阅读说明：本技术 一种新型快速高效的滤波器小样本建模及优化方法 (Novel rapid and efficient filter small sample modeling and optimizing method ) 是由 喻梦霞 陈林 汪家兴 李桂萍 于 2020-12-25 设计创作，主要内容包括：本发明公开了一种新型快速高效的滤波器小样本建模及优化方法,针对原有建模优化方法和实际滤波器电路设计中建模优化计算耗时长,且样本数据量有限的不足,本发明提出了一种新颖快速高效的滤波器小样本建模及模型优化方法,该方法是一种基于先验知识的高斯过程的滤波器正向建模及基于粒子群反向优化的方法(KBGP)。在正向建模时,该方法使用粗糙样本数据,训练ANNs粗糙网络；之后,将ANNs的输出向量作为先验知识注入GPR,可以减少GPR的样本数据,获得高精度电路模型；在反向优化时,使用粒子群优化(PSO)方法,快速优化模型输入向量获得最佳全局的输入向量了。(The invention discloses a novel rapid and efficient filter small sample modeling and optimizing method, and provides a novel rapid and efficient filter small sample modeling and model optimizing method aiming at the defects that the time consumption for modeling optimization calculation in the original modeling optimization method and the actual filter circuit design is long and the sample data volume is limited. In forward modeling, the method trains ANNs rough networks by using rough sample data; then, the output vector of the ANNs is used as priori knowledge to be injected into the GPR, sample data of the GPR can be reduced, and a high-precision circuit model is obtained; in the reverse optimization, the optimal global input vector is obtained by rapidly optimizing the model input vector by using a Particle Swarm Optimization (PSO) method.)

1. A novel rapid and efficient filter small sample modeling and optimization method is characterized by comprising the following steps: the method comprises the following steps:

step 1: establishing a rough model, namely firstly, determining an input vector and an output vector of the rough model, forming an input vector by using input variables and forming an output vector by using output variables; then, constructing an equivalent rough road model by using ADS, and quickly acquiring sample data of the rough network; training a network and detecting whether the precision of the rough network model meets the requirement;

step 2: establishing a GPR sample set, uniformly sampling by using CST software to obtain a small amount of full-wave electromagnetic simulation results, and combining the prediction results of the ANN as prior knowledge with the CST full-wave simulation results to form GPR sample data;

and step 3: calling GPR sample data and training a GPR model; calling CST to generate a test sample, and checking the reliability of the proposed method through the average absolute error to complete the modeling process of the model;

and 4, step 4: and (3) optimizing the KBGP model, namely taking the output result of the GPR model as PSO fitness, continuously updating the positions of the particles in the PSO algorithm, and finishing the optimization process of the whole KBGP model.

2. The method of claim 1 for modeling and optimizing a new fast and efficient filter small sample, comprising: the coarse model is a coarse model of an ANN-based microstrip band-stop filter.

3. The method of claim 1 for modeling and optimizing a new fast and efficient filter small sample, comprising: the GPR model trained in the step S3 adopts a GPR covariance function expression:

whereinσ_f，σ_mCharacteristic length and standard deviation of the signal, respectively.

4. The method of claim 1 for modeling and optimizing a new fast and efficient filter small sample, comprising: in the PSO algorithm in step S4, in the optimization process, the fit function is infinitely close to the minimum value.

5. The method of claim 1 for modeling and optimizing a new fast and efficient filter small sample, comprising: and after the KBGP model optimization process is completed in the step 4, performing model evaluation analysis in the step 5.

Technical Field

The invention relates to the field of simulation modeling and model optimization of a filter, in particular to a novel rapid and efficient filter small sample modeling and model optimization method, which is a method (KBGP) for forward modeling of a filter based on a Gaussian process of priori knowledge and reverse optimization based on particle swarm.

Background

With the development of wireless communication technology, various new technologies are emerging continuously, filters are also developing towards higher frequencies continuously, the performance and the circuit structure of the filters are more complex, and the original modeling optimization method can not meet the requirements of modern design gradually. Models for modern filter design need to be able to describe not only behavior characteristics, but also accurately reflect the changes in filter performance caused by the differences in variables. The modern precise modeling optimization technology of the filter is a full-wave analysis method based on an electromagnetic theory, an ideal circuit is obtained through professional electromagnetic simulation software (such as CST, HFSS and the like), but the full-wave analysis method consumes more computing resources and has long time, the structure is complex, the precision is improved, the computing times and the resource consumption are exponentially increased along with the increase of the working frequency of the filter, the software computing time is too long, the computing result is often not converged or the precision requirement is not met, and the actual complex engineering application is increasingly difficult, so that a rapid and efficient modeling optimization method is urgently needed for design to adapt to the continuously changing design requirements.

