CN112149351A - Microwave circuit physical dimension estimation method based on deep learning - Google Patents

Abstract

The invention discloses a method for estimating the physical size of a microwave circuit based on deep learning. The initial data set is used to construct a training sample set; step 2, the convolutional neural network model is trained according to the training sample set to obtain a neural network model for microwave circuit size estimation; step 3, the S parameters of the target microwave circuit are collected as input parameters, Input the neural network model for estimating the size of the microwave circuit to obtain the physical size of the target microwave circuit. The method for estimating the physical size of a microwave circuit based on deep learning provided by the invention has a high degree of automation and a high estimation accuracy, can effectively reduce the intermediate links and manual intervention in the estimation of the physical size parameters of the microwave circuit, application cost and complexity, and effectively improve the microwave circuit Accuracy and real-time performance of physical dimension parameter estimation.

Description

A method for estimating the physical size of microwave circuits based on deep learning

technical field

The invention belongs to the technical field of microwave circuit design simulation, and particularly relates to a method for estimating the physical size of a microwave circuit based on deep learning.

Background technique

In recent years, training neural networks to model the electrical properties of passive and active components/circuits for advanced simulation and design to provide quick answers to tasks has become an approach. Conventional methods of microwave circuit design in the industry today, such as numerical modeling methods, can be computationally expensive, and analytically new equipment that may be difficult to obtain, such as empirical models, may be limited in scope and accuracy. Therefore, neural network technology has been widely used in various microwave applications, such as embedded passive devices, transmission line components, coplanar waveguide (CPW) components, spiral inductors, FETs.

At present, the design method of microwave circuits in the industry is mainly based on theoretical modeling. After theoretical simulation, the approximate circuit size parameters are obtained. The actual simulation software is used to simulate within a certain parameter range. Finally, the comparison of the simulation results is used to select the effect that is closest to theory The size parameters of the microwave circuit corresponding to the simulation results, but this scheme has problems such as low efficiency, large workload, and no theoretical guidance, which cannot meet the efficiency requirements of today's industry.

SUMMARY OF THE INVENTION

The purpose of the present invention is a method for estimating the physical size of a microwave circuit based on deep learning. Through the deep learning technology, a regression analysis is performed on the corresponding relationship between the calculated theoretical electrical parameters and the circuit size parameters, so that the size parameters of the microwave circuit can be quickly and accurately obtained. .

The technical scheme provided by the present invention is:

A method for estimating the physical size of a microwave circuit based on deep learning, comprising the following steps:

Step 1, collecting the S parameters of multiple groups of microwave circuits and the physical dimensions of the multiple groups of microwave circuit parameters as an initial data set, and constructing a training sample set according to the initial data set;

Step 2, training the convolutional neural network model according to the training sample set to obtain a neural network model for microwave circuit size estimation;

Step 3: Collect the S parameters of the target microwave circuit as input parameters, input the size estimation neural network model of the microwave circuit, and obtain the physical size of the target microwave circuit.

Preferably, in the second step, the convolutional neural network model is trained by a gradient descent method to obtain the microwave circuit size estimation neural network model.

Preferably, in the first step, a training sample set is constructed according to the initial data set, including:

The size of the S-parameter of the microwave circuit is converted and adjusted to a 3×3×1 format; and a data enhancement method is used to expand the data amount of the initial data set, and a part of the expanded initial data set is taken as training sample set.

Preferably, the data in the expanded initial data set is divided into three parts, including:

The training sample set is 60%, the validation data set is 20%, and the test data set is 20%.

Preferably, the microwave circuit size estimation neural network model includes an input layer, a processing module, a dropout layer, a fully connected layer and a regression prediction output layer which are connected in sequence.

Preferably, the processing module includes at least four processing units connected in sequence, wherein,

The first to third processing units include sequentially connected convolution layers, batch normalized BN layers, modified linear units ReLU layers and pooling layers, and the fourth processing unit includes sequentially connected convolution layers, batch Normalized BN layer, modified linear unit ReLU layer.

