Multilayer composite sheet type acoustic metamaterial inverse design method and system based on neural network

Abstract

The invention provides a multilayer composite sheet type acoustic metamaterial inverse design method and system based on a neural network, a metamaterial design process is accelerated by using a neural network method, and the method comprises the following steps: constructing a metamaterial, obtaining a data set by using a COMSOL and MATLAB joint simulation method, dividing the data set, and carrying out normalization and standardization processing on related data; training a forward prediction network by taking the structural parameters as input and the sound insulation curve as output; training a reverse design network through a forward prediction network of frozen parameters; the function of reducing the time required by metamaterial design is achieved, the design efficiency is remarkably improved, and the design process of the metamaterial is accelerated. According to the method, the neural network algorithm is introduced into the inverse design of the multi-layer composite thin-plate type acoustic metamaterial, rich design experience is not needed, the professional requirement for a designer is lowered, metamaterial parameters meeting the sound insulation performance can be rapidly obtained, and powerful support is provided for customized design of the multi-layer composite thin-plate type acoustic metamaterial.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent application of neural network algorithms, and in particular relates to a method and system for inverse design of multi-layer composite thin plate acoustic metamaterials based on neural networks. Background Art

[0002] At present, metamaterials, as a new type of artificial composite materials with extraordinary physical properties, have attracted widespread attention in the fields of acoustics, optics, and electromagnetism in recent years. However, traditional noise control requires very thick materials to isolate long-wavelength low-frequency noise, which is not possible in some weight- and space-sensitive applications, such as aircraft and vehicles. Thin-plate acoustic metamaterials have the advantages of ultra-lightness and designable sound insulation frequency, and can expand the effective sound insulation bandwidth by stacking. Compared with single-layer metamaterials, multi-layer metamaterials have higher design freedom and significantly improve the designability of sound insulation performance. However, complex micro-acoustic structures and excessive design parameters increase the complexity of the design. Traditional design methods often rely on cumbersome parameter scanning and repeated trials, which are inefficient and require high professional skills of designers.

[0003] Neural network is an important branch of machine learning. It can simulate the connection and interaction between neurons in the human brain, build complex models through a large number of nodes and connections, and has demonstrated strong learning ability and adaptability in many fields. It is widely used in image recognition, speech recognition, natural language processing and other fields. Therefore, it is necessary to study how to introduce neural network algorithms into the reverse design of multi-layer composite thin plate acoustic metamaterials. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a multi-layer composite thin plate type acoustic metamaterial inverse design method and system based on a neural network, so as to reduce the time required for metamaterial design.

[0005] The technical solution adopted by the present invention to solve the above technical problems is: a multi-layer composite thin plate type acoustic metamaterial inverse design method based on neural network, comprising the following steps: S1: Modeling and simulating the multi-layer composite thin plate acoustic metamaterial to obtain a data set, and then dividing the data set into a training set and a test set for preprocessing; S2: Using structural parameters as input and sound insulation curve as output, build and train a forward prediction network; S3: Using the target sound insulation curve as input and the predicted structural parameters as output, a reverse design network is built; the trained forward prediction network is used to assist in training the reverse design network; S4: According to the required sound insulation curve, use the trained reverse design network to reverse design the corresponding multi-layer composite thin plate type acoustic metamaterial structure.

[0006] According to the above scheme, in step S1, the specific steps are: S11: construct single-layer and multi-layer composite thin plate acoustic metamaterial models, and select structural parameters and corresponding parameter ranges; S12: simulating the multi-layer composite thin plate type acoustic metamaterial model through a joint simulation method to obtain a data set including structural parameter data and corresponding sound insulation curve data; S13: Divide the data set into a training set and a test set according to a certain ratio; S14: Preprocess the dataset including normalization and standardization.

