Passive device synthesis method and related device based on convolutional neural network

Abstract

The present invention discloses a convolutional neural network-based passive device synthesis method and related apparatus, belonging to the field of millimeter wave and terahertz passive device design. The method comprises the following steps: pixelating multiple arbitrary passive network structures and performing electromagnetic simulation on the pixelated multiple arbitrary passive network structures to obtain a two-dimensional matrix and corresponding S parameters; utilizing a generative adversarial network to expand the data set corresponding to the S parameters and the two-dimensional matrix; constructing a function fitting relationship between the two-dimensional matrix and the S parameters using a relationship fitting model based on the two-dimensional matrix and the corresponding S parameters; and optimizing the S parameters using a global optimization algorithm based on the function fitting relationship between the two-dimensional matrix and the S parameters to obtain the optimal passive network structure, thereby realizing passive device synthesis. The present invention can solve the problem of difficulty in analyzing the relationship between the S parameters and the topological structure of a passive network.

Description

Technical Field

[0001] The present invention belongs to the field of millimeter wave and terahertz passive device design, and specifically relates to a passive device synthesis method based on convolutional neural networks and related devices. Background Art

[0002] In the millimeter-wave and terahertz frequency bands, the optimized design of passive components is crucial to the performance of millimeter-wave and terahertz integrated circuits (ICs). Traditional passive component design and optimization methods rely on classical passive network topologies, such as transmission lines, couplers, L-shaped structures, T-shaped structures, and transformers. These methods suffer from limited solution space and heavy reliance on designer experience. Patent application publication number CN118569199A proposes a deep learning-based inverse design method for passive integrated circuits. This method simulates specific examples to obtain the structural parameter limits of corresponding capacitors and inductors. Within these limits, the capacitor and inductor structural parameters are generated and combined with the simulated parameters as a dataset. A deep learning conditional generative adversarial network is then constructed and trained on the dataset to achieve inverse design of RFIC passive integrated circuits. The passive topology in this patent application is restricted, making it impossible to exploit the structural performance of arbitrary passive components and occupying a large area. Furthermore, the device structure generation process is limited by the existing dataset, making it impossible to generate a device structure with better performance than the existing dataset.

[0003] Therefore, under the constraints of limited gain of active devices and large loss of broadband passive devices, the existing technology finds it difficult to analyze the relationship between S parameters and topological structures, and it is difficult to develop millimeter-wave and terahertz integrated circuits with good insertion loss and phase shift accuracy. Summary of the Invention

[0004] The purpose of the present invention is to provide a passive device synthesis method and related devices based on convolutional neural networks to solve the problem of difficulty in analyzing the relationship between the S parameters and topological structure of the passive network.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, a passive device synthesis method based on a convolutional neural network comprises the following steps:

[0007] pixelating a plurality of arbitrary passive network structures, and performing electromagnetic simulation on the plurality of arbitrary passive network structures after pixelation to obtain a two-dimensional matrix and corresponding S parameters;

[0008] Expanding the data set of the two-dimensional matrix and the corresponding S parameters to obtain an expanded two-dimensional matrix and the corresponding S parameters;

[0009] Based on the expanded two-dimensional matrix and the corresponding S parameters, a function fitting relationship between the two-dimensional matrix and the S parameters is constructed using a relationship fitting model;

[0010] Based on the functional fitting relationship between the two-dimensional matrix and the S parameters, the S parameters are optimized by a global optimization algorithm to obtain an optimal passive network structure and realize passive device synthesis.

[0011] In some embodiments, the step of expanding the data set of the two-dimensional matrix and the corresponding S parameters specifically includes: using the two-dimensional matrix and the S parameters as the input and output of the generative adversarial network, respectively, to obtain the expanded two-dimensional matrix and the corresponding S parameters.

[0012] In some embodiments, the process of establishing the relationship fitting model is: using the two-dimensional matrix and S parameters as the input and output of a convolutional neural network, respectively, and training and verifying the convolutional neural network to obtain the relationship fitting model.

