Power converter control method, device and storage medium based on physical information machine learning

Abstract

The present application relates to the technical field of power converter, in particular to a power converter control method based on physical information machine learning, equipment and storage medium. Including the following steps: control system building; data collection, storage and pretreatment; model building and evaluation; model deployment; model application; power converter operation mode prediction model selection. The design of the application embeds physical constraint information into the loss function, uses machine learning algorithm to mine the relationship between control signal duty cycle, frequency and jump pulse number and input voltage, input current, output voltage, output current and temperature, and deploys the machine learning control model to FPGA to realize real-time prediction and control of control signal duty cycle, frequency and pulse jump number; Through the nonlinear ability of the machine learning model, accurate modeling can be realized, which can save a lot of cost and improve the accuracy of the model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power converter, in particular to a power converter control method and device based on physical information machine learning and a storage medium. BACKGROUND

[0002] Current power converter control technologies mainly include analog control technology and digital PID control technology. Analog control technology includes voltage mode control and current mode control, while digital PID control technology can support more complex algorithms, such as nonlinear control and adaptive control. Traditional switching power supplies usually adopt analog control technology, which uses comparators, error amplifiers, and analog power management chips to adjust the output voltage. However, this control method has many disadvantages, such as complex control circuit, large number of required components, and difficulty in modifying the control circuit once it is designed.

[0003] With the rapid development of microelectronics technology, power control technology has evolved from pure analog control to analog-digital hybrid control, and now to all-digital PID control. Digital PID control not only simplifies the control circuit and reduces the number of components, but also provides higher flexibility for later adjustment and optimization, greatly promoting the miniaturization and integration process of switching power supplies. The magnetic elements in the power converter (such as transformers and inductors), load changes, parasitic capacitance and parasitic diodes of switching devices, nonlinear elements in the feedback loop (such as the saturation characteristics of error amplifiers), and temperature changes (such as changes in the temperature of the heat sink, transformer, and device interior) all have nonlinear characteristics, resulting in nonlinear characteristics of switching power supply control. In addition, the switching action of the switching device (such as MOSFET or IGBT) at the core of the switching power supply is also nonlinear. These nonlinear characteristics work together to make the power converter a highly nonlinear system, thus requiring a nonlinear control strategy to ensure the stability of the output voltage.

[0004] The technical difficulties of power converter feedback loop design mainly include the design of compensation network, the suppression of noise, the response to temperature and environmental impact, the handling of multi-loop control, the stability of the system, the optimization of response speed and dynamic performance, and the adaptation to load changes and nonlinear characteristics, which require designers to have deep professional knowledge and rich practical experience. Traditional linear control methods, such as analog control technology and digital PID control technology, may not fully meet the requirements of high precision and stability by adjusting the duty cycle of PWM, the frequency of PFM or the number of pulse jumps of PSM according to the change of output voltage to control the output voltage. And the high nonlinearity of power converters makes it very difficult to establish an accurate mathematical model. Due to the complexity of system modeling, the dynamic adjustment requirement of parameter setting, the nonlinear compensation of control strategy design, and the difficulty of implementation and debugging, it is a challenge to design a nonlinear PID controller. In view of this, we propose a power converter control method, device and storage medium based on physical information machine learning. SUMMARY

[0005] The purpose of the present application is to provide a power converter control method, device and storage medium based on physical information machine learning to solve the problems raised in the background art.

[0006] To solve the above technical problems, one of the purposes of the present application is to provide a power converter control method based on physical information machine learning, comprising the following steps:

[0007] S1, control system building: building a control system architecture including power converter, data acquisition and processing module, machine learning control algorithm and control signal generator;

[0008] S2, data acquisition, storage and preprocessing: collecting physical information and control signal data of power converter operation, splicing and storing the collected data to facilitate the training and application of machine learning algorithm, and cleaning and normalizing the data for preprocessing;

[0009] S3, model building and evaluation: building an accurate mathematical control model related to power converter operation using machine learning, and optimizing the model based on loss function;

[0010] S4, model deployment: optimizing and compressing the model to reduce resource occupation and improve inference speed, compiling the trained model into a format suitable for MCU, ASIC and FPGA, and deploying it on an embedded system;

[0011] S5, model application: inputting real-time measured power converter operation parameters into the corresponding prediction model for real-time prediction;

[0012] S6, the prediction model selection of the power converter operation mode: the operation model of the power converter is intelligently selected through the operation efficiency and the prediction efficiency of the power converter during operation.

