Design method, system, device, storage medium for multi-dimensional resonator type sound absorber

Abstract

The application relates to a design method, system and equipment of a multi-dimensional resonant cavity type sound absorber, and a storage medium, the design method comprising the following steps: establishing a simulation model of the multi-dimensional resonant cavity type sound absorber and calculating sound absorption coefficients thereof, and constructing a machine learning data set; training an automatic encoder model, including an encoder and a decoder submodel; projecting features of the machine learning data set to a low-dimensional space by using the encoder submodel, and establishing a multi-dimensional sound absorption performance research model based on a random forest to generate a low-dimensional space sound absorption coefficient database; searching for low-dimensional structure parameters meeting design task requirements in the low-dimensional space sound absorption coefficient database, projecting the low-dimensional structure parameters back to a high-dimensional space by using the decoder submodel, and outputting structure design parameters and sound absorption coefficient data. Compared with the prior art, the application can design a multi-dimensional resonant cavity type sound absorber structure in a fast and accurate method, and can predict the acoustic performance of the multi-dimensional resonant cavity type sound absorber, so that the calculation cost of CAE processing of a complex structure is greatly reduced.

Description

Technical Field

[0001] This invention relates to the field of sound absorption and noise reduction technology, and in particular to a design method, system, device, and storage medium for a multidimensional resonant cavity sound absorber. Background Technology

[0002] In recent years, various simulation techniques have been widely applied to the acoustic performance prediction and structural design of acoustic metamaterials. Simulation technology can theoretically obtain the structural parameters of acoustic metamaterials with optimal sound absorption performance, reducing unnecessary experimental processes. However, when using simulation technology for acoustic metamaterial design, it is necessary to establish physical models and derive theoretical formulas through continuous trial and error adjustments. This involves tedious calculations to find the locally optimal structure that meets the target in the entire parameter space. While this is the most direct forward design method, its computational cost is exponentially high. Furthermore, calculating design schemes corresponding to different target values ​​using this method consumes significant computational resources and time, making it difficult to obtain the globally optimal solution in the search space.

[0003] Regarding structural design methods for materials, in existing technologies:

[0004] Publication number: CN105427853A, invention patent title: "Wideband Microperforated Plate Sound Absorber and Its Performance Prediction Method and Structural Design Method", proposes a performance prediction method for predicting the sound absorption coefficient spectrum of the sound absorber, and also proposes a structural design method for effectively realizing the wideband sound absorber.

[0005] Publication No. CN111611748A, entitled "A Data-Driven Material Reverse Design Method and System," discloses a data-driven material reverse design method and system. The method includes the following steps: First, acquiring material data from a pre-formed material sample; then, selecting machine learning models corresponding to the relationship between each performance parameter and design parameters based on the material data; using cross-validation to search and select parameters for each machine learning model based on the material data, obtaining modified machine learning models corresponding to the relationship between each performance parameter and design parameters, which serve as the fitness function for each performance parameter in the genetic algorithm; finally, using a genetic algorithm for optimization. This invention employs non-dominated sorting and crowding distance sorting in the genetic algorithm to achieve multi-objective optimization, eliminating the need for penalty factors. This overcomes the technical shortcomings of traditional grid search, where the search space cannot exhaust all possibilities and the introduction of penalty factors leads to inaccurate design results, thus improving the comprehensiveness and accuracy of material reverse design.

[0006] The paper, titled "Inverse Design of Acoustic Metamaterials Based on Machine Learning," constructs an inverse design method for acoustic metamaterials based on a genetic algorithm and applies it to a sound-absorbing structure composed of cascaded Helmholtz resonators, achieving low-frequency broadband sound absorption. It also proposes an inverse design method for acoustic metamaterials based on a Gaussian-Bayesian model.

[0007] With the improvement of computing power, data-driven machine learning algorithms are increasingly being used in sound-absorbing materials and designs. However, due to the large number of parameters and complex structure of multidimensional resonant cavity sound absorbers, existing technical solutions are mostly applied to micro-perforated plates, traditional porous sound-absorbing materials, or cascaded Helmholtz resonant cavities. There is very little research on the sound absorption performance prediction and structural design of multidimensional resonant cavity sound absorbers.

[0008] Therefore, there is an urgent need to propose a method that can simultaneously realize the structural parameter design and sound absorption coefficient curve prediction of multi-dimensional resonant cavity sound absorbers, so as to solve the problems of long prediction and design time and low efficiency in the existing simulation technology. Summary of the Invention

[0009] The purpose of this invention is to overcome the defects of the prior art by providing a design method, system, device, and storage medium for a multidimensional resonant cavity sound absorber. This invention solves the problems in the prior art where multidimensional resonant cavity sound absorbers have many parameters and complex structures, and it is difficult to predict the sound absorption performance and design the structure of multidimensional resonant cavity sound absorbers.

[0010] This invention utilizes a sound absorption performance database established by simulation software, which maps the structural parameters of functional primitive sequences of Helmholtz resonant materials in different dimensions to their corresponding sound absorption coefficient curves. Simultaneously, it employs an autoencoder and random forest to build an intelligent design model for a multidimensional resonant cavity sound absorber. This model can predict the sound absorption curves corresponding to projected low-dimensional features and generate a new low-dimensional spatial sound absorption curve database. Based on task requirements, it searches the two databases for sound absorption curves and their structural parameters that meet the requirements and outputs them.