[0013] In some embodiments, the convolutional layer in the convolutional neural network is sequentially connected to a normalization layer and an activation layer, and the fully connected layer in the convolutional neural network is sequentially connected to a normalization layer, an activation layer, and a random dropout layer, and a Leaky ReLU function is added to the activation layer.

[0014] In some implementations, optimizing the S parameters by a global optimization algorithm to obtain an optimal passive network structure specifically includes optimizing the S parameters by a genetic algorithm to obtain an optimal passive network structure.

[0015] In some embodiments, the step of optimizing the S parameters using a genetic algorithm to obtain an optimal passive network structure specifically includes:

[0016] Randomly generate a plurality of random two-dimensional matrices corresponding to arbitrary passive network structures to form a parent generation, and predict the S parameters corresponding to the random two-dimensional matrices using a relationship fitting model based on a function fitting relationship between the two-dimensional matrices and S parameters, the S parameters being referred to as predicted S parameters;

[0017] Calculating the error between the predicted S parameters and the expected S parameters using the mean square error between the predicted S parameters and the expected S parameters as a cost function, screening out a minimum error from the errors, and obtaining a plurality of two-dimensional matrices corresponding to the minimum errors;

[0018] Randomly select a number of two-dimensional matrices from the parent generation and select two two-dimensional matrices corresponding to the minimum error from them to cross and generate a child generation, repeating this step until the number of the child generation is equal to the number of the parent generation;

[0019] The parent generation is updated according to the two-dimensional matrices and children corresponding to the plurality of minimum errors, and the above steps are repeated until a preset number of updates is reached to obtain an optimal passive network structure.

[0020] In the second aspect, a passive component integrated system based on a convolutional neural network includes:

[0021] A passive network pixelation module is used to pixelate multiple arbitrary passive network structures, perform electromagnetic simulation on the pixelated multiple arbitrary passive network structures, and obtain a two-dimensional matrix and corresponding S parameters;

[0022] A data set expansion module, configured to expand the data set of the two-dimensional matrix and the corresponding S parameters to obtain an expanded two-dimensional matrix and the corresponding S parameters;

[0023] A structural parameter relationship fitting module is used to construct a function fitting relationship between the two-dimensional matrix and the S parameters using a relationship fitting model based on the expanded two-dimensional matrix and the corresponding S parameters;

[0024] The global optimization module is used to optimize the S parameters based on the function fitting relationship between the two-dimensional matrix and the S parameters through a global optimization algorithm to obtain the optimal passive network structure and realize passive device synthesis.

[0025] In a third aspect, an electronic device comprises a memory, a processor, and a computer program stored in the memory and executable in the processor, wherein the processor implements the steps of the passive device synthesis method based on convolutional neural networks when executing the computer program.

[0026] In a fourth aspect, a computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the passive device synthesis method based on a convolutional neural network.

[0027] In a fifth aspect, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the steps of the passive device synthesis method based on a convolutional neural network.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] This method performs pixelated decomposition on any passive network structure, constructs a two-dimensional matrix corresponding to the pixelated structure, then inputs the matrix into a relational fitting model to obtain the corresponding predicted S parameters. Finally, a global optimization algorithm is used to iteratively optimize the predicted S parameters to obtain the optimal passive network structure and synthesize passive components. This method has a large solution space and can produce passive components with excellent broadband performance, while also offering the advantages of low insertion loss and a small footprint. Furthermore, the entire process does not rely on electromagnetic simulation and is fast.

[0030] Furthermore, the present invention adopts a generative adversarial network to expand the data set consisting of the two-dimensional matrix corresponding to the passive network structure and its S parameters, which can reduce the time required for the passive device synthesis method in the electromagnetic simulation process and improve design efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a pixelated image of any passive network structure in an embodiment of the present invention;

[0032] Figure 2 An overall block diagram of a passive device synthesis method based on a convolutional neural network provided by an embodiment of the present invention;

[0033] Figure 3 A schematic diagram of the structure of a convolutional neural network used in an embodiment of the present invention;

[0034] Figure 4 A schematic diagram of a genetic algorithm flow chart used in an embodiment of the present invention;

[0035] Figure 5 Graphs showing comprehensive simulation results of broadband passive devices in an embodiment of the present invention, where (a) shows the passive network structure and (b) shows the S-parameter simulation curve.

