CN115902487A - Microwave filter debugging method for knowledge-driven particle swarm optimization - Google Patents
Microwave filter debugging method for knowledge-driven particle swarm optimization Download PDFInfo
- Publication number
- CN115902487A CN115902487A CN202211577956.9A CN202211577956A CN115902487A CN 115902487 A CN115902487 A CN 115902487A CN 202211577956 A CN202211577956 A CN 202211577956A CN 115902487 A CN115902487 A CN 115902487A
- Authority
- CN
- China
- Prior art keywords
- knowledge
- debugging
- model
- adjustable
- microwave filter
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 59
- 238000005457 optimization Methods 0.000 title claims abstract description 56
- 239000002245 particle Substances 0.000 title claims abstract description 36
- 230000008569 process Effects 0.000 claims abstract description 32
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 21
- 238000011156 evaluation Methods 0.000 claims abstract description 5
- 239000011159 matrix material Substances 0.000 claims description 30
- 230000004044 response Effects 0.000 claims description 19
- 238000012549 training Methods 0.000 claims description 11
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 238000011478 gradient descent method Methods 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 5
- 230000001133 acceleration Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 4
- 238000013527 convolutional neural network Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 230000001747 exhibiting effect Effects 0.000 claims description 2
- 230000006870 function Effects 0.000 abstract description 16
- 238000010276 construction Methods 0.000 abstract description 4
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 238000000605 extraction Methods 0.000 abstract 1
- 238000005065 mining Methods 0.000 abstract 1
- 238000002474 experimental method Methods 0.000 description 4
- 238000004088 simulation Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000002787 reinforcement Effects 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000011248 coating agent Substances 0.000 description 1
- 238000000576 coating method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000003754 machining Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000002310 reflectometry Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Control Of Motors That Do Not Use Commutators (AREA)
Abstract
The invention provides a microwave filter debugging method based on knowledge-driven particle swarm optimization, which mainly comprises the steps of data acquisition, data-driven knowledge model establishment, knowledge extraction and debugging, knowledge-driven particle swarm optimization, debugging process model construction and debugging according to an optimization result. The invention has the beneficial effects that: a series of evaluation functions are set for various performance indexes of the microwave filter, the performance of the filter is effectively and comprehensively evaluated, a knowledge model is built through a large amount of data, debugging knowledge is obtained through mining, the debugging knowledge is used for driving a particle swarm optimization algorithm, optimization can be effectively prevented from falling into local optimization through setting the optimization range of each adjustable parameter, and the debugging efficiency is remarkably improved.
Description
Technical Field
The invention relates to the technical field of microwave filter debugging, in particular to a microwave filter debugging method based on knowledge-driven particle swarm optimization.
Background
The construction of the 5G base station is the first of new capital construction in China, the microwave filter is a core frequency selection device in the 5G base station, and the filtering performance of the microwave filter has great influence on the frequency selection quality. In the production process, due to the existence of design errors and machining tolerances, the performance of the microwave filter usually needs to be met through debugging. The traditional manual debugging method has low efficiency and high cost. With the rapid development of machine learning and artificial intelligence, intelligent optimization and control have been applied to filter tuning processes, including fuzzy logic-based debugging, reinforcement learning-based debugging, and group intelligence optimization-based debugging. And (3) designing a debugging rule according to experience based on the debugging of the fuzzy logic, and changing the adjustable variables of the microwave filter one by one. The method ignores the coupling relation among variables and has low debugging efficiency. Continuous adjustable variables are discretized based on debugging of reinforcement learning, and debugging precision is low. The debugging technology based on the swarm intelligence optimization can be realized through the particle swarm optimization. The technology changes all adjustable variables simultaneously in a continuous space, and fully considers the coupling among all variables. However, the relation between the adjustable variable and the performance index is complex, the iteration times of the algorithm are multiple, and the improvement of the debugging efficiency is limited.
In the manual debugging process, an experienced debugging worker predicts the optimal variable variation range according to experience and carries out finer adjustment in the range. This mechanism, called the attention mechanism, can speed up debugging efficiency. The invention provides a microwave filter debugging method for knowledge-driven particle swarm optimization, which is used for learning debugging knowledge from debugging data and utilizing a knowledge-driven particle swarm optimization algorithm to debug, thereby greatly improving the debugging efficiency.