[0007] Furthermore, in the step S11, the specific steps are: S111: construct a single-layer thin plate acoustic metamaterial model, including a thin plate, a central large mass block, and four small mass blocks evenly distributed around the central large mass block, and two opposite small mass blocks are exactly the same; S112: constructing a multi-layer composite thin plate type acoustic metamaterial model, including a plurality of single-layer thin plate type acoustic metamaterials, and a frame and a porous material disposed between every two layers of the thin plate type acoustic metamaterials; S113: Selecting structural parameters and setting parameter ranges according to the performance of the metamaterial, including the width of the thin plate, the thickness of the thin plate of each layer, the height of the mass block, the radius of the large mass block, the radius and eccentricity of the first pair of small mass blocks, and the radius and eccentricity of the second pair of small mass blocks.

[0008] Furthermore, in step S12, the specific steps are: S121: Build a multi-layer composite thin plate acoustic metamaterial sound insulation calculation model in COMSOL, set structural parameters, establish a geometric model, set boundary conditions, and divide the network; S122: Calling the COMSOL interface through MATLAB to perform joint simulation, randomly sampling a set of structural parameters within the corresponding parameter range through MATLAB, importing the set of structural parameters into COMSOL to simulate and generate corresponding sound insulation curves, and saving the structural parameters and the sound insulation curves; S123: Repeat steps S121 to S122 to obtain a data set.

[0009] Furthermore, in step S14, the specific steps are: S141: scaling the structural parameter data to the range of 0 to 1 for normalization by using the maximum and minimum values ​​of the data; S142: Scale the sound insulation curve data to the mean by the mean and standard deviation of the data 0, standard deviation The distribution is 1 and standardized.

[0010] According to the above scheme, in step S2, the specific steps are: For the forward prediction network, the number of structural parameters is used as the number of neurons in the input layer, the dimension of the sound insulation curve is used as the number of neurons in the output layer, the number of hidden neuron layers and the number of neurons in each layer are determined by the accuracy on the test set, a ReLU activation function is set after each hidden layer, the loss function is set to the MSE mean square error function, and the training cycle is set according to the actual situation; through training, the loss function of the predicted sound insulation curve and the target sound insulation curve is minimized to obtain the final forward prediction network model.

[0011] Furthermore, in step S3, the specific steps are: S31: For the reverse design network, the dimension of the sound insulation curve is used as the number of neurons in the input layer, the number of structural parameters is used as the number of neurons in the output layer, the number of hidden neuron layers and the number of neurons in each layer are determined by the accuracy on the test set, a ReLU activation function is set after each hidden layer, the loss function is set to the MSE mean square error function, and the training cycle is set according to the actual situation; S32: Freeze the parameters of the trained forward prediction network, and then input the predicted structural parameters output by the reverse design network into the frozen forward prediction network to obtain a predicted sound insulation curve, so that the difference between the predicted sound insulation curve and the target sound insulation curve is minimized.

[0012] Furthermore, in step S4, the specific steps are: The desired sound insulation curve is input into the trained reverse design network; the reverse design network predicts the corresponding structural parameters, thereby reversely designing the corresponding multi-layer composite thin plate type acoustic metamaterial structure.

[0013] A neural network-based inverse design system for multi-layer composite thin-plate acoustic metamaterials. The data processing submodule is used to model and simulate the multi-layer composite thin plate acoustic metamaterial to obtain a data set, and then divide the data set into a training set and a test set for preprocessing; The forward network training submodule is used to build and train the forward prediction network with structural parameters as input and sound insulation curve as output; The reverse network training submodule is used to build a reverse design network with the target sound insulation curve as input and the predicted structural parameters as output; the trained forward prediction network is used to assist in training the reverse design network; The reverse design submodule is used to reverse design the corresponding multi-layer composite thin plate type acoustic metamaterial structure according to the required sound insulation curve using the trained reverse design network.

[0014] Furthermore, when the reverse network training submodule uses the trained forward prediction network to assist in training the reverse design network, the parameters of the trained forward prediction network are frozen, and then the predicted structural parameters output by the reverse design network are input into the frozen forward prediction network to obtain a predicted sound insulation curve, so that the difference between the predicted sound insulation curve and the target sound insulation curve is minimized.

