Knowledge transfer microwave filter optimization commissioning method, device, and storage device

Abstract

The application provides a microwave filter optimization debugging method and device based on knowledge transfer and a storage device, and comprises the following steps: for the initial microwave filter debugging task, establishing an initial knowledge generation model, determining an initial optimization range and optimizing and debugging the initial microwave filter, and based on the established initial knowledge generation model and the determined initial optimization range, optimizing and debugging the initial microwave filter; for the debugging task of other microwave filters, evaluating the difference degree, determining the transfer strength, establishing the knowledge generation model, determining the optimization range and optimizing and debugging, based on the difference degree and the transfer strength obtained in the difference degree evaluation stage, fine-tuning the initial knowledge generation model to obtain the knowledge generation model; and based on the knowledge generation model and the determined optimization range, optimizing and debugging the other initial microwave filter. The application has the beneficial effect of realizing efficient debugging of microwave filters between individual differences, transferring knowledge between different tasks and improving the overall debugging efficiency.

Description

Technical Field

[0001] This invention relates to the field of microwave filter optimization, and in particular to a microwave filter optimization and debugging method, equipment, and storage device based on knowledge transfer. Background Technology

[0002] Microwave filters are core frequency selection devices in microwave communication systems, capable of selecting useful signals and filtering out unwanted signals. With the vigorous promotion of "Network Power" and "Digital China," and the inclusion of 5G base station construction in my country's new infrastructure development, the market demand for microwave filters is enormous. Increased demand has led to an increase in batches; while different batches of microwave filters have the same structure, their material properties differ, resulting in significant individual differences in the material characteristics of microwave filters.

[0003] The production process of microwave filters includes three steps: design, manufacturing, and debugging. Design errors and manufacturing tolerances are difficult to avoid, and it is usually necessary to adjust the structural geometry parameters after manufacturing to change the filtering performance to meet the performance requirements. Currently, debugging mainly relies on experienced workers, but these workers have poor adaptability when dealing with microwave filters with significant individual differences, resulting in low debugging efficiency and difficulty in meeting the huge market demand.

[0004] Particle swarm optimization (PSO)-based tuning methods demonstrate excellent performance in tuning single microwave filters. However, when tuning multiple microwave filters, consistently starting from scratch results in low overall efficiency. For individual microwave filters with unique characteristics, while performance requirements may be identical, the mapping between structural geometry parameters and filtering performance differs, leading to feasible solutions for those parameters being distributed across different spatial domains. Precisely locating these different spatial domains before applying PSO can significantly improve tuning efficiency, but repeatedly locating different feasible solution spaces for structural geometry parameters from scratch each time is inefficient. The tuning process generates a large amount of data containing rich tuning knowledge, which is crucial for determining the feasible solution space for structural geometry parameters. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a knowledge transfer-based microwave filter optimization and debugging method. First, the differences in the mapping relationship between the structural geometric parameters and filtering performance of different microwave filters are analyzed. Then, a knowledge generation model is transferred based on these differences. The optimized range of the particle swarm optimization algorithm is determined using the transferred knowledge generation model, achieving efficient debugging of microwave filters with strong individual differences. This method mainly includes:

[0006] The debugging task for the initial microwave filter includes: establishing an initial knowledge generation model, determining the initial optimization range, and optimizing and debugging the initial microwave filter. Based on the established initial knowledge generation model and the determined initial optimization range, the initial microwave filter is optimized and debugged.

[0007] Commissioning tasks for other microwave filters include: difference assessment, determination of migration strength, establishment of knowledge generation model, determination of optimization range, and optimization commissioning;

[0008] In the stage of determining migration intensity, based on the degree of difference d t The migration strength was calculated;

[0009] Based on the degree of difference and the intensity of transfer obtained in the difference assessment stage, the initial knowledge generation model is fine-tuned to obtain the knowledge generation model.

[0010] Based on the knowledge generation model and the determined optimization range, other initial microwave filters are optimized and debugged.

[0011] Furthermore, the input to the initial knowledge generation model is the filtering performance, and the output is the structural geometric parameters, which include the resonant screw length and the coupling screw length.

[0012] Furthermore, in the initial optimization range determination stage, a groups of filtering performance parameters that meet the performance index requirements are sequentially input into the initial knowledge generation model M0 to obtain the corresponding a groups of predicted structural geometric parameters. Let represent the value of the m-th structural geometric parameter in the predicted solution of the a-th group of structural geometric parameters. Based on the predicted solutions of multiple groups of structural geometric parameters, the initial optimization range of the initial microwave filter is determined.