Disclosure of Invention
In order to solve the above problems, the present invention provides a microwave filter debugging method for knowledge-driven particle swarm optimization, which mainly comprises:
s1: multiple variation of the variable u of the adjustable component on the microwave filter * Obtaining S matrix S * Sampling and measuring S parameters S by using a vector network analyzer to construct a debugging data set;
s2: establishing a knowledge model based on a convolutional neural network according to the debugging data set, wherein the input of the knowledge model is a performance index I * Amplitude frequency response R a Sum phase frequency responseThe output is adjustable variable u * The input of the knowledge model is obtained by calculating an S parameter S, the knowledge model comprises a convolution layer, an activation layer, a pooling layer, a full-link layer and an output layer, and the knowledge model is trained by using a random gradient descent method to obtain a trained knowledge model;
s3: inputting a performance index I meeting requirements into a trained knowledge model * Amplitude-frequency response R a Sum phase frequency responseObtaining corresponding output to form debugging knowledge;
s4: using the debugging knowledge to determine the optimization range of the particle swarm algorithm, designing an evaluation function according to the optimization range, the performance index and the requirements thereof, and performing iterative optimization on the group scale, the inertia weight, the acceleration weight, the stop standard and the debugging and aligning parameters;
s5: will S matrix S * Conversion to Y parameter Y p A debugging process model is built by using the Elman neural network, and an optimization result of the adjustable variable is obtained by combining the iterative optimization result in the step S4 and is used for reflecting the performance index of the microwave filter;
s6: and after the debugging standard is met, debugging the actual microwave filter according to the optimization result of the step S5, and comprehensively evaluating the overall performance of the microwave filter.
Further, the loss function used to train the knowledge model is:
wherein N is the number of samples in the training set, and m is the adjustable variable u of the microwave filter * T denotes the t-th of the m variables, n denotes the n-th sample in the debug data, m, n and t are all positive integers greater than or equal to 1,the error of the t adjustable variable and the model output for the nth sample, and:
wherein,is the model output result for the t variable for the nth sample, based on the result of the model output, based on the value of the t variable in the nth sample>Is the sample value of the t variable of the nth sample;
when the value of the loss function is less than the allowable loss error, i.e. the maximum loss value delta loss Then, the knowledge model training is completed, namely:
Ll 2 loss<δl oss 。
furthermore, the debugging knowledge represents the adjustable variables corresponding to the performance and meeting the index requirements, which are obtained by the prediction of the knowledge model,is a knowledge matrix, is marked as &>Variable range of each adjustable variableEnclose and mark as->Wherein, b is expressed as knowledge quantity, namely the quantity of performance indexes meeting requirements input into the trained knowledge model, t is expressed as the tth adjustable variable, t =1, \ 8230, and m, m are the number of the adjustable variables.
Further, the range of variation of each of the adjustable variables is determined by the range of the corresponding adjustable variable in the output, i.e.b is expressed as knowledge quantity, namely the quantity of performance indexes meeting requirements input into the trained knowledge model, t is expressed as the tth adjustable variable, t =1, \ 8230, and m are the quantity of the adjustable variables.
Further, the Y matrix is represented as:
ω is angular frequency, i is imaginary unit, λ k Is the kth pole of the Y matrix, denoted as λ k =[λ 1 ,...,λ m ] T ,r 11 k Is Y 11 And Y 21 The k-th residue of (1) is respectively marked as r 11 =[r 11 1 ,...,r 11 m ] T And r 21 =[r 21 1 ,...,r 21 m ] T Y matrix is composed of k ,r 11 And r 12 Jointly determining, and obtaining Y parameter Y by vector fitting p =[y 1 ,y 2 ,y 3 ] T And is recorded as:
y 1 =imag(λ k ),y 2 =real(r 11 ),y 3 =real(r 21 ),
real and imag respectively represent a real part and an imaginary part, and the Y parameter is used as the input of the debugging process model.
Further, the debugging process model performs the debugging process as follows:
three mappings of tunable variables to imag (lambda) are established k ),real(r 11 ) And real (r) 21 ) The parameters of the three models are determined by using a gradient descent method, and the loss function is as follows:
wherein q =1,2,3, m is the number of adjustable variables, ns represents the number of times of sample input of the adjustable variables, n represents the sample input at the nth time, and k represents the k adjustable variable; y is p (k)=[y 1 (k),y 2 (k),y 3 (k)] T ,y p () And y p* () The expected output value and the predicted value of the debugging process model to the input sample are respectively, only one group of Y parameters can be obtained by each input, and the debugging process models M1, M2 and M3 are defined as follows:
M1:y 1 (n)=g 1 (u * (n)),M2:y 2 (n)=g 2 (u * (n)),M3:y 3 (n)=g 3 (u * (n));y 1 (n)、y 2 (n)、
y 3 (n) respectively represents expected output values y obtained by the nth sample input 1 、y 2 、y 3 ,u * (n) adjustable variable samples representing the nth input, i.e.When the following criteria are met or the number of times of debugging reaches an upper limit, the debugging process models M1, M2 and M3 stop training:
L q <δ Lq ,
δ Lq representing the allowable error of the loss function.