[0015] The beneficial effects of the present invention are: 1. A neural network-based multilayer composite thin plate acoustic metamaterial inverse design method and system of the present invention accelerates the metamaterial design process by using a neural network method, including: constructing a metamaterial, using a COMSOL and MATLAB joint simulation method to obtain a data set, dividing the data set and normalizing and standardizing the relevant data; using structural parameters as input and sound insulation curves as output to train a forward prediction network; and then training a reverse design network through a forward prediction network with frozen parameters; achieving the function of reducing the time required for metamaterial design, significantly improving design efficiency, and accelerating the design process of metamaterials.

[0016] 2. The present invention introduces a neural network algorithm into the inverse design of multi-layer composite thin plate acoustic metamaterials, which does not require extensive design experience and reduces the requirements for the designer's professionalism. The metamaterial parameters that meet the sound insulation performance can be quickly obtained, providing strong support for the customized design of multi-layer composite thin plate acoustic metamaterials.

[0017] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 is a flow chart of an embodiment of the present invention.

[0020] Figure 2 4 is a three-dimensional diagram of a thin plate type acoustic metamaterial according to an embodiment of the present invention.

[0021] Figure 3 4 is a top view of a thin plate type acoustic metamaterial according to an embodiment of the present invention.

[0022] Figure 4 It is a three-dimensional diagram of a multi-layer composite thin plate type acoustic metamaterial according to an embodiment of the present invention.

[0023] Figure 5It is a forward prediction network architecture diagram of an embodiment of the present invention.

[0024] Figure 6 It is a diagram of the reverse design network training architecture of an embodiment of the present invention. DETAILED DESCRIPTION

[0025] 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 accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0026] Example 1 See also Figure 1 , the specific steps of a neural network-based multi-layer composite thin plate acoustic metamaterial inverse design method are as follows: S1: Modeling and simulating the multi-layer composite thin plate acoustic metamaterial to obtain a data set, and then dividing the data set into a training set and a test set for preprocessing; S2: Using structural parameters as input and sound insulation curve as output, build and train a forward prediction network; S3: Using the target sound insulation curve as input and the predicted structural parameters as output, a reverse design network is built; the trained forward prediction network is used to assist in training the reverse design network; S4: According to the required sound insulation curve, use the trained reverse design network to reverse design the corresponding multi-layer composite thin plate type acoustic metamaterial structure.

[0027] Furthermore, in step S1, the specific steps are: S11: construct single-layer and multi-layer composite thin plate acoustic metamaterial models, and select structural parameters and corresponding parameter ranges; S12: simulating the multi-layer composite thin plate type acoustic metamaterial model through a joint simulation method to obtain a data set including structural parameter data and corresponding sound insulation curve data; S13: Divide the data set into a training set and a test set according to a certain ratio; S14: Preprocess the dataset including normalization and standardization.

[0028] Furthermore, in step S11, the specific steps are: S111: construct a single-layer thin plate acoustic metamaterial model, including a thin plate, a central large mass block, and four small mass blocks evenly distributed around the central large mass block, and two opposite small mass blocks are exactly the same; S112: constructing a multi-layer composite thin plate type acoustic metamaterial model, including a plurality of single-layer thin plate type acoustic metamaterials, and a frame and a porous material disposed between every two layers of the thin plate type acoustic metamaterials; S113: Selecting structural parameters and setting parameter ranges according to the performance of the metamaterial, including the width of the thin plate, the thickness of the thin plate of each layer, the height of the mass block, the radius of the large mass block, the radius and eccentricity of the first pair of small mass blocks, and the radius and eccentricity of the second pair of small mass blocks.

[0029] In step S12, the specific steps are: S121: Build a multi-layer composite thin plate acoustic metamaterial sound insulation calculation model in COMSOL, set structural parameters, establish a geometric model, set boundary conditions, and divide the network; S122: Calling the COMSOL interface through MATLAB to perform joint simulation, randomly sampling a set of structural parameters within the corresponding parameter range through MATLAB, importing the set of structural parameters into COMSOL to simulate and generate corresponding sound insulation curves, and saving the structural parameters and the sound insulation curves; S123: Repeat steps S121 to S122 to obtain a data set.