[0036] Figure 6 A flow chart of the passive device synthesis method based on convolutional neural network provided by the present invention;

[0037] Figure 7 This is a structural diagram of the passive device integrated system based on convolutional neural networks provided by the present invention. DETAILED DESCRIPTION

[0038] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings. The content described is intended to explain rather than limit the present invention.

[0039] It should be noted that the terms "including" and "having" and any variations thereof in the description and claims of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or are inherent to these processes, methods, systems, products or apparatus.

[0040] This embodiment provides a passive device synthesis method based on convolutional neural network, such as Figure 2 As shown in FIG, the overall concept of the above-mentioned passive device synthesis method is to generate a data set, namely the passive network structure (i, j) and the corresponding S parameters, by using electromagnetic simulation. The electromagnetic simulation data set is expanded by using a generative adversarial network to obtain an expanded data set, which involves using CNN to fit the functional relationship between the S parameters and the passive network structure, namely S=f(i, j). Finally, a genetic algorithm is used to find the optimal solution of the objective function for this functional relationship, and finally the optimal passive network structure is obtained.

[0041] like Figure 6 As shown, the above passive device synthesis method specifically includes the following steps:

[0042] S1: Using the idea of ​​image pixelation, any passive network structure is pixelated and decomposed. Electromagnetic simulation is performed on the pixelated multiple passive network structures to obtain a two-dimensional matrix and corresponding S parameters, such as Figure 1 As shown in Table 1, a two-dimensional matrix is ​​used to represent any passive network structure. MATLAB is used to randomly generate 540,000 18×18 two-dimensional matrices. Table 1 shows the two-dimensional matrix table corresponding to any passive network structure.

[0043] Table 1 Two-dimensional matrix table

[0044]

[0045] S2: The two-dimensional matrix and corresponding S parameters generated by the electromagnetic simulation in S1 are expanded to obtain an expanded two-dimensional matrix and corresponding S parameters. Specifically, a generative adversarial network is used to use the two-dimensional matrix and S parameters as the input and output of the generative adversarial network, respectively, to expand the obtained new two-dimensional matrix and corresponding S parameters. These two-dimensional matrix and corresponding S parameters are combined with the original two-dimensional matrix and corresponding S parameters to form the expanded two-dimensional matrix and corresponding S parameters, thereby reducing the time required for data generation. The generative adversarial network can take the form of a deep convolutional generative adversarial network (DCGAN), a progressive growing generative adversarial network (PGAN), a bulldozer distance generative adversarial network (WGAN), etc.

[0046] S3: On the PyTorch platform, the expanded two-dimensional matrix in S2 and the corresponding S parameters are used as the input and output of the convolutional neural network, and the convolutional neural network is trained and verified to finally obtain a relationship fitting model. This relationship fitting model can construct a functional fitting relationship between the S parameters and the two-dimensional matrix corresponding to any passive network structure. Therefore, the corresponding predicted S parameters can be obtained by inputting the two-dimensional matrix into the relationship fitting model.