Further, the overall performance of the filter is comprehensively evaluated as follows:
(1)f 1 : describing the actual center bandwidth w c And target center bandwidth w c * The difference between them was set as:
(2)f 2 : describing the actual center bandwidth W c And target centric bandwidth W c * The difference between them was set as:
(3)f 3 : describing actual return loss ζ and target return loss ζ * The difference between them is set as:
wherein alpha is 1 Is a scalable coefficient one;
(4)f 4 : exhibiting an amplitude-frequency response A s21 (ω) maximum in the pass band, set to:
where ω is the angular frequency, α 2 Is a scalable coefficient two;
(5)f 5 : the resonance state of the filter is reflected in relation to the number of peaks, and is set as follows:
nc is the number of wave troughs in amplitude-frequency response, when all resonant holes, namely the adjustable variables, reach good, the number of the wave troughs is m, namely the number of the wave troughs is equal to the number of the adjustable variables, so that f is more than or equal to 0 5 ≤1,f 5 The smaller the value is, the better the resonance state of the resonance hole is;
the objective function is set as:
wherein, w Nf Is f Nf Weight of (a), f Nf =1,2,3,4,5。
The technical scheme provided by the invention has the beneficial effects that: the optimization can be effectively prevented from being trapped in local optimization by setting the optimization range of each adjustable parameter, and the debugging efficiency is obviously improved.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
fig. 1 is a flowchart of a microwave filter debugging method for knowledge-driven particle swarm optimization according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of the structure of the elman neural network in the embodiment of the present invention.
Fig. 3 is a schematic diagram of building a debugging process model in the embodiment of the present invention.
FIG. 4 is a diagram of a debug platform emulation in an embodiment of the present invention.
FIG. 5 is a knowledge model output in an embodiment of the invention.
FIG. 6 is a debugging result of the disclosed method in an embodiment of the present invention.
FIG. 7 shows a debugging result of the knowledge-free particle swarm algorithm in the embodiment of the present invention.
Fig. 8 is an electromagnetic simulation result in the embodiment of the present invention.
Detailed Description
For a more clear understanding of the technical features, objects, and effects of the present invention, embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
Referring to fig. 1, fig. 1 is a flowchart of a microwave filter debugging method for knowledge-driven particle swarm optimization in an embodiment of the present invention, which mainly includes acquiring data, data-driven establishing a knowledge model, extracting debugging knowledge, knowledge-driven particle swarm optimization, building a debugging process model, and debugging according to an optimization result, and specifically includes the following steps:
firstly, the variable u of the adjustable component is changed several times on the microwave filter * Data acquisition is carried out to obtain an S matrix S * And acquiring an S parameter S by using a vector network analyzer, wherein a multi-feature fusion modeling method in patent CN113158541B is cited here to process data and construct a knowledge model. Wherein s is 11 =a 11 +b 11 i denotes the reflectivity of the input signal energy, s 21 =a 21 +b 21 i denotes the transmission rate of the input signal energy, a 11 And a 12 Are respectively s 11 And s 21 Real part of (b) 11 And b 21 Are respectively s 11 And s 21 The imaginary part of (a). Calculating according to the formula (1) and the formula (2) to obtain the amplitude and the phase, and further combining the sampling frequency f to obtain the amplitude-frequency responseAnd phase frequency response->
Then, setting a target performance index according to the communication requirement; the performance index includes a center frequency f c The bandwidth W and the return loss zeta are obtained through S parameter calculation; a. The s21 The sampling frequency corresponding to the time of = -3dB is the upper and lower cut-off frequency f 1 、f 2 Then the center frequency f c Comprises the following steps:
f c =(f 1 +f 2 )/2, (3)
the bandwidth is as follows:
W=f 1 -f 2 . (4)
the performance index meeting the requirement means that the actual performance index meets the following relationship:
|W-W * |≤δ W , (6)
ζ≤ζ * . (7)
wherein,W * and ζ * Is respectively a target center frequency, a target bandwidth and a target return loss->And delta w The allowable error of the center frequency and the bandwidth, respectively.