[0011] The objective of this invention can be achieved through the following technical solutions:

[0012] The first aspect of this invention provides a design method for a multidimensional resonant cavity type sound absorber, comprising the following steps:

[0013] S1. Establish a multi-dimensional resonant cavity sound absorption model based on finite element simulation, calculate the sound absorption coefficient, and construct a machine learning dataset;

[0014] S2. Train an autoencoder model with the sound absorber structural parameters as input and output, including encoder and decoder sub-models;

[0015] S3. Using the encoder sub-model, the features of the machine learning dataset are projected into a low-dimensional space, and a multi-dimensional sound absorption performance research model is established based on random forest to generate a low-dimensional space sound absorption coefficient database.

[0016] S4. Obtain the corresponding structural parameters in the design task, search for low-dimensional structural parameters that meet the design task requirements in the low-dimensional space sound absorption coefficient database, use the decoder sub-model to project the low-dimensional structural parameters back to the high-dimensional space, output the structural design parameters and sound absorption coefficient data, and complete the design task of the multi-dimensional resonant cavity sound absorber.

[0017] Furthermore, in S1, the multi-dimensional resonant cavity sound absorption model is composed of n 2 n-dimensional functional unit sequence structure composed of continuously parallel arrangement of resonant cavity type sound absorption units;

[0018] The structural parameters of the resonant cavity sound-absorbing unit include cavity length (xa), cavity width (xb), cavity height (xh), pore depth (kh), and pore diameter (kr);

[0019] The structural parameters include 5×n 2 There are n feature parameters, where n represents the number of dimensions.

[0020] Furthermore, in S1, the machine learning dataset contains:

[0021] 5×n multidimensional resonant cavity type sound absorber 2 Each structural parameter and its corresponding single-point frequency are used as features of the dataset.

[0022] Furthermore, the single-point absorption coefficient corresponding to each set of structural parameters and single-point frequency is used as the dataset label (y1, y2, y3, y4…y). m The structural parameters, single-point frequencies, and sound absorption coefficients are all in one-to-one correspondence.

[0023] Furthermore, S2 specifically includes the following steps:

[0024] S2-1. Reshuffle the sequences in the machine learning dataset, split them into training and validation sets according to a preset ratio, and preprocess the data.

[0025] S2-2. Establish an autoencoder model with structural parameters as input and output, and train the encoder sub-model and decoder sub-model of the autoencoder model.

[0026] Furthermore, in S2, the autoencoder model includes an encoder with multiple hidden layers, a bottleneck layer, and a decoder with multiple hidden layers;

[0027] In the autoencoder model, structural parameters are used as input data for encoding, realizing the transformation of data from high-dimensional space to low-dimensional space. Then, the encoding vector of the bottleneck layer is reconstructed to complete the decoding, and the reconstructed structural parameters are output. The output of the bottleneck layer is the low-dimensional feature space after dimensionality reduction.

[0028] Furthermore, S3 specifically includes the following steps:

[0029] S3-1, Encode 5×n using the encoder sub-model of the autoencoder model. 2 These feature parameters enable the transformation of structural parameters from high-dimensional space to low-dimensional space.

[0030] S3-2. Establish a multidimensional sound absorption performance research model based on random forest. Input the training set obtained in S2-1 into the multidimensional sound absorption performance research model for training, and use the data in the validation set for testing, validation and model evaluation.

[0031] S3-3, Using the GridSearch method to traverse the hyperparameter combination tuning and deployment of the multidimensional sound absorption performance research model in S3-2;

[0032] S3-4. Input the low-dimensional space obtained in S3-1 as a feature of the random forest into the multi-dimensional sound absorption performance research model obtained in S3-3, and predict the sound absorption coefficient curve in the corresponding frequency band of the low-dimensional feature space.

[0033] S3-5. A low-dimensional feature space database is constructed by predicting new low-dimensional feature spaces and corresponding sound absorption coefficient curves based on the multidimensional sound absorption performance research model based on random forest, and the database is stored in blocks.

[0034] Furthermore, the research content of the multidimensional resonant cavity sound absorber sound absorption performance research model in S3-2 is to use random forest to predict the sound absorption coefficient curves in the frequency bands corresponding to the structural parameters of the multidimensional resonant cavity sound absorber. The output result is a two-dimensional sound absorption curve that connects the sound absorption coefficients corresponding to each frequency band.

[0035] Furthermore, S4 specifically includes the following steps:

[0036] S4-1. Obtain the structural parameter design requirements. Search for the structural parameters and sound absorption coefficient curves that meet the requirements in the machine learning dataset mentioned in S1 and the low-dimensional feature space database in S3-5, respectively.

[0037] S4-2. If a sound absorption coefficient curve that meets the requirements is found in the machine learning dataset, the first five sets of structural parameters that meet the requirements and their corresponding sound absorption coefficient curves are directly output.