[0047] like Figure 3 As shown in Figure 2, the architecture of the convolutional neural network includes 12 convolutional layers and 5 fully connected layers, namely Figure 3 The 12 convolutional layers represent 12 convolutional layers, and the 5 fully connected layers represent 5 fully connected layers. Figure 3 The middle convolution layer 1 represents the first level of convolution layer. Similarly, each level of convolution layer is connected to the normalization layer and the activation layer in sequence. The first four levels of fully connected layers are connected to the normalization layer, the activation layer and the random dropout layer in sequence. The activation layer uses the Leaky ReLU function to prevent the gradient from disappearing to ensure a good learning effect of the convolutional neural network. The random dropout layer can reduce the possibility of overfitting in the convolutional neural network. The convolutional neural network has 102 outputs. Figure 3 Medium S parameter S 11 The real part (Re(S 11 ))、S parameter S 11 The imaginary part (Im(S 11 ))、S parameter S 21 The real part (Re(S 21 ))、S parameter S 21 The imaginary part (Im(S 21 ))、S parameter S22 The real part (Re(S 22 ))、S parameter S 22 The imaginary part (Im(S 22 )) is a partial diagram of 102 outputs. Figure 3 Where x, y, and z are coordinate axes. In this embodiment, the number of channels of each convolutional layer is 102, and the convolution kernel sizes of the first to thirteenth convolutional layers are 12×12, 10×10, 8×8, 6×6, 5×5, 5×5, 4×4, 4×4, 4×4, 3×3, 3×3, and 3×3, respectively; the number of neurons of the first to fifth fully connected layers is 800, 800, 800, 800, and 102, respectively, as shown in Tables 2 and 3 below:

[0048] Table 2 Convolution kernel size and number of channels of each convolution layer

[0049]

[0050] Table 3 Number of neurons in each fully connected layer

[0051]

[0052] Specifically, the 18×18 two-dimensional matrix in S1 is input into the above-mentioned CNN, where the first row, first column, eighteenth row, and eighteenth column of the two-dimensional matrix represent the positions of the excitation ports, respectively. The CNN output is the predicted S parameters of the electromagnetic simulation of the passive network structure represented by the two-dimensional matrix.

[0053] S4: Based on the function fitting relationship between the two-dimensional matrix and S parameters obtained in S3, the S parameters are optimized using a genetic algorithm to obtain the optimal passive network structure and implement passive device synthesis. This genetic algorithm can be replaced by other global optimization algorithms, such as the imperial competition algorithm, particle swarm optimization algorithm, simulated annealing algorithm, artificial bee colony algorithm, ant colony algorithm, firefly algorithm, gray wolf optimization algorithm, greedy algorithm, and domain search algorithm.

[0054] like Figure 4 As shown in Figure 2, the specific genetic algorithm optimization process is as follows:

[0055] S4.1, first initialize the population, randomly generate 4096 two-dimensional matrices as parents using MATLAB as described in S1, and use the CNN trained in S3 to predict the predicted S parameters of the passive network structure represented by the 4096 two-dimensional matrices;

[0056] S4.2, taking the expected S parameter as the target, the mean square error between the predicted S parameter and the expected S parameter is used as the cost function, as shown in the following formula:

[0057]

[0058]

[0059]

[0060] In the above formula, (Cost Function) is the cost function, To predict the S parameters, is the expected S parameter, To predict the function fitting relationship between S parameters and passive network structure.

[0061] The error between the predicted S parameters and the expected S parameters is calculated by the above cost function, and the minimum error between the predicted S parameters and the expected S parameters calculated by the cost function is used as the objective function, among which the two-dimensional matrices corresponding to the 16 minimum errors are directly inherited to the offspring.

[0062] S4.3. Randomly select 512 two-dimensional matrices from the parent generation and read the corresponding errors. Select the two two-dimensional matrices corresponding to the two smallest errors as parent individuals. After generating a progeny through crossover, repeat S4.3 until the number of progeny generated is equal to the number of parents described in S4.1.

[0063] S4.4, update the population through the offspring generated in S4.3 and the two-dimensional matrix corresponding to the 16 minimum errors in S4.2. The new round of population also includes the offspring generated by the crossover in S4.3 and the two-dimensional matrix corresponding to the 16 minimum errors in S4.2.

[0064] Repeat S4.2 to S4.4 until the preset number of iterations is 80, generating the 80th generation population. Finally, the two-dimensional matrix corresponding to the minimum error is selected as the optimal solution, and the optimal passive network structure is obtained.

