T-type filter insertion loss prediction method based on segmented Gaussian process regression

Abstract

The invention discloses a T-type filter insertion loss prediction method based on segmented Gaussian process regression. The method comprises the following steps: designing a T-type filter; determining the input parameters of the filter as capacitance, inductance and inter-element distance, and determining the output parameter as insertion loss; the vector network analyzer actually measures the insertion loss of the filter and preprocesses the actually measured data; compiling a frequency segmentation type prediction code based on Gaussian process regression; codes are input into the raspberry Pi, and raspberry Pi expansion equipment, data import and operation training are predicted; and evaluating the prediction data of the insertion loss by determining the coefficient evaluation index. According to the method, the insertion loss parameters of the designed filter are rapidly and reliably obtained through machine learning, and the design cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic compatibility technology for power electronic equipment, and specifically to a method for predicting the insertion loss of a T-type filter based on piecewise Gaussian process regression. Background Technology

[0002] In modern electronic systems, with the continuous increase in switching frequency, electromagnetic interference (EMI) has become a key challenge for the reliability and electromagnetic compatibility design of power electronic equipment. T-type filters, as a traditional EMI suppression solution, are widely used in various power electronic systems due to their advantages such as simple structure, high cost-effectiveness, and excellent filtering performance. However, with the expansion of the operating frequency range, especially in the high-frequency domain, there is a significant difference between the insertion loss prediction of traditional T-type filters based on ideal component models and the actual measurement results, which seriously affects the accuracy and optimization efficiency of filter parameter design.

[0003] This discrepancy stems primarily from two factors: First, the passive components (inductors and capacitors) in filters exhibit complex nonlinear characteristics at high frequencies, requiring their equivalent circuit models to account for parasitic parameters and coupling effects between components. Second, the approximation methods used in traditional theoretical modeling are insufficient to accurately describe actual filter performance, especially near the resonant point and at high frequencies. Studies have shown that when the frequency exceeds 1MHz, the insertion loss error between the theoretical model and actual measurements can reach over 20dB, making it difficult for design engineers to accurately predict filter performance. They are forced to rely on repeated experiments to adjust parameters, significantly reducing design efficiency. Summary of the Invention

[0004] Purpose of the invention: In order to overcome the shortcomings of the prior art, the present invention provides a method for predicting the insertion loss of a T-type filter based on piecewise Gaussian process regression. This method uses machine learning to quickly and reliably obtain the insertion loss parameters of the designed filter, thereby reducing design costs.

[0005] Technical solution: The insertion loss prediction method for T-type filters based on piecewise Gaussian process regression provided by this invention includes the following steps: S1. Design a T-type filter; S2. Determine the input parameters of the filter as capacitance, inductance, and distance between components, and the output parameter as insertion loss; S3. Measure the filter insertion loss using a vector network analyzer and preprocess the measured data; S4. Write frequency-based piecewise prediction code based on Gaussian process regression; S5. Input code into Raspberry Pi, expand the Raspberry Pi device, import data, and run training to make predictions; S6. Evaluate the predicted insertion loss data using the coefficient of determination evaluation index.

[0006] In S1, the T-type filter is either CLC or LCL type.

[0007] Furthermore, in the filter model, for CLC type, the two capacitors are set to have the same capacitance value, and for LCL type, the two inductors are set to have the same inductance value.

[0008] Furthermore, the preprocessing in S3 includes removing bad data, and the remaining qualified data is used as usable data and divided into training group and validation group.

[0009] Furthermore, S4 improves the prediction accuracy of low-sample frequency bands by dividing the frequency into high, medium, and low segments.

[0010] Furthermore, the kernel function ARD used in Gaussian process regression in S4 is:

[0011] Where χ i , χ j Given a D-dimensional input vector, the sample i and samples j , χ i，d , χ j，d For vector χ i , χ j In the d eigenvalues ​​of dimension The variance of the input signal. For the first d The length scale of the dimension, where D is the total number of dimensions of the input data.

[0012] Furthermore, in step S6, the coefficient of determination R... 2 Specifically:

[0013] in These are actual measured data. For predicted values, R is the average of all measured values. 2 It can reflect the difference between the model's predicted value and the actual value. The closer it is to 1, the better the model fits the data.