[0038] S4-3. If a sound absorption coefficient curve that meets the requirements is found in the low-dimensional feature space database, the low-dimensional features corresponding to the sound absorption coefficient curve that meets the requirements are first reconstructed using the decoder sub-model of the autoencoder, and then projected back to the high-dimensional features through the decoder. Finally, the first five sets of high-dimensional structural parameters and the corresponding frequency band sound absorption curves that meet the requirements are output.

[0039] A second aspect of the present invention provides a design system for a multidimensional resonant cavity sound absorber, comprising a first model building module, a second model building module, a dimensionality reduction processing module, and a dimensionality increase output module, wherein specifically:

[0040] The first model building module establishes a multi-dimensional resonant cavity sound absorption model based on finite element simulation, calculates the sound absorption coefficient, and builds a machine learning dataset.

[0041] The second model building module trains an autoencoder model with the sound absorber structural parameters as input and output, including encoder and decoder sub-models.

[0042] Dimensionality reduction module: Using the encoder sub-model, the features of the machine learning dataset are projected into a low-dimensional space, and a multi-dimensional sound absorption performance research model is established based on random forest to generate a low-dimensional space sound absorption coefficient database.

[0043] The up-dimensional output module obtains the corresponding structural parameters in the design task, searches for low-dimensional structural parameters that meet the design task requirements in the low-dimensional space sound absorption coefficient database, and uses the decoder sub-model to project the low-dimensional structural parameters back into the high-dimensional space, outputting the structural design parameters and sound absorption coefficient data to complete the design task of the multi-dimensional resonant cavity sound absorber.

[0044] A third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the processor is used to execute a program in the memory to implement the above-described design method for a multidimensional resonant cavity sound absorber.

[0045] A fourth aspect of the present invention provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, is used to execute the design method of the multidimensional resonant cavity type sound absorber described above.

[0046] Compared with the prior art, the present invention has the following technical advantages:

[0047] 1) The intelligent design method for multidimensional resonant cavity type sound absorbers constructed by the method of the present invention first applies 5×n 2The high-dimensional feature parameters are reduced in dimensionality and projected into a low-dimensional space. Then, the sound absorption curve is predicted in the low-dimensional space based on the multidimensional resonant cavity sound absorber sound absorption performance research model based on random forest. This prediction method solves the problem that it is difficult to predict the sound absorption performance of acoustic Helmholtz resonant materials through simulation technology due to the large number of parameters. It can quickly obtain accurate predicted values ​​of the sound absorption coefficients of Helmholtz resonant materials in different dimensions, thereby accurately evaluating the sound absorption performance of Helmholtz resonant materials.

[0048] 2) The intelligent design method for multidimensional resonant cavity sound absorbers constructed in this invention can achieve efficient structural parameter inverse design without relying on analytical models, based on the desired sound absorption coefficient requirements at the desired frequency. This intelligent design method does not require establishing a physical model or continuously adjusting the derived theoretical formulas through trial and error, nor does it require tedious calculations to find the locally optimal structure that meets the target in the entire parameter space. This intelligent design method only needs to input the task requirements, and can quickly output the structural parameters that meet the requirements and the predicted corresponding sound absorption coefficient curve, greatly saving computational resources and time costs.

[0049] 3) The multi-dimensional resonant cavity type sound absorber intelligent design method constructed in this invention refers to n 2 The cavities are arranged in parallel to form an n-dimensional (nxn type) functional unit sequence. The intelligent design method of multi-dimensional resonant cavity sound absorber can output the corresponding dimensional structural parameters to meet the sound absorption requirements of functional unit sequences of different dimensions. The practical application of the intelligent design method is extensive. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the design method of the multidimensional resonant cavity sound absorber of the present invention;

[0051] Figure 2 This is a schematic diagram of the two-dimensional resonant cavity type functional unit sequence structure in a specific embodiment of the present invention;

[0052] Figure 3 This is a schematic diagram of the five-dimensional resonant cavity type functional unit sequence structure in a specific embodiment of the present invention;

[0053] Figure 4 This is a flowchart illustrating the design method of a multidimensional resonant cavity sound absorber in a specific embodiment of the present invention;

[0054] Figure 5 This is a structural diagram of a multidimensional resonant cavity sound absorber design method based on an automatic encoder in a specific embodiment of the present invention;

[0055] Figure 6 This is a structural diagram of the prediction model for the performance study of a multidimensional resonant cavity sound absorber based on random forest, as described in a specific embodiment of the present invention.

[0056] Figure 7 This is a schematic diagram of the design system of the multidimensional resonant cavity sound absorber of the present invention;

[0057] Figure 8 This is a schematic diagram of the hardware structure of a computer device according to a specific embodiment of the present invention. Detailed Implementation

[0058] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Component models, material names, connection structures, control methods, algorithms, and other features not explicitly described in this technical solution are considered common technical features disclosed in the prior art.

[0059] Example 1

[0060] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any creative effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to the design, manufacturing, or production methods based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of this application.

