Noise control system based on metamaterial and machine learning method

Abstract

The invention discloses a noise control system based on a metamaterial and a machine learning method, and the system comprises a machine learning optimization module which comprises a deep learning unit and a finite element model correction unit, and a complex sound field inversion modeling method based on the deep learning and finite element model correction method. Accurately simulating and analyzing the noise characteristics of the complex sound field of the transformer substation; the acoustic metamaterial structure design module is combined with the machine learning optimization module to quickly predict the noise reduction performance of the acoustic metamaterial structures with different structure parameters in the current environment so as to realize customized design; and the noise control system construction module is used for applying the designed acoustic metamaterial structure to a noise control system. Through the machine learning optimization module, in combination with deep learning and a finite element model correction method, accurate simulation and analysis of complex sound fields of old substations and the like are achieved, meanwhile, reverse optimization is conducted on the acoustic metamaterial through a machine learning algorithm, the design efficiency is improved, and the design period is shortened.

Description

Technical Field

[0001] The present application belongs to the technical field of noise control, and specifically relates to a noise control system based on metamaterials and machine learning methods. Background Art

[0002] With the acceleration of modern industrialization and urbanization, noise pollution has become one of the important environmental issues that need to be solved urgently. Although traditional noise control methods, such as noise barriers and silencers, can reduce noise levels to a certain extent, they often have disadvantages such as limited noise reduction effect, long design cycle, and high cost. Especially in complex noise environments, such as industrial sites such as old substations, due to the diversity of noise sources and complex noise propagation paths, traditional noise control methods often fail to achieve ideal noise reduction effects.

[0003] In recent years, the rapid development of metamaterials has provided new ideas for noise control. Metamaterials are artificial composite materials with special physical properties. By precisely designing their microstructures, they can achieve precise control of sound waves, electromagnetic waves, etc.

[0004] The prior art discloses an adjustable acoustic metamaterial barrier system for the spatial distribution characteristics of transformer noise, which is composed of an acoustic metamaterial barrier (3) and a noise measurement and analysis system (2). The acoustic metamaterial barrier (3) is composed of a metamaterial support bracket (7) and a fractal structure acoustic metamaterial module (8). The metamaterial support bracket (7) is parallel to the transformer sound source surface (1). The fractal structure acoustic metamaterial module (8) is embedded in the metamaterial support bracket (4). The noise measurement and analysis system (2) measures the noise within the range of the acoustic metamaterial barrier (3) and analyzes to obtain the spatial distribution characteristics of the noise. According to the spatial distribution characteristics of the noise, fractal structure acoustic metamaterial modules (8) with different parameters are selected. When the sound wave is transmitted to the acoustic metamaterial barrier (3), since the parameters of the acoustic metamaterial module (8) are selected according to the noise distribution characteristics of the corresponding position, the noise amplitude in the target noise reduction area can be greatly reduced.

[0005] However, there are still many challenges in applying metamaterials to noise control. For example, how to quickly predict the noise reduction performance of acoustic metamaterials with different structural parameters, and how to build an efficient noise control system. Therefore, studying a noise control system based on metamaterials and machine learning methods has important practical significance and application value. Summary of the invention

[0006] The present application provides a noise control system based on metamaterials and machine learning methods, aiming to solve the problem that it is difficult to quickly predict the noise reduction performance of acoustic metamaterials with different structural parameters in the prior art.

[0007] A noise control system based on metamaterials and machine learning methods, comprising:

[0008] Machine learning optimization module: including deep learning unit and finite element model correction unit, complex sound field inversion modeling method based on deep learning and finite element model correction method, accurate simulation and analysis of noise characteristics of complex sound field in substation;

[0009] Acoustic metamaterial structure design module: Combined with the machine learning optimization module, it can quickly predict the noise reduction performance of acoustic metamaterial structures with different structural parameters in the current environment and realize customized design;

[0010] Specifically, the acoustic metamaterial structure design module determines the frequency range of low-frequency noise that needs to be controlled based on the simulation and analysis results in the machine learning optimization module, and then selects materials based on the target frequency and application scenario, and combines the design direction and adjusts the structural parameters to optimize the low-frequency sound absorption effect, thus realizing the acoustic metamaterial structure design;

[0011] Noise control system building module: Apply the designed acoustic metamaterial structure to the noise control system, specifically a comprehensive noise control system including acoustic metamaterial sound barrier unit and active noise control unit.