[0013]

[0014] in, Representation matrix The data in the q-th column, q = 1, 2, ..., m, where m is the number of structural geometric parameters.

[0015] Furthermore, in the difference evaluation phase, within the initial optimization range θ0, N on the t-th microwave filter and the initial microwave filter respectively. c Comparison points were collected at the same location;

[0016] Based on the t*N of the tasks in group t c Calculate the mean of the filtering performance for each of the n points. and standard deviation

[0017] Based on the calculated mean and standard deviation, the performance indices of the initial microwave filter and the t-th microwave filter are normalized as follows: and

[0018] The formula for normalization is:

[0019]

[0020] Then, the difference d between the t-th microwave filter and the initial microwave filter is calculated. t The calculation formula is:

[0021]

[0022] Where, N c This indicates the number of sampling points on the filter. This represents the normalized value of the first index of the initial microwave filter. This represents the normalized value of the first index of the t-th microwave filter. I represents the normalized value of the nth index of the t-th microwave filter. n This represents the nth indicator. This represents the t*N data collected. c The mean of the nth indicator among points, This represents the t*N data collected. c The standard deviation of the nth indicator among points, where n is a positive integer greater than or equal to 1.

[0023] Furthermore, the migration intensity of the t-th microwave filter The calculation formula is

[0024]

[0025] Where r() represents rounding up to the nearest integer, d t Let denot represent the difference between the t-th microwave filter and the initial microwave filter, η represent a confidence factor, and N0 represent the number of samples collected on the initial microwave filter.

[0026] Furthermore, in the knowledge generation model stage, N samples are first collected on the t-th microwave filter. t N sample data, using N t Fine-tuning the initial knowledge generation model M0 with sample data yields the knowledge generation model M0 for the t-th microwave filter. t ; Utilizing the knowledge generation model M t The output results determine the optimization range θ of the t-th microwave filter. t .

[0027] Furthermore, the knowledge generation model M is obtained. t Then, within the determined optimization range θ t Within the process, the particle swarm optimization algorithm is used to optimize the microwave filter, obtaining a feasible solution for the structural geometric parameters that meet the filtering performance requirements, and the debugging task of the t-th microwave filter is carried out based on the feasible solution.

[0028] A storage device that stores instructions and data for implementing a microwave filter optimization and debugging method based on knowledge transfer.

[0029] A microwave filter optimization and debugging device for knowledge transfer includes: a processor and a storage device; the processor loads and executes instructions and data in the storage device to implement a microwave filter optimization and debugging method for knowledge transfer.

[0030] Compared with the prior art, the beneficial effects of the technical solution provided by the present invention are:

[0031] (1) For the debugging task of microwave filters with obvious individual differences, a knowledge transfer optimization debugging framework is proposed, which includes difference assessment, transfer strength determination, knowledge generation model establishment, optimization range determination and optimization debugging. By transferring knowledge between different tasks, the overall debugging efficiency is improved.

[0032] (2) To address the unknown individual differences, a difference assessment method is proposed. By comparing the filtering performance of different microwave filters under the same structural geometric parameters, the difference between microwave filters is obtained.

[0033] (3) To address the problem of difficulty in determining migration intensity, a reasonable method for determining migration intensity was designed based on the differences between microwave filters, which effectively guides knowledge transfer. Attached Figure Description

[0034] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0035] Figure 1 This is a framework diagram of a microwave filter optimization and debugging method based on knowledge transfer, as described in an embodiment of the present invention.

[0036] Figure 2 This is a schematic diagram of the microwave filter simulation model established in HFSS in this embodiment of the invention.

[0037] Figure 3 This is a schematic diagram of the debugging process using M2 in an embodiment of the present invention.

[0038] Figure 4 M3 is used in this embodiment of the invention. ’ A diagram illustrating the debugging process.

[0039] Figure 5 This is a schematic diagram of the debugging process using M3 in an embodiment of the present invention.

[0040] Figure 6 This is a schematic diagram of the hardware device working in an embodiment of the present invention. Detailed Implementation

[0041] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0042] The embodiments of the present invention provide a microwave filter optimization and debugging method, device and storage device for knowledge transfer.

[0043] Please refer to Figure 1 , Figure 1 This is a framework diagram of a microwave filter optimization and debugging method based on knowledge transfer, as described in an embodiment of the present invention. The method includes:

[0044] The debugging task for the initial microwave filter is mainly divided into three stages: establishing an initial knowledge generation model, determining the initial optimization range, and optimizing and debugging the initial microwave filter.

