Electromagnetic topology simulation method and device based on convolutional neural network

Abstract

The invention provides an electromagnetic topology simulation method and device based on a convolutional neural network, and the method comprises the steps: obtaining a to-be-simulated electromagnetic topological structure, determining electromagnetic topological data information based on the to-be-simulated electromagnetic topological structure, carrying out the data preprocessing of the electromagnetic topological data information, obtaining the preprocessed electromagnetic topological data, and carrying out the simulation of the electromagnetic topological structure. The data preprocessing comprises the step of eliminating electromagnetic topological data information corresponding to a suspended metal piece in the electromagnetic topological structure to be simulated; inputting the preprocessed electromagnetic topological data into a trained electromagnetic topological simulation model, extracting topological characteristics of an electromagnetic topological structure by the electromagnetic topological simulation model, and determining scattering simulation parameters corresponding to the electromagnetic topological data based on the extracted topological characteristics; the electromagnetic topology simulation model comprises an input layer, a plurality of convolution layers, a ReLU activation layer, a cross-channel parameter pooling layer, a full connection layer and an output layer. According to the method, the electromagnetic simulation speed is increased, and rapid design and optimization of the electromagnetic topological structure are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of electromagnetic (EM) topology design, and in particular to an electromagnetic topology simulation method and device based on a convolutional neural network. Background Art

[0002] In the field of microwave engineering, electromagnetic simulation is a key step in evaluating the performance of microwave components. Although the commonly used electromagnetic simulation methods such as finite element method, finite difference time domain method, moment method, boundary element method, etc. can all realize electromagnetic simulation, traditional electromagnetic simulation methods have the disadvantages of large amount of calculation and long time consumption, especially when optimizing electromagnetic topology, a large number of candidate designs need to be evaluated. Existing electromagnetic simulation methods limit the speed and possibility of design innovation.

[0003] Therefore, how to improve the speed of electromagnetic simulation and thus achieve rapid optimization of electromagnetic topological structures is a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides an electromagnetic topology simulation method and device based on a convolutional neural network to eliminate or improve one or more defects existing in the prior art.

[0005] One aspect of the present invention provides an electromagnetic topology simulation method based on a convolutional neural network, the method comprising: Acquire the electromagnetic topological structure to be simulated, determine electromagnetic topological data information based on the electromagnetic topological structure to be simulated, and perform data preprocessing on the electromagnetic topological data information to obtain preprocessed electromagnetic topological data, wherein the data preprocessing includes eliminating the electromagnetic topological data information corresponding to the suspended metal parts in the electromagnetic topological structure to be simulated; The preprocessed electromagnetic topology data is input into a trained electromagnetic topology simulation model, and the electromagnetic topology simulation model extracts the topological features of the electromagnetic topological structure based on the received preprocessed electromagnetic topology data, and determines the scattering simulation parameters corresponding to the electromagnetic topology data based on the extracted topological features; the electromagnetic topology simulation model includes an input layer, multiple convolutional layers, a ReLU activation layer, a cross-channel parameter pooling layer, a fully connected layer, and an output layer.

[0006] In some embodiments of the present invention, the method further comprises: Constructing an initial electromagnetic topology simulation model, a sample data set, and a model loss function, wherein the sample data in the sample data set includes preprocessed electromagnetic topology data sample data and scattering simulation parameter sample data; The initial electromagnetic topology simulation model is pre-trained based on the sample data set and the model loss function to obtain a trained electromagnetic topology simulation model.

[0007] In some embodiments of the present invention, the model loss function is: ; Where n represents the total number of sample data. represents the true value of the scattering parameter, represents the predicted value of the scattering parameter.

[0008] In some embodiments of the present invention, the convolution kernel of the convolution layer is 3X3, and the pooling kernel of the cross-channel parameter pooling layer is 1X1; and / or, The method further comprises: Based on a breadth-first search algorithm, it is verified whether all metal units in the electromagnetic topological structure are connected to the outer wall of the waveguide structure.

[0009] In some embodiments of the present invention, pre-training the initial electromagnetic topology simulation model based on the sample data set and the model loss function to obtain a trained electromagnetic topology simulation model includes: Initializing the network weights of the initial electromagnetic topology simulation model based on the Kaiming initialization method; and / or, The model parameters of the initial electromagnetic topology simulation model are optimized based on the Adam optimization algorithm.