[0061] This invention aims to propose a method that can simultaneously realize the structural parameter design and sound absorption coefficient curve prediction of multidimensional resonant cavity sound absorbers. Specifically, it is based on a sound absorption performance database constructed from the structural parameters of functional primitive sequences of Helmholtz resonant materials of different dimensions and their corresponding sound absorption coefficient curves, established by simulation software. While using an autoencoder and random forest to build an intelligent design model of the multidimensional resonant cavity sound absorber, it can predict the sound absorption curves corresponding to the projected low-dimensional features and generate a new low-dimensional spatial sound absorption curve database. According to the task requirements, it searches for the sound absorption curves and their structural parameters that meet the requirements in the two databases and outputs them, thus solving the problems of long prediction and design time and low efficiency in the existing simulation technology.

[0062] In the first aspect, embodiments of this application provide a design method for a multi-dimensional resonant cavity type sound absorber, as shown in the schematic diagram below. Figure 1 As shown, the method includes:

[0063] S1. Finite element simulation (CAE) is used to establish a multi-dimensional resonant cavity sound absorption model and calculate the sound absorption coefficient, and to construct a machine learning dataset;

[0064] S2. Train an autoencoder model with the sound absorber structural parameters as input and output, including encoder and decoder sub-models;

[0065] S3. Using the encoder sub-model, the features of the dataset are projected into a low-dimensional space, and a multi-dimensional sound absorption performance research model is established based on random forest to generate a low-dimensional space sound absorption coefficient database.

[0066] S4. Search for low-dimensional structural parameters that meet the task requirements in the low-dimensional space database, and use the decoder sub-model to project the low-dimensional structural parameters back into the high-dimensional space to output the structural design parameters and sound absorption coefficient data.

[0067] In specific implementation, the machine learning dataset in step S1 includes: 5×n multidimensional resonant cavity sound absorbers. 2 Each structural parameter and its corresponding single-point frequency are used as features of the dataset. The single-point sound absorption coefficient corresponding to each set of structural parameters is used as the dataset label (y1, y2, y3, y4…y). m The structural parameters, single-point frequencies, and sound absorption coefficients are all in one-to-one correspondence.

[0068] In practical implementation, the functional unit sequence is derived by using CAE to build a multi-dimensional resonant cavity sound absorber and adjusting its structural parameters, such as... Figure 2 and Figure 3 As shown:

[0069] Among them, the multidimensional resonant cavity type functional unit sequence is n 2 An array of n rows and n columns of functional primitives is composed of individual functional primitives arranged in parallel. Each individual functional primitive includes an air cavity, a sound-absorbing duct, and a pipe wall. An opening is provided at the center of the top surface of the air cavity. A Boolean operation is performed at the center of the opening to establish a hard sound field boundary. An air column is generated at the center of the opening after the operation to serve as a sound-absorbing duct. The hollow cylinder formed by the difference between the opening and the air column serves as the pipe wall.

[0070] Specifically, the structural parameters of the CAE's adjustable overall functional primitive sequence include: cavity length (xa), cavity width (xb), cavity height (xh), aperture depth (kh), and aperture diameter (kr), totaling 5×n. 2 There are 3 characteristic parameters, where n represents the number of dimensions. By adjusting these structural parameters in CAE, the sound absorption performance of the functional primitive sequence can be changed.

[0071] The flowchart of the design method of the multi-dimensional resonant cavity sound absorber in the specific implementation of this invention is as follows: Figure 4 As shown, the specific design process is as follows:

[0072] The intelligent design model for multidimensional resonant cavity sound absorbers based on autoencoders and random forests mainly consists of three parts: First, establishing an autoencoder model, training encoder and decoder sub-models, and using the encoder sub-model to project high-dimensional features from the sound absorption curve database into a low-dimensional space; second, establishing a random forest prediction model to learn and verify the low-dimensional features and corresponding sound absorption curves in the low-dimensional space, and generating a new low-dimensional sound absorption curve database; third, searching for low-dimensional structural parameters that meet the requirements in the two databases and using the decoder sub-model in the autoencoder to project them back into the high-dimensional space, outputting the structural parameters and the sound absorption curves of the corresponding frequency bands.

[0073] The specific steps for implementation are as follows:

[0074] (1) The sound absorption curve database sequence is rewashed, the training set and the validation set are randomly divided according to the proportion, and the training set and the validation set are normalized by deviation standardization respectively, and the original values ​​are scaled to the [0,1] interval proportionally.

[0075] First, calculate the maximum and minimum values ​​of the n feature sequences in the training set. Then, calculate the proportion of the distance from an element to the minimum value in each feature sequence to the distance between the maximum and minimum values ​​of that column. The deviation standardization function used in this embodiment is as follows:

[0076]

[0077] Where, x i Let y be the i-th feature sequence. i For the i-th new feature sequence after normalization, max 1≤i≤n [x j} represents x i The minimum value of the number of the next j samples, min 1≤i≤n {x j} represents x i Find the minimum number of the next j samples.

[0078] (2) Establish an autoencoder model with structural parameters as input and output, and train its encoder and decoder sub-models;

[0079] (3) Encode 5×n using the encoder sub-model of the autoencoder model. 2 These feature parameters enable the transformation of structural parameters from high-dimensional space to low-dimensional space.

[0080] (4) Establish a multidimensional sound absorption performance research model based on random forest, input the training set of step (1) into the multidimensional sound absorption performance research model for training, and use the data in the validation set for testing, validation and model evaluation.

