according to an embodiment of the present invention; Figure 2 This is an architecture diagram of a high-frequency transformer design system provided in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0029] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0030] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0031] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0032] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0033] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0034] See Figure 1To address the issues of low accuracy and poor computational efficiency in optimizing the design of high-frequency transformers, this embodiment provides a design method for high-frequency transformers, the flowchart of which can be found in the provided text. Figure 1 ,include: Step S1: Obtain the design parameter data of the high-frequency transformer to be optimized and several standard historical performance data, and obtain the historical transformer design parameter dataset and the historical transformer actual performance index dataset based on the standard historical performance data; Step S2: Based on the historical transformer design parameter dataset, the historical transformer actual performance index dataset, and the preset initial deep learning model, obtain the convolutional neural multilayer perceptual deep learning model; Step S3: Based on the initial design parameter data of the high-frequency transformer to be optimized and the convolutional neural multilayer perception deep learning model, obtain the initial particle population, the initial velocity of several particles and the performance index data of several particles, and obtain the fitness of several predicted particles based on the particle performance index data. Step S4: If the predicted particle fitness is greater than the preset historical particle fitness, the initial particle population is updated based on the particle swarm performance index data corresponding to the predicted particle fitness, the initial particle velocity corresponding to the predicted particle fitness, and the preset physical constraints until the first preset convergence condition is met, and the first particle population is obtained, so as to obtain the optimization design parameter data of the high-frequency transformer to be optimized based on the first particle population. Step S5: Based on the optimization design parameter data of the high-frequency transformer to be optimized, obtain simulation data and experimental data, and obtain the simulation experiment deviation value based on the simulation experiment data and experimental data. If the simulation experiment deviation value is less than or equal to the preset simulation experiment deviation threshold, the design of the high-frequency transformer is completed.

[0035] In this embodiment, a preset initial deep learning model is trained using a historical transformer design parameter dataset and a historical transformer actual performance index dataset. This allows for precise mapping between transformer design parameters and transformer performance index data, enabling the resulting convolutional neural multilayer perception deep learning model to balance prediction accuracy and computational efficiency. This provides a high-efficiency performance prediction capability, avoiding the problems of low accuracy and poor computational efficiency encountered when optimizing the design of high-frequency transformers. Then, using the initial design parameter data of the high-frequency transformer to be optimized and the convolutional neural multilayer perception deep learning model, an initial particle population, initial particle velocity, and particle performance index data are generated, and the predicted particle fitness is calculated. This provides a performance evaluation basis for optimizing the high-frequency transformer design parameters and improves computational efficiency in the early stages of optimization. Subsequently, by comparing the predicted particle fitness with the preset historical particle fitness, and combining the initial particle velocity and preset physical constraints, the particle population is iteratively updated until the first preset convergence condition is met. The optimal particle is selected from the first particle population as the optimized design parameter data for the high-frequency transformer to be optimized. Specifically, this is done using the formula... Where v is the particle's velocity vector and X is the particle's position vector. To preset dynamic inertia weights, which can decrease with the number of iterations, , Preset learning factors; , For preset random coefficients, through Update particle positions; if the updated parameters exceed the physically feasible range of the design parameters... , For initial design parameters of high-frequency transformers, interval projection mapping is used: This method can efficiently screen high-frequency transformer design parameters that meet performance requirements, ensuring the optimization accuracy of the design parameters while avoiding the lengthy calculation process of traditional simulations. The first preset convergence condition refers to the fact that the global fitness change rate in the current particle population is less than the preset fitness change threshold, or the maximum number of iterations of the particle population is reached, or the particle population has not shown significant improvement for M consecutive generations. Finally, by obtaining simulation data and experimental data through the optimized high-frequency transformer design parameter data, calculating the simulation experiment deviation value, and judging whether the simulation experiment deviation value is less than or equal to the preset simulation experiment deviation threshold, the engineering feasibility of the optimized design parameters can be verified, ensuring that the optimized results have practical application value. At the same time, the design accuracy is further guaranteed by dual verification through simulation and experiment only in the final stage, achieving high-precision and high-efficiency optimization of high-frequency transformer design parameters. Furthermore, the high-frequency transformer design method provided in this embodiment has good scalability and can be extended to the design of other power electronic devices to meet the ever-evolving technological needs.

[0036] Furthermore, the process of acquiring the design parameter data of the high-frequency transformer to be optimized and several standard historical performance data, and acquiring a historical transformer design parameter dataset and a historical transformer actual performance index dataset based on the standard historical performance data, includes: Obtain the design parameter data of the high-frequency transformer to be optimized and some initial historical performance data; The initial historical performance data is cleaned to obtain clean historical performance data; The purified historical performance data is normalized to obtain standard historical performance data; Historical transformer design parameter datasets and historical transformer actual performance index datasets are obtained based on standard historical performance data.