[0010] In some embodiments of the present invention, the data preprocessing includes: The electromagnetic topology data information is normalized or binarized.

[0011] In some embodiments of the present invention, eliminating electromagnetic topological data information corresponding to the suspended metal parts in the electromagnetic topological structure to be simulated includes: Searching for metal cells in edge cells of electromagnetic topological structures based on edge cell priority search algorithm; Verifying connectivity between the metal units based on a breadth-first search algorithm, and determining the suspended metal parts in the electromagnetic topology structure based on the connectivity; The electromagnetic topological data information corresponding to the suspended metal parts in the electromagnetic topological structure is eliminated.

[0012] In some embodiments of the invention, the method comprises: Verifying the accuracy of simulation results of the pre-trained electromagnetic topology simulation model; Optimizing the model structure and model parameters of the electromagnetic topology simulation model based on the accuracy of the simulation results; Re-pretraining the electromagnetic topology simulation model that optimizes the model structure and model parameters; Among them, the model structure includes the number of convolutional layers and the number of cross-channel parameter pooling layers, and the model parameters include the convolution kernel size, the pooling kernel size, the number of channels and the fully connected layer dimension.

[0013] According to another aspect of the present invention, an electromagnetic topology simulation system based on a convolutional neural network is also disclosed. The system includes a processor, a memory, and a computer program stored in the memory. The processor is used to execute the computer program. When the computer program is executed, the system implements the steps of the method described in any of the above embodiments.

[0014] According to yet another aspect of the present invention, a computer-readable storage medium is disclosed, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in any of the above embodiments are implemented.

[0015] The electromagnetic topology simulation method and device based on convolutional neural network disclosed in the above-mentioned embodiment of the present invention first eliminates the electromagnetic topology data information corresponding to the suspended metal parts in the electromagnetic topology structure to be simulated, and then uses the electromagnetic topology data after data preprocessing to quickly and accurately obtain the scattering simulation parameters corresponding to the electromagnetic topology data through a trained electromagnetic topology simulation model. This method can quickly and accurately simulate the electromagnetic topology structure, improve the speed of electromagnetic simulation, and thus realize the efficient design and rapid optimization of the electromagnetic topology structure.

[0016] In addition, the electromagnetic topology simulation model of the present application utilizes multiple small convolution filters and introduces a cross-channel parameter pooling layer to enhance the network depth, effectively improve the model's generalization ability for microwave topology structure data, and reduce the number of model parameters and weights; thus, the method can accurately capture the key features of the microwave topology structure and achieve more accurate prediction of electromagnetic characteristics. In addition, the method reduces the demand for computing resources while maintaining high accuracy; therefore, the method can also run efficiently on platforms with limited computing power, thereby expanding the scope of application.

[0017] Additional advantages, purposes, and features of the present invention will be described in part in the following description, and will become apparent to those skilled in the art after studying the following, or may be learned from the practice of the present invention. The purposes and other advantages of the present invention may be achieved and obtained by the structures specifically indicated in the specification and the accompanying drawings.

[0018] Those skilled in the art will appreciate that the objectives and advantages that can be achieved with the present invention are not limited to the above specific description, and the above and other objectives that can be achieved by the present invention will be more clearly understood from the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of this application, and do not constitute a limitation of the present invention. The components in the drawings are not drawn to scale, but are only for illustrating the principles of the present invention. In order to facilitate the illustration and description of some parts of the present invention, the corresponding parts in the drawings may be enlarged, that is, they may become larger relative to other components in the exemplary device actually manufactured according to the present invention. In the drawings: Figure 1 A schematic diagram of the flow of an electromagnetic topology simulation method based on a convolutional neural network according to an embodiment of the present application.

[0020] Figure 2 Schematic diagram of the architecture of an electromagnetic topology simulation model according to an embodiment of the present application.

[0021] Figure 3 A schematic diagram of a convolution kernel of a convolution layer of an electromagnetic topology simulation model according to an embodiment of the present application.

[0022] Figure 4 This is a schematic diagram of electromagnetic topological data information corresponding to a suspended metal piece of an electromagnetic topological structure according to an embodiment of the present application.

[0023] Figure 5 A schematic diagram of the flow of an electromagnetic topology simulation method based on a convolutional neural network according to another embodiment of the present application.