[0081] (5) Use the GridSearch method to traverse the hyperparameter combination tuning and deploy the multidimensional sound absorption performance research model in the random forest model;

[0082] (6) Input the low-dimensional space obtained in step (3) as the feature of the random forest into the research model of the sound absorption performance of the multi-dimensional resonant cavity sound absorber, and predict the sound absorption coefficient curve in the corresponding frequency band of the low-dimensional feature space.

[0083] The research content of the multidimensional resonant cavity sound absorber sound absorption performance research model in steps (4) to (6) is to use random forest to predict the sound absorption coefficient curves in the frequency bands corresponding to the structural parameters of the multidimensional resonant cavity sound absorber. The output result is a two-dimensional sound absorption curve that connects the sound absorption coefficients corresponding to each frequency band.

[0084] (7) A new low-dimensional feature space and corresponding sound absorption coefficient curves are generated by the multidimensional sound absorption performance research model based on random forest as a new database, and the new database obtained by prediction is saved in blocks.

[0085] (8) Input the structural parameter design requirements, and simultaneously search for structural parameters and sound absorption coefficient curves that meet the requirements in the machine learning dataset in step (1) and the low-dimensional feature space database in step (7);

[0086] (9) If a sound absorption coefficient curve that meets the requirements is found in the machine learning dataset in step (1), the first five sets of structural parameters that meet the requirements and their sound absorption coefficient curves are directly output.

[0087] (10) If a sound absorption coefficient curve that meets the requirements is found in the low-dimensional feature space database in step (7), the low-dimensional features corresponding to the sound absorption coefficient curve that meets the requirements are reconstructed using the decoder sub-model of the autoencoder in step (2), the decoder projects it back to the high-dimensional features, and then outputs the first five sets of high-dimensional structural parameters and the corresponding frequency band sound absorption curves that meet the requirements.

[0088] The autoencoder model network structure constructed in this invention is as follows: Figure 5 As shown:

[0089] An autoencoder specifically comprises an encoder and a decoder sub-model. Defining the autoencoder model involves defining an encoder with multiple hidden layers, a bottleneck layer, and a decoder with multiple hidden layers, all composed of cascaded fully connected layers. The encoder encodes the original high-dimensional input feature X into a low-dimensional latent variable z = gθ1(X) = σ(W1X + b1); the decoder reconstructs the latent variable z from the hidden layers, attempting to restore it as close as possible to the original high-dimensional feature.

[0090] It should be noted that, in this embodiment of the invention, when defining the encoder and decoder, the autoencoder adds a BatchNormalization layer and a ReLU activation function to each hidden layer to address the problem of "gradient vanishing" in neurons.

[0091] Batch normalization layers maintain the activation input distribution of each hidden layer as a standard normal distribution with a mean of 0 and a variance of 1, thus mitigating gradient vanishing.

[0092]

[0093]

[0094]

[0095]

[0096] Where B is the input numerical set, m is the batch size, γ and β are the scale and shift parameters to be learned, and μ R The mean of the batch data. This represents the variance of the batch data. The specific operation of BatchNormalization is as follows: first, calculate the mean and variance of B; then, transform the mean and variance to 0 and 1 respectively; finally, multiply each element of B by Y and add β to obtain y. i Output.

[0097] Simultaneously, the ReLU activation function is used to introduce nonlinear features into the neural network, enabling it to approximate any nonlinear function and improving the optimization and generalization capabilities of the autoencoder to a certain extent. The expression for the activation function is:

[0098]

[0099] ReLU is a piecewise function that outputs 0 when the input x < 0 and outputs itself x when the input x ≥ 0. The gradient of ReLU is 1, therefore it avoids gradient explosion or gradient loss problems.

[0100] It should be noted that in this embodiment of the invention, when the autoencoder begins compilation, the Adam optimizer is used to dynamically adjust the learning rate using first-order and second-order moment estimates of the gradient. Gradient descent is implemented within a while loop, thereby improving the network's recognition accuracy. The optimization formula is as follows:

[0101] whileθ t No convergence

[0102] t←t+1

[0103]

[0104] m t ←β1·m t-1 +(1-β1)·g t

[0105]

[0106]

[0107]

[0108]

[0109] Where t represents the number of update steps, θ represents the parameters to be solved, and f t (θ) represents the loss function, g t Describe the objective function f t (θ) is the gradient obtained by differentiating θ, β1 represents the first-order moment decay coefficient, β2 represents the second-order moment decay coefficient, and m t The gradient g represents t The first moment, v t Describing the gradient g t The second moment, m t bias correction, Indicates v t Bias correction.

[0110] It should be noted that the objective function for dimensionality reduction of the autoencoder described in this invention is MinimizeLoss = dist(X, X). R ), where dist is the distance metric function between the two, and the autoencoder model described in this invention is metric by MSE.

[0111] The structure of the multidimensional resonant cavity sound absorber performance research model based on random forest constructed in this invention is as follows: Figure 6 As shown:

[0112] The random forest used in this embodiment is essentially an ensemble learning method in machine learning. This method can be applied to large-scale datasets, has higher prediction accuracy than most algorithms, and introduces random samples and random features, giving random forests a certain degree of noise resistance while making them less prone to overfitting, thus offering advantages over other algorithms. Therefore, this embodiment uses a random forest regression model to train the sound absorption performance of the multidimensional resonant cavity sound absorber.