[0030] In step S14, the specific steps are: S141: scaling the structural parameter data to the range of 0 to 1 for normalization by using the maximum and minimum values ​​of the data; S142: Scale the sound insulation curve data to the mean by the mean and standard deviation of the data 0, standard deviation The distribution is 1 and standardized.

[0031] In step S2, the specific steps are: For the forward prediction network, the number of structural parameters is used as the number of neurons in the input layer, the dimension of the sound insulation curve is used as the number of neurons in the output layer, the number of hidden neuron layers and the number of neurons in each layer are determined by the accuracy on the test set, a ReLU activation function is set after each hidden layer, the loss function is set to the MSE mean square error function, and the training cycle is set according to the actual situation; through training, the loss function of the predicted sound insulation curve and the target sound insulation curve is minimized to obtain the final forward prediction network model.

[0032] In step S3, the specific steps are: S31: For the reverse design network, the dimension of the sound insulation curve is used as the number of neurons in the input layer, the number of structural parameters is used as the number of neurons in the output layer, the number of hidden neuron layers and the number of neurons in each layer are determined by the accuracy on the test set, a ReLU activation function is set after each hidden layer, the loss function is set to the MSE mean square error function, and the training cycle is set according to the actual situation; S32: Freeze the parameters of the trained forward prediction network, and then input the predicted structural parameters output by the reverse design network into the frozen forward prediction network to obtain a predicted sound insulation curve, so that the difference between the predicted sound insulation curve and the target sound insulation curve is minimized.

[0033] In step S4, the specific steps are: The desired sound insulation curve is input into the trained reverse design network; the reverse design network predicts the corresponding structural parameters, thereby reversely designing the corresponding multi-layer composite thin plate type acoustic metamaterial structure.

[0034] This embodiment accelerates the metamaterial design process by using a neural network method, including: constructing a metamaterial, using a COMSOL and MATLAB joint simulation method to obtain a data set, dividing the data set and normalizing and standardizing the relevant data; using structural parameters as input and sound insulation curves as output to train a forward prediction network; and then training a reverse design network through a forward prediction network with frozen parameters; the function of reducing the time required for metamaterial design is achieved, the design efficiency is significantly improved, and the metamaterial design process is accelerated.

[0035] Example 2 The steps of this embodiment are the same as those of Embodiment 1, except that each step is applied to a specific example. Specifically, the following steps are included: S1: Model and simulate the multi-layer composite thin plate acoustic metamaterial to obtain a data set, divide the data set into a training set and a test set, and then pre-process it; the specific steps are: S11: Construct metamaterials and select appropriate structural parameters and corresponding parameter ranges; the specific steps are: S111: construct a thin plate acoustic metamaterial model; Single-layer thin plate acoustic metamaterials such as Figure 2 , 3 As shown, it includes a thin plate, a central large mass block and four small mass blocks around it, where mass block 1 is exactly the same as mass block 3, and mass block 2 is exactly the same as mass block 4. The thin plate material is PET board, and the mass block material is steel; Figure 2 , 3 As can be seen in the figure, the variable structural parameters include the sheet width L, the sheet thickness , mass block height , the radius of the large mass block R, the radius of mass block 1 and mass block 3 and eccentricity , the radius of mass 2 and mass 4 and eccentricity .

[0036] S112: constructing a multi-layer composite thin plate type acoustic metamaterial model; Multilayer composite thin plate type acoustic metamaterials such as Figure 4 As shown, it includes four layers of thin plate type acoustic metamaterials, a frame and a porous material are provided between every two layers of thin plate type acoustic metamaterials, and the frame material is steel; S113: Select structural parameters; including the width L of the thin plate, the thickness of the thin plate , mass block height , the radius of the large mass block R, the radius of mass block 1 and mass block 3 and eccentricity , the radius of mass 2 and mass 4 and eccentricity Note that the outer width of the frame is L, the inner width is L-2, and the height is set to 1.5mm. The width of the porous material is also L, and the height is set to 15mm. No variable structural parameters are set in the frame and the porous material.