[0024] Figure 6a Schematic diagram of the electromagnetic topology of a first-order microwave filter.

[0025] Figure 6b Schematic diagram of the electromagnetic topology of the third-order microwave filter.

[0026] Figure 7a This is a schematic diagram of the simulation results corresponding to the first electromagnetic topology structure of the first-order microwave filter.

[0027] Figure 7b Schematic diagram of simulation results corresponding to the second electromagnetic topology structure of the first-order microwave filter.

[0028] Figure 7c Schematic diagram of simulation results corresponding to the third electromagnetic topology structure of the first-order microwave filter.

[0029] Figure 7d Schematic diagram of simulation results corresponding to the fourth electromagnetic topology structure of the first-order microwave filter.

[0030] Figure 8a This is a schematic diagram of the simulation results corresponding to the first electromagnetic topology structure of the third-order microwave filter.

[0031] Figure 8b Schematic diagram of simulation results corresponding to the second electromagnetic topology structure of the third-order microwave filter.

[0032] Figure 8c Schematic diagram of simulation results corresponding to the third electromagnetic topology structure of the third-order microwave filter.

[0033] Figure 8d Schematic diagram of simulation results corresponding to the fourth electromagnetic topology structure of the third-order microwave filter.

[0034] Figure 8e This is a schematic diagram of the simulation results corresponding to the fifth electromagnetic topology structure of the third-order microwave filter.

[0035] Figure 8f This is a schematic diagram of the simulation results corresponding to the sixth electromagnetic topology structure of the third-order microwave filter.

[0036] Figure 8g Schematic diagram of simulation results corresponding to the seventh electromagnetic topology structure of the third-order microwave filter.

[0037] Figure 8h This is a schematic diagram of the simulation results corresponding to the eighth electromagnetic topology structure of the third-order microwave filter. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0039] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only structures and / or processing steps closely related to the solutions according to the present invention are shown in the accompanying drawings, while other details that are not closely related to the present invention are omitted.

[0040] It should be emphasized that the term “include / comprises” when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0041] It should also be noted here that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection with an intermediary, not only to a wired connection, but also to a wireless connection, and the specific connection can be changed based on the actual application scenario.

[0042] The electromagnetic topology simulation method and device based on convolutional neural network of the present application can quickly and accurately simulate the electromagnetic characteristics of the electromagnetic topology structure, so as to further evaluate and optimize the designed electromagnetic topology structure according to the simulation results. The electromagnetic topology simulation method and device of the present application can be applied to the electromagnetic topology structure design and optimization process, and can also be used in electromagnetic simulation scenarios.

[0043] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the accompanying drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0044] Figure 1 FIG. 1 is a flow chart of an electromagnetic topology simulation method based on a convolutional neural network according to an embodiment of the present application. Figure 1 As shown, the electromagnetic topology simulation method at least includes steps S10 and S20.

[0045] Step S10: Acquire the electromagnetic topological structure to be simulated, determine electromagnetic topological data information based on the electromagnetic topological structure to be simulated, and perform data preprocessing on the electromagnetic topological data information to obtain preprocessed electromagnetic topological data, wherein the data preprocessing includes eliminating the electromagnetic topological data information corresponding to the suspended metal parts in the electromagnetic topological structure to be simulated.

[0046] In this step, the electromagnetic topological structure to be simulated is first converted into electromagnetic topological data information, and then the electromagnetic topological data information is preprocessed to obtain preprocessed electromagnetic topological data. In the data preprocessing process, the electromagnetic topological data information corresponding to the suspended metal parts in the electromagnetic topological structure to be simulated is further eliminated. Figure 4 As shown, Figure 4 The content in the dotted box pointed by the middle arrow represents the electromagnetic topological data information corresponding to the suspended metal part; Figure 4 The schematic diagram shown shows whether the topology is connected or not, i.e. Figure 4 The connection between the metal unit and the outer wall of the waveguide is displayed in a visual way. Figure 6a and Figure 6b Schematic diagrams of the electromagnetic topological structures of first-order and third-order microwave filters are shown respectively, visualizing the key structural features and topological dimensions of different filters.