[0113] The construction steps of the random forest described in this embodiment specifically include:

[0114] (1) Random forest uses the bootstrap resampling technique to repeatedly and randomly sample N samples with replacement from the original training sample set N, one sample at a time, to form N samples. Then, a decision tree is trained based on the N samples generated by the bootstrap method, which are used as the samples at the root node of the decision tree.

[0115] (2) When each sample has M features, and each node of the decision tree needs to be split, m features are randomly selected from these M features, satisfying the condition m≤M. Then, information gain is used to select one feature from these m features as the splitting attribute of that node.

[0116] In step (2), information gain represents the degree to which the uncertainty of class Y is reduced by knowing information about feature X. Generally speaking, a larger information gain means a higher purity of information obtained by using feature A for partitioning. The mathematical expression is as follows: The information gain g(D,A) of feature A on the training dataset D is defined as the difference between the empirical entropy H(D) of set D and the empirical conditional entropy H(D|A) of D given feature A, i.e., g(D,A) = H(D) - H(D|A)

[0117] It should be noted that in this embodiment, the "max_features" parameter in scikit-learn is used to determine the maximum number of features to consider when training each tree. Features exceeding this limit are discarded. A smaller `max_features` parameter results in more diverse trees in the random forest, but too small a parameter can lead to difficulty in selecting which features to test during partitioning. In scikit-learn, `max_features` can be selected in several ways: "auto", "sqrt", "log2", and None. `auto` and `sqrt` take the square root of the total number of features, `log2` takes the logarithm of the total number of features, and `None` sets `max_features` to the total number of features. The default value for `max_features` is "auto".

[0118] (3) During the formation of the decision tree, each node must be split according to step (2) until it can no longer be split. No pruning is performed during the entire formation of the decision tree.

[0119] (4) A large number of decision trees were built according to steps (1) to (3), which constituted a random forest.

[0120] It should be noted that in this embodiment, the "n_estimators" parameter in scikit-learn is used to control the number of decision trees. If n_estimators is too small, it is easy to underfit; if it is too large, it cannot significantly improve the model. Therefore, in this embodiment, the number of decision trees with the highest training accuracy of n_estimators = 150 is selected within the acceptable memory / time limit.

[0121] (5) The output of a random forest depends on multiple decision trees. Therefore, when all decision trees are combined in parallel, the variance of the ensemble result is low because each decision tree is perfectly trained on its specific sample data. In the case of classification problems, the final output is obtained by using a majority-voting classifier. In the case of regression problems, the final output is the average of all outputs.

[0122] The random forest uses the low-dimensional space encoded by the autoencoder as the feature input, and a set of sound absorption coefficient curves corresponding to the low-dimensional space as labels. It dynamically adjusts the learning curve through grid search to determine the combination of random forest hyperparameters with the best training effect, deploys the random forest model, and generates a new low-dimensional feature space and the corresponding sound absorption coefficient curves of the frequency band as a new database. The new database obtained from the prediction is saved in blocks.

[0123] Secondly, embodiments of this application provide a multi-dimensional resonant cavity type sound absorber design system, such as... Figure 7 As shown, employing the above-mentioned multi-dimensional resonant cavity sound absorber design method, the intelligent design system for the multi-dimensional resonant cavity sound absorber includes:

[0124] The first model building module establishes a multi-dimensional resonant cavity sound absorption model based on finite element simulation, calculates the sound absorption coefficient, and builds a machine learning dataset.

[0125] The second model building module trains an autoencoder model with the sound absorber structural parameters as input and output, including encoder and decoder sub-models.

[0126] Dimensionality reduction module: Using the encoder sub-model, the features of the machine learning dataset are projected into a low-dimensional space, and a multi-dimensional sound absorption performance research model is established based on random forest to generate a low-dimensional space sound absorption coefficient database.

[0127] The up-dimensional output module obtains the corresponding structural parameters in the design task, searches for low-dimensional structural parameters that meet the design task requirements in the low-dimensional space sound absorption coefficient database, and uses the decoder sub-model to project the low-dimensional structural parameters back into the high-dimensional space, outputting the structural design parameters and sound absorption coefficient data to complete the design task of the multi-dimensional resonant cavity sound absorber.

[0128] In practical implementation, the above modules can be expanded with user interaction functions and their functions merged and simplified to form the following specific modules:

[0129] Account Login Module 10: This module includes image CAPTCHA configuration, setting the CAPTCHA content and image noise level. The account login interface is generated from a form. The user enters their nickname and password, triggering a POST request. Clicking the image CAPTCHA triggers an Ajax request. Djang dynamically refreshes the CAPTCHA image and creates the CAPTCHA information in the data table. After confirming the user's entered nickname, password, and CAPTCHA are correct, the user enters the main interface of the design system for a multi-dimensional resonant cavity sound absorber.

[0130] Performance Research Module 20: This module includes task creation, visualization of functional primitives, visualization of sound absorption curves, and task calculation information. It uses the aforementioned random forest-based multidimensional resonant cavity sound absorber sound absorption performance research model to predict the user-input structural parameters, outputting a visualized sound absorption coefficient curve.