[0012] Optionally, the modeling step in the deep learning unit includes:

[0013] S1: Collect sound field data;

[0014] S2: Data preprocessing: Clean and preprocess the collected data, and divide the data into training set, validation set, and test set to ensure that the model will not overfit during the training process and to effectively evaluate the performance of the model;

[0015] S3: Feature extraction: extract features that have an important impact on the sound field characteristics from the preprocessed data;

[0016] S4: Model design: Design the architecture of the recurrent neural network;

[0017] S5: Model training: Use the training data set to train the RNN model;

[0018] S6: Use the test data set to evaluate the trained RNN model, and deploy it to the machine learning optimization module after it meets the standards.

[0019] Optionally, the S4 includes:

[0020] S4.1: Data format conversion: convert the sound field sequence data into a vector sequence of fixed length and perform normalization or standardization on the data;

[0021] S4.2: Define the network structure: The input layer receives the converted vector sequence and converts it into a format that the model can process; the hidden layer contains multiple RNN units to capture the timing information and long-term dependencies in the sequence data, and selects a nonlinear activation function to process the output of the RNN unit; the output layer outputs a scalar representing the predicted value, and selects the Sigmoid activation function;

[0022] S4.3: Define loss function;

[0023] S4.4: Set training parameters: including learning rate, batch size, and number of iterations.

[0024] Optionally, the modeling step of the finite element model correction unit includes:

[0025] Step 1: Establish an initial finite element model: According to the geometric shape, material properties and boundary conditions of the acoustic metamaterial structure, an initial finite element model is established using finite element analysis software;

[0026] Step 2: Collect measured data: Collect measured data of acoustic metamaterial structure;

[0027] Step 3: Model parameterization: parameterize the geometric or physical parameters of the finite element model;

[0028] Step 4: Sensitivity analysis: Calculate the sensitivity of model parameters to the dynamic characteristics of the structure;

[0029] Step 5: Optimization solution: Based on the measured data and sensitivity analysis results, mathematical optimization methods are used to solve the modified model parameters;

[0030] Step 6: Compare and verify the revised finite element model with the measured data, evaluate the accuracy and reliability of the model, and deploy it after the evaluation meets the standards.

[0031] Optionally, the active noise control unit includes: a signal source, an inverse processor, a filter, an amplifier and an actuator.

[0032] Optionally, the design directions of the acoustic metamaterial structure design module include: film type, thin plate type, Helmholtz-like type and composite type.

[0033] Optionally, the system further comprises a system performance testing module, which tests and verifies the noise reduction performance of the constructed noise control system.

[0034] Optionally, the system further includes a system performance optimization module, which uses a machine learning optimization module to further optimize the acoustic metamaterial structure according to the test results.

[0035] Compared with the prior art, this application has at least the following beneficial effects:

[0036] This application uses a machine learning optimization module, combined with deep learning and finite element model correction method, to achieve accurate simulation and analysis of complex sound fields such as old substations. At the same time, it uses machine learning algorithms to reverse optimize acoustic metamaterials, thereby improving design efficiency and shortening the design cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic diagram of module connections of a noise control system based on metamaterials and machine learning methods provided in one embodiment of the present application;

[0038] Figure 2 A schematic diagram of a design process of an acoustic metamaterial structure design module for a noise control system based on metamaterials and machine learning methods provided in one embodiment of the present application; DETAILED DESCRIPTION

[0039] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0040] In one embodiment, Figure 1 As shown, a noise control system based on metamaterials and machine learning methods is provided, comprising:

[0041] Machine learning optimization module, acoustic metamaterial structure design module, noise control system construction module, system performance testing module and system performance optimization module;

[0042] Machine learning optimization module: including deep learning unit and finite element model correction unit. Based on deep learning and finite element model correction method, the inverse modeling method of complex sound field of old substation can realize accurate simulation and analysis of noise characteristics of complex sound field of substation. Machine learning algorithms, such as recurrent neural network (RNN), can be used to reversely optimize the designed acoustic metamaterials, improve design efficiency and shorten design cycle.

[0043] Specifically, the modeling steps in the deep learning unit include:

[0044] S1: Collect acoustic field data of old substations, including measuring noise levels at different locations and time periods, and possible spectrum analysis;

[0045] S2: Data preprocessing: The collected data may contain noise or outliers, and the collected data needs to be cleaned and preprocessed to ensure the accuracy and reliability of the data;

[0046] S3: Feature extraction: extract features that have an important impact on the sound field characteristics from the preprocessed data, such as spectrum features, time domain features, etc.;

[0047] S4: Model design: Design the architecture of the recurrent neural network (RNN), including the design of the input layer, hidden layer, and output layer, as well as the selection of activation function and loss function;

[0048] S5: Model training: Use the training data set to train the RNN model. During the training process, the model's prediction results are calculated through the forward propagation algorithm, and the gradient is calculated and the model parameters are updated through the back propagation algorithm. The model in the training process is evaluated using the validation data set, and the model is adjusted and optimized based on the evaluation results. The adjustment and optimization includes adjusting the network structure, hyperparameter settings, etc., to improve the generalization ability and performance of the model;

[0049] S6: Use the test data set to evaluate the trained RNN model to measure the performance of the model. Once the model meets the standards, it is deployed in the machine learning optimization module for use.