[0037] In this embodiment, by acquiring the design parameter data of the high-frequency transformer to be optimized and extracting some initial historical performance data from the monitoring system of the high-frequency transformer in operation or from a historical database, the basic parameters of the optimized high-frequency transformer are clarified, and raw data for subsequent model training is provided. The initial historical performance data includes temperature rise, losses (copper loss, iron loss, dielectric loss), efficiency, power density, etc. Next, the initial historical performance data is cleaned. The data cleaning includes handling missing values ​​and outliers. Specifically, statistical analysis methods are used to detect and delete data that does not meet the specifications, such as loss data that exceeds the normal range, thereby eliminating invalid data, improving the purity and effectiveness of the data, and avoiding interference from poor-quality data to subsequent model training and parameter optimization. Then, by normalizing the clean historical performance data, performance data of different dimensions and scales can be transformed into a unified scale. Z-score standardization is usually used to eliminate the influence of differences in dimensions between data, adapting to the training requirements of subsequent convolutional neural multilayer perceptron deep learning models, and obtaining standard historical performance data. Finally, historical transformer design parameter datasets and historical transformer actual performance index datasets were extracted from standard historical performance data. This enabled precise matching of design parameters and corresponding performance indices for high-frequency transformers, providing standardized and suitable input data for subsequent deep learning model training. This ensures that the model can effectively learn the correlation between design parameters and performance indices. The transformer design parameter data includes winding structural parameters (number of turns, copper thickness, skin depth, etc.), core parameters (core material, permeability, air gap size, etc.), and external working environment influencing factors (temperature, humidity, etc.).

[0038] Furthermore, the step of obtaining a convolutional neural multilayer perceptron deep learning model based on historical transformer design parameter datasets, historical transformer actual performance index datasets, and a preset initial deep learning model includes: Based on a preset ratio, the historical transformer design parameter dataset is divided into a model building subset, a performance evaluation subset, and an independent verification subset. Based on the historical transformer actual performance index dataset, the model building subset, the performance evaluation subset, and the independent verification subset, the first performance index dataset corresponding to the model building subset, the second actual performance index dataset corresponding to the performance evaluation subset, and the third actual performance index dataset corresponding to the independent verification subset are obtained. Based on the model construction subset, performance evaluation subset, independent testing subset, first performance index dataset, second actual performance index dataset, third actual performance index dataset, and preset initial deep learning model, a convolutional neural multilayer perceptual deep learning model is obtained.

[0039] In this embodiment, the historical transformer design parameter dataset is divided into a model construction subset, a performance evaluation subset, and an independent verification subset according to a preset ratio. Corresponding first performance index dataset, second actual performance index dataset, and third actual performance index dataset are then matched to establish a hierarchical dataset system for the training, evaluation, and verification of the deep learning model, ensuring the independence and specificity of data at each stage of model training. Next, by training, evaluating, and verifying the model using the model construction subset, performance evaluation subset, independent verification subset, first performance index dataset, second actual performance index dataset, and third actual performance index dataset with a preset initial deep learning model, the parameters of the model are fitted, validated, and verified. Finally, a convolutional neural multilayer perceptron deep learning model with high-precision performance prediction capabilities is obtained, thus providing a reliable performance evaluation foundation for subsequent optimization of high-frequency transformer design parameters. The preset initial deep learning model is pre-built based on convolutional neural networks (CNN) and multilayer perceptrons (MLP). The CNN part is responsible for extracting the spatial features between transformer design parameters, such as the relationship between winding structure and temperature rise distribution, while the MLP part is used to learn the nonlinear relationship between design parameters and transformer performance (temperature rise, loss, etc.). By jointly training data related to multiple physical fields (such as electromagnetic loss, heat conduction, mechanical stress, etc.), the complex relationship between them can be automatically mined to achieve comprehensive optimization of multiple physical fields and provide a more comprehensive design solution.

[0040] Furthermore, the step of obtaining a convolutional neural multilayer perceptron deep learning model based on a model construction subset, a performance evaluation subset, an independent testing subset, a first performance index dataset, a second actual performance index dataset, a third actual performance index dataset, and a preset initial deep learning model includes: Based on the model construction subset and the preset initial deep learning model, obtain the training performance index dataset corresponding to the model construction subset; An error function is constructed based on the first actual performance index dataset and the training performance index dataset. The preset initial deep learning model is updated by minimizing the error function until the second preset convergence condition is met, and the deep learning training model is obtained. Based on the performance evaluation subset and the deep learning training model, obtain the validation performance metric dataset corresponding to the performance evaluation subset; The model deviation fluctuation value is obtained based on the second actual performance index dataset and the validation performance index dataset. If the model deviation fluctuation value is less than the preset model deviation fluctuation threshold, the deep learning validation model is obtained. Based on the independent test subset and the deep learning verification model, obtain the test performance index dataset corresponding to the independent test subset; The indicator response curve is obtained based on the third actual performance indicator dataset and the test performance indicator dataset. If the indicator response curve meets the preset accuracy standard condition, the convolutional neural multilayer perceptual deep learning model is obtained.

[0041] In this embodiment, by inputting a subset of the model construction into a preset initial deep learning model to obtain a training performance index dataset corresponding to the subset, a prediction result reference can be provided for subsequent model parameter updates and error calculation. Next, an error function is constructed using the first actual performance index dataset and the training performance index dataset. The preset initial deep learning model is iteratively updated with the goal of minimizing the error function until a second preset convergence condition is met. This allows the model to initially learn the correlation between design parameters and performance indices, completing parameter fitting and optimization, and obtaining a deep learning training model. The second preset convergence condition refers to the convergence of the error function or reaching a preset training termination condition. Then, by inputting a performance evaluation subset into the deep learning training model for performance prediction, a validation performance index dataset corresponding to the performance evaluation subset is obtained, providing data support for verifying the model's prediction stability. Subsequently, the model bias fluctuation value is calculated using the second actual performance index dataset and the validation performance index dataset. Specifically, a rotating sample participation mechanism is used, allowing samples from different sub-regions to participate in the evaluation sequentially. This detects the model's prediction consistency across different design intervals and determines whether the model bias fluctuation value is less than a preset model bias fluctuation threshold, thus avoiding excessive local prediction bias and completing the evaluation of the deep learning training model. The model deviation fluctuation value specifically refers to the distribution of model deviation in different design subspaces. Next, by inputting an independent validation subset into the deep learning validation model for performance testing, a test performance index dataset corresponding to the independent validation subset is obtained. This provides experimental prediction data for verifying the generalization ability of the deep learning validation model to unknown samples. Finally, an index response curve is generated using the third actual performance index dataset and the test performance index dataset, and it is verified whether the index response curve meets the preset accuracy standard conditions. This completes the verification of the model's generalization ability and overall prediction accuracy, ultimately obtaining a convolutional neural multilayer perceptual deep learning model with both stable prediction ability and high-precision generalization ability, providing a reliable performance prediction foundation for subsequent optimization of high-frequency transformer design parameters.