[0045] In the initial knowledge generation model M0 stage, the structural geometric parameters x = [x1, x2, ..., x0] of the microwave filter are first randomly changed N0 times. m ], where m is the number of structural geometric parameters, and the corresponding filtering performance I = [I1, I2, ..., I n (n is the number of filtering performance parameters), forming a dataset containing N0 samples, used to train the knowledge generation model. The model's input is the filtering performance, and the output is the structural geometric parameters. The structural geometric parameters of the microwave filter include the resonant screw length and the coupling screw length. The model's structural design and parameter training are consistent with patent ZL202110128530.4.

[0046] In the initial optimization range determination stage, a sets of filtering performance that meet the performance index requirements are sequentially input into the initial knowledge generation model M0 to obtain the corresponding a sets of predicted structural geometric parameters. Let represent the value of the m-th structural geometric parameter in the predicted solution of the a-th group of structural geometric parameters. Based on the predicted solutions of multiple groups of structural geometric parameters, determine the initial optimization range of the initial microwave filter. The method of determination is

[0047]

[0048] in, Representation matrix The data in the qth column.

[0049] During the initial microwave filter optimization and debugging stage, within the initial optimization range θ0, the particle swarm optimization method in patent ZL201811627292.6 is used to optimize and obtain a feasible solution of structural geometric parameters that makes the filtering performance of the initial microwave filter meet the requirements, and the initial microwave filter is debugged based on the feasible solution.

[0050] After completing the initial microwave filter debugging, the debugging tasks for other microwave filters are divided into five parts: difference assessment, determination of migration strength, establishment of a knowledge generation model, determination of optimization range, and optimization debugging. For example... Figure 1 As shown, the debugging task for the t-th microwave filter (t is a positive integer greater than or equal to 1) (microwave filters have the same structure but different material properties) is mainly divided into five stages: difference assessment, determination of migration intensity, establishment of knowledge generation model, determination of optimization range, and optimization debugging.

[0051] First, a difference assessment is performed. Within the initial optimization range θ0, N is calculated for the t-th microwave filter and the initial microwave filter, respectively. c Comparison points were collected at the same locations. Based on t*N of the t-group tasks... c Calculate the mean of the filtering performance for each of the n points. and standard deviation For example, when the third microwave filter is tuned, there are 3 sets of tasks, 3N. c Based on the calculated mean and standard deviation, the performance metrics of the initial microwave filter and the t-th microwave filter are normalized to: and The formula for normalization is:

[0052]

[0053] Then, the difference d between the t-th microwave filter and the initial microwave filter is calculated. t The calculation formula is:

[0054]

[0055] Where, N c This indicates the number of sampling points on the filter. This represents the normalized value of the first index of the initial microwave filter. This represents the normalized value of the first index of the t-th microwave filter. I represents the normalized value of the nth index of the t-th microwave filter. n This represents the nth indicator. This represents the t*N data collected. c The mean of the nth index among n points (each point includes the structural geometric parameter x and the corresponding filtering performance I), This represents the t*N data collected. c The standard deviation of the nth indicator among points, where n is a positive integer greater than or equal to 1.

[0056] In the stage of determining migration intensity, the migration intensity is based on the degree of difference d. t The calculation is obtained. The calculation formula is:

[0057]

[0058] in, Let r represent the migration strength of the t-th microwave filter, r() represents rounding up to the nearest integer; η is a confidence factor, set through human experience.

[0059] Then, in the stage of establishing the knowledge generation model for the t-th microwave filter, N samples are first collected on the t-th microwave filter. t Based on the sample data, and according to the method in patent ZL202110130442.8, using N... t Fine-tuning the initial knowledge generation model M0 with sample data yields the knowledge generation model M0 for the t-th microwave filter. t .

[0060] In the optimization range determination phase, the same method as in the initial microwave filter debugging task is adopted, utilizing the knowledge generation model M. t The optimization range θ of the t-th microwave filter is determined based on its output. t ,Right now:

[0061] To the knowledge generation model M t By sequentially inputting the filtering performance of group a that meets the performance index requirements, the corresponding predicted solutions for the structural geometric parameters of group a are obtained. Let represent the value of the m-th structural geometric parameter in the predicted solution of the a-th group of structural geometric parameters. Based on the predicted solutions of multiple groups of structural geometric parameters, determine the optimization range of the t-th microwave filter. The method of determination is

[0062]

[0063] in, Representation matrix The data in the qth column.

