Method for optimizing dynamic response capability of LLC converter based on neural network

Abstract

The application relates to the technical field of power electronic converter control, in particular to an LLC converter dynamic response capability optimization control method based on a neural network, which comprises the following steps: establishing an LLC converter dynamic response mathematical model based on a frequency modulation strategy; selecting a differential evolution algorithm combined with the mathematical model to optimize the dynamic response process of the LLC converter during load switching, collecting k groups of frequency data of the best LLC dynamic performance of load switching; taking the collected k groups of frequency data of the best LLC dynamic performance of load switching as training data to train a pre-constructed neural network model; and applying the trained neural network model to an LLC converter controller to realize optimization of the dynamic response process. The application enables the LLC converter to have good dynamic response capability during the load switching process, so that the LLC converter can smoothly and quickly reach a new steady state.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power electronic converter control, and particularly relates to an LLC converter dynamic response capability optimization control method based on a neural network. BACKGROUND

[0002] Power electronics technology plays a key role in the development and utilization of new energy. As an important part of power electronic energy conversion, bidirectional DC / DC converters have been widely used in photovoltaic power generation, new energy vehicles and other new energy fields. LLC resonant converters are widely used in DC / DC converters due to their high transmission efficiency, wide voltage regulation range, good voltage regulation characteristics, and low switching tube voltage stress. However, in actual application scenarios, LLC converters are often affected by input voltage fluctuations or load mutations, and the traditional control method has a slow dynamic response, which becomes a bottleneck restricting its further application in high-performance fields. Therefore, how to optimize the dynamic response performance of LLC converters has become the current research target.

[0003] To improve the control performance of LLC resonant converters, various methods have been proposed in the prior art. For example, a fuzzy adaptive control method uses a fuzzy adaptive PI controller in the outer voltage loop, which automatically adjusts the PI parameters through voltage changes and their rates of change. The current inner loop uses a traditional PI controller. This method simplifies the parameter tuning process, improves control accuracy, and improves dynamic performance, but has high computational complexity. The hybrid modulation strategy based on sliding mode control has good dynamic performance and strong robustness, but can cause large output current ripple. The model predictive control method can effectively improve the dynamic response performance of the system. Compared with the traditional PI controller, which is limited by bandwidth, the model predictive control eliminates the bandwidth limitation through the prediction model, thereby improving the system stability. However, it requires an accurate state space average model, which has high modeling complexity and requires solving the derivative of the evaluation function in each control period, which is computationally intensive. The optimal trajectory control method requires drawing a state trajectory based on the time-domain equation of the LLC converter and designing a controller based on the existing state trajectory. By real-time sampling of state quantities to track the desired trajectory, the dynamic response capability is improved. However, this method is complex to design and difficult to implement in practice. Therefore, the existing control methods have their own advantages and limitations in improving the dynamic performance of LLC resonant converters. The challenge is to develop a control method that enables LLC converters to have good dynamic response capability during load switching, thereby smoothly and quickly reaching a new steady state. SUMMARY

[0004] The application provides a neural network-based LLC converter dynamic response capability optimization control method, so that the LLC converter has good dynamic response capability during load switching, thereby smoothly and quickly reaching a new steady state. The application provides the following technical solutions:

[0005] In a first aspect, the application provides a neural network-based LLC converter dynamic response capability optimization control method, which comprises:

[0006] establishing an LLC converter dynamic response mathematical model based on a variable frequency modulation strategy;

[0007] selecting a differential evolution algorithm combined with the mathematical model to optimize the dynamic response process of the LLC converter during load switching, collecting the frequency data of the LLC dynamic performance best for load switching;

[0008] the collected the frequency data of the LLC dynamic performance best for load switching as training data to train a pre-constructed neural network model;

[0009] applying the trained neural network model to an LLC converter controller to realize optimization of the dynamic response process.