[0131] Design Module 30: The design module includes new task requirements, structural design schemes, and task calculation information. The structural design scheme reduces the dimensionality of structural parameters based on the multidimensional resonant cavity absorber design model based on an autoencoder, projecting them into a low-dimensional space. Then, it uses the multidimensional resonant cavity absorber sound absorption performance research model based on a random forest to predict and generate a new low-dimensional space database. Finally, it searches and outputs several sets of structural parameters that meet the user's task requirements in both the original and new databases.

[0132] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the multidimensional resonant cavity type sound absorber design method as described above.

[0133] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the multidimensional resonant cavity type sound absorber design method as described above.

[0134] In addition, combined Figure 1 The multidimensional resonant cavity type sound absorber design method described in this application embodiment can be implemented by computer equipment. Figure 8 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of this application.

[0135] The computer device may include a processor 41 and a memory 42 storing computer program instructions.

[0136] Specifically, the processor 41 may include a central processing unit (CPU), a programmable logic device (PLD), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0137] The memory 42 may include a large-capacity storage device for data or instructions. For example, and not limitingly, the memory 42 may include a solid-state drive (SSD), a hybrid hard drive (HHD), or a traditional hard disk drive (HDD). Where appropriate, the memory 42 may be internal or external to the data processing device. In a particular embodiment, the memory 42 is non-volatile memory (NVM). In a particular embodiment, the memory 42 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM can be a mask-programmable ROM, a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be static random-access memory (SRAM) or dynamic random-access memory (DRAM), wherein the DRAM can be fast page mode dynamic random-access memory (FPMDRAM), extended data out dynamic random-access memory (EDODRAM), synchronous dynamic random-access memory (SDRAM), etc.

[0138] The memory 42 has a “memory” function and can use physical devices with two stable states, “0” and “1”, to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 41.

[0139] The processor 41 reads and executes the computer program instructions stored in the memory 42 to implement a multidimensional resonant cavity type sound absorber design method in the above embodiment.

[0140] In some embodiments, the computer device may further include a communication interface 43 and a bus 40. For example, Figure 8 As shown, the processor 41, memory 42, and communication interface 43 are connected through bus 40 and complete communication with each other.

[0141] The communication interface 43 can be a specification used to define the connection methods, electrical characteristics, protocols, and other details between communication devices; it can also be a hardware interface or a software interface. Hardware interfaces mainly include physical interfaces and electrical interfaces, used to specify the physical form of the connection between communication devices, electrical signals, and other specifications; software interfaces are mainly used to specify the communication protocols, data formats, and other information between communication devices.

[0142] Bus 40 includes hardware, software, or both, that couples components of a computer device together. Bus 40 includes, but is not limited to, at least one of the following: Data Bus, Address Bus, Control Bus, Expansion Bus, and Local Bus. The Data Bus (DB), Address Bus (AB), and Control Bus (CB) are collectively referred to as the system bus. For example, and not as a limitation, bus 40 may include an Accelerated Graphics Port (AGP), an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 40 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0143] Compared to existing technologies, this invention proposes a method based on multidimensional resonant cavity sound absorbers, autoencoders, and random forests. This method simultaneously enables the design of structural parameters and the prediction of sound absorption coefficient curves for multidimensional resonant cavity sound absorbers. Specifically, it utilizes a sound absorption coefficient database constructed from the structural parameters of different dimensions of resonant cavity sound-absorbing functional primitives established by CAE and their corresponding sound absorption coefficient curves. While using autoencoders and random forests to establish an intelligent design model for the multidimensional resonant cavity sound absorber, it can predict the sound absorption curves corresponding to the projected low-dimensional features and generate a new low-dimensional spatial sound absorption curve database. Based on task requirements, it searches the two databases for sound absorption curves and their structural parameters that meet the requirements and outputs them. This solves the problems of long prediction and design times and low efficiency in existing simulation technologies.

[0144] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

[0145] The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.

Claims

1. A design method for a multidimensional resonant cavity type sound absorber, characterized in that, Includes the following steps: S1. Establish a multi-dimensional resonant cavity sound absorption model based on finite element simulation, calculate the sound absorption coefficient, and construct a machine learning dataset; S2. Train the autoencoder model with the sound absorber structural parameters as input and output, including the encoder sub-model and the decoder sub-model; S3. Using the encoder sub-model, the features of the machine learning dataset are projected into a low-dimensional space, and a multi-dimensional sound absorption performance research model is established based on random forest to generate a low-dimensional space sound absorption coefficient database. S4. Obtain the corresponding structural parameters in the design task, search for low-dimensional structural parameters that meet the design task requirements in the low-dimensional space sound absorption coefficient database, use the decoder sub-model to project the low-dimensional structural parameters back to the high-dimensional space, output the structural design parameters and sound absorption coefficient data, and complete the design task of the multi-dimensional resonant cavity sound absorber.