Preferably, the size of the convolution kernels in the convolutional layers of the four processing units is all 2×2; the number of the convolution kernels of the four processing units increases in order of connection.

Preferably, in the third step, obtaining the physical size of the target microwave circuit includes the following steps:

Step 1. The four processing units in the processing module sequentially process and transmit the input data based on the gradient descent algorithm, until the target feature map is output in the four processing units;

Step 2, the fully connected layer converts the target feature map into a one-dimensional vector;

Step 3: The regression prediction output layer outputs the estimation result of the physical size of the target microwave circuit according to the one-dimensional vector.

Preferably, in the step 1, including:

The convolution layer extracts the feature map of the input data according to the following formula;

为卷积神经网络模型中第j个卷积层的第i个输出的特征图；m为卷积神经网络模型中第j个卷积层输入特征图的数量；In the formula,

is the feature map of the ith output of the jth convolutional layer in the convolutional neural network model; m is the number of input feature maps of the jth convolutional layer in the convolutional neural network model;

The batch normalization BN layer normalizes the feature map;

The normalized feature map of the ReLU layer is nonlinearly transformed;

The pooling layer performs size reduction processing on the received nonlinear transformed feature map according to the following formula;

为降采样函数；F为降采样滤波器大小；S为降采样步长。In the formula,

is the downsampling function; F is the downsampling filter size; S is the downsampling step size.

Preferably, in the step 3, the fully connected layer converts the target feature map into a one-dimensional vector according to the following formula;

v ^j =f(w ^j v ^j-1 +b ^j );

In the formula, v _j is the output one-dimensional vector of the j-th fully-connected layer; w _j is the weight matrix of the j-th fully-connected layer; b _j is the bias term of the j-th fully-connected layer; f( ) is Nonlinear activation function.

The beneficial effects of the present invention are:

The method for estimating the physical size of a microwave circuit based on deep learning provided by the invention has a high degree of automation and a high estimation accuracy, can effectively reduce the intermediate links and manual intervention in the estimation of the physical size parameters of the microwave circuit, application cost and complexity, and effectively improve the microwave circuit Accuracy and real-time performance of physical dimension parameter estimation.

Description of drawings

FIG. 1 is a flowchart of the method for estimating physical size parameters of microwave circuits based on a convolutional neural network according to the present invention.

FIG. 2 is a flow chart of establishing a convolutional neural network model according to the present invention.

FIG. 3 is a schematic structural diagram of the convolutional neural network model according to the present invention.

FIG. 4 is a flowchart of the method for estimating physical size parameters of microwave circuits based on a convolutional neural network according to the present invention.

FIG. 5 is a schematic flowchart of Embodiment 1 of the present invention.

FIG. 6 is a structural block diagram of a microwave circuit physical size parameter estimation system based on a convolutional neural network in Embodiment 1 of the present invention.

FIG. 7 is a schematic structural diagram of an electronic device in Embodiment 1 of the present invention.

FIG. 8 is a comparison diagram of the simulation results of the physical size parameters of the microwave circuit in theory and the simulation results of the physical size parameters of the microwave circuit obtained in Example 1. FIG.

Detailed ways

The present invention will be further described in detail below with reference to the accompanying drawings, so that those skilled in the art can implement it with reference to the description.

As shown in Figure 1, the present invention provides a method for estimating the physical size of a microwave circuit based on deep learning, including:

In step 101, the electrical parameters of the target microwave circuit are continuously collected, and a corresponding input data set is constructed according to the electrical parameters of the target microwave circuit.

In step 101, the microwave circuit electrical parameter acquisition unit in the convolutional neural network-based microwave circuit physical size parameter estimation system automatically receives the collected electrical parameters of the target microwave circuit, and the specific method for collecting the electrical parameters of the target microwave circuit The operation interface provided by the electromagnetic simulation software Sonnet can be invoked through MATLAB to perform automatic operation collection, and the collected electrical parameters are sent to the microwave circuit electrical parameter collection unit.