[0010] In a specific implementation scheme, the LLC converter dynamic response mathematical model based on the variable frequency modulation strategy comprises:

[0011] establishing a mathematical model containing the relationships of frequency , resonant current and output voltage , assuming that the dynamic process contains periods, the control variable to be solved is the frequency sequence , and the constraint conditions of the model include three aspects, first, the effective value of the resonant current and the output voltage after experiencing periods should respectively reach the initial values and of the new steady state after load switching to ensure smooth transition of the system to the new steady state:

[0012] ;

[0013] wherein, , respectively represent the effective value of the resonant current and the output voltage after experiencing periods; , represent the initial values of the new steady state after load switching;

[0014] Second, the frequency is defined within the allowed modulation range by the following equation :

[0015]

[0016] where , are the allowed minimum and maximum operating frequency boundaries, respectively;

[0017] Third, an optimization objective function is defined to quantitatively describe the output voltage deviation of the LLC converter during the dynamic response stage, which is mathematically expressed as follows:

[0018]

[0019] where is the desired output reference voltage, and the goal is to reduce the maximum deviation of the output voltage during the dynamic response stage by minimizing the objective function .

[0020] In one specific implementation, the selected differential evolution algorithm is combined with a mathematical model to optimize the dynamic response process of the LLC converter during load switching, and the frequency data of the LLC dynamic performance best for a set of load switching values are collected, including:

[0021] A sample set is established for different load switching conditions, and each condition corresponds to a set of load switching values , which are used as input conditions to initialize the frequency sequence , and a plurality of candidate frequency sequences are randomly generated to form an initial population;

[0022] The initial population is optimized and searched using the iterative mechanism of the differential evolution algorithm, and a set of different load switching conditions are optimized to obtain the optimal frequency sequence and corresponding load switching values under each condition, which is represented as , forming an optimal control sample set containing data .

[0023] In one specific implementation, the optimization and search of the initial population using the iterative mechanism of the differential evolution algorithm include:

[0024] In each iteration, a plurality of candidate sequences based on the current population generate new candidate frequency sequences according to the preset differential strategy, and a selection is made between the generated new sequences and the original sequences, comparing the objective function values output by the mathematical model , if the objective function value corresponding to the new sequence is smaller, then the new sequence is updated as the effective sequence in the current population;​​​

[0025] In each round of update, the constraint condition of the mathematical model needs to be met, and after the calculation is completed, it is judged whether the current iteration reaches the preset maximum iteration number or the target function convergence threshold. If the condition is not met, the next round of frequency adjustment and screening is continued; if the condition is met, the iteration is terminated and the frequency sequence with the optimal dynamic response capability under the working condition is output.

[0026] In a specific implementable embodiment, the collected frequency data of the LLC dynamic performance best for the group load switching is used as training data to train the pre-constructed neural network model. The frequency data of the LLC dynamic performance best for the group load switching is used as training data to train the pre-constructed neural network model.

[0027] The pre-constructed neural network model is a four-layer feedforward neural network, including an input layer, two hidden layers and an output layer, wherein the input layer receives the load switching value , the output layer outputs the frequency value corresponding to periods, and the hidden layer is used for feature extraction and nonlinear mapping.

[0028] In a specific implementable embodiment, the collected frequency data of the LLC dynamic performance best for the group load switching is used as training data to train the pre-constructed neural network model. The frequency data of the LLC dynamic performance best for the group load switching is used as training data to train the pre-constructed neural network model.

[0029] In the neural network training process, the input features are normalized, the optimization algorithm of the neural network is selected, and the activation function of the hidden layer is determined; the training data is divided into a training set, a validation set and a test set, so that the model training, validation and testing are independently carried out; the learning rate, the maximum iteration number and the early stopping mechanism are set, the network iteration process is monitored according to the output of the training set and the validation set, the training effect is judged, and a network model with good training convergence is selected to obtain a trained neural network model.

[0030] In a specific implementable embodiment, the trained neural network model is applied to the LLC converter controller to realize optimization of the dynamic response process.

[0031] When the controller receives the load switching signal, the load switching value is input into the trained neural network model;

[0032] The model outputs the optimal frequency value corresponding to periods according to the load and frequency relationship learned in the training process.

[0033] The controller adjusts the switching frequency of the LLC converter according to the optimal frequency value .