[0042] Furthermore, the process of obtaining initial particle population, initial particle velocities, and particle performance index data based on the initial design parameter data of the high-frequency transformer to be optimized and the convolutional neural multilayer perceptron deep learning model, and obtaining several predicted particle fitness values ​​based on the particle performance index data, includes: An initial particle population and several initial velocities of particles are generated based on the initial design parameter data of the high-frequency transformer to be optimized. Several particle performance index data are obtained based on the convolutional neural multilayer perceptual deep learning model and the initial particle population, and several predicted particle fitnesss are obtained based on the particle performance index data and the preset multi-objective weighted fitness function.

[0043] In this embodiment, the initial design parameters of the high-frequency transformer to be optimized are represented as a continuous variable vector. ,in, Indicates the number of turns on the primary side. Indicates the number of turns on the secondary side. Indicates copper thickness. Indicates the air gap length. Indicates the window width. Indicates the window height. This represents the magnetic permeability and presupposes a physically feasible range for each design parameter. ,in for The feasible lower bound, of which for The feasible upper limit is determined. An initial particle swarm and initial velocities of several particles are generated using the initial design parameters of the high-frequency transformer to be optimized. This transforms the optimization problem of the high-frequency transformer's design parameters into an optimization problem using a particle swarm optimization algorithm, thus establishing an initial parameter framework for the iterative optimization of the particle swarm optimization algorithm. Next, several particle performance index data are obtained by inputting the initial particle swarm into a convolutional neural multilayer perceptron deep learning model. Then, several predicted particle fitness values ​​are calculated using a preset multi-objective weighted fitness function, specifically: the position vector of the particle in the initial particle swarm. Inputting the convolutional neural multilayer perceptron deep learning model yields particle performance index data. , This indicates the temperature rise of the core components of a high-frequency transformer, such as the windings and core, relative to the ambient temperature under rated operating conditions. This represents the total power loss generated by the high-frequency transformer during operation. The energy transfer efficiency of the high-frequency transformer is represented by the predicted particle fitness calculated using a preset multi-objective weighted fitness function. ,in, , , Indicates the weighting coefficient. This indicates the temperature rise limit of a high-frequency transformer. As a preset value, This represents the temperature rise constraint penalty term, thereby enabling rapid and accurate prediction of the design parameter performance using a convolutional neural multilayer perceptual deep learning model. At the same time, the performance of particles is evaluated by a preset multi-objective weighted fitness function, providing a basis for updating the initial particle population. This ensures the optimization accuracy of the design parameters while avoiding the lengthy calculation process of traditional simulations.

[0044] Further, the step of obtaining simulation data and experimental data based on the optimization design parameter data of the high-frequency transformer to be optimized, and obtaining the simulation experiment deviation value based on the simulation experiment data and experimental data, and completing the design of the high-frequency transformer if the simulation experiment deviation value is less than or equal to a preset simulation experiment deviation threshold, includes: The optimization design parameters of the high-frequency transformer to be optimized are simulated to obtain simulation data; Based on the optimized design parameter data of the high-frequency transformer to be optimized, a high-frequency transformer prototype was obtained, and experiments were conducted on the high-frequency transformer prototype to obtain experimental data. The simulation experiment deviation value is obtained based on the simulation experiment data and the experimental data. If the simulation experiment deviation value is less than or equal to the preset simulation experiment deviation threshold, the design of the high-frequency transformer is completed.

[0045] In this embodiment, finite element analysis (FEA) software (such as COMSOL, ANSYS Maxwell, etc.) is used to perform detailed 3D modeling of the optimized design parameters of the high-frequency transformer to be optimized. This simulates the operating state of the high-frequency transformer under different load conditions. Simulations are conducted using the optimized design parameters, including temperature rise distribution, current density distribution, and magnetic field strength, with a focus on high-frequency losses, thermal effects, and nonlinear factors such as core saturation. The obtained simulation data theoretically verifies the transformer performance corresponding to the optimized design parameters, providing theoretical reference and a benchmark for subsequent experimental verification. Next, a high-frequency transformer prototype is fabricated using the optimized design parameters, and experiments are conducted. Temperature rise, losses, and efficiency are measured on an experimental platform, obtaining experimental data. This verifies the feasibility and actual performance of the optimized design parameters from an engineering perspective, overcoming the theoretical limitations of pure simulation analysis. Finally, the simulation-experiment deviation value is calculated using both simulation and experimental data, and it is determined whether the deviation value is less than or equal to a preset deviation threshold. This achieves dual verification of the optimized design parameters, ensuring that the optimized design parameters meet the requirements of engineering practicality and performance accuracy.