[0050] Specifically, S4 includes:

[0051] S4.1: Data format conversion: Convert the original data into a format that can be processed by RNN, including converting the sound field sequence data into a vector sequence of fixed length and normalizing or standardizing the data to accelerate the training process of the model and improve the convergence speed of the model;

[0052] The data is divided into training set, validation set and test set to ensure that the model does not overfit during training and to effectively evaluate the performance of the model;

[0053] S4.2: Define the network structure: The input layer receives the converted vector sequence and converts it into a format that the model can process; the hidden layer contains multiple RNN units to capture the timing information and long-term dependencies in the sequence data, and selects nonlinear activation functions (such as Tanh, ReLU, etc.) to process the output of the RNN unit; the output layer outputs a scalar representing the predicted value, and selects the Sigmoid activation function;

[0054] S4.3: Define loss function: Select an appropriate loss function according to the specific task to measure the difference between the predicted result of the model and the actual result. In a specific embodiment, the mean square error (MSE) loss function is used;

[0055] S4.4: Set training parameters: including learning rate, batch size, number of iterations, etc.

[0056] In a specific embodiment, the learning rate is selected to use the adaptive learning rate method RMSprop, so that the learning rate is automatically adjusted according to the gradient information during the training process; the batch size is set to 1024, and the number of iterations is selected to 10.

[0057] The modeling steps of the finite element model modification unit include:

[0058] Step 1: Establish an initial finite element model: According to the geometric shape, material properties and boundary conditions of the acoustic metamaterial structure, an initial finite element model is established using finite element analysis software;

[0059] Step 2: Collect measured data: Collect measured data such as modal parameters (such as natural frequency, modal vibration shape, etc.) and frequency response function of the acoustic metamaterial structure through experiments or field measurements;

[0060] Step 3: Model parameterization: parameterize the geometric or physical parameters of the finite element model for subsequent correction;

[0061] Step 4: Sensitivity analysis: Calculate the sensitivity of model parameters to structural dynamic characteristics to determine which parameters have a greater impact on model accuracy;

[0062] Step 5: Optimization solution: Based on the measured data and sensitivity analysis results, mathematical optimization methods (such as the least squares method) are used to solve the modified model parameters;

[0063] Step 6: Compare and verify the revised finite element model with the measured data, evaluate the accuracy and reliability of the model, and deploy it after the evaluation meets the standards.

[0064] Acoustic metamaterial structure design module: Combined with the machine learning optimization module, it can quickly predict the noise reduction performance of acoustic metamaterial structures with different structural parameters in the current environment and realize customized design. The design directions include film type, thin plate type, Helmholtz type and composite type.

[0065] Specifically, if Figure 2 As shown in the figure, the acoustic metamaterial structure design module determines the frequency range of the low-frequency noise that needs to be controlled according to the simulation and analysis results in the machine learning optimization module, and then selects suitable materials according to the target frequency and actual application scenarios, such as polymer materials, metal materials or composite materials, etc., and adjusts the structural parameters (such as the thickness and tension of the film, the size and shape of the thin plate, the volume of the resonance cavity, etc.) in combination with the design direction to optimize the low-frequency sound absorption effect and realize the acoustic metamaterial structure design.

[0066] Noise control system building module: Apply the designed acoustic metamaterial structure to the noise control system, including a comprehensive noise control system including an acoustic metamaterial sound barrier unit and an active noise control unit, with the main variable being the acoustic metamaterial structure;

[0067] The active noise control unit includes: a signal source, which is the starting point of the active noise control device and is responsible for collecting the original noise signal;

[0068] Reverse processor: It is one of the core parts of the active noise control device. Its main function is to process the original noise signal collected from the signal source to extract the characteristics of the noise and generate a reverse signal with the opposite phase and equal amplitude to the original noise.

[0069] Filter: used to further filter and optimize the reverse signal generated by the reverse processor. By adjusting the parameters of the filter, the reverse signal can be precisely controlled to ensure that it matches the original noise in phase and amplitude. The filter can also help eliminate possible clutter and interference in the reverse signal, thereby improving the accuracy and stability of noise control.

[0070] Amplifier: used to amplify the reverse signal generated by the reverse processor so that it has enough energy to offset the original noise;

[0071] Actuator: It is the output part of the active noise control device. The actuator converts the amplified and filtered reverse signal into sound waves (or mechanical vibrations) and releases them into the environment. When these sound waves (or mechanical vibrations) meet the original noise, destructive interference occurs, thereby eliminating or reducing the noise.