[0034] ​​In a second aspect, the application provides a neural network-based LLC converter dynamic response capability optimization control system, which adopts the following technical scheme:

[0035] A neural network-based LLC converter dynamic response capability optimization control system, comprising:

[0036] A model establishing module, configured to establish a LLC converter dynamic response mathematical model based on a frequency modulation strategy;

[0037] A data optimization module, configured to select a differential evolution algorithm in combination with the mathematical model to optimize a dynamic response process of the LLC converter during load switching, and collect frequency data of the LLC converter with the best dynamic performance during load switching of the group;

[0038] A model training module, configured to use the collected frequency data of the LLC converter with the best dynamic performance during load switching of the group as training data to train a pre-constructed neural network model;

[0039] A model application module, configured to apply the trained neural network model to an LLC converter controller to optimize the dynamic response process.

[0040] In a third aspect, the application provides an electronic device, which comprises a processor and a memory; the memory stores a program, which is loaded and executed by the processor to implement the neural network-based LLC converter dynamic response capability optimization control method according to the first aspect.

[0041] In a fourth aspect, the application provides a computer-readable storage medium, which stores a program, which is executed by a processor to implement the neural network-based LLC converter dynamic response capability optimization control method according to the first aspect.

[0042] In summary, the application has at least the following beneficial effects:

[0043] (1) The differential evolution algorithm possesses global optimization capabilities and adaptability to complex nonlinear problems. Through a population search and iterative update mechanism, this algorithm simultaneously explores multiple candidate solutions in the frequency sequence space, avoiding getting trapped in local optima and thus more reliably finding the frequency combination with the optimal dynamic response performance during load switching. Furthermore, the differential evolution algorithm does not require strict requirements on the derivative information or continuous differentiability of the problem, and can directly handle established dynamic response mathematical models containing nonlinear relationships and constraints. In addition, its iterative generation, screening, and updating mechanism can flexibly handle frequency constraints, resonant current, and steady-state conditions of output voltage, and achieves adaptive termination through convergence judgment, thereby automatically outputting the optimal frequency sequence under each operating condition. Compared with manual experience or traditional optimization methods, this method not only improves the optimization efficiency but also ensures the accuracy and repeatability of the optimization results.

[0044] (2) The design of the four-layer feedforward neural network can effectively represent the complex nonlinear relationship between load switching parameters and frequency sequence. The two hidden layers provide sufficient network capacity to capture the nonlinear influence of frequency modulation on resonant current and output voltage during dynamic response, enabling the network to learn the optimal frequency law under different load conditions and directly predict the frequency sequence suitable for the controller through the trained model, thereby realizing automated dynamic response adjustment. Compared with a single hidden layer or linear model, it can significantly improve prediction accuracy and adaptability to different operating conditions.

[0045] First, a frequency conversion modulation strategy is applied to the LLC converter to adjust its modal characteristics and adapt to dynamic requirements under different load conditions. Then, a differential evolution algorithm is used to globally optimize load switching under different operating conditions, obtaining the frequency sequence with the optimal dynamic response capability for each condition. Further, The load values ​​and corresponding optimal frequencies under different operating conditions are used as input and output data to train a four-layer feedforward neural network, resulting in a neural network model that can predict the optimal frequency. Finally, the trained neural network model is applied to the LLC converter controller to realize real-time prediction and adjustment of frequency during load switching, thereby significantly improving the dynamic response capability of the converter and enabling it to smoothly and quickly transition to a new steady state. This solves the technical problem in the prior art that it is difficult to balance response speed and steady-state stability.

[0046] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0047] Figure 1This is a flowchart illustrating the method for optimizing the dynamic response capability of an LLC converter based on a neural network, as described in this application.

[0048] Figure 2 This is a schematic diagram of the topology of the LLC converter in the embodiments of this application.

[0049] Figure 3 This is a schematic diagram illustrating the principle of the differential evolution algorithm in the embodiments of this application.

[0050] Figure 4 This is a schematic diagram of the structure of the pre-built neural network model in the embodiments of this application.

[0051] Figure 5 This is a block diagram of the dynamic response capability optimization control system for LLC converter based on neural networks in the embodiments of this application.