Step 102, the input data set is input into the trained convolutional neural network model, and the estimation result of the physical size parameter of the target microwave circuit is obtained; wherein, the trained convolutional neural network model is based on the training data set. Input, trained using gradient descent.

In step 102, after automatically receiving the collected electrical parameter data of the target microwave circuit, the microwave circuit electrical parameter acquisition unit sends all the electrical parameter data of the target microwave circuit to the convolutional neural network-based microwave circuit physics a convolutional neural network model establishment unit in the size parameter estimation system, the convolutional neural network model establishment unit receives the electrical parameters of the microwave circuit, and uses the input data set corresponding to the electrical parameters of the microwave circuit as the input of the model, According to characteristics such as the size of the input data set, a convolutional neural network model specially used for estimating the physical size parameters of the target microwave circuit is established. It can be understood that, in addition to the input layer, the fully connected layer and the output layer, which are sequentially connected, the convolutional neural network model is connected between the input layer and the fully connected layer. The setting method and the number of layers depend on characteristics such as the size of the input data set corresponding to the microwave electrical parameter data. Understandably, a Convolutional Neural Network (CNN) is a feedforward neural network whose artificial neurons can respond to surrounding units within a partial coverage.

On the other hand, this embodiment also provides a microwave circuit physical size parameter estimation system based on a convolutional neural network, including:

a microwave circuit electrical parameter and physical size parameter acquisition unit, configured to continuously collect the target microwave circuit and construct a corresponding input data set according to the electrical parameters of the target microwave circuit;

A parameter estimation unit, used for inputting the input data set into a trained convolutional neural network model to obtain an estimation result of the physical size parameter of the target microwave circuit; wherein, the trained convolutional neural network model is based on training The data set is the input and is trained by gradient descent.

This embodiment further includes a processor, a communication interface, a memory, and a bus, wherein the processor, the communication interface, and the memory communicate with each other through the bus, and the processor can call the deep learning training model stored in the memory to execute the first aspect Provides a method for estimating physical size parameters of microwave circuits based on convolutional neural networks.

Meanwhile, this embodiment also provides a non-transitory computer-readable storage medium, where the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the convolution-based method provided in the first aspect. Microwave circuit physical size parameter estimation method based on neural network.

The estimation unit of the microwave circuit physical size parameter in the microwave circuit physical size parameter estimation system based on the convolutional neural network performs model training, verification and testing on the convolutional neural network model according to the gradient descent algorithm and the test data set , to obtain the estimation result of the physical size parameter of the target microwave circuit.

It can be understood that the gradient descent algorithm is an optimization algorithm, also commonly referred to as the steepest descent method. The steepest descent method uses the negative gradient direction as the search direction. The closer the steepest descent method is to the target value, the smaller the step size and the slower the progress.

In this embodiment, as shown in FIG. 2 , before the input data set is input into the trained convolutional neural network model, and the estimation result of the physical size parameter of the target microwave circuit is obtained, the method further includes:

Step 201, using the input data set corresponding to the electrical parameters of the target microwave circuit as the input of the input layer of the convolutional neural network;

Step 202, establishing a processing module of the convolutional neural network model according to the size of the data in the input data set;

Step 203: Connect the input layer, the processing module, the Dropout layer for preventing overfitting of the output of the processing module, the fully connected layer for converting the output of the processing module into a one-dimensional vector, and the layer for A regression prediction output layer that outputs the estimation results of the physical size parameters of the target microwave circuit, and completes the establishment of the convolutional neural network model.

In the above steps 201 to 203, the overall architecture of the convolutional neural network model is shown in FIG. 3, the processing module includes at least four processing units connected in sequence, and the first processing unit to the third processing unit The units include successively connected convolution layers, batch normalization layers (BN layers), modified linear units ReLU layers and pooling layers, and the fourth processing unit includes the connected convolution layers, batch normalization BN layers layer, modified linear unit ReLU layer;

Among them, the size of the convolution kernels in each convolutional layer in the sequentially connected four processing units is unchanged, and the number of convolution kernels increases sequentially.