[0047] In the above steps, the electromagnetic topological data information corresponding to the suspended metal parts in the electromagnetic topological structure to be simulated is eliminated in order to improve the connectivity of the electromagnetic topological structure, that is, to exclude the non-physically meaningful structures in the electromagnetic topological structure, thereby improving the rationality and accuracy of the scattering simulation parameters of the electromagnetic topological data, and then better guiding the design and optimization of the electromagnetic topological structure based on the simulation results, reducing the number of optimization iterations in the manufacturing process, reducing costs, and maintaining or improving product performance.

[0048] Furthermore, the electromagnetic topology data information corresponding to the suspended metal parts in the electromagnetic topology structure to be simulated is eliminated, which specifically includes the following steps: searching for metal units in the edge units of the electromagnetic topology structure based on the edge unit priority search algorithm; verifying the connectivity between the metal units based on the breadth-first search algorithm, and determining the suspended metal parts in the electromagnetic topology structure based on the connectivity; and eliminating the electromagnetic topology data information corresponding to the suspended metal parts in the electromagnetic topology structure. This embodiment preprocesses the electromagnetic topology data information based on the breadth-first search (BFS) algorithm to avoid the appearance of suspended metal parts with non-physical meanings in the electromagnetic topology structure; and based on this method, the training data set of the electromagnetic topology simulation model is preprocessed, and the training data can also be reduced during the training process of the electromagnetic topology simulation model, thereby speeding up the design process of the electromagnetic topology structure.

[0049] The BFS algorithm is used in graph theory and computer science to traverse or search tree and graph structures. It starts from a specified source node and systematically explores the nodes in the graph in a width direction, first visiting all neighbor nodes, and then going deeper layer by layer to ensure the shortest path from the source node to all reachable nodes. In this application, the BFS algorithm is reconceived and applied to EM topology design. The optimized BFS algorithm is used to verify whether all metal units in the EM topology design are connected to the outer wall of the waveguide structure. Specifically, first search for metal units in the edge units of the design space, and put all edge metal units into the queue; by marking the visited units, the occurrence of loops and infinite loops is avoided; further, the order of node exploration is managed through the queue data structure to ensure that each node is visited exactly once. In the specific implementation process, starting from the edge unit of each rectangular plate, search for metal units in all edge units and add them to the queue; then, process the metal units in the queue one by one, for each metal unit, find the metal unit directly connected to it, and add the unvisited metal unit to the end of the queue; this process is gradually expanded until all metal units directly or indirectly connected to the first metal unit are found. Finally, check the unvisited units in the entire rectangular plate. If there are no metal units in these unvisited units, it means that all metal units are directly or indirectly connected to the edge metal units, that is, it means that the electromagnetic topology structure does not have suspended metal parts, so that the electromagnetic topology structure meets the manufacturability conditions; otherwise, if there are metal units that are not connected to the outer wall, it means that the electromagnetic topology structure has suspended metal parts, so that the electromagnetic topology structure does not meet the manufacturability conditions. It can be understood that the electromagnetic topology simulation method of the present application, in the actual electromagnetic simulation process and the training process of the electromagnetic topology simulation model, performs data preprocessing to eliminate the electromagnetic topology data information corresponding to the suspended metal parts, so that the data received by the electromagnetic topology simulation model is the data corresponding to the manufacturable electromagnetic topology structure, and the suspended metal parts that have no practical significance are ignored during the electromagnetic simulation process, thereby improving the training efficiency and prediction accuracy of the model.

[0050] Exemplarily, when the electromagnetic topology data information corresponding to the suspended metal parts in the electromagnetic topology structure is removed, the specific process includes the following: 1. Edge cell priority search: Metal cells are first searched among the edge cells in the design space. This is based on the recognition of the importance of edge cells on connectivity in microwave component design.

[0051] 2. Access mark optimization: By introducing the access mark mechanism, the explored nodes can be managed efficiently, avoiding the repeated access and infinite loop problems in traditional BFS.

[0052] 3. Connectivity verification logic: The connectivity of the metal unit is ensured to meet the manufacturability requirements through the connectivity verification logic based on the BFS algorithm. This logic is specially designed according to the physical characteristics and manufacturing process of microwave components.

[0053] 4. Queue management mechanism: The queue management mechanism is adopted to make the search and verification process of metal units more efficient.