[0064] During the optimization and debugging phase, the knowledge generation model M was obtained. t Then, within the optimization range θ t Within this framework, the method employs two phases—"determining the initial optimization range" and "optimizing and debugging the initial microwave filter"—in the initial microwave filter debugging task to efficiently complete the debugging task of the t-th microwave filter.

[0065] In this embodiment, a simulation and debugging platform was built based on the three-dimensional electromagnetic software HFSS and MATLAB. Three microwave filter simulation models with material dielectric constants of 35.5, 35.7, and 36.0 were established in HFSS, as shown below. Figure 2As shown. Each microwave filter simulation model has a total of 6 structural geometric parameters (e.g., ...). Figure 2 The light gray areas represent the cylindrical holes, while the dark gray areas represent the metal layer added beneath the two bottommost cylindrical holes. Figure 2 It is known that each microwave filter simulation model has two rows of cylindrical holes, with three holes in each row. However, since the dielectric filter simulation model has a symmetrical structure, it is only necessary to calculate the structural geometric parameters x = [x1, x2, x3] of one row of three holes, and the depth of the other row of three holes can be directly assigned according to the obtained values. The filtering performance includes the center frequency ω. c The bandwidth W and return loss target ψ, i.e., I1 is the center frequency, I2 is the bandwidth, and I3 is the return loss in the filtering performance, are used in this implementation. c =2.610GHz, W=0.193GHz and ψ=-20dB, the permissible errors for center frequency and bandwidth are respectively δ w =0.005GHz, trust factor η=0.01.

[0066] Using a dielectric filter with a dielectric constant of 35.5 as the initial microwave filter, its debugging task is τ. 1 The debugging task for a dielectric filter with a dielectric constant of 35.7 is τ. 2 The debugging task for a dielectric filter with a dielectric constant of 36.0 is τ. 3 Four data points were collected to calculate the degree of difference. The results are shown in the table below:

[0067] Table 13 shows the calculation results of the difference in microwave filters.

[0068]

[0069] In Table 1, d represents the degree of difference. t .

[0070] To verify the effectiveness of this method, two comparative experiments were conducted.

[0071] (1) Verify the effectiveness of the knowledge transfer optimization and debugging method.

[0072] In task τ 1 In the process, the knowledge generation model M1 was established using 200 samples. Finally, after three iterations, the particle swarm optimization algorithm found a feasible solution for the structural geometric parameters and completed the debugging task.

[0073] In task τ 2In this study, based on the method for determining the transfer intensity, a knowledge generation model M2 was established using six samples. Finally, after five iterations, the particle swarm optimization algorithm found a feasible solution for the structural geometric parameters, completing the debugging task and demonstrating that transferred knowledge can improve debugging efficiency. The debugging process is as follows: Figure 3 As shown.

[0074] (2) Verify the effectiveness of the difference assessment and migration strength determination methods.

[0075] In task τ 3 In the middle, the samples (28) obtained by the migration intensity determination method and the task τ were respectively used. 2 Knowledge generation models M3 and M3 are built using the same number of samples (6). ’ Using M3 ’ During debugging, completing the task requires 8 iterations. The debugging process is as follows: Figure 4 As shown; this result is greater than 5 iterations when debugging with M2, indicating that the accuracy of the transferred knowledge is insufficient. Debugging with M3 requires 2 iterations, and the debugging process is as follows: Figure 5 As shown; this result is also smaller than that using M3. ’ The effectiveness of the difference assessment and migration strength determination method was demonstrated by 8 trials and 5 trials using M2 debugging.

[0076] Please see Figure 6 , Figure 6 This is a schematic diagram of the hardware device in operation according to an embodiment of the present invention. The hardware device specifically includes: a microwave filter optimization and debugging device 401 for knowledge transfer, a processor 402, and a storage device 403.

[0077] A microwave filter optimization and debugging device 401 based on knowledge transfer: The microwave filter optimization and debugging device 401 based on knowledge transfer implements the microwave filter optimization and debugging method based on knowledge transfer.

[0078] Processor 402: The processor 402 loads and executes the instructions and data in the storage device 403 to implement the microwave filter optimization and debugging method for knowledge transfer.

[0079] Storage device 403: The storage device 403 stores instructions and data; the storage device 403 is used to implement the microwave filter optimization and debugging method for knowledge transfer.