It can be seen from the above description that the method for estimating physical size parameters of microwave circuits based on convolutional neural networks provided in this embodiment provides an effective and reliable method for establishing a convolutional neural network model for estimating physical size parameters of microwave circuits, Among them, the convolutional neural network model used for the physical size parameters of the microwave circuit has an accurate structure and strong pertinence, and can more accurately estimate the physical size parameters of the microwave circuit.

In this embodiment, as shown in FIG. 4 , the input data set is input into the trained convolutional neural network model, and the estimation result of the physical size parameter of the target microwave circuit is obtained, which specifically includes:

Step 401, the four processing units in the processing module sequentially process and transmit the input data set based on the gradient descent algorithm, until the target feature map is output in the four processing units; wherein, the processing module The convolution layer in extracts the received feature map of the input data set; the BN layer normalizes the received feature map; the ReLU layer normalizes the received The processed feature map is nonlinearly transformed; the pooling layer performs size reduction processing on the received nonlinearly transformed feature map.

In step 401, the convolutional layer in the processing module extracts the received feature map of the input data set, including:

The convolutional layer extracts the received feature map of the input data set according to formula 1:

为卷积神经网络模型中第j个卷积层的第i个输出的特征图；m为卷积神经网络模型中第j个卷积层输入特征图的数量；In formula (1),

is the feature map of the ith output of the jth convolutional layer in the convolutional neural network model; m is the number of input feature maps of the jth convolutional layer in the convolutional neural network model;

为第j个卷积层的第i个偏置项；f(·)为非线性激活函数；

is the i-th bias term of the j-th convolutional layer; f( ) is the nonlinear activation function;

Correspondingly, the pooling layer performs size reduction processing on the received nonlinear transformed feature map, including:

The pooling layer reduces the size of the received nonlinearly transformed feature map according to formula 2:

为降采样函数；F为降采样滤波器大小；S为降采样步长。In formula (2),

is the downsampling function; F is the downsampling filter size; S is the downsampling step size.

Step 402, the fully connected layer converts the target feature map into a one-dimensional vector, including:

The fully connected layer converts the target feature map into a one-dimensional vector according to formula 3:

v ^j =f(w ^j v ^j-1 +b ^j )#(3)

In formula (3), v _j is the output one-dimensional vector of the j-th fully-connected layer; w _j is the weight matrix of the j-th fully-connected layer; b _j is the bias term of the j-th fully-connected layer; f ( ) is a nonlinear activation function.

Step 403, the regression prediction layer outputs the estimation result of the physical size parameter of the target microwave circuit according to the one-dimensional vector.

Example 1

To further illustrate this solution, the present invention also provides an application example of a method for estimating physical size parameters of microwave circuits based on convolutional neural networks, see FIG. 5 , and specifically includes the following contents:

501. Obtain electrical parameter data of the microwave circuit, and preprocess the data;

Specifically, the preprocessing includes adjusting the size of the microwave circuit electrical parameter data, for example, adjusting to a 3×3×1 format. The electrical parameters of the preprocessed microwave circuit construct an initial data set, and a data enhancement method is used to expand the data volume of the initial data set;

The dataset is divided into three parts: training dataset, validation dataset and test dataset.

502. Build a convolutional neural network model structure, and train, verify and test the convolutional neural network model according to the expanded data set and gradient descent method;