[0072] System performance test module: test and verify the noise reduction performance of the constructed noise control system;

[0073] System performance optimization module: Based on the test results, the machine learning algorithm is used to further optimize the acoustic metamaterial structure to improve the noise reduction effect.

[0074] In the above-mentioned noise control system method based on metamaterials and machine learning methods, by combining metamaterials and machine learning technology, accurate simulation and analysis of complex noise environments and customized design of acoustic metamaterials are achieved, thereby significantly improving the noise reduction effect;

[0075] Using machine learning algorithms to reverse optimize acoustic metamaterials can quickly predict the noise reduction performance of acoustic metamaterials with different structural parameters, thereby significantly shortening the design cycle and reducing R&D costs.

[0076] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A noise control system based on metamaterials and machine learning methods, characterized in that: include: Machine learning optimization module: including deep learning unit and finite element model correction unit, complex sound field inversion modeling method based on deep learning and finite element model correction method, accurate simulation and analysis of noise characteristics of complex sound field in substation; Acoustic metamaterial structure design module: Combined with the machine learning optimization module, it can quickly predict the noise reduction performance of acoustic metamaterial structures with different structural parameters in the current environment and realize customized design; The acoustic metamaterial structure design module determines the frequency range of low-frequency noise that needs to be controlled based on the simulation and analysis results in the machine learning optimization module, selects materials based on the target frequency and application scenario, and optimizes the low-frequency sound absorption effect by combining the design direction and adjusting the structural parameters to achieve the acoustic metamaterial structure design; Noise control system building module: Apply the designed acoustic metamaterial structure to the noise control system, specifically a comprehensive noise control system including acoustic metamaterial sound barrier unit and active noise control unit.

2. A noise control system based on metamaterials and machine learning methods according to claim 1, characterized in that: The modeling steps in the deep learning unit include: S1: Collect sound field data; S2: Data preprocessing: Clean and preprocess the collected data, and divide the data into training set, validation set, and test set to ensure that the model will not overfit during the training process and to effectively evaluate the performance of the model; S3: Feature extraction: extract features that have an important impact on the sound field characteristics from the preprocessed data; S4: Model design: Design the architecture of the recurrent neural network; S5: Model training: Use the training data set to train the RNN model; S6: Use the test data set to evaluate the trained RNN model, and deploy it to the machine learning optimization module after it meets the standards.

3. A noise control system based on metamaterials and machine learning methods according to claim 2, characterized in that: The S4 includes: S4.1: Data format conversion: convert the sound field sequence data into a vector sequence of fixed length and perform normalization or standardization on the data; S4.2: Define the network structure: The input layer receives the converted vector sequence and converts it into a format that the model can process; the hidden layer contains multiple RNN units to capture the timing information and long-term dependencies in the sequence data, and selects a nonlinear activation function to process the output of the RNN unit; the output layer outputs a scalar representing the predicted value, and selects the Sigmoid activation function; S4.3: Define loss function; S4.4: Set training parameters: including learning rate, batch size, and number of iterations.

4. A noise control system based on metamaterials and machine learning methods according to claim 1, characterized in that: The modeling steps of the finite element model correction unit include: Step 1: Establish an initial finite element model: According to the geometric shape, material properties and boundary conditions of the acoustic metamaterial structure, an initial finite element model is established using finite element analysis software; Step 2: Collect measured data: Collect measured data of acoustic metamaterial structure; Step 3: Model parameterization: parameterize the geometric or physical parameters of the finite element model; Step 4: Sensitivity analysis: Calculate the sensitivity of model parameters to the dynamic characteristics of the structure; Step 5: Optimization solution: Based on the measured data and sensitivity analysis results, mathematical optimization methods are used to solve the modified model parameters; Step 6: Compare and verify the revised finite element model with the measured data, evaluate the accuracy and reliability of the model, and deploy it after the evaluation meets the standards.

5. The noise control system based on metamaterial and machine learning method according to claim 1, characterized in that: The active noise control unit includes: a signal source, an inverse processor, a filter, an amplifier and an actuator.

6. A noise control system based on metamaterials and machine learning methods according to claim 1, characterized in that: The design directions of the acoustic metamaterial structure design module include: film type, thin plate type, Helmholtz-like type and composite type.

7. The noise control system based on metamaterial and machine learning method according to claim 1, characterized in that: The system further comprises a system performance testing module, which tests and verifies the noise reduction performance of the constructed noise control system.

8. A noise control system based on metamaterials and machine learning methods according to claim 7, characterized in that: The system also includes a system performance optimization module, which uses a machine learning optimization module to further optimize the acoustic metamaterial structure according to the test results.

Images