[0046] Furthermore, the step of obtaining simulation data and experimental data based on the optimization design parameter data of the high-frequency transformer to be optimized, and obtaining the simulation experiment deviation value based on the simulation experiment data and experimental data, and completing the design of the high-frequency transformer if the simulation experiment deviation value is less than or equal to a preset simulation experiment deviation threshold, further includes: Based on the optimization design parameter data of the high-frequency transformer to be optimized, simulation data and experimental data are obtained, and simulation experimental deviation values ​​are obtained based on the simulation experimental data and experimental data. If the simulation experimental deviation value is greater than the preset simulation experimental deviation threshold, the convolutional neural multilayer perceptual deep learning model is updated based on the experimental data until the simulation experimental deviation value is less than or equal to the preset simulation experimental deviation threshold.

[0047] In this embodiment, when the simulation experiment deviation exceeds a preset simulation experiment deviation threshold, the parameters of the convolutional neural multilayer perception deep learning model are updated using experimental data. The model is then iteratively optimized and its accuracy calibrated again to compensate for the deficiencies in the initial training, making the newly obtained convolutional neural multilayer perception deep learning model more closely aligned with actual engineering scenarios. Then, the convolutional neural multilayer perception deep learning model is continuously iterated and updated until the simulation experiment deviation meets the preset simulation experiment deviation threshold, ensuring that the final optimized design parameters for the high-frequency transformer meet both the requirements of engineering practicality and performance accuracy.

[0048] This embodiment provides a design system for high-frequency transformers, such as... Figure 2 As shown, it includes a data acquisition module, a deep learning model training module, a particle fitness calculation module, a design parameter data acquisition module, and a simulation experiment verification module, specifically: The data acquisition module is used to acquire the design parameter data of the high-frequency transformer to be optimized and several standard historical performance data, and to acquire the historical transformer design parameter dataset and the historical transformer actual performance index dataset based on the standard historical performance data. The deep learning model training module is used to obtain a convolutional neural multilayer perceptual deep learning model based on the historical transformer design parameter dataset, the historical transformer actual performance index dataset, and a preset initial deep learning model. The particle fitness calculation module is used to obtain the initial particle population, several initial particle velocities, and several particle performance index data based on the initial design parameter data of the high-frequency transformer to be optimized and the convolutional neural multilayer perception deep learning model, and to obtain several predicted particle fitness based on the particle performance index data. The design parameter data acquisition module is used to update the initial particle population based on the particle swarm performance index data corresponding to the predicted particle fitness, the initial particle velocity corresponding to the predicted particle fitness, and the preset physical constraints if the predicted particle fitness is greater than the preset historical particle fitness, until the first preset convergence condition is met, and to acquire the first particle population, so as to acquire the optimization design parameter data of the high-frequency transformer to be optimized based on the first particle population. The simulation experiment verification module is used to obtain simulation data and experimental data based on the optimization design parameter data of the high-frequency transformer to be optimized, and to obtain the simulation experiment deviation value based on the simulation experiment data and experimental data. If the simulation experiment deviation value is less than or equal to the preset simulation experiment deviation threshold, the design of the high-frequency transformer is completed.

[0049] This embodiment provides a high-frequency transformer design system. In practical applications, it only requires a deep learning model training module. By training a pre-set initial deep learning model using historical transformer design parameter datasets and historical transformer actual performance index datasets, it can accurately map transformer design parameters and transformer performance index data. This allows the resulting convolutional neural multilayer perceptron deep learning model to balance prediction accuracy and computational efficiency, possessing high-efficiency performance prediction capabilities and avoiding the problems of low accuracy and poor computational efficiency encountered when optimizing the design of high-frequency transformers. Then, a particle fitness calculation module is used to generate initial particle populations, initial particle velocities, and particle performance index data using the initial design parameter data of the high-frequency transformer to be optimized and the convolutional neural multilayer perceptron deep learning model, and calculates the predicted particle fitness. This provides a performance evaluation basis for optimizing the high-frequency transformer design parameters and improves the computational efficiency in the early stages of optimization. Subsequently, a design parameter data acquisition module is employed. By comparing the predicted particle fitness with the preset historical particle fitness, and combining the initial particle velocity and preset physical constraints, the particle population is iteratively updated until the first preset convergence condition is met. This efficiently filters out the high-frequency transformer design parameters that meet the performance requirements, obtaining the optimized design parameter data for the high-frequency transformer to be optimized. This ensures the optimization accuracy of the design parameters while avoiding the lengthy calculation process of traditional simulations. Finally, a simulation experiment verification module is used. By acquiring simulation data and experimental data from the optimized design parameter data of the high-frequency transformer to be optimized, the simulation experiment deviation value is calculated, and it is determined whether the simulation experiment deviation value is less than or equal to the preset simulation experiment deviation threshold. This verifies the engineering feasibility of the optimized design parameters, ensuring that the optimized results have practical application value. Furthermore, the dual verification through simulation and experiment only in the final stage further guarantees the design accuracy, achieving high-precision and high-efficiency optimization of the high-frequency transformer design parameters.

[0050] Furthermore, the data acquisition module is used to acquire design parameter data of the high-frequency transformer to be optimized and several standard historical performance data, and to acquire a historical transformer design parameter dataset and a historical transformer actual performance index dataset based on the standard historical performance data, including: Obtain the design parameter data of the high-frequency transformer to be optimized and some initial historical performance data; The initial historical performance data is cleaned to obtain clean historical performance data; The purified historical performance data is normalized to obtain standard historical performance data; Historical transformer design parameter datasets and historical transformer actual performance index datasets are obtained based on standard historical performance data.