[0065] like Figure 5 As shown in Figure (a), 0mm, 0.1mm, and 0.2mm are the distance scale values ​​represented in the drawing after the passive network structure is pixelated during the simulation process. x, y, and z represent the coordinate axes. In Figure (b), m1, m2, m3, and m4 are the performance characteristic values ​​of the integrated passive components, where the horizontal axis (X axis) is the frequency axis (GHz) and the vertical axis (Y axis) is the dB value of the S parameter. dB(S(P1, P1)) is the |S of the S parameter. 11 |, represents the reflection of the input signal at port 1; dB(S(P2, P1)) is the |S of the S parameter 21 |, indicating the signal transmission from port 1 to port 2; dB(S(P2, P2)) is the |S of the S parameter 22| indicates 2-port matching. Curve info indicates curve information. Solve:sweep1 indicates solving through the process of sweep1 during electromagnetic simulation.

[0066] Electromagnetic simulation results show that within the frequency range of 160 GHz to 300 GHz, the insertion loss of the optimal passive network structure is less than 1.85 dB, and the output-to-output matching is less than 10 dB from 170 GHz to 300 GHz. These electromagnetic simulation results verify the feasibility and effectiveness of the passive component synthesis method based on convolutional neural networks and genetic algorithms provided in this embodiment for synthesizing broadband, high-performance passive components in the sub-terahertz frequency band.

[0067] The passive device synthesis method based on convolutional neural network provided in the above embodiment has the following advantages:

[0068] (1) Using the idea of ​​image pixelation, any passive network structure is pixelated and decomposed. A two-dimensional matrix is ​​used to represent the passive network structure. Then, a convolutional neural network is used to construct a fitting relationship between the passive network structure and the S parameters. The solution space is large, and finally, through genetic algorithm iteration, a passive device with excellent broadband performance can be obtained. At the same time, it has the advantages of low insertion loss and small footprint.

[0069] (2) Using generative adversarial networks to expand the data set formed by the passive network structure and its corresponding S parameters can reduce the time required for electromagnetic simulation and improve design efficiency;

[0070] (3) By modifying the genetic algorithm cost function according to the different requirements of different passive devices, the fitting relationship between the arbitrary pixelated passive network structure constructed by the convolutional neural network algorithm and the S parameters can be used to comprehensively generate the optimal passive device structure that meets the performance requirements. It has the advantages of strong applicability and efficient design.

[0071] like Figure 7 As shown, this embodiment also provides a passive component integrated system based on a convolutional neural network, including:

[0072] A passive network pixelation module is used to pixelate multiple arbitrary passive network structures, perform electromagnetic simulation on the pixelated multiple arbitrary passive network structures, and obtain a two-dimensional matrix and corresponding S parameters;

[0073] The data set expansion module is used to expand the data set of the two-dimensional matrix and the corresponding S parameters to obtain the expanded two-dimensional matrix and the corresponding S parameters; specifically, it is used to use the two-dimensional matrix and the S parameters as the input and output of the generative adversarial network respectively to obtain the expanded two-dimensional matrix and the corresponding S parameters.

[0074] A structural parameter relationship fitting module is used to construct a function fitting relationship between the two-dimensional matrix and the S parameters using a relationship fitting model based on the expanded two-dimensional matrix and the corresponding S parameters;

[0075] The structural parameter relationship fitting module is specifically configured to use the two-dimensional matrix and S parameters as the input and output of a convolutional neural network, respectively, and train and verify the convolutional neural network to obtain the relationship fitting model. The convolution layer in the convolutional neural network is sequentially connected to a normalization layer and an activation layer, and the connection layer in the convolutional neural network is sequentially connected to a normalization layer, an activation layer, and a random dropout layer, with a Leaky ReLU function added to the activation layer. The two-dimensional matrix is ​​input into the convolutional neural network to obtain the corresponding predicted S parameters.

[0076] The global optimization module is used to optimize the S parameters based on the function fitting relationship between the two-dimensional matrix and the S parameters through a global optimization algorithm to obtain the optimal passive network structure and realize passive device synthesis.