[0013] As a further improvement of the technical solution, in the control system architecture constructed in S1:

[0014] The power converter at least includes Buck, Boost, Buck-Boost, CUK, and resonant converters, and flyback converters, forward converters, push-pull converters, half-bridge converters, and full-bridge converters, and parallel topological architectures of these types of converters, and power converters using ZVS and ZCS technologies to reduce power device switching loss;

[0015] The data acquisition and processing module includes the acquisition of control signal duty cycle, frequency, and pulse jump number, the acquisition and A / D conversion of input voltage, input current, output voltage, output current, and temperature; At the same time, the maximum and minimum data for the inverse normalization process also need to be stored to convert the normalized data back to the scale of the original data;

[0016] The machine learning control algorithm at least includes SVR, RF, BP neural network, RBF, and LSTM; The prediction model selection algorithm of the power converter can also use these machine learning algorithms, and the prediction model selection algorithm selects whether the topology of the power converter is operated in PWM mode or in PFM / PSM mode according to the operation efficiency and the prediction efficiency of the power converter;

[0017] The control signal generator generates (pulse width modulation) PWM signals, (pulse frequency modulation) PFM signals, and (pulse step modulation) PSM signals according to the duty cycle, frequency, pulse jump number, and dead time predicted by the machine learning algorithm, and controls the switching of power MOS tubes or IGBT power devices through the driving circuit.

[0018] As a further improvement of the technical solution, the specific process of S2 includes the following steps:

[0019] S2.1, data acquisition: acquire power converter input and output voltage and current and environmental parameters, including input voltage, input current, output voltage, output current, and temperature data of the heat sink, transformer, and air, as machine learning features; Acquire control signal data, including control signal duty cycle, frequency, and pulse jump number data, as machine learning label data;

[0020] S2.2, data storage: sort the power converter data according to the acquisition time, use PWM duty cycle and PFM frequency as labels, use input voltage, input current, output voltage, output current, and temperature characteristics as model training data, and form a data set;

[0021] S2.3, data preprocessing: filling missing values in the collected power converter data, and normalizing the data set to accelerate model convergence.

[0022] As a further improvement of the technical solution, the data preprocessing in S2.3 specifically includes:

[0023] There are some missing values in the data set, which are supplemented by the data of the previous time point, or by the weighted average of the data before and after;

[0024] In the case of standard dimension, the input voltage, input current, output voltage, output current and temperature characteristic data of the power converter, and the data difference of the duty cycle, frequency, pulse jump number label data are large, the maximum and minimum value normalization preprocessing is carried out on the data, and the features and labels of different scales are scaled to the range of 0 to 1, so as to eliminate the influence of large data on the model, and the normalization formula is:

[0025]

[0026] In the formula, is the data to be normalized, i.e. the original data, is the normalized data, and respectively represent the maximum and minimum values of the data to be normalized, and finally the data is mapped between 0 and 1, and input into the model as feature data for training; and the processed data is divided into training set and test set in the ratio of 7:3, wherein the first 70% of the data set is used as the training set, and the last 30% is used as the test set.

[0027] As a further improvement of the technical solution, the specific process of S3 includes the following steps:

[0028] S3.1, model building and training: building a machine learning control model using SVR, RF, RBF, BP neural network machine learning algorithm, and using LSTM and other deep learning algorithms for complex situations; using machine learning as a control means, using the nonlinear mapping ability of machine learning to learn the relationship between input voltage, input current, temperature, output voltage, output current and pulse width, frequency, jump pulse number;

[0029] ​S3.2, Model evaluation: apply machine learning in the design of power converter control systems, while introducing physical constraint loss into the loss function of the machine learning model to reflect the penalty term of physical law or system characteristics; design a loss function including label loss, considering the error between output voltage and expected voltage, and efficiency loss term to ensure that the output voltage remains within the error range of a fixed value; the design of physical constraint loss aims to ensure that the behavior predicted by the model is consistent with the actual physical system.

[0030] As a further improvement of the technical solution, in S3.2, the machine learning is used to predict the duty cycle, frequency and pulse jump number, and the label is the duty cycle, frequency and pulse jump number, and the loss function is defined according to the label loss, output voltage loss and efficiency loss;

[0031] The mean square error (MSE) or absolute error (MAE) is used to measure the error between the output label and the actual label, and the fluctuation of the output voltage, the error between the output voltage and the expected voltage, and the efficiency are introduced as constraint conditions, which are part of the loss function, i.e. a penalty term related to the constraint condition is added to the original loss function; if the model violates the constraint condition, the penalty term will increase the loss value, thereby encouraging the model to adjust to meet the constraint, and guiding the model to learn in the direction that meets the condition through the penalty term of the loss function; wherein:

[0032] The mean square error (MSE) is used to measure the prediction error of the label

[0033]

[0034] wherein, is the predicted pulse width of the i-th sample, is the actual pulse width of the i-th sample, is the number of samples; In order to ensure the stability of the output voltage, an additional constraint term is introduced on the basis of the basic loss function, i.e. the output voltage fluctuation, the output voltage deviation and the efficiency constraint, and three penalty terms are introduced accordingly, i.e. the fluctuation penalty term, the voltage deviation penalty term and the efficiency penalty term;

[0035] A penalty term is defined to measure the fluctuation of the output voltage:

[0036]

[0037]

[0038] wherein, is the fluctuation function of the output voltage, i.e. the fluctuation penalty, is the fluctuation of the output voltage, ​​​Output voltage at each time point; It is the first Output voltage at each time point;

[0039] Define a penalty term to measure the deviation between the output voltage and the target voltage; use mean square error (MSE) to measure the error between the output voltage and the target voltage. This refers to the deviation penalty, which serves as a constraint on the output voltage.

