Intelligent analysis method for steady-state characteristics of direct-current converter based on machine learning

Abstract

The invention discloses an intelligent steady-state characteristic analysis method for a direct-current converter based on machine learning, and belongs to the field of direct-current converters. A plurality of groups of steady-state characteristic numerical solutions of the direct-current converter under different driving strategies and system parameters are automatically obtained by utilizing circuit simulation software, and a function relationship among the steady-state characteristics of the direct-current converter, the driving strategies and the system parameters is fitted through a machine learning method, so that intelligent analysis of comprehensive steady-state characteristics of the direct-current converter is realized. The method provided by the invention effectively solves the problem that steady-state characteristic analysis is difficult to carry out on the direct-current converter considering parasitic parameters or containing a multi-order resonance unit by using an analytical method, and meanwhile, the limitation that simulation needs to be continuously repeated when simulation software is used for obtaining comprehensive steady-state characteristics of the direct-current converter and a large amount of time is occupied is solved.

Description

Technical Field

[0001] The present invention relates to the field of direct current converters, and in particular to an intelligent analysis method for steady-state characteristics of direct current converters based on machine learning, which can quickly realize accurate analysis of comprehensive steady-state characteristics of complex direct current converters. Background Art

[0002] When designing a DC converter product, the converter steady-state characteristic analysis must be completed first as a theoretical basis for the design. Currently, the common DC converter steady-state characteristic analysis methods mainly include analytical methods and numerical simulation methods. Among them, the analytical method is only applicable to solving differential equations within the second order. Therefore, it is difficult to complete the steady-state characteristic analysis of a DC converter that takes into account component parasitic parameters and contains multi-order resonant units. To address this problem, researchers proposed an ultra-high frequency converter analysis method that integrates Kalman filtering technology ([1] Chen Yanfeng, Jiang Xinyi, Zhang Bo, Qiu Dongyuan, Xiao Wenxun, Xie Fan. Ultra-high frequency converter analysis method integrating Kalman filtering technology. CN 112507643 A) to quickly obtain the steady-state period analytical solution of the ultra-high frequency converter, so as to further solve the steady-state characteristics of the DC converter. However, the implementation of this method is relatively complex and requires the artificial application of Kalman filtering technology, equivalent small parameter method, harmonic balance method, etc. In contrast, it is relatively simple to obtain the steady-state characteristics of a DC converter under given working conditions and parameters through simulation software. However, each simulation can only obtain the numerical solution of component voltage and current under certain given operating conditions and parameters, making it difficult to know the comprehensive steady-state characteristics of the DC converter; therefore, how to simply and quickly achieve comprehensive and accurate analysis of the steady-state characteristics of complex DC converters urgently needs further research. Summary of the invention

[0003] The purpose of the present invention is to provide an intelligent analysis method of the steady-state characteristics of a DC converter based on machine learning, in order to address the problem that it is difficult to use analytical methods to analyze the steady-state characteristics of a DC converter that takes into account parasitic parameters or contains multi-order resonant units in the prior art. Through automated data acquisition and intelligent data analysis, the steady-state characteristics of the DC converter are fully understood and predicted, thereby realizing intelligent analysis of the comprehensive steady-state characteristics of the DC converter. At the same time, the limitation that repeated simulations are required to obtain the comprehensive steady-state characteristics of the DC converter using simulation software, which takes up a lot of time, is solved.

[0004] In order to achieve the above-mentioned invention object, the present invention provides the following technical solutions:

[0005] A method for intelligent analysis of steady-state characteristics of a DC converter based on machine learning, comprising the following steps:

[0006] 1) Data acquisition: Use circuit simulation software to automatically obtain multiple sets of steady-state characteristic numerical solutions of DC converters under different driving strategies and system parameters;

[0007] 2) Data preprocessing: Clean and organize the data obtained from the simulation to ensure the accuracy and consistency of the data;

[0008] 3) Machine learning model construction: Use machine learning methods to fit the functional relationship between the steady-state characteristics of the DC converter and the driving strategy and system parameters;

[0009] 4) Model training and optimization: Use the preprocessed data from step 2) to train the machine learning model constructed in step 3) and improve the model's fitting accuracy and generalization ability by adjusting model parameters and optimizing algorithms;

[0010] 5) Model verification and application: Use the test set data to verify the trained model and evaluate the prediction accuracy and reliability of the model. The verified model is applied to the steady-state characteristics analysis of the actual DC converter. By inputting different driving strategies and system parameters, the intelligent analysis of the comprehensive steady-state characteristics of the DC converter is realized.