2. The design method of a multidimensional resonant cavity type sound absorber according to claim 1, characterized in that, In S1, the multi-dimensional resonant cavity sound absorption model is composed of n 2 n-dimensional functional unit sequence structure composed of continuously parallel arrangement of resonant cavity type sound absorption units; The structural parameters of the resonant cavity type sound absorption unit include cavity length, cavity width, cavity height, hole depth, and hole diameter; The structural parameters include 5×n 2 There are n feature parameters, where n represents the number of dimensions.

3. The design method of a multidimensional resonant cavity type sound absorber according to claim 2, characterized in that, In S1, the machine learning dataset contains: Multi-dimensional resonant cavity type sound absorber Each structural parameter and its corresponding single-point frequency are used as features of the dataset. ; Furthermore, the single-point absorption coefficient corresponding to each set of structural parameters and single-point frequency is used as the dataset label. The structural parameters, single-point frequencies, and sound absorption coefficients are all in one-to-one correspondence.

4. The design method of a multidimensional resonant cavity type sound absorber according to claim 3, characterized in that, S2 specifically includes the following steps: S2-1. Reshuffle the sequences in the machine learning dataset, randomly split them into training and validation sets according to a preset ratio, and preprocess the data. S2-2. Establish an autoencoder model with structural parameters as input and output, and train the encoder sub-model and decoder sub-model of the autoencoder model.

5. The design method of a multidimensional resonant cavity type sound absorber according to claim 4, characterized in that, In S2, the autoencoder model includes an encoder with multiple hidden layers, a bottleneck layer, and a decoder with multiple hidden layers; In the autoencoder model, structural parameters are used as input data for encoding, realizing the transformation of data from high-dimensional space to low-dimensional space. Then, the encoding vector of the bottleneck layer is reconstructed to complete the decoding, and the reconstructed structural parameters are output. The output of the bottleneck layer is the low-dimensional feature space after dimensionality reduction.

6. The design method of a multidimensional resonant cavity type sound absorber according to claim 4, characterized in that, S3 specifically includes the following steps: S3-1, Encoding via the encoder sub-model of the autoencoder model. These feature parameters enable the transformation of structural parameters from high-dimensional space to low-dimensional space. S3-2. Establish a multidimensional sound absorption performance research model based on random forest. Input the training set obtained in S2-1 into the multidimensional sound absorption performance research model for training, and use the data in the validation set for testing, validation and model evaluation. S3-3, Using the GridSearch method to traverse the hyperparameter combination tuning and deployment of the multidimensional sound absorption performance research model in S3-2; S3-4. Input the low-dimensional space obtained in S3-1 as a feature of the random forest into the multi-dimensional sound absorption performance research model obtained in S3-3, and predict the sound absorption coefficient curve in the corresponding frequency band of the low-dimensional feature space. S3-5. Using the multidimensional sound absorption performance research model based on random forest, a new low-dimensional feature space and the corresponding sound absorption coefficient curve are predicted and generated. This is used to construct a low-dimensional feature space database, which is then stored in blocks.

7. The design method of a multidimensional resonant cavity type sound absorber according to claim 5, characterized in that, S4 specifically includes the following steps: S4-1. Obtain the structural parameter design requirements. Search for the structural parameters and sound absorption coefficient curves that meet the requirements in the machine learning dataset mentioned in S1 and the low-dimensional feature space database in S3-5, respectively. S4-2. If a sound absorption coefficient curve that meets the requirements is found in the machine learning dataset, the first five sets of structural parameters that meet the requirements and their corresponding sound absorption coefficient curves are directly output. S4-3. If a sound absorption coefficient curve that meets the requirements is found in the low-dimensional feature space database, the low-dimensional features corresponding to the sound absorption coefficient curve that meets the requirements are first reconstructed using the decoder sub-model of the autoencoder, and then projected back to the high-dimensional features through the decoder. Finally, the first five sets of high-dimensional structural parameters and the corresponding frequency band sound absorption curves that meet the requirements are output.

8. A design system for a multidimensional resonant cavity type sound absorber, characterized in that, include: The first model building module establishes a multi-dimensional resonant cavity sound absorption model based on finite element simulation, calculates the sound absorption coefficient, and builds a machine learning dataset. The second model building module trains an autoencoder model with the sound absorber structural parameters as input and output, including encoder and decoder sub-models. Dimensionality reduction module: Using the encoder sub-model, the features of the machine learning dataset are projected into a low-dimensional space, and a multi-dimensional sound absorption performance research model is established based on random forest to generate a low-dimensional space sound absorption coefficient database. The up-dimensional output module obtains the corresponding structural parameters in the design task, searches for low-dimensional structural parameters that meet the design task requirements in the low-dimensional space sound absorption coefficient database, and uses the decoder sub-model to project the low-dimensional structural parameters back into the high-dimensional space, outputting the structural design parameters and sound absorption coefficient data to complete the design task of the multi-dimensional resonant cavity sound absorber.

9. An electronic device, comprising a memory and a processor, characterized in that, The processor is used to execute the program in the memory to implement the design method of the multidimensional resonant cavity sound absorber as described in any one of claims 1 to 7.

10. A storage medium containing computer-executable instructions, characterized in that, The storage medium of the computer-executable instructions, when executed by a computer processor, is used to perform the design method of the multidimensional resonant cavity type sound absorber as described in any one of claims 1 to 7.

Images