[0052] Figure 6 This is a block diagram of an electronic device for optimizing the dynamic response capability of an LLC converter based on a neural network, as described in this application embodiment. Detailed Implementation

[0053] The specific embodiments of this application will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate this application, but are not intended to limit the scope of this application.

[0054] Optionally, this application uses the method for optimizing the dynamic response capability of an LLC converter based on neural networks provided in various embodiments as an example for application in an electronic device. The electronic device is a terminal or a server. The terminal can be a computer, tablet computer, etc. This embodiment does not limit the type of electronic device.

[0055] Reference Figure 1 This is a flowchart illustrating a neural network-based method for optimizing the dynamic response capability of an LLC converter according to an embodiment of this application. The method includes at least the following steps:

[0056] Step S101: Establish a mathematical model of the dynamic response of the LLC converter based on the frequency conversion modulation strategy.

[0057] In step S101, a frequency conversion modulation strategy is used to control the LLC converter, that is, by periodically adjusting the operating frequency during the response process. This allows for alteration of the converter's modal characteristics, thereby enabling dynamic adjustment of the resonant parameters at different frequencies. (See attached image.) Figure 2 As shown, the topology of an LLC converter includes key components such as a resonant inductor, resonant capacitor, and transformer. The resonant current and output voltage at different frequencies exhibit different dynamic response characteristics. At time... When load switching occurs, the LLC converter enters a dynamic adjustment stage from the original steady state, and the changes of the resonant current and the output voltage are affected by the selected frequency sequence, and finally reach a new steady state after experiencing switching cycles.

[0058] To describe the dynamic process, a mathematical model containing the relationships among the frequency, the resonant current and the output voltage is established to characterize the dynamic response characteristics of the converter under load mutation. Specifically, the dynamic process contains cycles, and the control variable to be solved is the frequency sequence . The constraint conditions of the model include the following three aspects: first, the effective value of the resonant current and the output voltage should respectively reach the initial values and of the new steady state after experiencing cycles to ensure smooth transition of the system to the new steady state:

[0059] ;

[0060] wherein , respectively represent the effective value of the resonant current and the output voltage after experiencing cycles; , represent the initial values of the new steady state after load switching.

[0061] Secondly, the frequency is limited within the allowed modulation range by the following formula to meet the working constraints of the LLC converter:

[0062] ;

[0063] wherein , are respectively the minimum and maximum working frequency boundaries.

[0064] Finally, the optimization objective function is defined to quantitatively describe the output voltage deviation of the LLC converter in the dynamic response stage. The function takes the maximum deviation of the output voltage from the reference voltage as the measurement index, and its mathematical expression is as follows:

[0065] ;

[0066] wherein, To obtain the desired output reference voltage, the target is to reduce the maximum deviation of the output voltage in the dynamic response stage by minimizing the objective function , thereby improving the voltage stabilization capability and dynamic response performance of the converter during load mutation.

[0067] In step S102, the differential evolution algorithm is selected to optimize the dynamic response process of the LLC converter during load switching in combination with the mathematical model, and the frequency data of the LLC dynamic performance of the load switching group are collected.

[0068] In step S102, in order to obtain the optimal control frequency sequence of the dynamic response capability of the LLC converter under different load switching conditions, global optimization is performed on the established dynamic response mathematical model based on the differential evolution algorithm. As shown in Figure 3 , the differential evolution algorithm optimizes the frequency sequence through population search and parameter iteration, so that the output voltage deviation objective function is minimized, thereby obtaining the optimal frequency combination of the dynamic response performance under each working condition.

[0069] Specifically, first, a sample set is established for different working conditions of load switching, each working condition corresponds to a group of load switching values , and the frequency sequence is initialized as the input condition , which is the control variable to be optimized. In the initial stage of the algorithm, multiple candidate frequency sequences are randomly generated to form an initial population, and each individual represents a possible frequency modulation mode of the LLC converter under the working condition.

[0070] Subsequently, the differential evolution algorithm is used to optimize the search of the above population. In each iteration, first, multiple candidate sequences based on the current population are generated according to the preset differential strategy, which can be regarded as perturbation or adjustment of the frequency parameters; then, the new sequences and the original sequences are selected and compared in terms of the objective function value, i.e. the output voltage deviation index , under the mathematical model. If the objective function value corresponding to the new sequence is smaller, the sequence is updated as the effective sequence in the current population.