[0054] Step S20: input the preprocessed electromagnetic topology data into a trained electromagnetic topology simulation model, the electromagnetic topology simulation model extracts the topological features of the electromagnetic topological structure based on the received preprocessed electromagnetic topology data, and determines the scattering simulation parameters corresponding to the electromagnetic topology data based on the extracted topological features; the electromagnetic topology simulation model includes an input layer, multiple convolutional layers, a ReLU activation layer, a cross-channel parameter pooling layer, a fully connected layer and an output layer.

[0055] In this step, the preprocessed electromagnetic topology data is used as input data of the electromagnetic topology simulation model, so that the electromagnetic topology simulation model outputs scattering simulation parameter prediction results. Figure 2 The connection method of the input layer, multiple convolutional layers, ReLU activation layer, cross-channel parameter pooling layer, fully connected layer and output layer of the electromagnetic topology simulation model is shown, and the depth and complexity of the model are clearly expressed.

[0056] The electromagnetic topology simulation model in the present application is suitable for processing two-dimensional or three-dimensional data of microwave topological structures. The electromagnetic topology simulation model includes a cross-channel parameter pooling layer, which allows the network to increase depth and improve performance without losing key spatial information. Since the accurate representation of the spatial relationship of the electromagnetic topological structure is crucial to the accuracy of the prediction, the electromagnetic topology simulation model of the present application introduces a cross-channel parameter pooling layer design, which is suitable for processing complex electromagnetic problems.

[0057] Furthermore, the electromagnetic topology simulation model of the present application may use multiple small convolution filters to capture subtle features in the electromagnetic topology structure; illustratively, the convolution kernel of the convolution layer is 3X3 (refer to Figure 3 ), the pooling kernel of the cross-channel parameter pooling layer is 1X1. In this embodiment, a small convolution filter with a convolution kernel of 3X3 is used to improve the electromagnetic topology simulation model's ability to capture the details of the microwave topology structure, thereby improving the model's expression ability and computational efficiency, and further improving the accuracy of the scattering simulation parameters predicted by the electromagnetic topology simulation method.

[0058] In the electromagnetic topology simulation model of the above embodiment, multiple small convolution filters are used to enhance the feature extraction capability of the model; a cross-channel parameter pooling layer is introduced to increase the network depth, improve the model performance and generalization capability, reduce the number of model parameters and weights, and optimize the computational efficiency of the model. In addition, in the above electromagnetic topology simulation model, a ReLU activation function layer is connected after the convolution layer to enhance the nonlinear expression capability of the network; a cross-channel parameter pooling layer with a pooling kernel of 1X1 is used to maintain spatial resolution and capture interactions between features.

[0059] In some embodiments of the present application, the electromagnetic topology simulation method based on the convolutional neural network also includes the following steps: constructing an initial electromagnetic topology simulation model, a sample data set, and a model loss function, wherein the sample data in the sample data set includes preprocessed electromagnetic topology data sample data and scattering simulation parameter sample data; pre-training the initial electromagnetic topology simulation model based on the sample data set and the model loss function to obtain a trained electromagnetic topology simulation model. This embodiment is used to realize the construction of an electromagnetic topology simulation model. In this embodiment, an initial electromagnetic topology simulation model is first constructed, and the electromagnetic topology simulation model includes an input layer, multiple convolutional layers, a ReLU activation layer, a cross-channel parameter pooling layer, a fully connected layer, and an output layer; further collecting electromagnetic topological structures and S (scattering) parameters of different microwave components, and performing data preprocessing on the collected electromagnetic topological structures and S (scattering) parameters of different microwave components to form a sample data set; the sample data after data preprocessing meets the input requirements of the initial electromagnetic topology simulation model. Exemplarily, data preprocessing may specifically include removing electromagnetic topological data information corresponding to suspended metal parts, and normalizing or binarizing the electromagnetic topological data information.

[0060] In the process of constructing the sample data set, the electromagnetic topological data information corresponding to the suspended metal parts is eliminated, so that the electromagnetic topological structures corresponding to the sample data are all manufacturable structures. The electromagnetic topological simulation model is trained based on the preprocessed sample data, which can reduce the number of model iterations and avoid the influence of the electromagnetic topological simulation model on the simulation results due to the design defects of the topological structure in the electromagnetic simulation process.