[0040]

[0041] in, It is the target output voltage. It is the output voltage;

[0042] Introducing efficiency constraints helps machine learning models better understand and predict system behavior:

[0043]

[0044] in, It is the output power efficiency constraint function, i.e., power penalty; , They are the first Input voltage and input current at each time point , They are the first Output voltage and output current at each time point;

[0045] The basic loss function and the stability constraint term are combined to form the comprehensive loss function:

[0046]

[0047] in, It is the first The actual output voltage of each sample is a normalized value; It is the desired output voltage, a normalized value, which is a fixed value.

[0048] in, It is a hyperparameter used to control the weight of the fluctuation penalty; It is a hyperparameter used to control the weight of the bias penalty; It is a hyperparameter used to control the weight of the power penalty.

[0049] As a further improvement of the technical solution, in S5, the input voltage, input current, output voltage, output current and temperature measured in real time are input into the duty cycle, frequency and pulse jump number prediction model to perform real-time prediction of the duty cycle, frequency and pulse jump number, and perform reverse normalization operation; wherein the reverse normalization formula is:

[0050]

[0051] wherein, is the original data, is the normalized data, and are the minimum and maximum values in the original data, respectively;

[0052] The reverse normalization process is to convert the normalized data back to the scale of the original data.

[0053] As a further improvement of the technical solution, in S6, the efficiency of the power converter when running and the predicted efficiency of the other two modes are used to intelligently select whether the power converter is running in PWM mode, PFM mode or PSM mode; the input voltage , output voltage , output current , cycle length are used as features, and the input current is used as a label to train the input current prediction model under PWM, PFM and PSM modulation; if the power converter is now running in PWM mode, , , , can be measured and obtained, and the running efficiency can be measured as:

[0054]

[0055] At this time, through the input current prediction model of PFM / PSM, the input current in PFM / PSM mode is predicted through the measured , , , , and the efficiency in this mode can be obtained:

[0056]

[0057] wherein, is the measured value of the input current, is the predicted value of the input current;

[0058] By comparing , , select the operation mode of the power converter.

[0059] The second object of the present application is to provide a power converter control device based on physical information machine learning, comprising a processor, a memory and a computer program stored in the memory and running on the processor, the processor being configured to implement the steps of the power converter control method based on physical information machine learning as described above when executing the computer program.

[0060] The third object of the present application is to provide a computer readable storage medium storing a computer program, the computer program being configured to implement the steps of the power converter control method based on physical information machine learning as described above when executed by a processor.

[0061] Compared with the prior art, the present application has the following advantages:

[0062] 1. In the power converter control method, device and storage medium based on physical information machine learning, the duty cycle, frequency and pulse jump number of the control signal are collected, the input voltage, input current, output voltage, output current and temperature are collected and A / D converted, these information are spliced and stored for training the machine learning control model, in order to adapt the machine learning control model to high frequency control of the signal, the physical constraint information is embedded into the loss function, the relationship between the duty cycle, frequency and jump pulse number of the control signal and the input voltage, input current, output voltage, output current and temperature is mined by using the machine learning algorithm, and the machine learning control model is deployed on the FPGA to predict and control the duty cycle, frequency and pulse jump number of the control signal in real time, the information and dead time are sent to the PWM signal generator, PFM signal generator and PSM signal generator respectively, the switching tube of the power converter is controlled, and the stable output of the output voltage is realized.

[0063] 2. In the power converter control method, device and storage medium based on physical information machine learning, since machine learning is based on statistics, probability theory and the like, it has a relatively strong mathematical foundation. Traditional machine learning models such as linear regression, logistic regression and decision tree have fewer parameters and the meaning of each parameter is clear, easy to understand and explain. The scheme preferentially adopts SVR, RF, RBF and BP neural network machine learning algorithms, and uses LSTM and other deep learning algorithms for complex conditions. Through the nonlinear capability of the machine learning model, the nonlinear influence of input voltage, input current, output voltage, output current and temperature on the duty cycle, frequency and pulse jump number of the control signal can be accurately modeled, compared with modeling by mathematical methods, a large amount of cost can be saved, and the accuracy of the model can be improved. Machine learning combines physical laws and machine learning technology, uses known physical laws to guide the design and training process of the machine learning model, and fuses physical constraints and data driving to improve the prediction ability, generalization ability and interpretability of the model. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 The example power converter control system overall scheme diagram in the application;

[0065] Figure 2 The example power converter control model overall architecture diagram in the application;

[0066] Figure 3 The example power converter data acquisition block diagram in the application;

[0067] Figure 4 The example model training into loss function construction illustration in the application;

[0068] Figure 5 The example model deployment flowchart in the application;

[0069] Figure 6 The example pulse width frequency pulse jump number prediction model selection algorithm diagram in the application;

[0070] Figure 7 The example electronic computer platform device architecture diagram in the application. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0072] Example 1

[0073] like Figures 1-6 As shown, this embodiment provides a power converter control method based on physical information machine learning, characterized by the following steps:

[0074] S1. Control System Construction: Construct a control system architecture that includes a power converter, a data acquisition and processing module, a machine learning control algorithm, and a control signal generator.