[0080] The beneficial effects of this invention are:

[0081] (1) For the debugging task of microwave filters with obvious individual differences, a knowledge transfer optimization debugging framework is proposed, which includes difference assessment, transfer strength determination, knowledge generation model establishment, optimization range determination and optimization debugging. By transferring knowledge between different tasks, the overall debugging efficiency is improved.

[0082] (2) To address the unknown individual differences, a difference assessment method is proposed. By comparing the filtering performance of different microwave filters under the same structural geometric parameters, the difference between microwave filters is obtained.

[0083] (3) To address the problem of difficulty in determining migration intensity, a reasonable method for determining migration intensity was designed based on the differences between microwave filters, which effectively guides knowledge transfer.

[0084] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A microwave filter optimization and debugging method based on knowledge transfer, characterized in that: include: The debugging task for the initial microwave filter includes: establishing an initial knowledge generation model, determining the initial optimization range, and optimizing and debugging the initial microwave filter. Based on the established initial knowledge generation model and the determined initial optimization range, the initial microwave filter is optimized and debugged. Commissioning tasks for other microwave filters include: difference assessment, determination of migration strength, establishment of knowledge generation model, determination of optimization range, and optimization commissioning; In the stage of determining migration intensity, based on the degree of difference d t The migration strength was calculated; Based on the difference degree and transfer strength obtained in the difference degree assessment stage, the initial knowledge generation model is fine-tuned to obtain the knowledge generation model. Based on the knowledge generation model and the determined optimization range, other microwave filters are optimized and debugged. The input to the initial knowledge generation model is the filtering performance, and the output is the structural geometric parameters, which include the resonant screw length and the coupling screw length. In the initial optimization range determination stage, a sets of filtering performance that meet the performance index requirements are sequentially input into the initial knowledge generation model M0 to obtain the corresponding a sets of predicted structural geometric parameters. , Let represent the value of the m-th structural geometric parameter in the predicted solution of the a-th group of structural geometric parameters. Based on the predicted solutions of multiple groups of structural geometric parameters, the initial optimization range of the initial microwave filter is determined. , (1) in, Representation matrix The data in the q-th column, q=1,2,…,m, where m is the number of structural geometric parameters; During the difference assessment phase, within the initial optimization range θ0, N is calculated for the t-th microwave filter and the initial microwave filter, respectively. c Comparison points were collected at the same location, where t is a positive integer greater than 1; According to the tasks in group t Calculate the mean of the filtering performance for each of the n points. and standard deviation ; Based on the calculated mean and standard deviation, the performance indices of the initial microwave filter and the t-th microwave filter are normalized as follows: and ; The formula for normalization is: (2) Then, the difference d between the t-th microwave filter and the initial microwave filter is calculated. t The calculation formula is: (3) Where, N c This indicates the number of sampling points on the filter. This represents the normalized value of the first index of the initial microwave filter. This represents the normalized value of the first index of the t-th microwave filter. I represents the normalized value of the nth index of the t-th microwave filter. n This represents the nth indicator. Indicates collection The mean of the nth indicator among points, Indicates collection The standard deviation of the nth indicator among the points, where n is a positive integer greater than or equal to 1; The migration intensity of the t-th microwave filter The calculation formula is: (4) Where r() represents rounding up to the nearest integer, d t η represents the difference between the t-th microwave filter and the initial microwave filter, η represents a confidence factor, and N0 represents the number of samples collected on the initial microwave filter. In the knowledge generation model stage, N samples are first collected on the t-th microwave filter. t N sample data, using N t Fine-tuning the initial knowledge generation model M0 with sample data yields the knowledge generation model M0 for the t-th microwave filter. t ; Utilizing the knowledge generation model M t The output results determine the optimization range θ of the t-th microwave filter. t ; Obtain the knowledge generation model M t Then, within the determined optimization range θ t Within the process, the particle swarm optimization algorithm is used to optimize the microwave filter, obtaining a feasible solution for the structural geometric parameters that meet the filtering performance requirements, and the debugging task of the t-th microwave filter is carried out based on the feasible solution.

2. The microwave filter optimization and debugging method based on knowledge transfer as described in claim 1, characterized in that: The filtering performance includes center frequency, bandwidth, and return loss targets.

3. A storage device, characterized in that: The storage device stores instructions and data for implementing the microwave filter optimization and debugging method for knowledge transfer as described in any one of claims 1 to 2.

4. A microwave filter optimization and debugging device based on knowledge transfer, characterized in that: include: A processor and a storage device; the processor loads and executes instructions and data in the storage device to implement the microwave filter optimization and debugging method for knowledge transfer as described in any one of claims 1 to 2.

Images