[0037] S114: Setting parameter range; the final structural parameters include the width L of the composite material and the thickness of each layer of the thin plate type acoustic metamaterial. , , R, , , and ,Here the corresponding structural parameters in the four layers of metamaterials vary independently, so there are a total of 29 structural parameters, and the variation range of each parameter is selected by the metamaterial performance and experience.

[0038] S12: Simulate the multi-layer composite thin plate acoustic metamaterial by a joint simulation method to obtain a data set; the specific steps are: S121: Build a multi-layer composite thin plate acoustic metamaterial sound insulation calculation model in COMSOL, set structural variable parameters, establish a geometric model, set boundary conditions, divide the network, etc.; Note that except for L, other structural parameters vary independently among the four-layer thin plate acoustic metamaterial; S122: Then, the COMSOL interface is called through MATLAB to perform joint simulation, and a set of structural parameters is randomly sampled within the corresponding parameter range through MATLAB, and the parameters are imported into COMSOL to simulate and generate corresponding sound insulation curves. In this embodiment, the sound insulation curve data is taken from 1 Hz to 1000 Hz, and one is taken every 1 Hz; S123: Then save the structural parameters and sound insulation curve, and repeat the above steps to obtain the final required data set. In order to facilitate subsequent training, the data set is normalized and standardized.

[0039] S13: Divide the structural parameter data and the corresponding sound insulation curve data into a training set and a test set in a ratio of 4:1.

[0040] S14: Preprocess the original data; the specific steps are: S141: The data normalization refers to scaling the data to the range of 0 to 1 through the maximum and minimum values ​​of the data, which can reduce the impact of excessive differences in different data and facilitate subsequent analysis. The selected structural parameters have clear upper and lower limits, so it is suitable to use normalized data processing. The transformation formula is: (1) in is the maximum value of the data, is the minimum value of the data; S142: The data standardization refers to scaling the data to a distribution with a mean of 0 and a standard deviation of 1 through the mean and standard deviation of the data. The sound insulation curve data has no fixed upper and lower limits, so it is suitable for processing using standardization. The formula is: (2) in is the data mean, is the standard deviation of the data; S2: Build and train the forward prediction network; the specific steps are: like Figure 5 As shown in the figure, the structural parameters are used as input, and its dimension is 1 29. The sound insulation curve data is used as output, and its dimension is 1 1000. That is, the number of neurons in the input layer is 29, and the number of neurons in the output layer is 1000. The number of hidden neuron layers and the number of neurons in each layer are determined by the accuracy on the test set. There is a ReLU activation function after each hidden layer. The loss function is set to the MSE mean square error function. The training cycle is set according to the actual situation. Through training, the loss function of the predicted sound insulation curve and the target sound insulation curve is minimized. The model with the smallest loss function is the final model.

[0041] S3: Freeze the trained forward prediction network parameters, build a reverse design network, and train the reverse design network based on the forward prediction network; the specific steps are: S31: Figure 6 As shown, for the reverse design network, the sound insulation curve is used as input and the structural parameters are used as output, that is, the number of input neurons is 1000, the number of output neurons is 29, the number of hidden neuron layers and the number of neurons in each layer are determined by the accuracy on the test set, each hidden layer is followed by a ReLU activation function, the loss function is set to the MSE mean square error function, and the training cycle is set according to the actual situation; S32: Due to the possible non-uniqueness problem, that is, the sound insulation curves corresponding to different structural parameters are relatively similar, a pre-trained forward prediction network is used to assist in training; the specific process is to freeze the parameters of the pre-trained forward prediction network, and then input the sound insulation curve data into the reverse design network to output the predicted structural parameters, and then input the predicted structural parameters into the pre-trained forward prediction network to obtain the predicted sound insulation curve. The loss is the difference between the predicted sound insulation curve and the input sound insulation curve. Since the parameters of the forward neural network are frozen, only the parameters of the reverse design network will be changed during the back propagation process.