[0061] Furthermore, in the pre-training stage of the initial electromagnetic topology simulation model, the network weights of the initial electromagnetic topology simulation model can be initialized based on the Kaiming initialization method; and / or, the model parameters of the initial electromagnetic topology simulation model can be optimized based on the Adam optimization algorithm. This embodiment performs weight initialization of the model based on the Kaiming initialization technology, avoids a series of problems such as gradient disappearance or explosion during the model training process, and can also ensure that the model can converge quickly and make the simulation results have high accuracy.

[0062] In one embodiment, the electromagnetic topology simulation model uses the root mean square error as the loss function, and the model loss function can be specifically: ; Where n represents the total number of sample data. represents the true value of the scattering parameter, represents the predicted value of the scattering parameter.

[0063] Furthermore, the electromagnetic topology simulation model can use the relative root mean square error as an evaluation function, and the model evaluation function can be specifically: ; Where n represents the total number of sample data. represents the true value of the scattering parameter, represents the predicted value of the scattering parameter, represents the maximum predicted value of the scattering parameter, Represents the minimum predicted value of the scattering parameter.

[0064] Through the above embodiments, it can be found that the present application pre-trains the initial electromagnetic topology simulation model based on the preprocessed sample data set. During the training process, Kaiming initialization technology is used to avoid the gradient disappearance or explosion problem, and appropriate loss functions and optimization algorithms are used to improve the performance of the model, so that the prediction accuracy, calculation speed and generalization ability of the trained electromagnetic topology simulation model are better.

[0065] In addition, in the pre-training process of the model, in order to ensure the performance of the model, in some embodiments, the electromagnetic topology simulation method also includes the following steps: optimizing the model structure and model parameters of the electromagnetic topology simulation model based on the accuracy of the simulation results; re-pre-training the electromagnetic topology simulation model with the optimized model structure and model parameters; wherein the model structure includes the number of convolutional layers and the number of cross-channel parameter pooling layers, and the model parameters include the convolution kernel size, the pooling kernel size, the number of channels, and the dimension of the fully connected layer. In this embodiment, the simulation results of the trained electromagnetic topology simulation model are verified to evaluate the training effect. If the simulation results of the trained electromagnetic topology simulation model do not meet the use requirements, the model is further adjusted and trained until an electromagnetic topology simulation model that meets the use requirements is obtained.

[0066] In addition, the accuracy and reliability of the simulation results of the electromagnetic topology simulation method of the present application can be further verified through experimental results, that is, the performance of the electromagnetic topology simulation method in different microwave topology designs can be verified, including accuracy, calculation speed and generalization ability.

[0067] When verifying the accuracy, the simulation results obtained based on the electromagnetic topology simulation method of the present application can be compared with the simulation results obtained by the existing electromagnetic simulation method. The test includes predicting the S parameters of microwave components with different configurations, and evaluating the accuracy of the electromagnetic topology simulation method of the present application by comparing with high-fidelity electromagnetic simulation results. The test results show that the electromagnetic topology simulation method of the present application can predict electromagnetic characteristics with high accuracy, and has significant performance improvement compared with the electromagnetic simulation method of the prior art.

[0068] When verifying the calculation speed, the running time can be tested under different hardware configurations to evaluate the response speed of the electromagnetic topology simulation method in practical applications; the results show that even in an environment with limited computing resources, the electromagnetic topology simulation method of this application can quickly complete the prediction of electromagnetic characteristics, greatly shortening the design evaluation time.

[0069] When verifying the generalization ability of the electromagnetic topology simulation model in the electromagnetic topology simulation method, a batch of data that did not participate in the model training was used for testing; the experimental results proved that the model not only performed well on the training data, but also could provide accurate predictions on unseen data, showing good generalization ability. This generalization ability can enable the electromagnetic topology simulation method of the present application to adapt to changing design requirements and application scenarios.