[0071] In each round of update, the following constraints must be met:

[0072] (1) Ensure that the resonant current and the output voltage reach the new steady-state value after periods;

[0073] (2) Ensure that the frequency value is always within the allowed range ;

[0074] (3) The time interval corresponding to the calculation of the objective function J ​The dynamic response stage should be completely covered to ensure the effectiveness of the optimization results.

[0075] After the above calculation is completed, it is determined whether the current iteration reaches the preset maximum iteration number or the target function convergence threshold. If the conditions are not met, the next round of frequency adjustment and screening is continued. If the conditions are met, the iteration is terminated and the frequency sequence with the optimal dynamic response capability under the working condition is output.

[0076] Through the above process, the optimal frequency sequence and the corresponding load switching value under each group of working conditions can be obtained. , thereby forming an optimal control sample set containing .

[0077] The optimal control sample set is completely represented as follows:

[0078] ;

[0079] Through the above optimization steps, the frequency control parameters with the optimal dynamic response performance under different working conditions can be automatically obtained without relying on artificial experience, providing a reliable data basis for subsequent adaptive control models based on neural networks.

[0080] In addition, the main advantage of selecting the differential evolution algorithm is that it has global optimization capability and adaptability to complex nonlinear problems. Through the group search and iterative update mechanism, the algorithm simultaneously explores multiple candidate solutions in the frequency sequence space, which can avoid falling into local optimum, and thus more reliably find the frequency combination with the optimal dynamic response performance in the load switching process. At the same time, the differential evolution algorithm does not require derivative information or strict continuity of the problem, and can directly process the dynamic response mathematical model established in step S101 containing nonlinear relationships and constraint conditions. In addition, its iterative generation, screening and updating mechanism can flexibly handle frequency constraints, resonant current and output voltage steady-state conditions, and realize adaptive termination through convergence determination, thereby automatically outputting the optimal frequency sequence under each working condition. Compared with artificial experience or traditional optimization methods, this method not only improves the optimization efficiency, but also ensures the accuracy and repeatability of the optimization results, providing a reliable data basis for subsequent neural network model training.

[0081] Step S103, the frequency data collected from the LLC dynamic performance of the load switching under the group of working conditions is used as training data to train the pre-constructed neural network model.

[0082] In step S103, the frequency data collected from the LLC dynamic performance of the load switching under the ​The group load switching value and the corresponding optimal frequency sequence are taken as the training data of the neural network, to establish a mapping relationship between the load switching and the frequency control parameter, so as to realize the optimization of the dynamic response capability in the load switching process.

[0083] Specifically, referring to Figure 4 , a four-layer feedforward neural network (BP neural network) is constructed, including an input layer, two hidden layers and an output layer, wherein the input layer receives the load switching value , the output layer outputs the frequency value corresponding to periods , and the hidden layer is used for feature extraction and nonlinear mapping. Through the design of the network structure, the complex nonlinear relationship between the load switching parameter and the frequency sequence can be effectively represented, wherein the two hidden layers provide sufficient network capacity to capture the nonlinear influence of the frequency modulation on the resonant current and the output voltage in the dynamic response process, so that the network can learn the optimal frequency law under different load conditions, and directly predict the frequency sequence suitable for the controller through the trained model, to realize the automatic dynamic response adjustment. Compared with the single hidden layer or the linear model, the prediction accuracy and the adaptability to different conditions can be significantly improved.

[0084] In the neural network training process, the input features are normalized to ensure consistent numerical range; the optimization algorithm of the neural network is selected, and the activation function of the hidden layer is determined; the training data is divided into training set, validation set and test set to ensure independent model training, validation and testing; the learning rate, the maximum number of iterations and the early stopping mechanism are set to control the network convergence process and avoid overfitting. According to the output of the training set and the validation set, the network iteration process is monitored to judge the training effect and select the network model with good training convergence. After the above steps are completed, the trained neural network model is obtained.

[0085] Alternatively, the training of the neural network in the present application is realized by using MATLAB, and other ways can also be used to realize it.