[0011] In step 1), the circuit simulation software may be Matlab / Simulink, PSIM, PLECS or other commonly used power electronic circuit simulation software, and the circuit simulation software is used to automatically modify parameters and implement batch simulation, thereby quickly generating a large number of numerical solutions for steady-state characteristics of DC converters.

[0012] In step 1), the different driving strategies may adopt PWM, PFM, etc., and the on and off times of different switch tubes may be freely changed.

[0013] In step 1), the system parameters include switching frequency, switch tube conduction duty cycle, input voltage, parameters of different components and system parasitic parameters, etc.; through automated simulation, a large number of numerical solutions for steady-state characteristics of the DC converter are generated, including output voltage average value, output voltage ripple value, inductor current average value, inductor current ripple value, inductor current effective value, capacitor voltage average value, capacitor voltage ripple value, switch tube current effective value, diode current average value, etc.

[0014] In step 2), the data preprocessing includes data cleaning, data organization, feature extraction, etc.; data cleaning is used to remove outliers and noise in the simulation data to ensure the accuracy and consistency of the data; data organization is used to organize the simulation data into a format suitable for processing by machine learning algorithms, such as CSV, Excel, etc.; feature extraction is used to extract key parameters and indicators that have an important impact on the steady-state characteristics of the DC converter, such as the average value, effective value, ripple amount, etc. of voltage and current, as input features of the machine learning model.

[0015] In step 3), the machine learning method is used to intelligently fit the function mapping relationship between multi-input and multi-output variables. Machine learning methods such as neural networks can be used to design a reasonable model structure, including the number of neurons in the input layer, hidden layer and output layer, activation function, etc.

[0016] In step 3), the steady-state characteristics of the DC converter include electrical indicators such as the average value, effective value, and ripple of the voltage and current of each component.

[0017] Compared with the prior art, the present invention has the following outstanding technical effects and advantages:

[0018] The present invention uses circuit simulation software to automatically obtain multiple groups of steady-state characteristic numerical solutions of DC converters under different driving strategies and system parameters, and fits the functional relationship between the DC converter steady-state characteristics and the driving strategy and system parameters through a machine learning method, thereby realizing intelligent analysis of the comprehensive steady-state characteristics of the DC converter. Through the automatic simulation of the circuit simulation software and the rapid prediction of the machine learning algorithm, the analysis efficiency of the DC converter steady-state characteristics can be greatly improved. The machine learning algorithm can intelligently fit complex functional relationships, thereby realizing accurate prediction of the DC converter steady-state characteristics. By covering a wide range of operating conditions and parameter combinations, the analysis of the comprehensive steady-state characteristics of the DC converter can be realized, providing strong support for optimization design and fault diagnosis. The present invention effectively solves the problem that it is difficult to use analytical methods to analyze the steady-state characteristics of DC converters that take into account parasitic parameters or contain multi-order resonant units, and at the same time solves the limitation that the simulation needs to be repeated continuously when the comprehensive steady-state characteristics of the DC converter are learned using simulation software, which takes up a lot of time. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention provides an intelligent analysis method for steady-state characteristics of a DC converter based on machine learning;

[0020] Figure 2 This is an example of a Buck converter taking parasitic parameters into account;

[0021] Figure 3 for Figure 2 Training results of the average output voltage of the Buck converter in the middle;

[0022] Figure 4 for Figure 2 Training results of the average value of the inductor current of the Buck converter;

[0023] Figure 5 for Figure 2 The average error rate of different electrical indicators in the steady-state characteristic analysis of Buck converter. DETAILED DESCRIPTION

[0024] To explain more clearly Figure 1 An intelligent analysis method for steady-state characteristics of a DC converter based on machine learning is described in the following. The principle of this method is described in detail in conjunction with the accompanying drawings and specific implementation methods; it is worth noting that the specific examples described here are only used to explain the present invention and are not used to limit the present invention.

[0025] The embodiment of the present invention uses circuit simulation software to automatically obtain multiple sets of steady-state characteristic numerical solutions of the DC converter under different driving strategies and system parameters, and fits the functional relationship between the DC converter steady-state characteristics and the driving strategy and system parameters through a machine learning method, thereby realizing intelligent analysis of the comprehensive steady-state characteristics of the DC converter.

[0026] The circuit simulation software can automatically modify parameters to achieve batch simulation, such as Matlab / Simulink simulation software. The system parameters include switching frequency, switch duty cycle, input voltage, parameters of different components and system parasitic parameters.

[0027] The steady-state characteristics of the DC converter include electrical indicators such as the average value, effective value, and ripple of the voltage and current of each component.