[0070] In addition, when verifying the electromagnetic topology simulation method of the present application, two different filters can be selected to verify the practicality of the method in the design of microwave topology structures; that is, the prediction results obtained by the electromagnetic topology simulation method of the present application are applied to the actual microwave component manufacturing process, and the practicality of the electromagnetic topology simulation method of the present application is proved by testing the impact of the model output on manufacturing accuracy and cost. Figure 7a , 7b 7c and 7d are schematic diagrams of simulation results corresponding to different electromagnetic topological structures of first-order microwave filters, Figure 8a , 8b 8c, 8d, 8e, 8f, 8g and 8h are schematic diagrams of simulation results corresponding to different electromagnetic topological structures of third-order microwave filters respectively; according to the above schematic diagrams of simulation results, it can be seen that the scattering simulation parameters predicted by the electromagnetic topological simulation method of the present application are basically close to the actual scattering parameters, thereby proving the accuracy, reliability, generalization ability and prediction accuracy of the electromagnetic topological simulation method based on convolutional neural network of the present application in different designs. Therefore, the electromagnetic topological simulation method of the present application can be directly applied to the design scenario of the electromagnetic topological structure by simulating the electromagnetic characteristics of the electromagnetic topological structure, providing an efficient and fast electromagnetic topological simulation method for electromagnetic topological design, thereby improving the efficiency and accuracy of microwave component design.

[0071] In some embodiments of the present invention, after obtaining the scattering simulation parameters corresponding to the electromagnetic topological structure, it is further verified based on the breadth-first search algorithm whether all metal units in the electromagnetic topological structure are connected to the outer wall of the waveguide structure. This method avoids the electromagnetic topological structure containing non-physical suspended metal parts, prevents the electromagnetic topological structure from having non-physical structures, ensures the manufacturability of the microwave component, and reduces the cost.

[0072] In addition, when the electromagnetic topology simulation method based on convolutional neural network of the present application is applied in the design process of electromagnetic topology structure, the designer can explore more design options in the same time, which greatly improves the design efficiency, which provides more possibilities for innovative design and helps to develop microwave components with better performance and more compact structure. And the simulation method reduces the dependence on expensive hardware simulation resources and time consumption, which helps to reduce the overall R&D cost, especially in the early stages of product development, this cost saving is particularly significant. In addition, the electromagnetic topology simulation method based on the present application completes the design of electromagnetic topology structure, reduces iteration and rework, helps to reduce the manufacture of physical prototypes, thereby reducing material consumption and environmental impact, and conforms to the concept of green design and sustainable development. In addition to the above, the present application demonstrates the application potential of deep learning technology in the field of electromagnetics, provides new ideas and tools for the integration of technology between different disciplines, and helps to promote interdisciplinary research and innovation; with the development of 5G and future communication technology, the demand for high-performance microwave components is growing, and the electromagnetic topology simulation method of the present application provides an effective design method to meet these needs, laying a solid foundation for the development of future technology. To sum up, the advantages of the electromagnetic topology simulation method of this application are not only reflected in the breakthroughs at the technical level, but also in promoting industry development, reducing costs, improving design quality, and promoting technological innovation.

[0073] Figure 5 FIG. 1 is a flow chart of an electromagnetic topology simulation method based on a convolutional neural network according to another embodiment of the present application. In this embodiment, the construction and optimization process of the electromagnetic topology simulation model is shown. Figure 5As shown, first, the data variables in the topological structure are defined, and a data set is randomly generated. The data variables can also be understood as electromagnetic topological data information; further, the restricted EM topological design method is used to filter the data (preprocess the data), and the data is randomly divided into a training set and a test set; the neural network structure is further designed and adjusted; the model is trained using the training set, and if the training error is less than θ, the model is tested using the test set, and if the test error is less than ε, the process stops; if the training error is not less than θ, the neural network structure is redesigned and adjusted; and if the test error is not less than ε, it is determined whether the test error is decreasing, and if it is decreasing, the neural network structure is redesigned and adjusted, and if the test error is not decreasing, the data is re-added, and the restricted EM topological design method is continued to be used to filter the data, and the data is randomly divided into a training set and a test set.

[0074] Correspondingly, the present invention also provides an electromagnetic topology simulation system based on a convolutional neural network, the system comprising a processor, a memory and a computer program stored in the memory, the processor is used to execute the computer program, and when the computer program is executed, the system implements the steps of the method described in any of the above embodiments.

[0075] Exemplarily, the electromagnetic topology simulation system based on convolutional neural network may specifically include a data generation module, a model training module and a performance verification module, wherein the data generation module is used to generate EM topology data that meets the design constraints, the model training module is used to train the electromagnetic topology simulation model and optimize the model parameters, and the performance verification module is used to verify the accuracy and reliability of the model through experimental results. Furthermore, the data generation module may also use the finite element method (FEM) to perform EM simulation to generate metal and non-metal topology data in the design space.