[0075] In the control system architecture constructed in this step:

[0076] Power converters include at least Buck, Boost, Buck-Boost, CUK and resonant converters, as well as flyback converters, forward converters, push-pull converters, half-bridge converters and full-bridge converters, and parallel topologies of these types of converters, and power converters that use ZVS and ZCS technologies to reduce the switching losses of power devices.

[0077] The data acquisition and processing module includes the acquisition of control signal duty cycle, frequency and pulse jump number, as well as the acquisition and A / D conversion of input voltage, input current, output voltage, output current and temperature; it also needs to store the maximum and minimum data for the inverse normalization process, so as to convert the normalized data back to the scale of the original data.

[0078] Machine learning control algorithms include at least SVR, RF, BP neural network, RBF and LSTM; the prediction model selection algorithm for power converters can also use these machine learning algorithms. The prediction model selection algorithm selects whether the power converter topology is to operate in PWM mode or PFM / PSM mode based on the power converter's operating efficiency and prediction efficiency.

[0079] The control signal generator generates PWM (Pulse Width Modulation), PFM (Pulse Frequency Modulation), and PSM (Pulse Step Modulation) signals based on the duty cycle, frequency, number of pulse transitions, and dead time predicted by the machine learning algorithm. These signals are then used to control the switching of power MOSFETs or IGBT power devices via the drive circuit.

[0080] The core of power converter control implementation includes data acquisition and preprocessing, construction and training of a predictive model for control signals (pulse width, frequency, and number of pulse transitions), and model deployment, such as... Figure 2 As shown.

[0081] S2, data collection, storage and preprocessing: collect the physical information and control signal data of the power converter operation, splice and store the collected data to facilitate the training and application of machine learning algorithms, and perform cleaning and normalization preprocessing on the data; the specific process includes the following steps:

[0082] S2.1, data collection: collect the input and output voltage and current of the power converter and the environmental parameters, including input voltage, input current, output voltage, output current and temperature data of the heat sink, transformer and air, as the features of machine learning; collect the control signal data, including the duty cycle, frequency and pulse jump number data of the control signal, as the label data of machine learning; specifically as shown in Figure 3 ;

[0083] Wherein, when collecting the jump pulse number data, multiple factors need to be considered, including the response time of the system, the frequency of load change, the requirements of the control algorithm, the stability of the output voltage and the computing power of the system, etc. A suitable measurement period length can optimize the performance of the PSM system.

[0084] Since the power converter uses high-frequency signal driving, a high-frequency signal synchronous data collection scheme is required to ensure the collection of control signal duty cycle and frequency, and the time synchronization of input voltage, input current, output voltage, output current and temperature collection, and the accuracy of data collection.

[0085] For pulse width and frequency sampling, the rising edge or falling edge of the pulse can be captured by the timer of STM32, and the timer interrupt is used to calculate the pulse width and frequency, and further calculate the pulse jump number.

[0086] For sampling labels, suppose the sampling period For a 500kHz PWM signal, it includes 500 PWM pulse signals, and the machine learning model predicts the pulse width of each of the 500 PWM signals; for a 500kHz PSM (pulse jump modulation) signal, it also includes 500 potential pulses in 1 millisecond, and the machine learning model predicts the number of actual jumps (i.e. the appearance of pulses) in the 500 potential pulses; for a PFM (pulse frequency modulation) signal, the machine learning model predicts the interval time between pulses in 1 millisecond, and the number of pulses is calculated according to the pulse width and total time.

[0087] In addition, data collection also includes the following constraints:

[0088] First, data collection for modeling:

[0089] The data from the start of the power converter to a certain period of subsequent work must be covered;

[0090] Particular attention is paid to data collection when switching between PWM and PFM / PSM to ensure the accuracy of inference under different modulation modes;

[0091] The initial environment at startup should cover seasonal temperature changes.

[0092] Second, data collection for inference:

[0093] Data from a period before inference must be included to ensure sufficient context information;

[0094] Initial environmental data is included so that the model can be accurately initialized;

[0095] Particular attention is paid to data collection when switching between PWM and PFM / PSM to ensure the accuracy of inference under different modulation modes;

[0096] Data collection for inference can only retain a small amount of data to speed up inference.

[0097] S2.2, Data Storage: Sort the power converter data by collection time, tag with PWM duty cycle and PFM frequency, and use input voltage, input current, output voltage, output current and temperature characteristics as model training data to form a data set, as shown in Table 1 below. Data collection for training mode must ensure the integrity of data collection:

[0098] Table 1 Power Converter Data Table Structure

[0099]

[0100] S2.3, Data Preprocessing: Fill in missing values of collected power converter data, and normalize the data set to speed up model convergence.