Specifically, the convolutional neural network model of the present invention takes the electrical parameters of the microwave circuit as input, and takes the estimated physical size parameters of the microwave circuit as the output, and includes five modules in total. The first to fourth modules all include 1 convolutional layer, 1 BN layer, 1 RELU layer and 1 pooling layer. The kernel size of the convolutional layer is 2×2, which is used to obtain the features of the input image. The RELU layer non-linearly transforms the feature map generated by the convolutional layer to realize the sparseness of the network, reduce the interdependence of parameters, and alleviate the overfitting problem. The RELU layer does not change the size of the feature map. The size of the convolution kernel of the pooling layer is 2 × 2, and the stride is 1, both of which are maximum pooling, and the size of the feature map is reduced to 1/2 of the original. The number of convolution kernels in the four modular convolutional layers gradually increases, which are 32, 64, 128, 256 in turn. The size of the convolution kernel of the fifth module and the functions of each layer are the same as those of the first four modules, including 1 convolution layer, 1 BN layer, and 1 RELU layer. The number of convolution kernels in the convolution layer is 512. The last module includes 2 fully-connected layers and 1 regression output layer. The first fully-connected layer of the 2 fully-connected layers contains 500 neurons, and the second fully-connected layer contains 1 neuron. The regression output layer estimates the physical size parameters of the microwave circuit.

503. Perform physical size parameter estimation on the input microwave circuit data set according to the convolutional neural network model;

The method for estimating the physical size parameters of a microwave circuit based on a convolutional neural network provided by the present invention can directly use the electrical parameters of the microwave circuit as the input of the estimation model, expand the amount of input data through data enhancement, and construct a convolutional neural network model for training and verification. And test, realize the rapid estimation of the physical size parameters of the microwave circuit, improve the accuracy of the estimation, and realize the practical application of the method for estimating the physical size parameters of the microwave circuit.

As shown in FIG. 6 , the microwave circuit physical size parameter estimation system based on the convolutional neural network includes: a microwave circuit electrical parameter acquisition unit 601 and a microwave circuit physical size parameter estimation unit 602 . in:

The microwave circuit electrical parameter collection unit 601 is configured to continuously collect electrical parameters of the target microwave circuit, and construct a corresponding input data set according to the electrical parameters of the target microwave circuit. The microwave circuit physical size parameter estimation unit 602 is configured to input the input data set into a trained convolutional neural network model to obtain an estimation result of the target microwave circuit physical size parameter; wherein, the trained convolutional neural network The model takes the training dataset as input and is trained by gradient descent.

Specifically, the acquisition unit 601 may acquire microwave circuit electrical parameter data, and perform preprocessing on the microwave electrical parameter data, where the preprocessing includes adjusting the size of the input data.

Further, an initial data set is constructed according to the preprocessed data, and a data enhancement method is used to expand the size of the initial data set. The data set is divided into three parts: training data set, validation data set and test data set, the proportions are: 60%, 20% and 20% respectively.

The microwave circuit physical size parameter estimation unit 602 is used to construct the convolutional neural network structure of the present invention. The convolutional neural network model of the present invention takes the electrical parameter data of the microwave circuit as the input, and takes the estimated physical size parameter of the microwave circuit as the output, and includes five modules in total. The first to fourth modules all include 1 convolutional layer, 1 BN layer, 1 RELU layer and 1 pooling layer. The kernel size of the convolutional layer is 2×2, which is used to obtain the features of the input image. The RELU layer non-linearly transforms the feature map generated by the convolutional layer to realize the sparseness of the network, reduce the interdependence of parameters, and alleviate the overfitting problem. The RELU layer does not change the size of the feature map. The size of the convolution kernel of the pooling layer is 2 × 2, and the stride is 2, both of which are maximum pooling, which reduces the size of the feature map to 1/2 of the original. The number of convolution kernels in the four modular convolutional layers gradually increases, which are 32, 64, 128, 256 in turn. The size of the convolution kernel of the fifth module and the function of each layer are the same as those of the first four modules, including 1 convolution layer, 1 BN layer, and 1 RELU layer. The number of convolution kernels in the convolution layer is 512. The last module includes 2 dropout layers, 2 fully connected layers, and 1 regression output layer. The dropout rates of the 2 dropout layers are both 0.5. The first fully connected layer of the 2 fully connected layers contains 500 neurons. Two fully connected tests contain 1 neuron. The regression output layer estimates the physical size parameters of the microwave circuit. The microwave circuit physical size parameter estimation unit 602 performs the training, verification and testing of the convolutional neural network based on the expanded data set and the gradient descent algorithm, and based on the tested convolutional neural network model, the input microwave circuit The electrical parameters are used to estimate the physical size parameters of the microwave circuit.