[0076] The embodiments of the present invention also provide a computer-readable storage medium and a computer program product, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described in any of the above embodiments are implemented. The computer-readable storage medium can be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the technical field.

[0077] It should be understood by those skilled in the art that the exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0078] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0079] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with features of other embodiments or replace features of other embodiments.

[0080] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An electromagnetic topology simulation method based on convolutional neural network, characterized in that: The method comprises: Acquire the electromagnetic topological structure to be simulated, determine electromagnetic topological data information based on the electromagnetic topological structure to be simulated, and perform data preprocessing on the electromagnetic topological data information to obtain preprocessed electromagnetic topological data, wherein the data preprocessing includes eliminating the electromagnetic topological data information corresponding to the suspended metal parts in the electromagnetic topological structure to be simulated; The preprocessed electromagnetic topology data is input into a trained electromagnetic topology simulation model, and the electromagnetic topology simulation model extracts the topological features of the electromagnetic topological structure based on the received preprocessed electromagnetic topology data, and determines the scattering simulation parameters corresponding to the electromagnetic topology data based on the extracted topological features; the electromagnetic topology simulation model includes an input layer, multiple convolutional layers, a ReLU activation layer, a cross-channel parameter pooling layer, a fully connected layer, and an output layer.

2. The electromagnetic topology simulation method based on convolutional neural network according to claim 1, characterized in that: The method further comprises: Constructing an initial electromagnetic topology simulation model, a sample data set, and a model loss function, wherein the sample data in the sample data set includes preprocessed electromagnetic topology data sample data and scattering simulation parameter sample data; The initial electromagnetic topology simulation model is pre-trained based on the sample data set and the model loss function to obtain a trained electromagnetic topology simulation model.

3. The electromagnetic topology simulation method based on convolutional neural network according to claim 2, characterized in that: The model loss function is: ； Where n represents the total number of sample data. represents the true value of the scattering parameter, represents the predicted value of the scattering parameter.

4. The electromagnetic topology simulation method based on convolutional neural network according to claim 1, characterized in that: The convolution kernel of the convolution layer is 3X3, and the pooling kernel of the cross-channel parameter pooling layer is 1X1; and / or, The method further comprises: Based on a breadth-first search algorithm, it is verified whether all metal units in the electromagnetic topological structure are connected to the outer wall of the waveguide structure.

5. The electromagnetic topology simulation method based on convolutional neural network according to claim 2, characterized in that: Pre-training the initial electromagnetic topology simulation model based on the sample data set and the model loss function to obtain a trained electromagnetic topology simulation model, including: Initializing the network weights of the initial electromagnetic topology simulation model based on the Kaiming initialization method; and / or, The model parameters of the initial electromagnetic topology simulation model are optimized based on the Adam optimization algorithm.

6. The electromagnetic topology simulation method based on convolutional neural network according to claim 1, characterized in that: The data preprocessing includes: The electromagnetic topology data information is normalized or binarized.

7. The electromagnetic topology simulation method based on convolutional neural network according to claim 1, characterized in that: Eliminating electromagnetic topological data information corresponding to the suspended metal parts in the electromagnetic topological structure to be simulated, including: Searching for metal cells in edge cells of electromagnetic topological structures based on edge cell priority search algorithm; Verifying connectivity between the metal units based on a breadth-first search algorithm, and determining the suspended metal parts in the electromagnetic topology structure based on the connectivity; The electromagnetic topological data information corresponding to the suspended metal parts in the electromagnetic topological structure is eliminated.

8. The electromagnetic topology simulation method based on convolutional neural network according to claim 2, characterized in that: The method comprises: Verifying the accuracy of simulation results of the pre-trained electromagnetic topology simulation model; Optimizing the model structure and model parameters of the electromagnetic topology simulation model based on the accuracy of the simulation results; Re-pretraining the electromagnetic topology simulation model that optimizes the model structure and model parameters; Among them, the model structure includes the number of convolutional layers and the number of cross-channel parameter pooling layers, and the model parameters include the convolution kernel size, the pooling kernel size, the number of channels and the fully connected layer dimension.

9. An electromagnetic topology simulation system based on a convolutional neural network, the system comprising a processor, a memory and a computer program stored in the memory, characterized in that: The processor is used to execute the computer program. When the computer program is executed, the system implements the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images