[0086] In step S104, the trained neural network model is applied to the LLC converter controller to realize the optimization of the dynamic response process.

[0087] In step S104, the trained neural network model is integrated into the controller of the LLC converter, to calculate and output the corresponding frequency sequence in real time in the actual load switching process.

[0088] Specifically, when the controller receives the load switching signal, the load switching value is input into the neural network model, and the model outputs the optimal frequency value corresponding to periods according to the load and frequency relationship learned in the training process The controller then adjusts the switching frequency of the LLC converter according to the frequency sequence, realizing smooth transition of the converter from the original steady state to the new steady state. Through this step, the controller can adaptively select the frequency sequence with the optimal dynamic response under different load conditions without manual intervention, so as to realize fast steady-state response of the LLC converter during load switching and ensure stability of the output voltage and resonant current during the dynamic process.

[0089] In summary, by first adopting the variable-frequency modulation strategy for the LLC converter, the modal characteristics of the converter are adjusted by adjusting the operating frequency to adapt to the dynamic requirements under different load conditions; then, the differential evolution algorithm is used to globally optimize the load switching under different conditions to obtain the frequency sequence with the optimal dynamic response capability under each condition; further, the load values under different conditions and the corresponding optimal frequencies are used as input and output data to train the four-layer feedforward neural network to obtain a neural network model capable of predicting the optimal frequency; finally, the trained neural network model is applied to the LLC converter controller to realize real-time prediction and adjustment of the frequency during load switching, thereby significantly improving the dynamic response capability of the converter, enabling it to smoothly and quickly transition to a new steady state, and solving the technical problem of difficulty in balancing response speed and steady-state smoothness in the prior art. In summary, by first adopting the variable-frequency modulation strategy for the LLC converter, the modal characteristics of the converter are adjusted by adjusting the operating frequency to adapt to the dynamic requirements under different load conditions; then, the differential evolution algorithm is used to globally optimize the load switching under different conditions to obtain the frequency sequence with the optimal dynamic response capability under each condition; further, the load values under different conditions and the corresponding optimal frequencies are used as input and output data to train the four-layer feedforward neural network to obtain a neural network model capable of predicting the optimal frequency; finally, the trained neural network model is applied to the LLC converter controller to realize real-time prediction and adjustment of the frequency during load switching, thereby significantly improving the dynamic response capability of the converter, enabling it to smoothly and quickly transition to a new steady state, and solving the technical problem of difficulty in balancing response speed and steady-state smoothness in the prior art.

[0090] Figure 5 is a structural block diagram of an LLC converter dynamic response capability optimization control system based on a neural network provided by an embodiment of the present application. The system includes at least the following modules:

[0091] A model establishment module is configured to establish a dynamic response mathematical model of the LLC converter based on a variable-frequency modulation strategy.

[0092] A data optimization module is configured to use a differential evolution algorithm in combination with the mathematical model to optimize the dynamic response process of the LLC converter during load switching, collect frequency data of the LLC converter with the best dynamic performance during load switching;

[0093] A model training module is configured to use the collected frequency data of the LLC converter with the best dynamic performance during load switching as training data to train a pre-constructed neural network model.

[0094] A model application module is configured to apply the trained neural network model to the LLC converter controller to realize optimization of the dynamic response process.

[0095] For related details, refer to the method embodiments described above.

[0096] Figure 6is a block diagram of an electronic device provided by an embodiment of the present application. The device at least includes a processor 601 and a memory 602.

[0097] The processor 601 can include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 601 can be implemented in at least one of a hardware form of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), a PLA (Programmable Logic Array). The processor 601 can also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also known as a CPU (Central Processing Unit). The coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 601 can be integrated with a GPU (Graphics Processing Unit) that is responsible for rendering and drawing the content required to be displayed by the display screen. In some embodiments, the processor 601 can also include an AI (Artificial Intelligence) processor for processing machine learning-related computing operations.

[0098] The memory 602 can include one or more computer-readable storage media that can be non-transitory. The memory 602 can also include a high-speed random access memory, and a non-volatile memory such as one or more disk storage devices, flash storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 602 is used to store at least one instruction for being executed by the processor 601 to implement the neural network-based LLC converter dynamic response capability optimization control method provided by the method embodiment of the present application.