The advent of machine learning provides a new idea for modeling optimization of filters. If the electromagnetic simulation is replaced by the model generated by machine learning, the modeling optimization of the modern filter can be efficiently carried out by utilizing the characteristics of high machine learning precision, short calculation time and less occupied memory. Artificial neural networks (ans) and Gaussian Process Regression (GPR) are two efficient methods of machine learning modeling optimization that have been increasingly applied in recent years to the modeling and optimization of filters. ANNs can utilize a large amount of sample data to establish ANNs models, and can approximate complex relationships which are difficult to describe between geometric structure dimensions and filter electrical parameters. However, the network structure of the ANNs is difficult to determine, especially, the selection of the number of network layers and the number of neurons lacks of theoretical support, too many training samples are needed, the time consumption is too long, and the rapid modeling is difficult to realize. The intellectual neural network (KBNN) is an improved method for ANNs, and although training samples can be reduced to some extent, the effect of the intellectual neural network is limited when the training samples are extremely insufficient for accurate training. The GPR is used as a powerful tool, not only is the model training time short and the precision high, but also the GPR can be applied to modeling optimization of complex circuits, and the defects of ANNs can be just made up. Therefore, the need for rapid and accurate modeling optimization of modern filters cannot be met by using a machine learning method alone.

Aiming at the problems that the modeling optimization calculation consumes long time and the sample data amount is limited in the original modeling optimization method and the actual filter circuit design, a novel, quick and efficient filter small sample modeling and model optimization method is provided, and the method is a filter forward modeling and particle swarm reverse optimization method (KBGP) based on the prior knowledge Gaussian process. The new method (KBGP) of the invention has the advantages of small number of used samples, high modeling optimization speed and high accuracy, can solve the problem of modeling optimization of the original circuit, and realizes rapid, efficient and high-accuracy modeling optimization of the filter.

Disclosure of Invention

The invention aims to: the invention provides a novel rapid and efficient filter small sample modeling and optimizing method, and provides a novel rapid and efficient filter small sample modeling and model optimizing method aiming at the defects that the time consumption for modeling optimization calculation in the original modeling optimization method and the actual filter circuit design is long and the sample data amount is limited. In forward modeling, the method trains ANNs rough networks by using rough sample data; then, the output vector of the ANNs is used as priori knowledge to be injected into the GPR, sample data of the GPR can be reduced, and a high-precision circuit model is obtained; in the reverse optimization, the optimal global input vector is obtained by rapidly optimizing the model input vector by using a Particle Swarm Optimization (PSO) method.

The technical scheme adopted by the invention is as follows:

the KBGP method provided by the invention has two stages: forward modeling and reverse optimization, see fig. 1. In forward modeling, the rough model is a low-precision model, namely a multilayer perceptive neural network (MLPANN), and training sample data of the rough model is derived from simulation software ADS. The ADS is electromagnetic software based on a road, the simulation speed is high, the precision is low, enough sample data can be obtained in a very short time by utilizing the ADS, and a rough model is constructed. Then, by utilizing GPR and combining the priori knowledge output by the rough model, the mapping between the results obtained by the rough model and the full-wave electromagnetic simulation is accelerated, the precise convergence of GPR is realized, and the forward modeling process of KBGP is completed. During reverse optimization, the output result of the GPR is used as the fitness by utilizing the PSO global optimization characteristic, the positions of particles in the PSO algorithm are updated, the global optimal solution can be obtained only within a few seconds, and the KBGP reverse optimization process is completed.

In order to achieve the purpose of the invention, the technical scheme of the invention is as follows:

a novel fast and efficient filter small sample modeling and model optimization method comprises the following steps:

step 1: and establishing a rough model. Firstly, determining an input vector and an output vector of a rough model, forming input variables into the input vector, and forming output variables into the output vector; then, constructing an equivalent rough road model by using the ADS, and quickly acquiring sample data of the rough network; finally, training the network and detecting whether the precision of the rough network model meets the requirement;

step 2: a GPR sample set is established. And uniformly sampling by using CST software to obtain a small amount of full-wave electromagnetic simulation results. Taking the prediction result of the ANN as prior knowledge, and combining a CST full-wave simulation result to form GPR sample data;

and step 3: the GPR model is trained. Calling GPR sample data and training a GPR model; then, calling CST to generate a test sample, and checking the reliability of the proposed method through the average absolute error to complete the modeling process of the model;

and 4, step 4: and (5) optimizing the KBGP model. And taking the output result of the GPR model as PSO fitness, continuously updating the positions of the particles in the PSO algorithm, and completing the optimization process of the whole KBGP model only in a few seconds.

In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:

(1) the modeling technology of the KBGP method can get rid of the limitation of the original neural network model to the huge number of samples, greatly reduce the number of samples, and has high calculation precision and calculation efficiency particularly in solving the problem of a small sample rapid filter model;

(2) the forward modeling process of the KBGP method provided by the invention comprises a rough model and a Gaussian Process (GPR). The method combines the characteristics of ANNs and GPRs, and small sample modeling can be quickly realized under the condition that the model precision can be ensured;

(3) the PSO model is used independently, so that local convergence is easy to happen, and the calculation speed is low; according to the method, the output result of the KBGP model is used as the fitness, the positions of particles in the PSO algorithm are updated, the reverse optimization of the model can be realized, the optimal input vector value can be quickly searched, and the local optimization is avoided;

(4) the KBGP method provided by the invention can reduce the training sample and modeling time of modeling, can also improve the prediction precision of the model, and shows that the KBGP method has important value in filter design.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and that for those skilled in the art, other relevant drawings can be obtained according to the drawings without inventive effort, wherein:

FIG. 1 is a schematic diagram of the whole process of modeling optimization of the novel KBGP;

FIG. 2 is a forward modeling flow diagram of the novel KBGP;

FIG. 3 is a schematic structural diagram of a band-stop filter;

FIG. 4 is the | S of different methods for the same test sample₂₁An | curve;

fig. 5 shows the filter optimization result based on the KBGP method.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the detailed description and specific examples, while indicating the preferred embodiment of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.

Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.

It is noted that relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

The features and properties of the present invention are described in further detail below with reference to examples.

Example one

Based on the technical indexes of the band-stop filter, the process and the modeling schematic diagram shown in fig. 1 and fig. 2 are adopted in the process, the circuit structure shown in fig. 3 is adopted, and the following steps are optimized in a specific modeling mode:

step 1: and establishing a rough model of the micro-strip band-stop filter based on the ANN. The index requirements of the band elimination filter are as follows:

|S21|≥0.92(-0.72dB)，for 2.8GHz≤f≤3.6GHz

|S21|≤0.05(-26dB)，for 4.1GHz≤f≤4.9GHz

|S21|≥0.92(-0.72dB)，for 5.4GHz≤f≤6.3GHz

determining input vectors for a coarse model(The ith frequency point representing the jth geometric structure), see table 1 for sample data of the microstrip band-stop filter, the output vector is | S₂₁|。

TABLE 1 sample data for microstrip band reject filters

The rough model selects Multilayer Neural Network Structures (MLPANNs), selects 100 groups of ADS simulation results of geometric structures as sample data by using a full-factor orthogonal experiment method, and conducts rough MLPANN model training, wherein the set test error precision is 0.7 dB.

Step 2: a GPR sample set is established.

And generating an input vector, and calling CST to perform full-wave electromagnetic simulation. Taking the prediction result of the ANN as prior knowledge, and combining with a CST simulation result to form a GPR sample set;

and step 3: the GPR model is trained.

GPR covariance function expression:

whereinσ_f，σ_mCharacteristic length and standard deviation of the signal, respectively.

Firstly, selecting different numbers of GPR sample sets, training GPRs, then testing generalized errors by using test samples, and comparing the generalized errors with the original method, wherein the average errors of the novel KBGP method and the original method are shown in FIG. 4 and Table 2 when the samples are different in number;

number of samples

6

9

12

15

18

21

24

KBNN

20.47％

9.47％

3.54％

2.28％

1.37％

1.01％

0.76％

GPR

2.39％

1.93％

1.49％

1.36％

1.32％

1.14％

1.05％

Novel KBGP

1.28％

1.15％

0.76％

0.68％

0.61％

0.56％

0.48％

TABLE 2 mean error of the new KBGP method from the original method for different sample numbers

And 4, step 4: and optimizing the model, and searching the optimal input vector of the circuit model.

And (4) combining a Particle Swarm Optimization (PSO) algorithm, and optimizing the input parameter vector of the circuit. According to the index requirement, the fitness function is set as:

fit＝3*S₂₁i1_{Pass band}+0.8*(0.93-S₂₁I1_{Low stop band})+1.2*(0.93-S₂₁I1_{High stop band})

In the optimization process of the PSO algorithm, the fit function should be minimized. After optimization and optimization, the input vectors of the available circuit model and the optimization result are shown in fig. 5.

And 5: and (6) evaluating and analyzing the model.

As can be seen from fig. 4 and table 2, compared with the original methods such as the knowledge neural network (KBNN) and the gaussian regression process (GPR), the degree of fitting between the KBGP result and the CST result is the best, and when the number of samples is 12, the error reaches 0.76%, which indicates that the KBGP modeling method has the highest degree of fitting, the KBGP error is the lowest, the accuracy is the highest, and the samples are the least. The method has high modeling precision and small sample number, and is particularly suitable for actual complex engineering.

As can be seen from fig. 5, after the optimization by the method, the KBGP result is completely consistent with the full-wave CST accurate result, thereby further reducing the model error and improving the model accuracy; therefore, the KBGP modeling and optimization method can reduce training samples and modeling time for modeling, can also ensure model accuracy, has higher prediction accuracy, and shows that the modeling optimization method provided by the invention has important value in filter design optimization, as shown in the consumption time of the novel KBGP method and the original method in the table 3, the method provided by the invention replaces the original filter modeling optimization method, and can quickly and accurately establish a reliable model by using a small amount of sample data, so that the modeling time is greatly reduced, and the aim of efficient and rapid modeling optimization is fulfilled, and the method is shown in the table 2 and the table 3.

Algorithm	KBNN	GPR	Novel KBGP
				Time(s)	9.46	22.4	3.13

TABLE 3 consumption time of the novel KBGP method and the original method

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and should not be taken as limiting the scope of the present invention, and any modifications, equivalents and improvements made by those skilled in the art within the spirit and principle of the present invention should be included in the scope of the present invention.