[0014] Beneficial effects: Compared with the prior art, the significant advantage of this invention is that it uses machine learning to quickly and reliably predict the insertion loss of the T-type filter composed of given components, thereby reducing the time and material costs consumed by iterative design. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0016] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0017] like Figure 1 The T-type filter insertion loss prediction method based on piecewise Gaussian process regression shown includes the following steps: S1. Design a T-type filter, which can be either CLC or LCL type.

[0018] S2. Determine the input parameters of the filter as capacitor C, inductor L, and distance d between components, and the output parameter as insertion loss in -dB. To simplify the filter model, for CLC type, set the two capacitors to have the same capacitance value; for LCL type, set the two inductors to have the same inductance value.

[0019] S3. The vector network analyzer measures the filter insertion loss and preprocesses the measured data to remove bad data. The remaining qualified data is used as usable data and divided into training group and validation group.

[0020] S4. Write frequency segmented prediction code based on Gaussian process regression to improve the prediction accuracy of low-sample frequency bands by dividing the frequency into three segments: high, medium and low.

[0021] The kernel function ARD used in Gaussian process regression is:

[0022] Where χ i , χ j Given a D-dimensional input vector, the sample i and samples j , χ i，d , χ j，d For vector χ i , χ j In the d eigenvalues ​​of dimension The variance of the input signal. For the first d The length scale of the dimension, where D is the total number of dimensions of the input data.

[0023] S5. Input code into the Raspberry Pi, extend the screen and buttons on the Raspberry Pi for easy operation, import data, and run training for prediction.

[0024] S6. Evaluate the predicted insertion loss data using the coefficient of determination evaluation index.

[0025] Coefficient of determination R 2 Specifically:

[0026] in These are actual measured data. For predicted values, R is the average of all measured values. 2 It can reflect the difference between the model's predicted value and the actual value. The closer it is to 1, the better the model fits the data.

Claims

1. A method for predicting the insertion loss of a T-type filter based on piecewise Gaussian process regression, characterized in that, Includes the following steps: S1. Design a T-type filter; S2. Determine the input parameters of the filter as capacitance, inductance, and distance between components, and the output parameter as insertion loss; S3. Measure the filter insertion loss using a vector network analyzer and preprocess the measured data; S4. Write frequency-based piecewise prediction code based on Gaussian process regression; S5. Input code into Raspberry Pi, expand the Raspberry Pi device, import data, and run training to make predictions; S6. Evaluate the predicted insertion loss data using the coefficient of determination evaluation index.

2. The method for predicting insertion loss of a T-type filter based on piecewise Gaussian process regression according to claim 1, characterized in that: The T-type filter in S1 is either CLC or LCL type.

3. The method for predicting insertion loss of a T-type filter based on piecewise Gaussian process regression according to claim 2, characterized in that: In the filter model, for CLC type, the two capacitors are set to have the same capacitance value, and for LCL type, the two inductors are set to have the same inductance value.

4. The method for predicting insertion loss of a T-type filter based on piecewise Gaussian process regression according to claim 1, characterized in that: Preprocessing in S3 includes removing bad data, and the remaining qualified data is used as usable data and divided into training group and validation group.

5. The method for predicting the insertion loss of a T-type filter based on piecewise Gaussian process regression according to claim 1, characterized in that: S4 improves the prediction accuracy of low-sample frequency bands by dividing the frequency into three segments: high, medium, and low.

6. The method for predicting the insertion loss of a T-type filter based on piecewise Gaussian process regression according to claim 1, characterized in that, The kernel function ARD used in Gaussian process regression in S4 is: ， Where χ i , χ j Given a D-dimensional input vector, the sample i and samples j , χ i，d , χ j，d For vector χ i , χ j In the d eigenvalues ​​of dimension The variance of the input signal. For the first d The length scale of the dimension, where D is the total number of dimensions of the input data.

7. The method for predicting the insertion loss of a T-type filter based on piecewise Gaussian process regression according to claim 1, characterized in that, The coefficient of determination R in step S6 2 Specifically: ， in These are actual measured data. For predicted values, R is the average of all measured values. 2 It can reflect the difference between the model's predicted value and the actual value. The closer it is to 1, the better the model fits the data.

Images