[0101] In this step, data preprocessing specifically includes:

[0102] The data set has some missing values, which are supplemented with data from the previous time point, or supplemented with weighted average of the previous and subsequent data;

[0103] In the case of standard dimension, the data difference of input voltage, input current, output voltage, output current and temperature characteristic data of power converter, and duty cycle, frequency, pulse jump number tag data is large, the maximum and minimum value normalization preprocessing is carried out on the data, and the features and labels of different scales are scaled to the range of 0 to 1, to eliminate the influence of large and small data on the model, and the normalization formula is:

[0104] ​

[0105] wherein, is the data to be normalized, i.e. the raw data, is the normalized data, and respectively represent the maximum and minimum values in the data to be normalized, and after finally mapping the data between 0 and 1, the data is input into the model for training as feature data; and the processed data is divided into a training set and a test set in a ratio of 7:3, wherein the first 70% of the data set is used as the training set and the last 30% is used as the test set.

[0106] S3, model building and evaluation: a precise power converter operation-related mathematical control model is built using machine learning, and the model is evaluated and optimized based on a loss function; the specific process includes the following steps:

[0107] S3.1, model building and training: a machine learning control model using SVR, RF, RBF, BP neural network machine learning algorithm is constructed, and deep learning algorithms such as LSTM are used for complex situations; in the design of the power converter, the high nonlinearity of the power converter makes it very difficult to establish an accurate mathematical model, resulting in the difficulty of nonlinear PID design and the difficulty of debugging, and machine learning is used as a control means to learn the relationship between input voltage, input current, temperature, output voltage, output current and pulse width, frequency, and the number of jump pulses using the nonlinear mapping capability of machine learning;

[0108] S3.2, model evaluation: machine learning is applied in the design of the power converter control system, and a physical constraint loss is introduced into the loss function of the machine learning model to reflect a penalty term for physical laws or system characteristics; the loss function includes a label loss, considers the error between the output voltage and the expected voltage, and an efficiency loss term to ensure that the output voltage remains within a fixed error range; the design of the physical constraint loss aims to ensure that the behavior predicted by the model is consistent with the behavior of the actual physical system.

[0109] In this step, machine learning predicts duty cycle, frequency, and pulse count, with these three parameters as labels. The loss function is defined based on label loss, output voltage loss, and efficiency loss. Output voltage and output power efficiency are included in the loss function because duty cycle adjustment directly affects output voltage quality, and output voltage and output power efficiency are the most important performance indicators in practical applications. Therefore, even if the model directly predicts duty cycle, frequency, and pulse count, by including output voltage fluctuations, output voltage deviations, and changes in input and output power in the loss function, the trained model can consider output voltage fluctuations, deviations, and efficiency. Selecting parameters that better reflect the final system performance in the loss function helps optimize the model's practical application performance. Figure 4 As shown;

[0110] Mean squared error (MSE) or absolute error (MAE) is used to measure the error between the output label and the actual label. Output voltage fluctuations, the error between the output voltage and the expected voltage, and efficiency are introduced as constraints. These constraints are incorporated into the loss function, i.e., a penalty term related to the constraints is added to the original loss function. If the model violates the constraints, the penalty term increases the loss value, thereby encouraging the model to adjust to meet the constraints. The penalty term in the loss function guides the model to learn in a direction that satisfies the conditions. Where:

[0111] The mean squared error (MSE) is used to measure the prediction error of the labels. :

[0112]

[0113] in, It is the first The predicted pulse width for each sample, It is the first The actual pulse width of each sample It is the sample size;

[0114] To ensure the stability of the output voltage, additional constraint terms are introduced on the basis of the basic loss function, namely output voltage fluctuation, output voltage deviation and efficiency constraints, and three penalty terms are introduced accordingly, namely fluctuation penalty term, voltage deviation penalty term and efficiency penalty term.

[0115] Define a penalty term to measure the fluctuation of the output voltage: for example, the change in output voltage between consecutive time points can be calculated and incorporated into the loss function.

[0116]

[0117] in, is a fluctuation function of the output voltage, i.e., a fluctuation penalty, is the output voltage at the th time point; is the output voltage at the th time point;

[0118] A penalty term is defined to measure the deviation between the output voltage and the target voltage; the mean square error (MSE) is used to measure the error between the output voltage and the target voltage , i.e., a deviation penalty, as a constraint condition for the output voltage:

[0119]

[0120] wherein, is the target output voltage, is the output voltage;

[0121] The introduction of the efficiency constraint helps the machine learning model better understand and predict the behavior of the system:

[0122]

[0123] wherein, is an output power efficiency constraint function, i.e., a power penalty; , are the input voltage and input current at the th time point, respectively, , are the output voltage and output current at the th time point, respectively;

[0124] The basic loss function and the stability constraint term are combined to form a comprehensive loss function:

[0125]

[0126] wherein, is the actual output voltage of the th sample, which is a normalized value; is the expected output voltage, which is a normalized value, and this value is a fixed value;

[0127] wherein, is a hyperparameter for controlling the weight of the fluctuation penalty; is a hyperparameter for controlling the weight of the deviation penalty; is a hyperparameter for controlling the weight of the power penalty.