[0028] The machine learning method can intelligently fit the function mapping relationship between multiple input and output variables, such as machine learning methods such as neural networks.

[0029] like Figure 1 , an intelligent analysis method for the steady-state characteristics of a DC converter based on machine learning is provided. First, multiple groups of parameters of the DC converter system are set, including switching frequency, switch tube conduction duty cycle, input voltage, parameters of different components and system parasitic parameters, etc., and then circuit simulation software (such as the "sim" function of Matlab / Simulink, etc.) is used to batch simulate the above-set parameters to obtain the corresponding electrical quantities such as the average value, effective value, ripple amount and so on of the voltage and current of each component, and they are used as the input and output of the machine learning algorithm for training.

[0030] by Figure 1 Taking the neural network machine learning algorithm given as an example, a neural network structure consisting of an input layer, two hidden layers and an output layer is constructed. The input layer is responsible for receiving external inputs, such as various parameters of the DC converter, including switching frequency, switch duty cycle, input voltage, parameters of different components and system parasitic parameters, etc., respectively represented by x1, x2, …, x n Representation; The hidden layer has the characteristics of feature abstraction and nonlinear mapping, which is used to capture the input data x1, x2, …, x n The output layer is used to output the final prediction results P1, P2, …, Pm , such as the average value and ripple value of the output voltage of the converter or the inductor current. In the forward propagation stage, the input data is processed by multiple neurons in each layer, and forward calculations are performed layer by layer from the input layer to the output layer, and prediction results are generated; in the reverse propagation stage, according to the prediction results P1, P2,…, P m and the expected results P'1, P'2, …, P' m Based on the gradient descent strategy, the weight w in the network is quantized in the negative gradient direction of the target. ij The forward propagation and backward propagation iterative process is repeated until the network's mean square error mse drops below a set threshold ε, or the number of iterations reaches the upper limit. At this time, the network's predicted output {P1, P2, …, P m} will be very close to the expected result {P'1, P'2, …, P' m After training, the neural network can realize intelligent prediction of the steady-state characteristics of the DC converter. By inputting new DC converter parameters, the neural network can quickly output prediction results. These prediction results can be used to analyze, optimize and predict faults of the DC converter.

[0031] The following is a specific application example, in which the proposed intelligent analysis method of DC converter steady-state characteristics based on machine learning is applied to the Buck converter with parasitic parameters ( Figure 2 )middle.

[0032] 1. Parameter setting and simulation

[0033] Set the converter input voltage V in is 36V~66V, the inductor L is 15μH~25μH, and the inductor parasitic resistance R L is 0.05Ω~0.055Ω, the output capacitance C is 370nF~470nF, and the parasitic resistance R C is 0.05Ω~0.055Ω, the load resistance R is 6Ω~16Ω, and the on-resistance R of the switch tube S1 is S is 0.05Ω~0.055Ω, the diode’s conduction voltage V F The voltage is 0.4V to 1.2V, the duty cycle of the switch tube varies from 0.3 to 0.8, and the switching frequency is 1MHz. Figure 2 In the figure, V0 represents the output voltage; i L Indicates current.

[0034] 1000 sets of data are randomly selected within the above parameter range, and the 1000 sets of data are batch simulated using Matlab / Simulink simulation software to obtain the voltage and current values ​​of each component, thereby obtaining 1000 sets of voltage and current values ​​of each component.

[0035] 2. Machine Learning Algorithm Training

[0036] 900 sets of data were used as training sets, and the remaining 100 sets of data were used as validation sets.

[0037] Figure 3 exhibit Figure 2 The training results of the average output voltage of the Buck converter in the figure show that the fitting degrees of the training set, validation set, test set and comprehensive set are all approximately 1, which shows that the neural network can well fit and predict the changing trend of the output voltage. Figure 4 exhibit Figure 2 The training results of the average value of the inductor current of the Buck converter in the figure show that the fitting degrees of the training set, validation set, test set and comprehensive set are all approximately 1, which also shows that the neural network has a high accuracy in predicting the inductor current.

[0038] 3. Verification and error analysis

[0039] By using the remaining 100 sets of validation set data for prediction, we can obtain the average error rate of different electrical indicators in the steady-state characteristic analysis of the Buck converter, compare it with the simulation results, and calculate the average error rate of different electrical indicators.