[0077] The global optimization module is specifically used to calculate the error between the predicted S parameters and the expected S parameters using the mean square error between the predicted S parameters and the expected S parameters as a cost function, screen out the minimum error based on the error and obtain the two-dimensional matrices corresponding to several minimum errors; randomly select several two-dimensional matrices in the parent generation and screen out two two-dimensional matrices corresponding to the minimum errors to cross-generate a child generation, repeating this step until the number of the child generations is equal to the number of the parent generations; update the parent generation based on the two-dimensional matrices corresponding to the several minimum errors and the child generations until a preset number of iterations is reached to obtain the optimal passive network structure.

[0078] The module division in the embodiments of the present invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in various embodiments of the present invention may be integrated into a single processor, exist physically as separate modules, or two or more modules may be integrated into a single module. The integrated modules may be implemented in either hardware or software functional modules.

[0079] This embodiment also provides a computer device, which includes a processor and a memory, wherein the memory is used to store a computer program (in this embodiment, the computer program includes a computing component and an iterative component, and can perform model calculation and model update), the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the passive device synthesis method based on convolutional neural networks.

[0080] This embodiment also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space that stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the passive device synthesis method based on the convolutional neural network in the above embodiment.

[0081] This embodiment also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the corresponding steps of the passive component synthesis method based on convolutional neural network in the above embodiment.

[0082] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0083] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0084] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A passive device synthesis method based on convolutional neural network, characterized in that: The following steps are involved: Pixelating a plurality of arbitrary passive network structures using a two-dimensional matrix, performing electromagnetic simulation on the plurality of arbitrary passive network structures after pixelation, and obtaining a two-dimensional matrix of passive network structure parameters and corresponding electromagnetic simulation S parameters; Expanding the data set of the two-dimensional matrix of the passive network structure parameters and the corresponding S parameters of the electromagnetic simulation to obtain the expanded two-dimensional matrix of the passive network structure parameters and the corresponding S parameters of the electromagnetic simulation; Based on the expanded two-dimensional matrix of the passive network structure parameters and the corresponding electromagnetic simulation S parameters, a function fitting relationship between the two-dimensional matrix of the passive network structure parameters and the S parameters is constructed using a relationship fitting model; Based on the functional fitting relationship between the two-dimensional matrix of the passive network structure parameters and the S parameters, the S parameters are optimized by a global optimization algorithm to obtain the optimal passive network structure and realize the passive device synthesis; The step of expanding the data set of the two-dimensional matrix of the passive network structure parameters and the corresponding S parameters specifically includes: using the two-dimensional matrix of the passive network structure parameters and the S parameters as the input and output of the generative adversarial network, respectively, to obtain the expanded two-dimensional matrix of the passive network structure parameters and the corresponding S parameters; The process of establishing the relationship fitting model is as follows: using the two-dimensional matrix of the passive network structural parameters and the S parameters as the input and output of the convolutional neural network respectively, and training and verifying the convolutional neural network to obtain the relationship fitting model; The optimizing the S parameters by a global optimization algorithm to obtain an optimal passive network structure is specifically: optimizing the S parameters by a genetic algorithm to obtain an optimal passive network structure; The step of optimizing the S parameters using a genetic algorithm to obtain an optimal passive network structure specifically includes: Randomly generate 4096 two-dimensional matrices of random passive network structure parameters corresponding to any passive network structure to form a parent generation, and based on the function fitting relationship between the two-dimensional matrix of the passive network structure parameters and the S parameter, use the relationship fitting model to predict the S parameter corresponding to the two-dimensional matrix of the random passive network structure parameters, the S parameter is called the predicted S parameter; Calculating the error between the predicted S parameters and the expected S parameters using the mean square error between the predicted S parameters and the expected S parameters as a cost function, screening out the minimum error from the errors, and obtaining a two-dimensional matrix of passive network structure parameters corresponding to 16 minimum errors; Randomly select 512 two-dimensional matrices of passive network structure parameters in the parent generation and select two two-dimensional matrices corresponding to the minimum error to cross and generate a child generation, repeating this step until the number of the child generations is equal to the number of the parent generations; Update the parent generation according to the two-dimensional matrix of the passive network structure parameters corresponding to the 16 minimum errors and the child generation, repeat the above steps until a preset number of updates is reached, and obtain the optimal passive network structure; The convolution layer in the convolutional neural network is sequentially connected to a normalization layer and an activation layer, and the fully connected layer in the convolutional neural network is sequentially connected to a normalization layer, an activation layer, and a random dropout layer, and a Leaky ReLU function is added to the activation layer.