[0128] S4, model deployment: optimize and compress the model to reduce resource occupation and improve inference speed, compile the trained model into a format suitable for MCU, ASIC and FPGA, and deploy it on an embedded system; as Figure 5 shown;

[0129] S5, model application: input the real-time measured power converter operating parameters into the corresponding prediction model for real-time prediction;

[0130] In this step, the real-time measured input voltage, input current, output voltage, output current and temperature are input into the duty cycle, frequency and pulse jump number prediction model for real-time prediction of duty cycle, frequency and pulse jump number, and inverse normalization operation is performed; wherein the inverse normalization formula is:

[0131]

[0132] wherein, is the original data, is the normalized data, and are the minimum and maximum values in the original data, respectively;

[0133] The inverse normalization process is to convert the normalized data back to the original data scale. The specific operation depends on the normalization method you used in the preprocessing stage. Make sure to save the normalization parameters during training and use them for inverse normalization during prediction to obtain the prediction results in the original scale.

[0134] When generating the number of jump pulses of the PSM, the distribution of the jump pulses needs to be considered. The distribution of the jump pulses depends on the specific control strategy and system requirements, and there are uniform distribution, random distribution, concentrated distribution and other distribution modes. When implementing ZVS for the power switch tube, the number of jump pulses needs to be in a concentrated distribution mode, and the number of jump pulses needs to be even to ensure ZVS turn-off of the power switch tube. Therefore, when detecting that the number of jump pulses is odd, the number of pulses needs to be increased by 1 or decreased by 1.

[0135] S6, prediction model selection of power converter operating mode: intelligently select the operating model of the power converter through the operating efficiency and prediction efficiency of the power converter during operation;

[0136] In this step, as Figure 6 shown, through the efficiency of the power converter during operation and the prediction efficiency of the other two modes, it is intelligently selected whether the power converter is operating in PWM mode, PFM mode or PSM mode; the selection is based on input voltage , output voltage , output current , cycle length is the input current is the input current, the input current prediction model under PWM, PFM and PSM modulation is trained; it is assumed that the power converter is now running in PWM mode, , , , The measured efficiency during operation can be measured, and the measured efficiency during operation can be measured.

[0137]

[0138] At this time, through the input current prediction model of PFM / PSM, through the measured , , , , the input current in PFM / PSM mode can be predicted, and the efficiency in this mode can be obtained.

[0139]

[0140] wherein, is the measured value of the input current, is the predicted value of the input current;

[0141] By comparing , , the operating mode of the power converter is selected.

[0142] The technical solution proposes a power converter control method based on physical information machine learning; comprehensively considers the factors affecting the duty cycle and frequency of the control signal, i.e. input voltage, input current, output voltage, output current and temperature, wherein the temperature includes radiator temperature, transformer temperature and air temperature inside the device; proposes that when measuring and generating the number of jump pulses of PSM, the sampling period and pulse distribution theory need to be considered; it can be used alone in PWM-based power converter control method and PFM / PSM-based power converter control method, and also can be used in PWM and PFM / PSM-based power converter hybrid control method; proposes a design scheme of digital feedback loop and a design method of loss function based on power loss, uses machine learning algorithm for high-frequency fine control in high-frequency application scenarios, and outputs the expected output voltage The loss function is introduced.

[0143] Compared with the prior art, the technical solution has the following advantages:

[0144] (1) Through the digital way, the research and development of the power converter control circuit is innovated, especially for the research and development of wide voltage input, wide load output and high-power power converter, a new idea is provided;

[0145] (2) Adaptive control: The parameters of a PID controller are usually fixed or adjusted through simple adaptive algorithms; this makes it difficult for a PID controller to adjust in time to maintain optimal performance when system parameters change (such as load changes, temperature changes, etc.); machine learning models can learn and adapt to changes in system parameters, such as load changes, temperature changes, etc.; this means that even when system conditions change, the controller can maintain good performance;

[0146] (3) Nonlinear processing: Power converters often exhibit highly nonlinear behavior, and PID controllers may have difficulty handling nonlinear systems; machine learning-based control methods can better capture and handle these nonlinear relationships;

[0147] (4) Dynamic response: The response of a PID controller is usually based on linear assumptions and may not respond quickly and accurately to dynamic changes in the system, especially when system parameters change significantly or external disturbances exist; machine learning models, especially deep neural networks, can learn and predict the dynamic behavior of the system, providing faster and more accurate control responses;

[0148] (5) Robustness: PID controllers have poor robustness to model errors, measurement noise, and external disturbances; once the system deviates from the predetermined operating point, the PID controller may not be able to effectively respond, resulting in performance degradation; machine learning-based controllers can handle uncertainties such as model errors, measurement noise, and external disturbances through learning, thereby improving the robustness of the system;

[0149] (6) Global optimization: PID controller parameter adjustment is usually based on local optimization methods and may not find the global optimal solution; in contrast, machine learning can find a better control strategy through global search;

[0150] (7) Accurate modeling: PID controllers require an accurate system model for parameter tuning; for complex systems that are difficult to model or have uncertain models, the effectiveness of the PID controller may be affected; machine learning-based control methods can learn from large amounts of data, using historical data to improve control strategies; this is particularly important in the era of big data, where machine learning models do not require accurate modeling, but rather learn the input-output relationship of the system from data, which is particularly useful for complex systems that are difficult to model;