[0051] In this embodiment, by acquiring the design parameter data of the high-frequency transformer to be optimized and several initial historical performance data, the basic parameters of the optimized high-frequency transformer are clarified, and raw data for subsequent model training is provided. Next, by cleaning the initial historical performance data, invalid data is removed, improving the purity and effectiveness of the data and preventing poor-quality data from interfering with subsequent model training and parameter optimization. Then, by normalizing the clean historical performance data, performance data of different dimensions and scales can be transformed into a unified scale, eliminating the influence of dimensional differences between data and adapting to the training requirements of the subsequent convolutional neural multilayer perceptron deep learning model, resulting in standard historical performance data. Finally, by extracting the standard historical performance data, a historical transformer design parameter dataset and a historical transformer actual performance index dataset are obtained, achieving accurate matching between the design parameters and corresponding performance indicators of the high-frequency transformer. This provides standardized and suitable input data for the subsequent training of the deep learning model, ensuring that the model can effectively learn the correlation between design parameters and performance indicators.

[0052] Furthermore, the deep learning model training module is used to obtain a convolutional neural multilayer perceptron deep learning model based on a historical transformer design parameter dataset, a historical transformer actual performance index dataset, and a preset initial deep learning model, including: Based on a preset ratio, the historical transformer design parameter dataset is divided into a model building subset, a performance evaluation subset, and an independent verification subset. Based on the historical transformer actual performance index dataset, the model building subset, the performance evaluation subset, and the independent verification subset, the first performance index dataset corresponding to the model building subset, the second actual performance index dataset corresponding to the performance evaluation subset, and the third actual performance index dataset corresponding to the independent verification subset are obtained. Based on the model construction subset, performance evaluation subset, independent testing subset, first performance index dataset, second actual performance index dataset, third actual performance index dataset, and preset initial deep learning model, a convolutional neural multilayer perceptual deep learning model is obtained.

[0053] In this embodiment, the historical transformer design parameter dataset is divided into a model construction subset, a performance evaluation subset, and an independent verification subset according to a preset ratio. Corresponding first performance index dataset, second actual performance index dataset, and third actual performance index dataset are then matched to establish a hierarchical dataset system for the training, evaluation, and verification of the deep learning model, ensuring the independence and specificity of data at each stage of model training. Next, by training, evaluating, and verifying the model using the model construction subset, performance evaluation subset, independent verification subset, first performance index dataset, second actual performance index dataset, and third actual performance index dataset with a preset initial deep learning model, the parameters of the model are fitted, validated, and verified. Finally, a convolutional neural multilayer perceptron deep learning model with high-precision performance prediction capabilities is obtained, thus providing a reliable performance evaluation foundation for subsequent optimization of high-frequency transformer design parameters.

[0054] Furthermore, the step of obtaining a convolutional neural multilayer perceptron deep learning model based on a model construction subset, a performance evaluation subset, an independent testing subset, a first performance index dataset, a second actual performance index dataset, a third actual performance index dataset, and a preset initial deep learning model includes: Based on the model construction subset and the preset initial deep learning model, obtain the training performance index dataset corresponding to the model construction subset; An error function is constructed based on the first actual performance index dataset and the training performance index dataset. The preset initial deep learning model is updated by minimizing the error function until the second preset convergence condition is met, and the deep learning training model is obtained. Based on the performance evaluation subset and the deep learning training model, obtain the validation performance metric dataset corresponding to the performance evaluation subset; The model deviation fluctuation value is obtained based on the second actual performance index dataset and the validation performance index dataset. If the model deviation fluctuation value is less than the preset model deviation fluctuation threshold, the deep learning validation model is obtained. Based on the independent test subset and the deep learning verification model, obtain the test performance index dataset corresponding to the independent test subset; The indicator response curve is obtained based on the third actual performance indicator dataset and the test performance indicator dataset. If the indicator response curve meets the preset accuracy standard condition, the convolutional neural multilayer perceptual deep learning model is obtained.

[0055] In this embodiment, by inputting a subset of the model construction into a preset initial deep learning model to obtain a training performance index dataset corresponding to the subset, a prediction result reference can be provided for subsequent model parameter updates and error calculation. Next, an error function is constructed using the first actual performance index dataset and the training performance index dataset. The preset initial deep learning model is iteratively updated with the goal of minimizing the error function until a second preset convergence condition is met. This allows the model to initially learn the correlation between design parameters and performance indices, completing parameter fitting and optimization, and obtaining a deep learning training model. The second preset convergence condition refers to the convergence of the error function or reaching a preset training termination condition. Then, by inputting a performance evaluation subset into the deep learning training model for performance prediction, a validation performance index dataset corresponding to the performance evaluation subset is obtained, providing data support for verifying the model's prediction stability. Subsequently, the model bias fluctuation value is calculated using the second actual performance index dataset and the validation performance index dataset. Specifically, a rotating sample participation mechanism is used, allowing samples from different sub-regions to participate in the evaluation sequentially. This detects the model's prediction consistency across different design intervals and determines whether the model bias fluctuation value is less than a preset model bias fluctuation threshold, thus avoiding excessive local prediction bias and completing the evaluation of the deep learning training model. The model deviation fluctuation value specifically refers to the distribution of model deviation in different design subspaces. Next, by inputting an independent validation subset into the deep learning validation model for performance testing, a test performance index dataset corresponding to the independent validation subset is obtained. This provides experimental prediction data for verifying the generalization ability of the deep learning validation model to unknown samples. Finally, an index response curve is generated using the third actual performance index dataset and the test performance index dataset, and it is verified whether the index response curve meets the preset accuracy standard conditions. This completes the verification of the model's generalization ability and overall prediction accuracy, ultimately obtaining a convolutional neural multilayer perceptual deep learning model with both stable prediction ability and high-precision generalization ability, providing a reliable performance prediction foundation for subsequent optimization of high-frequency transformer design parameters.