[0040] Figure 5 exhibit Figure 2 The average error rate of different electrical indicators in the steady-state characteristic analysis of the Buck converter (including the effective value of the diode current I Frms , average diode current I Favg , Switching tube current effective value I Srms , the average value of the switch tube current I Savg , the average value of inductor current I Lavg , inductor current effective value I Lrms , inductor current peak-to-peak value I Lpk-pk , output voltage average value U oavg , output voltage effective value U orms , output voltage peak-to-peak value U opk-pk ), it can be seen that in addition to the peak-to-peak value of the inductor current I Lpk-pk In addition, the error rates of the calculated results of other voltage and current characteristics are all within 0.1%, which shows that the neural network has high accuracy and reliability in predicting the steady-state characteristics of the Buck converter.

[0041] This method can accurately predict the output voltage and inductor current of the converter and other key electrical indicators, providing strong support for the design, optimization and fault prediction of the converter. This method can be further extended to other types of DC converters and more complex circuit systems, contributing to the intelligent development of power electronics.

[0042] In summary, the intelligent analysis method of the steady-state characteristics of a DC converter based on machine learning proposed in the present invention has the advantages of simple implementation and high accuracy. It can effectively overcome the shortcomings of traditional analytical methods or simulation methods, and is of great significance for the theoretical design of DC converters. It provides new ideas and methods for the intelligent analysis of power electronic systems.

[0043] The above embodiments are only preferred embodiments of the present invention and cannot be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A method for intelligent analysis of steady-state characteristics of a DC converter based on machine learning, characterized in that The following steps are involved: 1) Data acquisition: Use circuit simulation software to automatically obtain multiple sets of steady-state characteristic numerical solutions of DC converters under different driving strategies and system parameters; 2) Data preprocessing: Clean and organize the data obtained from the simulation to ensure the accuracy and consistency of the data; 3) Machine learning model construction: Use machine learning methods to fit the functional relationship between the steady-state characteristics of the DC converter and the driving strategy and system parameters; 4) Model training and optimization: Use the preprocessed data from step 2) to train the machine learning model constructed in step 3) and improve the model's fitting accuracy and generalization ability by adjusting model parameters and optimizing algorithms; 5) Model verification and application: Use the test set data to verify the trained model and evaluate the prediction accuracy and reliability of the model. The verified model is applied to the steady-state characteristics analysis of the actual DC converter. By inputting different driving strategies and system parameters, the intelligent analysis of the comprehensive steady-state characteristics of the DC converter is realized.

2. The method for intelligent analysis of steady-state characteristics of a DC converter based on machine learning as claimed in claim 1, characterized in that In step 1), the circuit simulation software uses Matlab / Simulink, PSIM, PLECS power electronic circuit simulation software, and the circuit simulation software is used to automatically modify parameters and realize batch simulation, so as to quickly generate a large number of numerical solutions for the steady-state characteristics of the DC converter.

3. The method for intelligent analysis of steady-state characteristics of a DC converter based on machine learning as claimed in claim 1, characterized in that In step 1), the different driving strategies adopt PWM or PFM, and the on and off times of different switch tubes can be freely changed.

4. The method for intelligent analysis of steady-state characteristics of a DC converter based on machine learning as claimed in claim 1, characterized in that In step 1), the system parameters include switching frequency, switch conduction duty cycle, input voltage, parameters of different components and system parasitic parameters.

5. The method for intelligent analysis of steady-state characteristics of a DC converter based on machine learning as claimed in claim 1, characterized in that In step 1), the plurality of groups of steady-state characteristic numerical solutions include output voltage average value, output voltage ripple value, inductor current average value, inductor current ripple value, inductor current effective value, capacitor voltage average value, capacitor voltage ripple value, switch tube current effective value, and diode current average value.

6. The method for intelligent analysis of steady-state characteristics of a DC converter based on machine learning as claimed in claim 1, characterized in that In step 2), the data preprocessing includes data cleaning, data sorting, and feature extraction; data cleaning is used to remove outliers and noise in the simulation data to ensure the accuracy and consistency of the data; Data organization is used to organize simulation data into a format suitable for processing by machine learning algorithms, such as CSV and Excel. Feature extraction is used to extract key parameters and indicators that have an important impact on the steady-state characteristics of the DC converter, such as the average value, effective value, and ripple of voltage and current, as input features of the machine learning model.

7. The method for intelligent analysis of steady-state characteristics of a DC converter based on machine learning as claimed in claim 1, characterized in that In step 3), the machine learning method is used to intelligently fit the function mapping relationship between multi-input and multi-output variables, and a neural network machine learning method is used to design a reasonable model structure, including the number of neurons and activation functions in the input layer, hidden layer and output layer.

8. The method for intelligent analysis of steady-state characteristics of a DC converter based on machine learning as claimed in claim 1, characterized in that In step 3), the steady-state characteristics of the DC converter include the average value, effective value, and ripple of the voltage and current of each component.

Images