[0151] (8) Manual parameter adjustment: PID controller parameter adjustment usually requires professional knowledge and experience, which increases the workload and complexity of manual parameter adjustment; machine learning models can automatically adjust parameters through training, reducing the workload and complexity of manual parameter adjustment;

[0152] (9) Continuous optimization: over time, the model can improve control performance through continuous learning and optimization, and as computing power and data volume increase, machine learning models can be continuously improved and expanded to cope with more complex control tasks; when system parameters change or environmental conditions change, frequent adjustment of PID parameters is required, increasing maintenance costs and time;

[0153] (10) Flexibility: PID controllers are usually designed for specific systems and applications, lacking flexibility; if the system topology or application requirements change, the controller may need to be redesigned and adjusted; machine learning-based control methods can easily adapt to different power converter topologies and application requirements.

[0154] As shown in Figure 7 The embodiment also provides a power converter control device based on physical information machine learning, which includes a processor, a memory, and a computer program stored in the memory and running on the processor.

[0155] The processor includes one or more processing cores, and the processor is connected to the memory through a bus. The memory is used to store program instructions, and the processor executes the program instructions in the memory to implement the steps of the power converter control method based on physical information machine learning.

[0156] Optionally, the memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0157] In addition, the embodiment also provides a computer readable storage medium, which stores a computer program. When the computer program is executed by the processor, the steps of the power converter control method based on physical information machine learning are implemented.

[0158] Optionally, the present application also provides a computer program product containing instructions, which, when executed on a computer, causes the computer to perform the steps of the power converter control method based on physical information machine learning.

[0159] Those skilled in the art can understand that the process of implementing all or part of the steps of the above-mentioned embodiments can be completed by hardware, or by program to instruct relevant hardware, and the program can be stored in a computer readable storage medium, and the above-mentioned storage medium can be read-only memory, magnetic disk or optical disk, etc.

[0160] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Various changes and improvements can be made to the present application without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A power converter control method based on physical information machine learning, characterized in that, Includes the following steps: S1. Control System Construction: Construct a control system architecture that includes a power converter, a data acquisition and processing module, a machine learning control algorithm, and a control signal generator. Power converters include at least Buck, Boost, Buck-Boost, CUK and resonant converters, as well as flyback converters, forward converters, push-pull converters, half-bridge converters and full-bridge converters, and parallel topologies of these types of converters, and power converters that use ZVS and ZCS technologies to reduce the switching losses of power devices. The data acquisition and processing module includes the acquisition of control signal duty cycle, frequency and pulse jump number, as well as the acquisition and A / D conversion of input voltage, input current, output voltage, output current and temperature; Machine learning control algorithms include at least SVR, RF, BP neural network, RBF and LSTM; the prediction model selection algorithm for power converters can also use these machine learning algorithms. The prediction model selection algorithm selects whether the power converter topology is to operate in PWM mode or PFM / PSM mode based on the power converter's operating efficiency and prediction efficiency. The control signal generator generates pulse width modulation (PWM), pulse frequency modulation (PFM), and pulse step modulation (PSM) signals based on the duty cycle, frequency, number of pulse transitions, and dead time predicted by the machine learning algorithm. These signals are then used to control the switching of power MOSFETs or IGBT power devices via the drive circuit. S2. Data Acquisition, Storage and Preprocessing: Acquire physical information and control signal data of the power converter operation, splice and store the acquired data, and perform data cleaning and normalization preprocessing. S3. Model Building and Evaluation: Utilize machine learning to build an accurate mathematical control model for the operation of the power converter, and evaluate and optimize the model based on the loss function. The specific process of S3 includes the following steps: S3.1 Model Building and Training: Construct a machine learning control model using SVR, RF, RBF, and BP neural network machine learning algorithms, and use LSTM deep learning algorithm for complex cases; use machine learning as the control method, and use the nonlinear mapping capability of machine learning to learn the relationship between input voltage, input current, temperature, output voltage, output current and pulse width, frequency and number of transition pulses. S3.2 Model Evaluation: Apply machine learning to the design of the power converter control system, and introduce physical constraint loss into the loss function of the machine learning model to reflect the penalty term of physical laws or system characteristics; design a loss function including label loss, considering the error between the output voltage and the desired voltage, and efficiency loss term to ensure that the output voltage is kept within the error range of a fixed value; In S3.2, when machine learning predicts duty cycle, frequency, and number of pulse transitions, the labels are duty cycle, frequency, and number of pulse transitions, and the loss function is defined based on label loss, output voltage loss, and efficiency loss. Mean Squared Error (MSE) or Mean Absolute Error (MAE) is used to measure the error between the output label and the actual label. Output voltage fluctuation, the error between the output voltage and the expected voltage, and efficiency are introduced as constraints. These constraints are incorporated into the loss function, i.e., a penalty term related to the constraints is added to the original loss function. If the model violates the constraints, the penalty term increases the loss value, thereby encouraging the model to adjust to meet the constraints. The penalty term in the loss function guides the model to learn in a direction that satisfies the conditions. Where: The mean squared error (MSE) is used to measure the prediction error of the labels. : ； in, It is the first The predicted pulse width for each sample, It is the first The actual pulse width of each sample It is the sample size; To ensure the stability of the output voltage, additional constraint terms are introduced on the basis of the basic loss function, namely output voltage fluctuation, output voltage deviation and efficiency constraints, and three penalty terms are introduced accordingly, namely fluctuation penalty term, voltage deviation penalty term and efficiency penalty term. Define a penalty term to measure the fluctuation of the output voltage: ； in, It is a function of the output voltage fluctuation, i.e., the fluctuation penalty. It is the first Output voltage at each time point; It is the first Output voltage at each time point; Define a penalty term to measure the deviation between the output voltage and the target voltage; use mean square error (MSE) to measure the error between the output voltage and the target voltage. This refers to the deviation penalty, which serves as a constraint on the output voltage. ； in, It is the target output voltage. It is the output voltage; Introducing efficiency constraints helps machine learning models better understand and predict system behavior: ； in, It is the output power efficiency constraint function, i.e., power penalty; , They are the first Input voltage and input current at each time point , They are the first Output voltage and output current at each time point; The basic loss function and the stability constraint term are combined to form the comprehensive loss function: ； in, It is the first The actual output voltage of each sample is a normalized value; It is the desired output voltage, a normalized value, which is a fixed value. in, It is a hyperparameter used to control the weight of the fluctuation penalty; It is a hyperparameter used to control the weight of the bias penalty; It is a hyperparameter used to control the weight of the power penalty; S4. Model Deployment: Optimize and compress the model, compile the trained model into a format suitable for MCU, ASIC and FPGA, and deploy it on the embedded system; S5. Model Application: Input the real-time measured power converter operating parameters into the corresponding prediction model for real-time prediction; S6. Predictive model selection for power converter operation mode: The power converter operation mode is intelligently selected based on the operating efficiency and predictive efficiency during operation.