And then, establishing a knowledge model based on the convolutional neural network according to the debugging data set, wherein the knowledge model comprises an input layer, a convolutional layer, a pooling layer, an activation function layer, a full connection layer and an output layer.
Establishing a knowledge model, wherein a loss function used for training the knowledge model is as follows:
wherein N is the number of samples in the training set, and m is the adjustable variable u of the microwave filter * T denotes the t-th of the m variables, n denotes the n-th sample in the debug data, m, n and t are all positive integers greater than or equal to 1,error of the t adjustable variable and the model output for the nth sample:
wherein,is the model output result for the t variable for the nth sample, based on the result of the model output, based on the value of the t variable in the nth sample>Is the sample value of the t variable of the nth sample.
When the value of the loss function is less than the allowable loss error, i.e. the maximum loss value delta loss Then, the knowledge model training is completed, namely:
thereafter, debug knowledge is acquired. Inputting a large amount of performance indexes I meeting the requirements into a trained model * Amplitude frequency response R a Sum phase frequency responseAnd obtaining corresponding output to form debugging knowledge. The debugging knowledge represents adjustable variables which are obtained by prediction of the knowledge model and correspond to the performance meeting the index requirement, and the adjustable variables are judged to be greater than or equal to the standard value>Is a knowledge matrix, is marked as &>The range of variation of each adjustable variable is recorded as->Wherein, b is expressed as knowledge quantity, namely the number of performance indexes meeting requirements input into the trained knowledge model, t is expressed as the tth adjustable variable, t =1, \8230, and m are the number of the adjustable variables.
Optimization using debug knowledge for determining particle swarm algorithmsRange, optimum range being the range over which all adjustable variables in the debugging knowledge varyDecision in which the range of change of each adjustable variable is determined by &>To determine, i.e. that
b is expressed as knowledge quantity, namely the quantity of performance indexes meeting requirements input into the trained knowledge model, t is expressed as the tth adjustable variable, t =1, \ 8230, and m are the quantity of the adjustable variables.
In order to improve the debugging efficiency, a neural network is used for establishing a debugging process model, and the debugging process model is established by a method for establishing an electromechanical characteristic model of a microwave cavity filter in the patent CN 109783905B. Based on S matrix S * And an adjustable variable u * Build a debugging Process model due to the S matrix S * The method is a high-dimensional and complex matrix, can only reflect partial electromechanical characteristics of the microwave filter, and is difficult to directly establish a model by using the S matrix. Thus, the S matrix S * Equivalent admittance matrix (Y matrix) Y = [ Y matrix) converted into microwave filter 11 ,Y 21 ]To describe the voltage and current relationship of the equivalent node of the filter, the conversion relationship is as follows:
wherein E is and s 11 Dimension numberThe same and all elements are a matrix of 1. The Y matrix can also be expressed as:
ω is angular frequency, i is imaginary unit, λ k Is the kth pole of the Y matrix, denoted as λ k =[λ 1 ,...,λ m ] T 。r 11 k Is Y 11 And Y 21 The k-th residue of (1) is respectively marked as r 11 =[r 11 1 ,…,r 11 m ] T And r 21 =[r 21 1 ,...,r 21 m ] T Y matrix is formed by k ,r 11 And r 12 And (4) jointly determining. Then obtaining Y parameter Y by vector fitting p =[y 1 ,y 2 ,y 3 ] T And is recorded as:
y 1 =imag(λ k ),y 2 =real(r 11 ),y 3 =real(r 21 ), (16)
where real and imag represent the real and imaginary parts, respectively. The Y-parameter contains the mechanical properties of the microwave filter and is lower in dimension than the Y-matrix and the S-matrix, and therefore the Y-parameter is more suitable as an input for a model of the debugging process.
Due to the Y parameter Y p And adjustable variable u * And a nonlinear relationship between the two, a neural network is used for establishing a debugging process model. Moreover, debugging is a continuous decision process, and different Y parameters and adjustable variables are closely related to continuity, so that as shown in fig. 2, an Elman neural network with memory is used, local feedback is connected with an artificial neural network, and the input of the current hidden layer consists of the output of the last hidden layer and the input of the current network. In addition, the activation function accompanied with the hidden layer can promote the networkNon-linear mapping capability.