2. A passive component integrated system based on convolutional neural network, characterized in that: include: A passive network pixelation module is used to pixelate multiple arbitrary passive network structures using a two-dimensional matrix, perform electromagnetic simulation on the multiple arbitrary passive network structures after pixelation, and obtain a two-dimensional matrix of passive network structure parameters and corresponding electromagnetic simulation S parameters; A data set expansion module is used to expand the data set of the two-dimensional matrix of the passive network structure parameters and the corresponding S parameters of the electromagnetic simulation to obtain the expanded two-dimensional matrix of the passive network structure parameters and the corresponding S parameters of the electromagnetic simulation; A structural parameter relationship fitting module is used to construct a function fitting relationship between the two-dimensional matrix of the passive network structural parameters and the S parameters using a relationship fitting model based on the two-dimensional matrix of the expanded passive network structural parameters and the corresponding electromagnetic simulation S parameters; A global optimization module is used to optimize the S parameters based on the function fitting relationship between the two-dimensional matrix of the passive network structure parameters and the S parameters through a global optimization algorithm to obtain the optimal passive network structure and realize passive device synthesis; The step of expanding the data set of the two-dimensional matrix of the passive network structure parameters and the corresponding S parameters specifically includes: using the two-dimensional matrix of the passive network structure parameters and the S parameters as the input and output of the generative adversarial network, respectively, to obtain the expanded two-dimensional matrix of the passive network structure parameters and the corresponding S parameters; The process of establishing the relationship fitting model is as follows: using the two-dimensional matrix of the passive network structural parameters and the S parameters as the input and output of the convolutional neural network respectively, and training and verifying the convolutional neural network to obtain the relationship fitting model; The optimizing the S parameters by a global optimization algorithm to obtain an optimal passive network structure is specifically: optimizing the S parameters by a genetic algorithm to obtain an optimal passive network structure; The step of optimizing the S parameters using a genetic algorithm to obtain an optimal passive network structure specifically includes: Randomly generate 4096 two-dimensional matrices of random passive network structure parameters corresponding to any passive network structure to form a parent generation, and based on the function fitting relationship between the two-dimensional matrix of the passive network structure parameters and the S parameter, use the relationship fitting model to predict the S parameter corresponding to the two-dimensional matrix of the random passive network structure parameters, the S parameter is called the predicted S parameter; Calculating the error between the predicted S parameters and the expected S parameters using the mean square error between the predicted S parameters and the expected S parameters as a cost function, screening out the minimum error from the errors, and obtaining a two-dimensional matrix of passive network structure parameters corresponding to 16 minimum errors; Randomly select 512 two-dimensional matrices of passive network structure parameters in the parent generation and select two two-dimensional matrices corresponding to the minimum error to cross and generate a child generation, repeating this step until the number of the child generations is equal to the number of the parent generations; Update the parent generation according to the two-dimensional matrix of the passive network structure parameters corresponding to the 16 minimum errors and the child generation, repeat the above steps until a preset number of updates is reached, and obtain the optimal passive network structure; The convolution layer in the convolutional neural network is sequentially connected to a normalization layer and an activation layer, and the fully connected layer in the convolutional neural network is sequentially connected to a normalization layer, an activation layer, and a random dropout layer, and a Leaky ReLU function is added to the activation layer.

3. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable in the processor, wherein the processor implements the steps of the passive device synthesis method based on convolutional neural network according to claim 1 when executing the computer program.

4. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the passive device synthesis method based on convolutional neural network according to claim 1 are implemented.

5. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, the steps of the passive component synthesis method based on convolutional neural network according to claim 1 are implemented.

Images