2. The power converter control method based on physical information machine learning according to claim 1, characterized in that, The specific process of S2 includes the following steps: S2.1 Data Acquisition: Collect the input and output voltage and current of the power converter and the environmental parameters, including input voltage, input current, output voltage, output current, and temperature data of the heat sink, transformer, and air, as features for machine learning; Collect control signal data, including control signal duty cycle, frequency, and pulse transition count, as label data for machine learning; S2.2 Data storage: The power converter data is sorted according to the acquisition time, labeled with PWM duty cycle and PFM frequency, and the input voltage, input current, output voltage, output current and temperature features are used as model training data to form a dataset; S2.3 Data Preprocessing: Fill missing values ​​in the collected power converter data and normalize the dataset.

3. The power converter control method based on physical information machine learning according to claim 2, characterized in that: In step S2.3, the data preprocessing specifically includes: The dataset contains some missing values. These can be filled by using data from the previous time point or by using a weighted average of the data from the previous and previous time points. Under standard dimensions, the input voltage, input current, output voltage, output current, and temperature characteristic data of the power converter, as well as the duty cycle, frequency, and pulse transition number label data, exhibit significant differences. Therefore, a maximum-minimum value normalization preprocessing is performed on the data to scale the features and labels from different scales to a minimum value. Within a certain range, to eliminate the influence of large or small data on the model, the normalization formula is: ； In the formula, The data to be normalized, i.e., the original data. For the normalized data, and These represent the maximum and minimum values ​​in the data to be normalized, respectively. After mapping the data to between 0 and 1, it is used as feature data to input into the model for training. The processed data is then divided into a training set and a test set in a 7:3 ratio, with the first 70% of the dataset used as the training set and the last 30% used as the test set.

4. The power converter control method based on physical information machine learning according to claim 1, characterized in that: In step S5, the real-time measured input voltage, input current, output voltage, output current, and temperature input duty cycle, frequency, and pulse jump number prediction model are used to predict the duty cycle, frequency, and pulse jump number in real time, and then an inverse normalization operation is performed; wherein, the inverse normalization formula is: ； in, It is the raw data. It is normalized data. and These are the minimum and maximum values ​​in the original data, respectively. The process of denormalization is to convert the normalized data back to the scale of the original data.

5. The power converter control method based on physical information machine learning according to claim 4, characterized in that: In step S6, based on the efficiency of the power converter during operation and the predicted efficiency of the other two modes, the system intelligently selects whether the power converter operates in PWM mode, PFM mode, or PSM mode; and selects the input voltage... Output voltage Output current Cycle duration Characterized by input current Using labels, train input current prediction models under PWM, PFM, and PSM modulation; assume the power converter is currently operating in PWM mode. , , , All of these can be measured, and the runtime efficiency can be measured as follows: ； At this point, the input current prediction model of PFM / PSM is used to predict the current based on the measured current. , , , By predicting the input current in PFM / PSM mode, the efficiency of that mode can be obtained: ； in, This is the measured value of the input current. This is the predicted value of the input current; By comparison , Select the operating mode of the power converter.

6. A power converter control device based on physical information machine learning, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the power converter control method based on physical information machine learning as described in any one of claims 1-5.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, it implements the steps of the power converter control method based on physical information machine learning as described in any one of claims 1-5.

Images