The mapping of the adjustable variables and the S-parameters and the construction of the debugging process model are shown in fig. 3. Due to imag (λ), real (r) 11 ) And real (r) 21 ) Are different from each other, in order to obtain a more accurate mapping relation, three adjustable variables are established to be respectively mapped to imag (lambda) k ),real(r 11 ) And real (r) 21 ) The parameters of the three models are determined by using a gradient descent method, and the loss function is as follows:
wherein q =1,2,3, m is the number of adjustable variables, ns represents the number of times of sample input of the adjustable variables, n represents the sample input at the nth time, and k represents the k adjustable variable; y is p (k)=[y 1 (k),y 2 (k),y 3 (k)] T ,y p () And y p* () The expected output value and the predicted value of the debugging process model to the input sample are respectively, only one group of Y parameters can be obtained by each input, and the debugging process models M1, M2 and M3 are defined as follows:
M1:y 1 (n)=g 1 (u * (n)),M2:y 2 (n)=g 2 (u * (n)),M3:y 3 (n)=g 3 (u * (n));y 1 (n)、y 2 (n)、
y 3 (n) respectively represents the expected output values y obtained by the k-th sample input 1 、y 2 、y 3 ,u * (n) adjustable variable samples representing the kth input, i.e.When the following criteria are met or the number of times of debugging reaches an upper limit, the debugging process models M1, M2 and M3 stop training:
δ Lq representing the allowable error of the loss function. Next, the Y matrix is calculated from the output of the debugging process model by equations (14) and (15), and the S parameter S is converted from the Y matrix by equations (12) and (13). Finally, the debugging process model can quickly and accurately evaluate the performance index of the microwave filter.
Besides, a series of evaluation functions are set, and the overall performance of the filter is comprehensively evaluated.
(1)f 1 : describing the actual center bandwidth w c And target center bandwidth w c * The difference between them is set as:
(2)f 2 : describing the actual center bandwidth W c And target centric bandwidth W c * The difference between them is set as:
(3)f 3 : describing actual return loss ζ and target return loss ζ * The difference between them is set as:
wherein alpha is 1 The parameter value can be set manually for the first scalable coefficient;
(4)f 4 : presentation A s21 (ω) maximum in the pass band, set to:
where ω is the angular frequency, α 2 The parameter value can be set manually for the second scalable coefficient;
(5)f 5 : the resonance state of the filter is reflected in relation to the number of peaks, and is set as follows:
nc is the number of wave troughs in amplitude-frequency response, when all resonant holes, namely the adjustable variables, reach good, the number of the wave troughs is m, namely the number of the wave troughs is equal to the number of the adjustable variables, so that f is more than or equal to 0 5 ≤1,f 5 The smaller the value, the better the resonance state of the resonance hole is shown;
the objective function is set as:
wherein, w Nf Is f Nf The weight of (a) can be set artificially, f Nf =1,2,3,4,5。
The improved algorithm flow is as follows:
wherein the stop criterion (stop criterion) is that the iteration number reaches an upper limit, and the tuning criterion (tuning criterion) is f 3 Less than a set value delta f3 ,f 5 Is 0, i.e.:
f 3 ≤δ f3 ,f 5 =0. (18)
as shown in FIG. 4, the present invention establishes an electromagnetic simulation model of a microwave filter based on three-dimensional electromagnetic simulation software (High-Frequency Structure Simulator), and calculates S parameters. And (3) establishing a microwave filter debugging knowledge model and a debugging knowledge model by using MATLAB, and realizing a knowledge-driven particle swarm optimization algorithm.
The microwave filter in the simulation model has six resonance holes, and the structure of the microwave filter is a symmetrical structure, namelyThen the adjustable variable->The performance indicators of the filter include:
(1) Center frequency omega c * =2.610GHz, allowable error δ ωc =0.001GHz。
(2) Bandwidth W * =0.1930GHz, allowable error δ W =0.0005GHz。
(3) Return loss ζ * =-20dB。
Firstly, a debugging knowledge model is constructed based on a convolutional neural network. Knowledge model Structure As shown in Table 1, debugging data { u }is used * S, the knowledge model is simulated.
TABLE 1 knowledge model Structure
Given 21 inputs to the knowledge model, the parameters are changed as shown in FIG. 5. The optimum ranges for all variables are as follows: θ ° (1) = [0.0467mm,0.0643mm ], θ ° (2) = [ -0.0064mm, -0.0237mm ], θ ° (3) = [ -0.0302mm, -0.0449mm ].
The parameter optimization is set as follows:
(1) Population size: the population size, i.e., the number of particles, was set to 50.
(2) Inertial weight c 0 : inertial weight range of [0.4,0.9]The larger the set value is, the more global the research is. Setting to 0.9 can avoid falling into local optimum and can improve the global search capability.