It can be seen from the above description that the estimation of the physical size parameters of a microwave circuit based on a convolutional neural network provided by the embodiments of the present invention can directly use the electrical parameters of the microwave circuit as the input of the recognition model, expand the input data volume through data enhancement, and construct a convolutional neural network. The network model is trained, verified and tested to realize the estimation of the physical size parameters of the microwave circuit, which improves the accuracy of the estimation and realizes the practical application of the estimation of the physical size parameters of the microwave circuit.

As shown in FIG. 7 , the electronic device includes: a processor, a communication interface (CommunicationsInterface), a memory (memory), and a bus, wherein the processor, the communication interface, and the memory communicate with each other through the bus. The processor can invoke the logic instructions in the memory to perform the following method, for example, including: continuously collecting electrical parameters of the target microwave circuit, and constructing a corresponding input data set according to the electrical parameters of the target microwave circuit; Input the trained convolutional neural network model to obtain the estimation result of the physical size parameter of the target microwave circuit; wherein, the trained convolutional neural network model takes the training data set as input, and is obtained by using the gradient descent method to train .

The above-mentioned logic instructions in the memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product in essence, or the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: U disk, removable hard disk, Read-Only Memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes.

All or part of the steps of implementing the above method embodiments can be completed by program instructions related to hardware, the aforementioned program can be stored in a computer-readable storage medium, and when the program is executed, it executes the steps including the above method embodiments; The aforementioned storage medium includes various media that can store program codes, such as ROM, RAM, magnetic disk, or optical disk.

The above-described embodiments of communication devices and the like are only illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in One place, or it can be distributed over multiple network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

From the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and certainly can also be implemented by hardware. Based on this understanding, the above-mentioned technical solutions can be embodied in the form of software products in essence or the parts that make contributions to the prior art, and the computer software products can be stored in computer-readable storage media, such as ROM/RAM, magnetic Disks, optical discs, etc., include instructions for causing a computer device (which may be a personal computer, server, or network device, etc.) to perform the methods of various embodiments or portions of embodiments.

As shown in Table 1, it is the result table obtained by comparing the accuracy of the microwave circuit physical size parameter estimation method based on deep learning with other regression prediction methods such as linear Regression and SVM RF Regression in Example 1. It can be seen that in MSE, MAE, and R ² are the three commonly used evaluation criteria for regression analysis.

Table 1. Accuracy comparison table of microwave circuit physical size results estimated by different methods

Figure 8 is a comparison diagram of the theoretical simulation results of the physical size parameters of the circuit and the results obtained by the simulation of the physical size parameters predicted by this method. It can be seen that the two are basically consistent at the equal ripple, which means that the circuit is based on this method. Preparations based on the physical size parameters predicted by the method can obtain properties that are close to those obtained by theoretical calculations.

To sum up, the estimation method provided by the present invention includes automatically collecting the electrical parameters of the target microwave circuit, and constructing an input data set; A convolutional neural network model of the physical size parameters of the target microwave circuit; and performing model training, verification and testing on the convolutional neural network model according to a gradient descent algorithm and a test data set, to obtain an estimate of the physical size parameters of the target microwave circuit result. The invention has a high degree of automation and high estimation accuracy, can effectively reduce the intermediate links and artificial uncertainty of microwave circuit design simulation, reduce the application cost and complexity of microwave circuit design, and effectively improve the accuracy of microwave circuit physical size parameter estimation.

Although the embodiment of the present invention has been disclosed as above, it is not limited to the application listed in the description and the embodiment, and it can be applied to various fields suitable for the present invention. For those skilled in the art, it can be easily Therefore, the invention is not limited to the specific details and illustrations shown and described herein without departing from the general concept defined by the appended claims and the scope of equivalents.