[0056] Furthermore, the particle fitness calculation module is used to obtain an initial particle population, several initial particle velocities, and several particle performance index data based on the initial design parameter data of the high-frequency transformer to be optimized and the convolutional neural multilayer perceptron deep learning model, and to obtain several predicted particle fitness values ​​based on the particle performance index data, including: An initial particle population and several initial velocities of particles are generated based on the initial design parameter data of the high-frequency transformer to be optimized. Several particle performance index data are obtained based on the convolutional neural multilayer perceptual deep learning model and the initial particle population, and several predicted particle fitnesss are obtained based on the particle performance index data and the preset multi-objective weighted fitness function.

[0057] In this embodiment, the initial design parameters of the high-frequency transformer to be optimized are represented as a continuous variable vector. ,in, Indicates the number of turns on the primary side. Indicates the number of turns on the secondary side. Indicates copper thickness. Indicates the air gap length. Indicates the window width. Indicates the window height. This represents the magnetic permeability and presupposes a physically feasible range for each design parameter. ,in for The feasible lower bound, of which for The feasible upper limit is determined. An initial particle swarm and initial velocities of several particles are generated using the initial design parameters of the high-frequency transformer to be optimized. This transforms the optimization problem of the high-frequency transformer's design parameters into an optimization problem using a particle swarm optimization algorithm, thus establishing an initial parameter framework for the iterative optimization of the particle swarm optimization algorithm. Next, several particle performance index data are obtained by inputting the initial particle swarm into a convolutional neural multilayer perceptron deep learning model. Then, several predicted particle fitness values ​​are calculated using a preset multi-objective weighted fitness function, specifically: the position vector of the particle in the initial particle swarm. Inputting the convolutional neural multilayer perceptron deep learning model yields particle performance index data. , This indicates the temperature rise of the core components of a high-frequency transformer, such as the windings and core, relative to the ambient temperature under rated operating conditions. This represents the total power loss generated by the high-frequency transformer during operation. The energy transfer efficiency of the high-frequency transformer is represented by the predicted particle fitness calculated using a preset multi-objective weighted fitness function. ,in, , , Indicates the weighting coefficient. This indicates the temperature rise limit of a high-frequency transformer. As a preset value, This represents the temperature rise constraint penalty term, thereby enabling rapid and accurate prediction of the design parameter performance using a convolutional neural multilayer perceptual deep learning model. At the same time, the performance of particles is evaluated by a preset multi-objective weighted fitness function, providing a basis for updating the initial particle population. This ensures the optimization accuracy of the design parameters while avoiding the lengthy calculation process of traditional simulations.

[0058] Furthermore, the simulation experiment verification module is used to acquire simulation data and experimental data based on the optimization design parameter data of the high-frequency transformer to be optimized, and to acquire the simulation experiment deviation value based on the simulation experiment data and experimental data. If the simulation experiment deviation value is less than or equal to a preset simulation experiment deviation threshold, the design of the high-frequency transformer is completed, including: The optimization design parameters of the high-frequency transformer to be optimized are simulated to obtain simulation data; Based on the optimized design parameter data of the high-frequency transformer to be optimized, a high-frequency transformer prototype was obtained, and experiments were conducted on the high-frequency transformer prototype to obtain experimental data. The simulation experiment deviation value is obtained based on the simulation experiment data and the experimental data. If the simulation experiment deviation value is less than or equal to the preset simulation experiment deviation threshold, the design of the high-frequency transformer is completed.

[0059] In this embodiment, finite element analysis (FEA) software (such as COMSOL, ANSYS Maxwell, etc.) is used to perform detailed 3D modeling of the optimized design parameters of the high-frequency transformer to be optimized. This simulates the operating state of the high-frequency transformer under different load conditions. Simulations are conducted using the optimized design parameters, including temperature rise distribution, current density distribution, and magnetic field strength, with a focus on high-frequency losses, thermal effects, and nonlinear factors such as core saturation. The obtained simulation data theoretically verifies the transformer performance corresponding to the optimized design parameters, providing theoretical reference and a benchmark for subsequent experimental verification. Next, a high-frequency transformer prototype is fabricated using the optimized design parameters, and experiments are conducted. Temperature rise, losses, and efficiency are measured on an experimental platform, obtaining experimental data. This verifies the feasibility and actual performance of the optimized design parameters from an engineering perspective, overcoming the theoretical limitations of pure simulation analysis. Finally, the simulation-experiment deviation value is calculated using both simulation and experimental data, and it is determined whether the deviation value is less than or equal to a preset deviation threshold. This achieves dual verification of the optimized design parameters, ensuring that the optimized design parameters meet the requirements of engineering practicality and performance accuracy.

[0060] Furthermore, the simulation experiment verification module is also used to obtain simulation data and experimental data based on the optimization design parameter data of the high-frequency transformer to be optimized, and to obtain the simulation experiment deviation value based on the simulation experiment data and experimental data. If the simulation experiment deviation value is greater than the preset simulation experiment deviation threshold, the convolutional neural multilayer perceptual deep learning model is updated based on the experimental data until the simulation experiment deviation value is less than or equal to the preset simulation experiment deviation threshold.

[0061] In this embodiment, when the simulation experiment deviation exceeds a preset simulation experiment deviation threshold, the parameters of the convolutional neural multilayer perception deep learning model are updated using experimental data. The model is then iteratively optimized and its accuracy calibrated again to compensate for the deficiencies in the initial training, making the newly obtained convolutional neural multilayer perception deep learning model more closely aligned with actual engineering scenarios. Then, the convolutional neural multilayer perception deep learning model is continuously iterated and updated until the simulation experiment deviation meets the preset simulation experiment deviation threshold, ensuring that the final optimized design parameters for the high-frequency transformer meet both the requirements of engineering practicality and performance accuracy.

[0062] This embodiment also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the functions of the system as described above.

[0063] It is understood that the above system embodiments correspond to the method embodiments of the present invention, and can implement the design method of a high-frequency transformer provided by any of the above method embodiments of the present invention.

[0064] It should be noted that the system embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0065] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A design method for a high-frequency transformer, characterized in that, include: Obtain design parameter data of the high-frequency transformer to be optimized and several standard historical performance data, and obtain historical transformer design parameter dataset and historical transformer actual performance index dataset based on the standard historical performance data; Based on the historical transformer design parameter dataset, the historical transformer actual performance index dataset, and the preset initial deep learning model, a convolutional neural multilayer perceptual deep learning model is obtained. Based on the initial design parameter data of the high-frequency transformer to be optimized and the convolutional neural multilayer perception deep learning model, the initial particle population, the initial velocity of several particles and the performance index data of several particles are obtained, and the fitness of several predicted particles is obtained based on the particle performance index data. If the predicted particle fitness is greater than the preset historical particle fitness, the initial particle population is updated based on the particle swarm performance index data corresponding to the predicted particle fitness, the initial particle velocity corresponding to the predicted particle fitness, and the preset physical constraints until the first preset convergence condition is met, and the first particle population is obtained, so as to obtain the optimization design parameter data of the high-frequency transformer to be optimized based on the first particle population. Based on the optimization design parameter data of the high-frequency transformer to be optimized, simulation data and experimental data are obtained, and simulation experimental deviation values ​​are obtained based on the simulation experimental data and experimental data. If the simulation experimental deviation value is less than or equal to the preset simulation experimental deviation threshold, the design of the high-frequency transformer is completed.

2. The design method for a high-frequency transformer according to claim 1, characterized in that, The process of acquiring the design parameter data of the high-frequency transformer to be optimized and several standard historical performance data, and acquiring a historical transformer design parameter dataset and a historical transformer actual performance index dataset based on the standard historical performance data, includes: Obtain the design parameter data of the high-frequency transformer to be optimized and some initial historical performance data; The initial historical performance data is cleaned to obtain clean historical performance data; The purified historical performance data is normalized to obtain standard historical performance data; Historical transformer design parameter datasets and historical transformer actual performance index datasets are obtained based on standard historical performance data.

3. The design method for a high-frequency transformer according to claim 1, characterized in that, The process of obtaining a convolutional neural multilayer perceptron deep learning model based on historical transformer design parameter datasets, historical transformer actual performance index datasets, and a pre-set initial deep learning model includes: Based on a preset ratio, the historical transformer design parameter dataset is divided into a model building subset, a performance evaluation subset, and an independent verification subset. Based on the historical transformer actual performance index dataset, the model building subset, the performance evaluation subset, and the independent verification subset, the first performance index dataset corresponding to the model building subset, the second actual performance index dataset corresponding to the performance evaluation subset, and the third actual performance index dataset corresponding to the independent verification subset are obtained. Based on the model construction subset, performance evaluation subset, independent testing subset, first performance index dataset, second actual performance index dataset, third actual performance index dataset, and preset initial deep learning model, a convolutional neural multilayer perceptual deep learning model is obtained.

4. The design method for a high-frequency transformer according to claim 3, characterized in that, The process of obtaining a convolutional neural multilayer perceptron deep learning model based on a model construction subset, a performance evaluation subset, an independent testing subset, a first performance index dataset, a second actual performance index dataset, a third actual performance index dataset, and a preset initial deep learning model includes: Based on the model construction subset and the preset initial deep learning model, obtain the training performance index dataset corresponding to the model construction subset; An error function is constructed based on the first actual performance index dataset and the training performance index dataset. The preset initial deep learning model is updated by minimizing the error function until the second preset convergence condition is met, and the deep learning training model is obtained. Based on the performance evaluation subset and the deep learning training model, obtain the validation performance metric dataset corresponding to the performance evaluation subset; The model deviation fluctuation value is obtained based on the second actual performance index dataset and the validation performance index dataset. If the model deviation fluctuation value is less than the preset model deviation fluctuation threshold, the deep learning validation model is obtained. Based on the independent test subset and the deep learning verification model, obtain the test performance index dataset corresponding to the independent test subset; The indicator response curve is obtained based on the third actual performance indicator dataset and the test performance indicator dataset. If the indicator response curve meets the preset accuracy standard condition, the convolutional neural multilayer perceptual deep learning model is obtained.