(3) Acceleration weight c 1 、c 2 : the acceleration weight indicates how fast the speed changes. The greater the weight, the faster the speed change. To improve global search capability, c 1 And c 2 Set to 1.5 and 2, respectively.
(4) Stopping standard: the maximum number of iterations is 100.
(5) Tuning standard: delta. For the preparation of a coating f3 =0.6。
(6) Can be used forCoefficient of expansion: alpha (alpha) ("alpha") 1 =3;α 2 =10;α 3 =0.3;α 4 =10。
The amplitude-frequency response of the algorithm proposed in this patent for debugging is shown in fig. 6, and for comparison, the debugging result of the knowledge-free particle swarm algorithm is shown in fig. 7. The optimization range of the knowledge-driven particle swarm algorithm is [2.15,2.35], which is larger than the knowledge-driven optimization particle swarm algorithm provided by the invention. In addition, other initial settings of the two algorithms are the same, and the comparison results are shown in fig. 6 and 7:
in fig. 6, (a) is the amplitude-frequency response after the first iteration. And (b) the amplitude-frequency response after the second iteration. And (c) the amplitude-frequency response after the third iteration. In fig. 7, (a) is the amplitude-frequency response after the first iteration. And (b) the amplitude-frequency response after the second to the fifth iterations. And (c) the amplitude-frequency response after the sixth iteration. And (d) the amplitude-frequency response after the seventh iteration.
As can be seen from fig. 6, after the first iteration, the frequency response index has satisfied the requirement, which is the result of the action of the debugging knowledge. Therefore, the particle swarm optimization algorithm provided by the patent has higher efficiency than a knowledge-driven particle swarm optimization algorithm.
To make the conclusions universal and reduce contingency, we debugged the filter simulation model ten times with each of the two algorithms, respectively, and the results are shown in fig. 8. When the particle swarm optimization algorithm without knowledge driving is used for debugging, three successful experiments and seven failed experiments are carried out. In these experiments with several failures, the positions of the particle clusters all fell into a locally optimal situation after the same number of iterations. And the improved algorithm is used for debugging, all experiments are successful, and the iteration times and time consumption are far less than those of the knowledge-free particle swarm algorithm. Therefore, the algorithm of knowledge-driven particle swarm optimization greatly improves debugging precision and debugging efficiency.
The invention has the beneficial effects that: a series of evaluation functions are set for various performance indexes of the microwave filter, the performance of the filter is effectively and comprehensively evaluated, a knowledge model is built through a large amount of data, debugging knowledge is obtained through excavation, the debugging knowledge is used for driving a particle swarm optimization algorithm, optimization can be effectively prevented from falling into local optimization through setting the optimization range of each adjustable parameter, and the debugging efficiency is remarkably improved.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (7)
1. A microwave filter debugging method for knowledge-driven particle swarm optimization is characterized in that: the method comprises the following steps:
s1: multiple variation of the variable u of the adjustable component on the microwave filter * Obtaining S matrix S * Sampling and measuring an S parameter S by using a vector network analyzer, and constructing a debugging data set;
s2: establishing a knowledge model based on a convolutional neural network according to the debugging data set, wherein the input of the knowledge model is a performance index I * Amplitude frequency response R a Sum phase frequency responseThe output is adjustable variable u * The input of the knowledge model is obtained by calculating an S parameter S, the knowledge model comprises a convolution layer, an activation layer, a pooling layer, a full-link layer and an output layer, and the knowledge model is trained by using a random gradient descent method to obtain a trained knowledge model;
s3: inputting performance index I meeting requirements into a trained knowledge model * Amplitude-frequency response R a Sum phase frequency responseObtaining corresponding output to form debugging knowledge;
s4: using the debugging knowledge to determine the optimization range of the particle swarm algorithm, designing an evaluation function according to the optimization range, the performance index and the requirements thereof, and performing iterative optimization on the group scale, the inertia weight, the acceleration weight, the stop standard and the debugging and aligning parameters;
s5: will be provided withS matrix S * Conversion to Y parameter Y p A debugging process model is built by using the Elman neural network, and an optimization result of the adjustable variable is obtained by combining the iterative optimization result in the step S4 and is used for reflecting the performance index of the microwave filter;
s6: and after the debugging standard is met, debugging the actual microwave filter according to the optimization result of the step S5, and comprehensively evaluating the overall performance of the microwave filter.