[0042] S4: According to the required sound insulation curve, use the trained reverse design network model to reversely design the corresponding multi-layer composite thin plate type acoustic metamaterial structure; the specific steps are: The expected sound insulation curve is input into the trained reverse design network model, and the corresponding structural parameters are obtained through prediction by the reverse design network, so as to reversely design the corresponding multi-layer composite thin plate type acoustic metamaterial structure.

[0043] This embodiment introduces a neural network algorithm into the inverse design of multi-layer composite thin plate type acoustic metamaterials. It does not require extensive design experience, reduces the requirements for the professionalism of the designer, and can quickly obtain metamaterial parameters that meet the sound insulation performance, providing strong support for the customized design of multi-layer composite thin plate type acoustic metamaterials.

[0044] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0045] The environment in which this embodiment runs Example 3 This embodiment is used to implement the principle construction system of the above method embodiment, including a data processing submodule, a forward network training submodule, a reverse network training submodule and a reverse design submodule; The data processing submodule is used to model and simulate the multi-layer composite thin plate acoustic metamaterial to obtain a data set, and then divide the data set into a training set and a test set for preprocessing; The forward network training submodule is used to build and train the forward prediction network with structural parameters as input and sound insulation curve as output; The reverse network training submodule is used to build a reverse design network with the target sound insulation curve as input and the predicted structural parameters as output; the trained forward prediction network is used to assist in training the reverse design network; when the reverse network training submodule uses the trained forward prediction network to assist in training the reverse design network, the parameters of the trained forward prediction network are frozen, and then the predicted structural parameters output by the reverse design network are input into the frozen forward prediction network to obtain a predicted sound insulation curve, so that the difference between the predicted sound insulation curve and the target sound insulation curve is minimized.

[0046] The reverse design submodule is used to reverse design the corresponding multi-layer composite thin plate type acoustic metamaterial structure according to the required sound insulation curve using the trained reverse design network.

[0047] Each sub-module is mainly used to implement each step of the method embodiment, which will not be described in detail here.

[0048] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0049] The above embodiments are only used to illustrate the design ideas and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design ideas disclosed by the present invention are within the protection scope of the present invention.

Claims

1. A neural network-based inverse design method for multi-layer composite thin plate acoustic metamaterials, characterized in that: The following steps are involved: S1: Modeling and simulating the multi-layer composite thin plate acoustic metamaterial to obtain a data set, and then dividing the data set into a training set and a test set for preprocessing; S2: Using structural parameters as input and sound insulation curve as output, build and train a forward prediction network; S3: Using the target sound insulation curve as input and the predicted structural parameters as output, a reverse design network is built; the trained forward prediction network is used to assist in training the reverse design network; S4: According to the required sound insulation curve, use the trained reverse design network to reverse design the corresponding multi-layer composite thin plate type acoustic metamaterial structure.

2. The neural network-based multi-layer composite thin plate acoustic metamaterial inverse design method according to claim 1, characterized in that: In the step S1, the specific steps are: S11: construct single-layer and multi-layer composite thin plate acoustic metamaterial models, and select structural parameters and corresponding parameter ranges; S12: simulating the multi-layer composite thin plate type acoustic metamaterial model through a joint simulation method to obtain a data set including structural parameter data and corresponding sound insulation curve data; S13: Divide the data set into a training set and a test set according to a certain ratio; S14: Preprocess the dataset including normalization and standardization.

3. The neural network-based multi-layer composite thin plate acoustic metamaterial inverse design method according to claim 2, characterized in that: In the step S11, the specific steps are: S111: construct a single-layer thin plate acoustic metamaterial model, including a thin plate, a central large mass block, and four small mass blocks evenly distributed around the central large mass block, and two opposite small mass blocks are exactly the same; S112: constructing a multi-layer composite thin plate type acoustic metamaterial model, including a plurality of single-layer thin plate type acoustic metamaterials, and a frame and a porous material disposed between every two layers of the thin plate type acoustic metamaterials; S113: Selecting structural parameters and setting parameter ranges according to the performance of the metamaterial, including the width of the thin plate, the thickness of the thin plate of each layer, the height of the mass block, the radius of the large mass block, the radius and eccentricity of the first pair of small mass blocks, and the radius and eccentricity of the second pair of small mass blocks.