5. The design method for a high-frequency transformer according to claim 1, characterized in that, The process involves obtaining initial particle population, initial particle velocities, and performance index data based on the initial design parameter data of the high-frequency transformer to be optimized and a convolutional neural multilayer perceptron deep learning model. Based on the particle performance index data, several predicted particle fitness values ​​are then obtained, including: An initial particle population and several initial velocities of particles are generated based on the initial design parameter data of the high-frequency transformer to be optimized. Several particle performance index data are obtained based on the convolutional neural multilayer perceptual deep learning model and the initial particle population, and several predicted particle fitnesss are obtained based on the particle performance index data and the preset multi-objective weighted fitness function.

6. The design method for a high-frequency transformer according to claim 1, characterized in that, The process involves obtaining simulation and experimental data based on the optimized design parameters of the high-frequency transformer to be optimized, and obtaining simulation deviation values ​​based on the simulation and experimental data. If the simulation deviation value is less than or equal to a preset simulation deviation threshold, the design of the high-frequency transformer is completed. This includes: The optimization design parameters of the high-frequency transformer to be optimized are simulated to obtain simulation data; Based on the optimized design parameter data of the high-frequency transformer to be optimized, a high-frequency transformer prototype was obtained, and experiments were conducted on the high-frequency transformer prototype to obtain experimental data. The simulation experiment deviation value is obtained based on the simulation experiment data and the experimental data. If the simulation experiment deviation value is less than or equal to the preset simulation experiment deviation threshold, the design of the high-frequency transformer is completed.

7. The design method for a high-frequency transformer according to claim 1, characterized in that, The process of obtaining simulation data and experimental data based on the optimization design parameter data of the high-frequency transformer to be optimized, and obtaining the simulation experiment deviation value based on the simulation experiment data and experimental data, and completing the design of the high-frequency transformer if the simulation experiment deviation value is less than or equal to a preset simulation experiment deviation threshold, further includes: Based on the optimization design parameter data of the high-frequency transformer to be optimized, simulation data and experimental data are obtained, and simulation experimental deviation values ​​are obtained based on the simulation experimental data and experimental data. If the simulation experimental deviation value is greater than the preset simulation experimental deviation threshold, the convolutional neural multilayer perceptual deep learning model is updated based on the experimental data until the simulation experimental deviation value is less than or equal to the preset simulation experimental deviation threshold.

8. A design system for a high-frequency transformer, characterized in that, It includes a data acquisition module, a deep learning model training module, a particle fitness calculation module, a design parameter data acquisition module, and a simulation experiment verification module, specifically: The data acquisition module is used to acquire the design parameter data of the high-frequency transformer to be optimized and several standard historical performance data, and to acquire the historical transformer design parameter dataset and the historical transformer actual performance index dataset based on the standard historical performance data. The deep learning model training module is used to obtain a convolutional neural multilayer perceptual deep learning model based on the historical transformer design parameter dataset, the historical transformer actual performance index dataset, and a preset initial deep learning model. The particle fitness calculation module is used to obtain the initial particle population, several initial particle velocities, and several particle performance index data based on the initial design parameter data of the high-frequency transformer to be optimized and the convolutional neural multilayer perception deep learning model, and to obtain several predicted particle fitness based on the particle performance index data. The design parameter data acquisition module is used to update the initial particle population based on the particle swarm performance index data corresponding to the predicted particle fitness, the initial particle velocity corresponding to the predicted particle fitness, and the preset physical constraints if the predicted particle fitness is greater than the preset historical particle fitness, until the first preset convergence condition is met, and to acquire the first particle population, so as to acquire the optimization design parameter data of the high-frequency transformer to be optimized based on the first particle population. The simulation experiment verification module is used to obtain simulation data and experimental data based on the optimization design parameter data of the high-frequency transformer to be optimized, and to obtain the simulation experiment deviation value based on the simulation experiment data and experimental data. If the simulation experiment deviation value is less than or equal to the preset simulation experiment deviation threshold, the design of the high-frequency transformer is completed.

9. A design system for a high-frequency transformer according to claim 8, characterized in that, The data acquisition module is used to acquire design parameter data of the high-frequency transformer to be optimized and several standard historical performance data, and to acquire a historical transformer design parameter dataset and a historical transformer actual performance index dataset based on the standard historical performance data, including: Obtain the design parameter data of the high-frequency transformer to be optimized and some initial historical performance data; The initial historical performance data is cleaned to obtain clean historical performance data; The purified historical performance data is normalized to obtain standard historical performance data; Historical transformer design parameter datasets and historical transformer actual performance index datasets are obtained based on standard historical performance data.

10. A design system for a high-frequency transformer according to claim 8, characterized in that, The deep learning model training module is used to obtain a convolutional neural multilayer perceptron deep learning model based on a historical transformer design parameter dataset, a historical transformer actual performance index dataset, and a preset initial deep learning model, including: Based on a preset ratio, the historical transformer design parameter dataset is divided into a model building subset, a performance evaluation subset, and an independent verification subset. Based on the historical transformer actual performance index dataset, the model building subset, the performance evaluation subset, and the independent verification subset, the first performance index dataset corresponding to the model building subset, the second actual performance index dataset corresponding to the performance evaluation subset, and the third actual performance index dataset corresponding to the independent verification subset are obtained. Based on the model construction subset, performance evaluation subset, independent testing subset, first performance index dataset, second actual performance index dataset, third actual performance index dataset, and preset initial deep learning model, a convolutional neural multilayer perceptual deep learning model is obtained.

Images