2. The microwave filter debugging method of knowledge-driven particle swarm optimization according to claim 1, wherein: in step S2, the loss function used for training the knowledge model is:
wherein N is the number of samples in the training set, and m is the adjustable variable u of the microwave filter * T denotes the t-th of the m variables, n denotes the n-th sample in the debug data, m, n and t are all positive integers greater than or equal to 1,the error of the t adjustable variable and the model output for the nth sample, and:
wherein,is the model output result of the tth variable of the nth sample>Is the sample value of the t variable of the nth sample;
when the value of the loss function is smallAt an allowable loss error, i.e. maximum loss value delta loss Then, the knowledge model training is completed, namely:
Ll 2 loss<δ loss 。
3. the method for debugging a microwave filter for knowledge-driven particle swarm optimization according to claim 1, wherein the method comprises the following steps: in step S3, the debugging knowledge represents adjustable variables which are obtained by prediction of the knowledge model and correspond to the performance meeting the index requirements,is a knowledge matrix and is marked as->The variation range of each adjustable variable is recorded as>Wherein, b is expressed as knowledge quantity, namely the quantity of performance indexes meeting requirements input into the trained knowledge model, t is expressed as the tth adjustable variable, t =1, \ 8230, and m, m are the number of the adjustable variables. />
4. The method for debugging a microwave filter for knowledge-driven particle swarm optimization according to claim 1, wherein the method comprises the following steps: in step S4, the variation range of each adjustable variable is determined by the range of the corresponding adjustable variable in the output, i.e.b is expressed as knowledge quantity, namely the quantity of performance indexes meeting requirements input into the trained knowledge model, t is expressed as the tth adjustable variable, t =1, \ 8230, and m are the quantity of the adjustable variables.
5. The method for debugging a microwave filter for knowledge-driven particle swarm optimization according to claim 1, wherein the method comprises the following steps: in step S5, the Y matrix is represented as:
ω is angular frequency, i is imaginary unit, λ k Is the kth pole of the Y matrix, denoted as λ k =[λ 1 ,...,λ m ] T ,r 11 k Is Y 11 And Y 21 The kth residue of (1) is respectively marked as r 11 =[r 11 1 ,...,r 11 m ] T And r 21 =[r 21 1 ,...,r 21 m ] T Y matrix is composed of k ,r 11 And r 12 Jointly determining, and obtaining Y parameter Y by vector fitting p =[y 1 ,y 2 ,y 3 ] T And is recorded as:
y 1 =imag(λ k ),y 2 =real(r 11 ),y 3 =real(r 21 ),
real and imag respectively represent a real part and an imaginary part, and the Y parameter is used as the input of the debugging process model.
6. The method for debugging a microwave filter for knowledge-driven particle swarm optimization according to claim 1, wherein the method comprises the following steps: the debugging process model carries out the debugging process as follows:
three mappings of tunable variables to imag (lambda) are established k ),real(r 11 ) And real (r) 21 ) The parameters of the three models are determined by using a gradient descent method, and the loss function is as follows:
wherein q =1,2,3,m isThe number of the adjustable variables, ns represents the input times of the adjustable variable sample, n represents the nth input sample, and k represents the kth adjustable variable; y is p (k)=[y 1 (k),y 2 (k),y 3 (k)] T ,y p () And y p* () The expected output value and the predicted value of the debugging process model to the input sample are respectively, only one group of Y parameters can be obtained by each input, and the debugging process models M1, M2 and M3 are defined as follows:
M1:y 1 (n)=g 1 (u * (n)),M2:y 2 (n)=g 2 (u * (n)),M3:y 3 (n)=g 3 (u * (n));y 1 (n)、y 2 (n)、
y 3 (n) respectively represents expected output values y obtained by the nth sample input 1 、y 2 、y 3 ,u * (n) adjustable variable samples representing the nth input, i.e.When the following criteria are met or the number of times of debugging reaches an upper limit, the debugging process models M1, M2 and M3 stop training:
L q <δ Lq ,
δ Lq representing the allowable error of the loss function.