4. The neural network-based multi-layer composite thin plate acoustic metamaterial inverse design method according to claim 2, characterized in that: In the step S12, the specific steps are: S121: Build a multi-layer composite thin plate acoustic metamaterial sound insulation calculation model in COMSOL, set structural parameters, establish a geometric model, set boundary conditions, and divide the network; S122: Calling the COMSOL interface through MATLAB to perform joint simulation, randomly sampling a set of structural parameters within the corresponding parameter range through MATLAB, importing the set of structural parameters into COMSOL to simulate and generate corresponding sound insulation curves, and saving the structural parameters and the sound insulation curves; S123: Repeat steps S121 to S122 to obtain a data set.

5. The neural network-based multi-layer composite thin plate acoustic metamaterial inverse design method according to claim 2, characterized in that: In the step S14, the specific steps are: S141: scaling the structural parameter data to the range of 0 to 1 for normalization by using the maximum and minimum values ​​of the data; S142: Scale the sound insulation curve data to the mean by the mean and standard deviation of the data 0, standard deviation The distribution is 1 and standardized.

6. The neural network-based multi-layer composite thin plate acoustic metamaterial inverse design method according to claim 1, characterized in that: In the step S2, the specific steps are: For the forward prediction network, the number of structural parameters is used as the number of neurons in the input layer, the dimension of the sound insulation curve is used as the number of neurons in the output layer, the number of hidden neuron layers and the number of neurons in each layer are determined by the accuracy on the test set, a ReLU activation function is set after each hidden layer, the loss function is set to the MSE mean square error function, and the training cycle is set according to the actual situation; through training, the loss function of the predicted sound insulation curve and the target sound insulation curve is minimized to obtain the final forward prediction network model.

7. The neural network-based multi-layer composite thin plate acoustic metamaterial inverse design method according to claim 6, characterized in that: In the step S3, the specific steps are: S31: For the reverse design network, the dimension of the sound insulation curve is used as the number of neurons in the input layer, the number of structural parameters is used as the number of neurons in the output layer, the number of hidden neuron layers and the number of neurons in each layer are determined by the accuracy on the test set, a ReLU activation function is set after each hidden layer, the loss function is set to the MSE mean square error function, and the training cycle is set according to the actual situation; S32: Freeze the parameters of the trained forward prediction network, and then input the predicted structural parameters output by the reverse design network into the frozen forward prediction network to obtain a predicted sound insulation curve, so that the difference between the predicted sound insulation curve and the target sound insulation curve is minimized.

8. The neural network-based multi-layer composite thin plate acoustic metamaterial inverse design method according to claim 7, characterized in that: In the step S4, the specific steps are: The desired sound insulation curve is input into the trained reverse design network; the reverse design network predicts the corresponding structural parameters, thereby reversely designing the corresponding multi-layer composite thin plate type acoustic metamaterial structure.

9. A multi-layer composite thin plate acoustic metamaterial inverse design system based on neural network, characterized by: The data processing submodule is used to model and simulate the multi-layer composite thin plate type acoustic metamaterial to obtain a data set, and then divide the data set into a training set and a test set for preprocessing; The forward network training submodule is used to build and train the forward prediction network with structural parameters as input and sound insulation curve as output; The reverse network training submodule is used to build a reverse design network with the target sound insulation curve as input and the predicted structural parameters as output; the trained forward prediction network is used to assist in training the reverse design network; The reverse design submodule is used to reverse design the corresponding multi-layer composite thin plate type acoustic metamaterial structure according to the required sound insulation curve using the trained reverse design network.

10. The neural network-based multi-layer composite thin plate acoustic metamaterial inverse design system according to claim 9, characterized in that: When the reverse network training submodule uses the trained forward prediction network to assist in training the reverse design network, the parameters of the trained forward prediction network are frozen, and then the predicted structural parameters output by the reverse design network are input into the frozen forward prediction network to obtain a predicted sound insulation curve, so that the difference between the predicted sound insulation curve and the target sound insulation curve is minimized.

Images