[0099] In some embodiments, the electronic device can also optionally include a peripheral device interface and at least one peripheral device. The processor 601, the memory 602, and the peripheral device interface can be connected through a bus or a signal line. Each peripheral device can be connected to the peripheral device interface through a bus, a signal line, or a circuit board. Illustratively, the peripheral device includes, but is not limited to, a radio frequency circuit, a touch display screen, an audio circuit, and a power supply, etc.

[0100] Of course, the electronic device can also include fewer or more components, which are not limited in the present embodiment.

[0101] Optionally, the present application also provides a computer readable storage medium, the computer readable storage medium stores a program, the program is loaded and executed by a processor to implement the neural network-based LLC converter dynamic response capability optimization control method of the above method embodiment.

[0102] Optionally, the present application also provides a computer product, the computer product includes a computer readable storage medium, the computer readable storage medium stores a program, the program is loaded and executed by a processor to implement the neural network-based LLC converter dynamic response capability optimization control method of the above method embodiment.

[0103] The technical features of the above embodiments can be combined in any manner. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present disclosure.

[0104] The above embodiments only express several implementation manners of the present application, and the description is specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, some modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for optimizing the dynamic response capability control of an LLC converter based on a neural network, characterized in that, The method includes: A mathematical model of the dynamic response of an LLC converter is established based on a frequency conversion modulation strategy; the mathematical model of the dynamic response of an LLC converter based on a frequency conversion modulation strategy includes: Establish frequency inclusion Resonant current and output voltage A mathematical model of the relationship, assuming the dynamic process includes The control variable to be determined is a frequency sequence with 1 cycle. The model's constraints include three aspects. First, they are limited by the following formula when experiencing... After one cycle, the effective value of the resonant current With output voltage The initial values ​​that should reach the new steady state are respectively and To ensure a smooth transition of the system to the new steady state: ； in, , They respectively represent the experience The effective value of the resonant current and the output voltage after one cycle; , This represents the initial value of the new steady state after load switching; Second, the frequency is expressed by the following formula. Limited to the allowed modulation range Inside: ； in , These are the minimum and maximum allowable operating frequency boundaries, respectively. Third, define the optimization objective function. This is used to quantitatively describe the output voltage deviation of an LLC converter during the dynamic response phase, and its mathematical expression is as follows: ； in, To obtain the desired output reference voltage, the objective is to minimize the objective function. Reduce the maximum deviation of the output voltage during the dynamic response phase; The differential evolution algorithm combined with a mathematical model is used to optimize the dynamic response process of the LLC converter during load switching, and data is collected. Frequency data for optimal LLC dynamic performance during load switching; Collected The frequency data of the LLC dynamic performance with the best group load switching is used as training data to train a pre-built neural network model; The trained neural network model is applied to the LLC converter controller to optimize the dynamic response process.

2. The method for optimizing the dynamic response capability control of an LLC converter based on a neural network according to claim 1, characterized in that, The differential evolution algorithm combined with a mathematical model is used to optimize the dynamic response process of the LLC converter during load switching, and data is collected. The frequency data for optimal LLC dynamic performance during load switching includes: A sample set is established for different load switching conditions, with each condition corresponding to a set of load switching values. And use this as input to initialize the frequency sequence. Multiple candidate frequency sequences are randomly generated to form an initial population; The iterative mechanism of the differential evolution algorithm is used to optimize the initial population. Optimization was performed on different load switching conditions to obtain the optimal frequency sequence and corresponding load switching values ​​for each condition. The results are expressed as follows: , forming a collection Optimal control sample set for group data .