7. The method for debugging a microwave filter for knowledge-driven particle swarm optimization according to claim 1, wherein the method comprises the following steps: the overall performance of the filter is comprehensively evaluated as follows:
(1)f 1 : describing the actual center bandwidth w c And target center bandwidth w c * The difference between them is set as:
(2)f 2 : describing the actual center bandwidth W c And target center bandwidth W c * The difference between them is set as:
(3)f 3 : describing actual and target return loss ζ * The difference between them is set as:
wherein alpha is 1 Is a scalability factor one;
(4)f 4 : exhibiting an amplitude-frequency response A s21 (ω) a maximum value in the pass band set to:
where ω is the angular frequency, α 2 Is a scalable coefficient two;
(5)f 5 : the resonance state of the filter is reflected in relation to the number of peaks, and is set as:
nc is the number of wave troughs in amplitude-frequency response, when all resonant holes, namely the adjustable variables, reach good, the number of the wave troughs is m, namely the number of the wave troughs is equal to the number of the adjustable variables, so that f is more than or equal to 0 5 ≤1,f 5 The smaller the value, the better the resonance state of the resonance hole is shown;
the objective function is set as:
wherein, w Nf Is f Nf Weight of (f) Nf =1,2,3,4,5。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211577956.9A CN115902487A (en) | 2022-12-09 | 2022-12-09 | Microwave filter debugging method for knowledge-driven particle swarm optimization |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211577956.9A CN115902487A (en) | 2022-12-09 | 2022-12-09 | Microwave filter debugging method for knowledge-driven particle swarm optimization |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115902487A true CN115902487A (en) | 2023-04-04 |
Family
ID=86486694
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211577956.9A Pending CN115902487A (en) | 2022-12-09 | 2022-12-09 | Microwave filter debugging method for knowledge-driven particle swarm optimization |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115902487A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117330939A (en) * | 2023-11-20 | 2024-01-02 | 芯火微测(成都)科技有限公司 | ATE-based switched capacitor filter adjustment and measurement method, system and storage medium |
CN117521564A (en) * | 2024-01-04 | 2024-02-06 | 中铁电气化铁路运营管理有限公司 | Optimized design method and system for microwave circuit |
-
2022
- 2022-12-09 CN CN202211577956.9A patent/CN115902487A/en active Pending
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117330939A (en) * | 2023-11-20 | 2024-01-02 | 芯火微测(成都)科技有限公司 | ATE-based switched capacitor filter adjustment and measurement method, system and storage medium |
CN117521564A (en) * | 2024-01-04 | 2024-02-06 | 中铁电气化铁路运营管理有限公司 | Optimized design method and system for microwave circuit |
CN117521564B (en) * | 2024-01-04 | 2024-03-26 | 中铁电气化铁路运营管理有限公司 | Optimized design method and system for microwave circuit |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN115902487A (en) | Microwave filter debugging method for knowledge-driven particle swarm optimization | |
CN109783905B (en) | Intelligent microwave cavity filter debugging method based on particle swarm optimization algorithm | |
CN112016167B (en) | Aircraft aerodynamic shape design method and system based on simulation and optimization coupling | |
CN111426957B (en) | SOC estimation optimization method for power battery under simulated vehicle working condition | |
CN101478069A (en) | Microwave filter assistant debugging method based on nuclear machine learning | |
CN112395777B (en) | Engine calibration parameter optimization method based on automobile exhaust emission simulation environment | |
CN113065268A (en) | Rapid multi-objective space reflector optimization design method based on proxy model | |
CN115169235A (en) | Super surface unit structure inverse design method based on improved generation of countermeasure network | |
CN107562990A (en) | BAW filter design methods, device and equipment based on BVD models | |
CN114330067A (en) | Soft foundation sluice finite element model correction method | |
CN117350170B (en) | Nonlinear oscillation analysis method based on KOOPHAN deep neural network | |
CN113221278B (en) | Light weight method for vehicle-mounted photoelectric platform calibration frame | |
Liu et al. | An efficient data-driven optimization framework for designing graded cellular structures | |
CN116933386A (en) | Aircraft pneumatic data fusion method based on MCOK proxy model | |
CN117037140A (en) | Perovskite property prediction method based on deep learning | |
CN110137967B (en) | Large-scale power system power flow convergence adjusting method for key nodes | |
CN111240198B (en) | Piezoelectric ceramic actuator hysteresis analysis method | |
CN113342734B (en) | Plasma density distribution real-time computing device based on microwave reflectometer | |
CN115481554A (en) | Thermal diffusion digital twin model, temperature field real-time optimization control model and method in explosive fusion casting and curing process | |
CN111522226B (en) | Multi-objective optimization high-type PID optimal controller design method for servo turntable | |
CN112966423A (en) | Construction method of IPSO-IAGA-BPNN algorithm-based space primary mirror structure response prediction model | |
CN113158541B (en) | Multi-feature fusion modeling debugging method for microwave filter | |
CN112952315B (en) | Medium filter migration modeling debugging method based on mechanism guidance data acquisition | |
CN105550424A (en) | Screening method of interpolating sequences based on RBF grid deformation | |
Zhou et al. | Short-term traffic flow prediction based on the IMM-BP-UKF model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |