CN116738820A - Electromagnetic response optimization system, method and product for passive device - Google Patents

Electromagnetic response optimization system, method and product for passive device Download PDF

Info

Publication number
CN116738820A
CN116738820A CN202310518521.5A CN202310518521A CN116738820A CN 116738820 A CN116738820 A CN 116738820A CN 202310518521 A CN202310518521 A CN 202310518521A CN 116738820 A CN116738820 A CN 116738820A
Authority
CN
China
Prior art keywords
layer
electromagnetic response
training
neural network
passive device
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
Application number
CN202310518521.5A
Other languages
Chinese (zh)
Inventor
郑小平
任一民
邓晓娇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tsinghua University
Original Assignee
Tsinghua University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tsinghua University filed Critical Tsinghua University
Priority to CN202310518521.5A priority Critical patent/CN116738820A/en
Publication of CN116738820A publication Critical patent/CN116738820A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Geometry (AREA)
  • Computer Hardware Design (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides an electromagnetic response optimization system, method and product for a passive device, and relates to the technical field of electromagnetic simulation. In the embodiment of the invention, the feature extraction module is used for extracting the features of the target electromagnetic response through the pole remainder transfer function to obtain poles and remainder corresponding to the target electromagnetic response; the processing module is used for inputting poles and residues corresponding to the target electromagnetic response into the pre-trained multilayer convolutional neural network and outputting geometric design variables of the broadband device; according to the embodiment of the invention, the transfer function is introduced into the inverse model, so that the problem of high input dimension of the traditional inverse model can be solved, the accurate representation of electromagnetic response is realized, and the frequency resonance phenomenon possibly occurring in the direct training of S parameters is avoided. In addition, the embodiment of the invention also specifically determines a proper transfer function form and builds a proper neural network structure aiming at the simulation design of the broadband high-precision device so as to obtain smaller training errors and higher model accuracy.

Description

Electromagnetic response optimization system, method and product for passive device
Technical Field
The embodiment of the invention relates to the technical field of electromagnetic simulation, in particular to an electromagnetic response optimization system, an electromagnetic response optimization method and an electromagnetic response optimization product for a passive device.
Background
As the precision requirements on high-performance microwave passive devices (such as filters and antennas) are higher and higher, the requirements on high simulation precision and low modeling cost are expected to be met in the simulation optimization design process, so that the rapid and accurate structural design of the devices is realized. The diversified design of the device can be realized by depending on commercial software, but the consumed electromagnetic simulation cost and the structure of the complex device are mutually restricted. The invention aims to realize high-efficiency and high-precision electromagnetic response optimization of the passive device.
Traditional optimization methods include analytical methods, numerical calculations, and empirical modeling. The electromagnetic response calculated based on the analytical method requires a designer to have very strong expertise, and the method is difficult to be applied to complex new devices. Electromagnetic simulation results can be obtained depending on numerical calculations in commercial software. However, the method needs to determine a proper initial structure according to design experience and theoretical knowledge of a designer, and combines a large number of repeated parameter scanning and manual parameter adjustment to approach a design target, and the process consumes a large amount of computing resources and may be difficult to obtain an optimal solution. The experience model has the problems of limited application range and insufficient precision.
With the rapid development of the artificial intelligence field, the circuit design can be accelerated by introducing a machine learning algorithm into the electromagnetic simulation process, and a computer-aided model is constructed to replace the traditional simulation optimization process. The nonlinear relation between the electromagnetic simulation result and the geometric design variable of the passive device can be learned by using the neural network. The trained model can accurately and rapidly predict the electromagnetic behaviors of the microwave element, and effectively lighten the calculation burden of full-wave electromagnetic simulation.
The input of the existing neural network parameterized model is geometric and physical parameters, and the output of the model is electrical parameters. However, the actual circuit design is a process of solving the geometric variables reversely according to the given target electrical parameters, namely, reversely extracting the geometric variables. The trained model needs to be repeatedly called by selecting a proper optimization algorithm until a result is obtained, and once the design target changes, the model needs to be optimized again.
It follows that a new method for optimizing electromagnetic response to passive devices is currently needed.
Disclosure of Invention
The embodiment of the invention provides an electromagnetic response optimization system, an electromagnetic response optimization method and an electromagnetic response optimization product for a passive device, which aim to at least partially solve the problems in the related art.
An embodiment of the present invention provides an electromagnetic response optimization system for a passive device, where the system includes:
the feature extraction module is used for extracting features of the target electromagnetic response through a pole remainder transfer function to obtain poles and remainder corresponding to the target electromagnetic response;
the processing module is used for inputting poles and residues corresponding to the target electromagnetic response into a pre-trained multilayer convolutional neural network and outputting geometric design variables of a broadband device;
the multi-layer convolutional neural network includes: at least 1 convolution layer, a pooling layer and a full connection layer, wherein in each convolution layer, different characteristics of input parameters are extracted by using a plurality of convolution kernels, and different characteristic mapping is obtained; nonlinear mapping is carried out on the output of the convolution layer by adopting a ReLU activation function; adopting a maximum pooling layer to realize a downsampling process; after multiple convolution calculations, the transition from the convolution layer to the full connection layer is realized by utilizing the flat layer; the fully connected layer uses a sigmoid function as an activation function; the output layer is the geometric design variable of the target device.
Optionally, introducing a BN layer to perform standardization treatment after the multi-layer convolutional neural network convolutional layer; and introducing a dropout layer after the full connection layer, randomly deleting part of neurons with target probability, and inhibiting the overfitting phenomenon.
Optionally, the system further comprises:
the training data construction module is used for completing the construction of a training set and a testing set by utilizing commercial software to carry out full-wave electromagnetic simulation and carrying out data preprocessing on geometric design variables;
and the model training module is used for training the multi-layer convolutional neural network according to the training set, continuously optimizing and adjusting the super parameters in the network according to the learning condition of the network in the training process, and ending the training when the model meets the precision requirement.
Optionally, the training data construction module is further configured to: normalization processing is carried out on the training set and the testing set, and the convergence speed of the network is improved and the learning difficulty is reduced by normalizing all variables distributed in different intervals into the same interval; the normalization processing adopts a Z-score normalization method, and processed data are in standard normal distribution, so that the influence of data magnitude is eliminated.
The second aspect of the embodiment of the invention provides an electromagnetic response optimization method for a passive device, which comprises the following steps:
extracting characteristics of a target electromagnetic response through a pole remainder transfer function to obtain poles and remainder corresponding to the target electromagnetic response;
inputting poles and residues corresponding to the target electromagnetic response into a pre-trained multi-layer convolutional neural network, and outputting geometric design variables of a broadband device;
the multi-layer convolutional neural network includes: at least 1 convolution layer, a pooling layer and a full connection layer, wherein in each convolution layer, different characteristics of input parameters are extracted by using a plurality of convolution kernels, and different characteristic mapping is obtained; nonlinear mapping is carried out on the output of the convolution layer by adopting a ReLU activation function; adopting a maximum pooling layer to realize a downsampling process; after multiple convolution calculations, the transition from the convolution layer to the full connection layer is realized by utilizing the flat layer; the fully connected layer uses a sigmoid function as an activation function; the output layer is the geometric design variable of the target device.
Optionally, introducing a BN layer to perform standardization treatment after the multi-layer convolutional neural network convolutional layer; and introducing a dropout layer after the full connection layer, randomly deleting part of neurons with target probability, and inhibiting the overfitting phenomenon.
Optionally, the method further comprises:
the method comprises the steps of constructing a training set and a testing set by using commercial software to carry out full-wave electromagnetic simulation, and carrying out data preprocessing on geometric design variables;
and training the multi-layer convolutional neural network according to the training set, continuously optimizing and adjusting super parameters in the network according to the learning condition of the network in the training process, and ending the training when the model meets the precision requirement.
Optionally, the method further comprises:
normalization processing is carried out on the training set and the testing set, and the convergence speed of the network is improved and the learning difficulty is reduced by normalizing all variables distributed in different intervals into the same interval; the normalization processing adopts a Z-score normalization method, and processed data are in standard normal distribution, so that the influence of data magnitude is eliminated.
A third aspect of the embodiments of the present invention provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps in the electromagnetic response optimization method for a passive device according to the second aspect of the present invention when executed.
A fourth aspect of the embodiments of the present invention provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of the passive device oriented electromagnetic response optimization method according to the second aspect of the present invention.
In the embodiment of the invention, the characteristic extraction of the target electromagnetic response is realized by selecting the transfer function with low sensitivity and high robustness in the form of pole remainder, and the corresponding transfer function coefficient, namely the pole and the remainder, is obtained by a vector fitting mode. This process reduces the input dimension of the model. Further, the pole remainder is used as an input variable to be fed into a pre-trained multi-layer convolutional neural network, and the output of the network is the geometric design variable corresponding to the target electromagnetic response. Therefore, the embodiment of the invention can directly obtain the geometric structure meeting the requirement according to the set target electric parameter. In addition, the embodiment of the invention is particularly suitable for parameterized modeling of the broadband device, and meets the requirement of low modeling cost while ensuring high simulation precision.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments of the present invention will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a block diagram of a passive device oriented electromagnetic response optimization system in accordance with an embodiment of the present invention;
FIG. 2 is a flow chart of steps of a method for optimizing electromagnetic response of a passive device according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a model training flow in an electromagnetic response optimization method for passive devices according to an embodiment of the present invention;
fig. 4 is a schematic diagram of a training data processing flow in an electromagnetic response optimization method for a passive device according to an embodiment of the present invention.
Detailed Description
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
The neural network model based on the transfer function belongs to one of forward models, wherein the transfer function is used for realizing feature extraction of electromagnetic response, and the neural network is used for learning a nonlinear relation between transfer function coefficients and geometric design variables of passive devices. The introduction of the transfer function can reduce the highly nonlinear relation between the frequency and the electromagnetic response and can also improve the convergence speed of the neural network. Therefore, the embodiment of the invention provides that the transfer function is introduced into the inverse model, so that the problem of high input dimension of the traditional inverse model can be solved, the accurate representation of electromagnetic response is realized, and the frequency resonance phenomenon possibly occurring when the electromagnetic parameter (such as S parameter) is directly trained is avoided. In addition, the embodiment of the invention also specifically determines a proper transfer function form and builds a proper neural network structure aiming at the simulation design of the broadband high-precision device so as to obtain smaller training errors and higher model accuracy.
Specifically, the embodiment of the invention provides an electromagnetic response optimization system for a passive device. Referring to fig. 1, a structural block diagram of an electromagnetic response optimization system for a passive device according to an embodiment of the present invention is shown, and as shown in the drawing, the electromagnetic response optimization system for a passive device provided by the embodiment of the present invention includes:
the feature extraction module 101 is configured to perform feature extraction on a target electromagnetic response through a pole-remainder transfer function, so as to obtain a pole and a remainder corresponding to the target electromagnetic response.
And the processing module 102 is used for inputting poles and residuals corresponding to the target electromagnetic response into the pre-trained multilayer convolutional neural network and outputting geometric design variables of the broadband device.
The multi-layer convolutional neural network includes: at least 1 convolution layer, a pooling layer and a full connection layer, wherein in each convolution layer, different characteristics of input parameters are extracted by using a plurality of convolution kernels, and different characteristic mapping is obtained; nonlinear mapping is carried out on the output of the convolution layer by adopting a ReLU activation function; adopting a maximum pooling layer to realize a downsampling process; after multiple convolution calculations, the transition from the convolution layer to the full connection layer is realized by utilizing the flat layer; the fully connected layer uses a sigmoid function as an activation function; the output layer is the geometric design variable of the target device.
Compared with a single hidden layer full-connection network adopted by the traditional method, the convolutional neural network can expand the depth and complexity of the network by adding the convolutional layer, the pooling layer and the full-connection layer, so that the method has better feature extraction capability and generalization capability and better robustness. In each convolution layer, a plurality of convolution kernels (typically 16, 32, 64, etc.) are used to extract different features of the input parameters, resulting in different feature maps.
In the embodiment of the invention, the ReLU activation function forcedly corrects the input value smaller than 0 to 0, so that the gradient disappearance problem caused by the sigmoid function when the hidden layer is too many is avoided.
In the embodiment of the invention, the maximum pooling layer is selected to realize the down sampling process. The process reduces the input dimension of the subsequent network layer, and improves the calculation speed and the robustness of the feature mapping process; after multiple convolution calculations, the transition from the convolution layer to the full connection layer is realized by utilizing the flat layer; the fully connected layer uses a sigmoid function as the activation function. The output layer is a geometric design variable of the device.
Specifically, in the embodiment of the invention, a BN layer can be introduced after the multi-layer convolutional neural network convolutional layer for standardization treatment; and introducing a dropout layer after the full connection layer, randomly deleting part of neurons with target probability, and inhibiting the overfitting phenomenon.
In the embodiment of the invention, in order to improve the robustness of the model, a BN layer can be introduced after a convolution layer to perform standardization processing, so that the output distribution of a network intermediate layer is more stable, and the convergence of a network is facilitated. In addition, a dropout layer is introduced after the full connection layer, partial neurons are deleted randomly with a certain probability, the overfitting phenomenon can be restrained, and the generalization capability of the model is improved to a certain extent.
Specifically, in an embodiment of the present invention, the system further includes:
the training data construction module is used for constructing a training set and a testing set by full-wave electromagnetic simulation through commercial software and carrying out data normalization preprocessing on geometric design variables;
the model training module is used for training the multi-layer convolutional neural network according to the training set, continuously optimizing and adjusting the super parameters in the network according to the learning condition of the network in the training process, and ending the training when the training error of the super parameter model of the network is not reduced any more.
Specifically, the training data construction module is further configured to: normalization processing is carried out on the training set and the testing set, and the convergence speed of the network is improved and the learning difficulty is reduced by normalizing all variables distributed in different intervals into the same interval; the normalization processing adopts a Z-score normalization method, and processed data are in standard normal distribution, so that the influence of data magnitude is eliminated.
In the embodiment of the invention, an inverse modeling method based on a transfer function and a convolutional neural network is established, and the high-precision and rapid optimization design of the electromagnetic response of the passive device is realized. Feature extraction of electromagnetic response is achieved through a pole residue transfer function with high robustness, and the problem of high-dimensional input of a broadband device is solved. The convolutional layer is introduced to realize stronger feature extraction, the robustness of the model is improved through the design of the internal structure of the convolutional neural network, the training error of the model is reduced, and the accuracy of the model is improved. The inverse model takes coefficients of the transfer function as inputs and geometric design variables as model outputs. Once an accurate overall model is established, the geometric structure can be directly provided, and accurate and rapid prediction of electromagnetic response is realized.
In the embodiment of the invention, corresponding models can be trained for different devices, once an accurate overall model is established, the geometric structure can be directly provided, and accurate and rapid prediction of electromagnetic response is realized.
For the broadband device, due to the characteristic of high-dimensional input, on the basis that the input of the traditional inverse model is the electromagnetic response corresponding to the discrete frequency points, huge input dimension can be generated when the frequency variation range is too large, and the complexity of the model is increased, so that the application of the inverse model in the problem of optimizing the electromagnetic response of the broadband device is limited. In the embodiment of the invention, the introduction of the transfer function can reduce the high nonlinear relation between the frequency and the electromagnetic response, improve the convergence rate of the neural network, and greatly reduce the coefficient number of the transfer function to be smaller than the number of discrete frequencies when a broadband device is modeled. Therefore, the embodiment of the invention can solve the problem of high-dimensional input of the broadband device.
Based on the same inventive concept, the embodiment of the invention also provides an electromagnetic response optimization method for a passive device, which comprises the following steps:
s201, extracting characteristics of a target electromagnetic response through a pole remainder transfer function to obtain poles and remainder corresponding to the target electromagnetic response;
s202, inputting poles and residues corresponding to the target electromagnetic response into a pre-trained multilayer convolutional neural network, and outputting geometric design variables of a broadband device;
the multi-layer convolutional neural network includes: at least 1 convolution layer, a pooling layer and a full connection layer, wherein in each convolution layer, different characteristics of input parameters are extracted by using a plurality of convolution kernels, and different characteristic mapping is obtained; nonlinear mapping is carried out on the output of the convolution layer by adopting a ReLU activation function; adopting a maximum pooling layer to realize a downsampling process; after multiple convolution calculations, the transition from the convolution layer to the full connection layer is realized by utilizing the flat layer; the fully connected layer uses a sigmoid function as an activation function; the output layer is the geometric design variable of the target device.
Optionally, introducing a BN layer to perform standardization treatment after the multi-layer convolutional neural network convolutional layer; and introducing a dropout layer after the full connection layer, randomly deleting part of neurons with target probability, and inhibiting the overfitting phenomenon.
Optionally, the method further comprises:
the method comprises the steps of constructing a training set and a testing set by using commercial software to carry out full-wave electromagnetic simulation, and carrying out data preprocessing on geometric design variables;
and training the multi-layer convolutional neural network according to the training set, continuously optimizing and adjusting super parameters in the network according to the learning condition of the network in the training process, and ending the training when the model meets the precision requirement.
Optionally, the method further comprises:
normalization processing is carried out on the training set and the testing set, and the convergence speed of the network is improved and the learning difficulty is reduced by normalizing all variables distributed in different intervals into the same interval; the normalization processing adopts a Z-score normalization method, and processed data are in standard normal distribution, so that the influence of data magnitude is eliminated.
For easy understanding, the model training flow in the electromagnetic response optimization method for a passive device provided by the embodiment of the invention is explained with reference to fig. 3, and as shown in fig. 3, a schematic diagram of the model training flow in the electromagnetic response optimization method for a passive device in the embodiment of the invention is shown.
Specifically, the model training process in the embodiment of the invention comprises the following steps:
s1, determining a training set and a testing set.
S2, full-wave electromagnetic simulation results.
In the embodiment of the invention, full-wave electromagnetic simulation can be performed by using commercial software to complete the construction of a training set and a testing set, and data preprocessing is performed on geometric design variables.
And S3, obtaining corresponding transfer function coefficients based on the electromagnetic parameters.
In the embodiment of the invention, the characteristic extraction of the electromagnetic parameters of the training sample is realized by selecting the transfer function in the form of pole remainder with low sensitivity and high robustness, and the corresponding transfer function coefficient, namely the pole and the remainder, is obtained by a vector fitting mode. This process reduces the input dimension of the model.
And S4, training the neural network and testing.
In the embodiment of the invention, the pole remainder is fed into the neural network as an input variable, and the output of the network is a geometric design variable corresponding to each electromagnetic response, so that the neural network is trained and tested.
In the embodiment of the invention, the main structure of the neural network is the multilayer convolutional neural network.
And S5, judging whether the precision requirement is met, if so, turning to S7, and if not, turning to S6.
S6, adding a training sample correction model structure, and turning to the step S2.
S7, completing model construction, and being used for device design and optimization.
In the embodiment of the invention, firstly, an electromagnetic simulation result (for example, S parameter) is obtained according to the simulation, secondly, a proper transfer function (for example, a pole remainder transfer function) is selected, and a corresponding pole remainder (namely, a coefficient of the transfer function) is obtained through a vector fitting method. This process enables feature extraction of the electromagnetic response by obtaining transfer function coefficients.
In the embodiment of the invention, once model training is completed, the geometric design variable meeting the requirements can be directly obtained according to the set target electromagnetic parameters. In the embodiment of the invention, the performance of the model can be verified by using test data, and the advancement and effectiveness of the model in realizing device geometric structure prediction in microwave and terahertz frequency bands are verified. The electromagnetic response optimization method provided by the embodiment of the invention is particularly suitable for parameterized modeling of the broadband device, and meets the requirement of low modeling cost while ensuring high simulation precision.
For easy understanding, the training data processing flow of the electromagnetic response optimization method for a passive device provided by the embodiment of the invention is explained with reference to fig. 4, and as shown in fig. 4, a schematic diagram of the training data processing flow in the electromagnetic response optimization method for a passive device in the embodiment of the invention is shown.
Specifically, the training data processing flow in the embodiment of the invention comprises the following steps:
s11, determining a full-wave electromagnetic simulation result as training data;
s12, performing transfer function vector fitting on the training data to obtain poles and residuals;
and S13, taking poles and residues as network inputs, outputting most of geometric design variables by using a network, and training the multi-layer convolutional neural network.
In the embodiment of the invention, in the training stage, full-wave electromagnetic simulation is performed by using commercial software to complete the construction of a training set and a testing set, the full-wave electromagnetic simulation is used as training data, the electromagnetic response in the training data is subjected to feature extraction, poles and residues are used as network inputs, geometrical design variables are mostly network outputs, and the multi-layer convolutional neural network is trained. After model training is completed, in an application stage, feature extraction can be directly carried out on the target electromagnetic response by utilizing a pole remainder transfer function to obtain poles and remainder corresponding to the target electromagnetic response, the poles and remainder corresponding to the target electromagnetic response are input into a multi-layer convolutional neural network after the training is completed, and geometric design variables of the broadband device are directly obtained.
In the embodiment of the present invention, an exemplary embodiment is further provided to exemplarily explain the electromagnetic response optimization method for a passive device, which is described above:
an example one, a parameterized model was developed for a three-pole H-plane filter operating in the microwave frequency band. The model has 4 geometric design parameters as output, and the full-wave electromagnetic simulation result is S 11 The frequency range is 11.6-12.4GHz. 49 training samples and 49 test samples were defined. The order of the pole-remainder transfer function is set to 10, and the input dimension of the model is reduced from 161 dimensions to 40 dimensions. The test error of the model is 4.80E-03, and when the training samples are increased, the test error of the model is further reduced to be 2.95E-03.
Example two, a parameterized model was developed for a compact microstrip resonator element low pass filter operating in the terahertz frequency band. The model has 6 geometric design parameters as output, and the full-wave electromagnetic simulation result is S= [ S ] 11 ,S 21 ]The frequency range is 50-270GHz. 243 training samples and 50 test samples were defined. The order of the pole-remainder transfer function is set to 10, and the input dimension of the model is reduced from 442 dimension to 80 dimension. The test error of the model is 1.97E-04, and the training effect of the model is good.
Based on the same inventive concept, an embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the steps in the electromagnetic response optimization method for a passive device described in any one of the embodiments are implemented when the processor executes the computer program.
Based on the same inventive concept, an embodiment of the present invention provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements the steps in the electromagnetic response optimization method for a passive device according to any one of the above embodiments.
In this specification, each embodiment is described in a progressive manner, and each embodiment is mainly described by differences from other embodiments, and identical and similar parts between the embodiments are all enough to be referred to each other.
It will be apparent to those skilled in the art that embodiments of the present invention may be provided as a method, apparatus, or computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the invention may take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
Embodiments of the present invention are described with reference to flowchart illustrations and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable passive device oriented electromagnetic response optimizing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable passive device oriented electromagnetic response optimizing terminal device, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable passive device-oriented electromagnetic response optimization terminal device to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable passive device-oriented electromagnetic response optimization terminal device to cause a series of operational steps to be performed on the computer or other programmable terminal device to produce a computer implemented process such that the instructions which execute on the computer or other programmable terminal device provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiment and all such alterations and modifications as fall within the scope of the embodiments of the invention.
Finally, it is further noted that relational terms such as first and second, and the like are 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. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal 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 terminal. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or terminal device comprising the element.
The electromagnetic response optimizing system, method and product for passive device provided by the invention are described in detail, and specific examples are applied to illustrate the principle and implementation of the invention, and the description of the above examples is only used for helping to understand the method and core idea of the invention; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in accordance with the ideas of the present invention, the present description should not be construed as limiting the present invention in view of the above.

Claims (10)

1. A passive device oriented electromagnetic response optimization system, the system comprising:
the feature extraction module is used for extracting features of the target electromagnetic response through a pole remainder transfer function to obtain poles and remainder corresponding to the target electromagnetic response;
the processing module is used for inputting poles and residues corresponding to the target electromagnetic response into a pre-trained multilayer convolutional neural network and outputting geometric design variables of a broadband device;
the multi-layer convolutional neural network includes: the method comprises the steps of extracting different characteristics of input parameters by using a plurality of convolution kernels in each convolution layer to obtain different characteristic maps; nonlinear mapping is carried out on the output of the convolution layer by adopting a ReLU activation function; adopting a maximum pooling layer to realize a downsampling process; after multiple convolution calculations, the transition from the convolution layer to the full connection layer is realized by utilizing the flat layer; the fully connected layer uses a sigmoid function as an activation function; the output layer is the geometric design variable of the target device.
2. The passive device-oriented electromagnetic response optimization system of claim 1, wherein a BN layer is introduced after the multi-layer convolutional neural network convolutional layer for normalization; and introducing a dropout layer after the full connection layer, randomly deleting part of neurons with target probability, and inhibiting the overfitting phenomenon.
3. The passive device-oriented electromagnetic response optimization system of claim 1, further comprising:
the training data construction module is used for constructing a training set and a testing set by full-wave electromagnetic simulation through commercial software and carrying out data normalization preprocessing on geometric design variables;
the model training module is used for training the multi-layer convolutional neural network according to the training set, continuously optimizing and adjusting the super parameters in the network according to the learning condition of the network in the training process, and ending the training when the training error of the super parameter model of the network is not reduced any more.
4. The passive device-oriented electromagnetic response optimization system of claim 3, wherein the training data construction module is further configured to: normalization processing is carried out on the training set and the testing set, and the convergence speed of the network is improved and the learning difficulty is reduced by normalizing all variables distributed in different intervals into the same interval; the normalization processing adopts a Z-score normalization method, and processed data are in standard normal distribution, so that the influence of data magnitude is eliminated.
5. The electromagnetic response optimization method for the passive device is characterized by comprising the following steps of:
extracting characteristics of a target electromagnetic response through a pole remainder transfer function to obtain poles and remainder corresponding to the target electromagnetic response;
inputting poles and residues corresponding to the target electromagnetic response into a pre-trained multi-layer convolutional neural network, and outputting geometric design variables of a broadband device;
the multi-layer convolutional neural network includes: the method comprises the steps of extracting different characteristics of input parameters by using a plurality of convolution kernels in each convolution layer to obtain different characteristic maps; nonlinear mapping is carried out on the output of the convolution layer by adopting a ReLU activation function; adopting a maximum pooling layer to realize a downsampling process; after multiple convolution calculations, the transition from the convolution layer to the full connection layer is realized by utilizing the flat layer; the fully connected layer uses a sigmoid function as an activation function; the output layer is the geometric design variable of the target device.
6. The electromagnetic response optimization method for the passive device according to claim 5, wherein a BN layer is introduced after the multi-layer convolutional neural network convolutional layer for normalization processing; and introducing a dropout layer after the full connection layer, randomly deleting part of neurons with target probability, and inhibiting the overfitting phenomenon.
7. The passive device oriented electromagnetic response optimization method of claim 5, further comprising:
the method comprises the steps of constructing a training set and a testing set by using commercial software to carry out full-wave electromagnetic simulation, and carrying out data normalization preprocessing on geometric design variables;
and training the multi-layer convolutional neural network according to the training set, continuously optimizing and adjusting the super parameters in the network according to the learning condition of the network in the training process, and ending the training when the training error of the super parameter model of the network is not reduced any more.
8. The method for optimizing electromagnetic response of a passive device according to claim 7, wherein the performing data normalization preprocessing on the geometric design variables comprises:
normalization processing is carried out on the training set and the testing set, and the convergence speed of the network is improved and the learning difficulty is reduced by normalizing all variables distributed in different intervals into the same interval; the normalization processing adopts a Z-score normalization method, and processed data are in standard normal distribution, so that the influence of data magnitude is eliminated.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the passive device oriented electromagnetic response optimization method of any of claims 5-8 when the computer program is executed by the processor.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the passive device oriented electromagnetic response optimization method of any of the claims 5-8.
CN202310518521.5A 2023-05-09 2023-05-09 Electromagnetic response optimization system, method and product for passive device Pending CN116738820A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310518521.5A CN116738820A (en) 2023-05-09 2023-05-09 Electromagnetic response optimization system, method and product for passive device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310518521.5A CN116738820A (en) 2023-05-09 2023-05-09 Electromagnetic response optimization system, method and product for passive device

Publications (1)

Publication Number Publication Date
CN116738820A true CN116738820A (en) 2023-09-12

Family

ID=87901971

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310518521.5A Pending CN116738820A (en) 2023-05-09 2023-05-09 Electromagnetic response optimization system, method and product for passive device

Country Status (1)

Country Link
CN (1) CN116738820A (en)

Similar Documents

Publication Publication Date Title
CN111860982A (en) Wind power plant short-term wind power prediction method based on VMD-FCM-GRU
CN109886464B (en) Low-information-loss short-term wind speed prediction method based on optimized singular value decomposition generated feature set
Gonzalez et al. Optimizing loss functions through multi-variate taylor polynomial parameterization
CN108182316B (en) Electromagnetic simulation method based on artificial intelligence and electromagnetic brain thereof
CN110852417B (en) Single-depth neural network model robustness improving method for application of Internet of things
CN113553755B (en) Power system state estimation method, device and equipment
CN114399032A (en) Method and system for predicting metering error of electric energy meter
CN111950711A (en) Second-order hybrid construction method and system of complex-valued forward neural network
CN112380762A (en) Power transmission line short-circuit fault diagnosis method based on VMD-WOA-LSSVM
CN110765704A (en) Novel automatic deep neural network modeling method applied to microwave device
Regazzoni et al. A physics-informed multi-fidelity approach for the estimation of differential equations parameters in low-data or large-noise regimes
CN116738820A (en) Electromagnetic response optimization system, method and product for passive device
CN111460734A (en) Microwave device automatic modeling method combining advanced adaptive sampling and artificial neural network
CN115906303A (en) Planar microwave filter design method and device based on machine learning
CN115204036A (en) Ionized layer foF2 prediction method and device based on CNN-LSTM and storage medium
CN114638421A (en) Method for predicting requirement of generator set spare parts
CN111475931B (en) PCI rapid analysis method and device based on neural network, electronic equipment and storage medium
Yuan et al. A novel fault diagnosis method for second-order bandpass filter circuit based on TQWT-CNN
CN116595883B (en) Real-time online system state correction method for numerical reactor
Peng et al. An Efficient Antenna Optimization Framework Based on Inverse Radial Basis Function Network
Zhang et al. Application of CNN Based on Improved Activation Function in Image Classification Processing
Gao et al. Rolling Prediction Model of Closing Price Based on EEMD Data Noise Reduction and HGS-DELM
CN117113592A (en) Antenna housing electrical property prediction method based on compensation network
CN116540190A (en) End-to-end self-supervision intelligent interference suppression method and device and electronic equipment
Zhao et al. Research on No-reference Image Quality Assessment Algorithm Based on Generative Adversarial Networks

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