3. The method for optimizing the dynamic response capability control of an LLC converter based on a neural network according to claim 2, characterized in that, The iterative search for optimizing the initial population using the differential evolution algorithm includes: In each iteration, new candidate frequency sequences are generated from multiple candidate sequences in the current population using a pre-defined differencing strategy. The generated new sequences are then compared with the original sequences, and their objective function values ​​output by the mathematical model are analyzed. If the objective function value corresponding to the new sequence is smaller, then update it to the effective sequence in the current population; In each round of updates, the constraints of the mathematical model must be met. After the calculation is completed, it is determined whether the current iteration has reached the preset maximum number of iterations or the convergence threshold of the objective function. If the conditions are not met, the next round of frequency adjustment and screening will continue. If the conditions are met, the iteration will be terminated and the frequency sequence with the best dynamic response capability under this working condition will be output.

4. The method for optimizing the dynamic response capability control of an LLC converter based on a neural network according to claim 2, characterized in that, The collected The frequency data that best represents the dynamic performance of LLC group load switching is used as training data to train a pre-built neural network model, including: The pre-built neural network model is a four-layer feedforward neural network, including one input layer, two hidden layers, and one output layer, wherein the input layer receives the load switching value. Output layer output Frequency value corresponding to each cycle Hidden layers are used for feature extraction and nonlinear mapping.

5. The method for optimizing the dynamic response capability control of an LLC converter based on a neural network according to claim 4, characterized in that, The collected The frequency data that best represents the dynamic performance of LLC group load switching is used as training data to train the pre-built neural network model, which also includes: During neural network training, the input features are normalized, the optimization algorithm of the neural network is selected, and the activation function of the hidden layer is determined. The training data is divided into training set, validation set, and test set so that model training, validation, and testing are carried out independently. The learning rate, maximum number of iterations, and early stopping mechanism are set. The network iteration process is monitored based on the output of the training set and validation set to judge the training effect and select the network model with good training convergence, thus obtaining the trained neural network model.

6. The method for optimizing the dynamic response capability control of an LLC converter based on a neural network according to claim 4, characterized in that, The process of applying the trained neural network model to the LLC converter controller to optimize the dynamic response includes: When the controller receives the load switching signal, it will set the load switching value. Input the trained neural network model; The model outputs a corresponding value based on the load and frequency relationship learned during training. Optimal frequency value for each cycle ; The controller uses the optimal frequency value Adjust the switching frequency of the control LLC converter.

7. A neural network-based dynamic response capability optimization control system for LLC converters, characterized in that, include: The model building module is used to build a mathematical model of the dynamic response of the LLC converter based on the frequency conversion modulation strategy. The mathematical model for the dynamic response of the LLC converter based on the frequency conversion modulation strategy includes: Establish frequency inclusion Resonant current and output voltage A mathematical model of the relationship, assuming the dynamic process includes The control variable to be determined is a frequency sequence with 1 cycle. The model's constraints include three aspects. First, they are limited by the following formula when experiencing... After one cycle, the effective value of the resonant current With output voltage The initial values ​​that should reach the new steady state are respectively and To ensure a smooth transition of the system to the new steady state: ； in, , They respectively represent the experience The effective value of the resonant current and the output voltage after one cycle; , This represents the initial value of the new steady state after load switching; Second, the frequency is expressed by the following formula. Limited to the allowed modulation range Inside: ； in , These are the minimum and maximum allowable operating frequency boundaries, respectively. Third, define the optimization objective function. This is used to quantitatively describe the output voltage deviation of an LLC converter during the dynamic response phase, and its mathematical expression is as follows: ； in, To obtain the desired output reference voltage, the objective is to minimize the objective function. Reduce the maximum deviation of the output voltage during the dynamic response phase; The data optimization module is used to optimize the dynamic response process of the LLC converter during load switching by selecting the differential evolution algorithm combined with a mathematical model, and to collect data. Frequency data for optimal LLC dynamic performance during load switching; The model training module is used to process the collected data. The frequency data of the LLC dynamic performance with the best group load switching is used as training data to train a pre-built neural network model; The model application module is used to apply the trained neural network model to the LLC converter controller to optimize the dynamic response process.

8. An electronic device, characterized in that, The device includes a processor and a memory; the memory stores a program, which is loaded and executed by the processor to implement a neural network-based method for optimizing the dynamic response capability of an LLC converter as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The storage medium stores a program, which, when executed by a processor, is used to implement a neural network-based method for optimizing the dynamic response capability of an LLC converter as described in any one of claims 1 to 6.

Images