Claims (10)

1. A microwave circuit physical dimension estimation method based on deep learning is characterized by comprising the following steps:

step one, collecting S parameters of a plurality of groups of microwave circuits and physical dimensions of the plurality of groups of microwave circuit parameters as an initial data set, and constructing a training sample set according to the initial data set;

training the convolutional neural network model according to the training sample set to obtain a microwave circuit size estimation neural network model;

and step three, collecting S parameters of the target microwave circuit as input parameters, and inputting the S parameters into the microwave circuit size estimation neural network model to obtain the physical size of the target microwave circuit.

2. The microwave circuit physical dimension estimation method based on deep learning of claim 1, wherein in the second step, the convolutional neural network model is trained through a gradient descent method to obtain the microwave circuit dimension estimation neural network model.

3. The deep learning-based microwave circuit physical dimension estimation method according to claim 2, wherein in the step one, a training sample set is constructed from the initial data set, and the method comprises:

adjusting the size of the S parameter of the microwave circuit to a 3 x 1 format; and expanding the data volume of the initial data set by adopting a data enhancement method, and taking a part of the expanded initial data set as a training sample set.

4. The deep learning based microwave circuit physical dimension estimation method of claim 3, wherein the data in the expanded initial data set is divided into three parts, including:

training sample set 60%, validation data set 20% and test data set 20%.

5. The microwave circuit physical dimension estimation method based on deep learning of claim 3 or 4, wherein the microwave circuit dimension estimation neural network model comprises sequentially connected: the device comprises an input layer, a processing module, a Dropout layer, a full connection layer and a regression prediction output layer.

6. The deep learning based microwave circuit physical dimension estimation method of claim 5, wherein the processing module comprises at least four processing units connected in sequence, wherein,

the first processing unit to the third processing unit respectively comprise a convolution layer, a batch standardization BN layer, a correction linear unit ReLU layer and a pooling layer which are connected in sequence, and the fourth processing unit comprises a convolution layer, a batch standardization BN layer and a correction linear unit ReLU layer which are connected in sequence.

7. The deep learning-based microwave circuit physical size estimation method according to claim 6, wherein the sizes of convolution kernels in the convolution layers of the four processing units are all 2 x 2; the number of convolution kernels of the four processing units is increased in sequence according to the connection order.

8. The method for estimating the physical size of the microwave circuit based on the deep learning of claim 7, wherein in the third step, obtaining the physical size of the target microwave circuit comprises the following steps:

step 1, sequentially processing and transmitting input data in four processing units in the processing module based on a gradient descent algorithm until a target characteristic diagram is output in the four processing units;

step 2, the full connection layer converts the target characteristic diagram into a one-dimensional vector;

and 3, outputting a physical size estimation result of the target microwave circuit by the regression prediction output layer according to the one-dimensional vector.

9. The deep learning-based microwave circuit physical dimension estimation method according to claim 8, wherein in the step 1, the method comprises:

the convolutional layer extracts a characteristic diagram of the input data according to the following formula;

in the formula (I), the compound is shown in the specification,

a feature map of the ith output of the jth convolutional layer in the convolutional neural network model; m is the number of the input characteristic graphs of the jth convolutional layer in the convolutional neural network model;

the batch standardization BN layer is used for carrying out normalization processing on the characteristic diagram;

carrying out nonlinear transformation on the normalized characteristic diagram of the ReLU layer;

the pooling layer performs size reduction processing on the received nonlinear-converted characteristic diagram according to the following formula;

in the formula (I), the compound is shown in the specification,

is a down-sampling function; f is the size of the down-sampling filter; and S is a down-sampling step length.

10. The deep learning-based microwave circuit physical dimension estimation method according to claim 9, wherein in the step 3, the full connection layer converts the target feature map into a one-dimensional vector according to the following formula;

v^j＝f(w^jv^j-1+b^j)；

in the formula, v_jOutputting a one-dimensional vector for the jth fully-connected layer; w is a_jThe weight matrix of the jth full connection layer; b_jBias term of jth fully-connected layer; f (-) is a nonlinear activation function.

Images