Methods, devices, equipment, and storage media for rapid prediction of underwater vehicle flow noise

Abstract

This application proposes a method, apparatus, device, and storage medium for rapid prediction of underwater vehicle flow noise, relating to the field of underwater radiated noise prediction technology. The method includes: acquiring far-field radiated noise data of multi-configuration underwater vehicles using numerical simulation, and preprocessing the data to obtain training and testing datasets; training a hybrid prediction model using the training and testing datasets to obtain a target hybrid prediction model, which includes a kernel ridge regression prediction sub-model and a radial basis function neural network prediction sub-model. The prediction result of the hybrid prediction model is obtained by weighted summation of the prediction results of the two prediction sub-models based on the coefficient of determination, and subtracting the average deviation of the training dataset; inputting the configuration data of the underwater vehicle to be predicted into the target hybrid prediction model to obtain the flow noise prediction result of the underwater vehicle to be predicted. This application can achieve rapid and accurate prediction of the flow noise characteristics of underwater vehicles through the above method.

Description

Technical Field

[0001] This application relates to the field of underwater radiated noise prediction technology for ships, and in particular to a method, apparatus, equipment and storage medium for rapid prediction of underwater vehicle flow noise. Background Technology

[0002] In the field of shipbuilding and ocean engineering, flow noise is a core indicator for evaluating the acoustic stealth performance of underwater vehicles. For a long time, numerical simulation methods based on computational fluid dynamics (CFD) and acoustic analogy theory (such as the FW-H equation), especially high-precision unsteady flow field simulation techniques such as large eddy simulation (LES), have been considered the mainstream and benchmark methods for predicting the flow noise characteristics of underwater vehicles.

[0003] This technical approach simulates flow field details (especially turbulent fluctuations) by constructing a detailed 3D model of the underwater vehicle in a computer, dividing it into computational grids of millions to tens of millions of pixels, and directly solving complex fluid dynamics control equations numerically, thereby calculating its radiated noise. Its greatest advantage lies in the clear physical mechanism, high accuracy and reliability of the prediction results, and the ability to reveal in detail the generation and propagation mechanism of flow noise.

[0004] However, predicting underwater vehicle flow noise using numerical calculations requires enormous computing resources and time, making it inefficient when dealing with a large number of underwater vehicle flow noise prediction scenarios. Summary of the Invention

[0005] In view of this, this application proposes a method, apparatus, device and storage medium for rapid prediction of underwater vehicle flow noise.

[0006] In a first aspect, this application provides a method for rapid prediction of underwater vehicle flow noise, including: Numerical simulation was used to obtain far-field radiated noise data of a multi-configuration underwater vehicle, and the far-field radiated noise data was preprocessed to obtain training datasets and test datasets. The hybrid prediction model is trained using the training dataset and the test dataset to obtain the target hybrid prediction model. The hybrid prediction model includes a kernel ridge regression prediction sub-model and a radial basis function neural network prediction sub-model. The prediction result of the hybrid prediction model is obtained by weighting and summing the prediction results of the two prediction sub-models based on the coefficient of determination, and then subtracting the average deviation of the training dataset. The configuration data of the underwater vehicle to be predicted is input into the target hybrid prediction model to obtain the flow noise prediction result of the underwater vehicle to be predicted.

[0007] In one embodiment, the step of acquiring far-field radiated noise data of a multi-configuration underwater vehicle using numerical simulation methods includes: The computational domain is set as a cylinder, and the inflow surface and side surface of the computational domain are set as velocity inlets, and the outflow surface is set as pressure outlets; The computational domain is divided into grids using a structured grid. Based on the mesh generation results, the instantaneous flow field and radiated noise of the multi-configuration underwater vehicle in the computational domain are calculated using large eddy simulation and acoustic analogy methods, and the far-field radiated noise data of the multi-configuration underwater vehicle are obtained.

[0008] In one embodiment, based on the mesh generation results, the instantaneous flow field and radiated noise of the multi-configuration underwater vehicle in the computational domain are calculated using large eddy simulation and acoustic analogy methods to obtain the far-field radiated noise data of the multi-configuration underwater vehicle, including: Using a cylindrical computational domain after meshing as the simulation space, the Large Eddy Simulation (LES) method is first used to calculate the unsteady flow field at a first time step until the flow field reaches a steady state. After the flow field stabilizes, the calculation is switched to a second time step to calculate the instantaneous flow field data. The first time step is longer than the second time step. The LES method uses a LES turbulence model combined with a subgrid-scale model as the calculation model, and the separation scheme adopts a bounded-central difference scheme. Using the instantaneous flow field data as the input source, the acoustic analogy method is used to solve the radiated noise of the multi-configuration underwater vehicle at multiple preset field points; The total sound level and sound pressure level spectrum data of each underwater vehicle configuration were collected at each preset field point, and the far-field radiated noise data of the multi-configuration underwater vehicle were obtained by summarizing them.

[0009] In one embodiment, the structured mesh is controlled by a three-layer volume mesh transition, with the size of the volume mesh increasing sequentially from the inside out.

[0010] In one embodiment, the number of preset field points is six; The reference coordinate system is set at the bow of the underwater vehicle, and the coordinate positions of the six preset field points are (-100,0,0), (100,0,0), (0,-100,0), (0,100,0), (0,0,-100), and (0,0,100).

[0011] In one embodiment, before acquiring far-field radiated noise data of a multi-configuration underwater vehicle using numerical simulation methods, the method further includes: Obtain the design parameter types and corresponding parameter ranges, wherein the design parameter types include total length, total width, inlet section length, outlet section length, longitudinal profile coefficient in the inlet section, longitudinal profile coefficient in the transition section, and longitudinal profile coefficient in the outlet section; Multiple underwater vehicle models with different configurations are constructed based on the design parameter types and corresponding parameter ranges.

[0012] In one embodiment, training the hybrid prediction model using the training dataset and the test dataset to obtain the target hybrid prediction model includes: The kernel ridge regression prediction sub-model and radial basis function neural network prediction sub-model in the hybrid prediction model are trained using the training dataset, and the prediction accuracy of the hybrid prediction model is verified using the test dataset. If the prediction accuracy reaches the preset target, output the target hybrid prediction model; If the prediction accuracy does not reach the preset target, the accuracy correlation parameters in the hybrid prediction model are adjusted and the training is repeated. The accuracy correlation parameters include regularization parameters, kernel width, and the number of hidden layer nodes.

[0013] Secondly, this application also provides a rapid prediction device for underwater vehicle flow noise, comprising: The acquisition module is used to acquire far-field radiated noise data of a multi-configuration underwater vehicle using numerical simulation methods, and to preprocess the far-field radiated noise data to obtain a training dataset and a test dataset. The training module is used to train the hybrid prediction model using the training dataset and the test dataset to obtain the target hybrid prediction model. The hybrid prediction model includes a kernel ridge regression prediction sub-model and a radial basis function neural network prediction sub-model. The prediction result of the hybrid prediction model is obtained by weighting and summing the prediction results of the two prediction sub-models based on the coefficient of determination, and then subtracting the average deviation of the training dataset. The forecasting module is used to input the configuration data of the underwater vehicle to be predicted into the target hybrid prediction model to obtain the flow noise forecasting result of the underwater vehicle to be predicted.

[0014] Thirdly, this application also provides an electronic device, including a processor and a memory; the memory has a stored computer program, wherein the computer program, when executed by the processor, implements the rapid prediction method for underwater vehicle flow noise as described in the first aspect.

[0015] Fourthly, this application also provides a non-transitory computer storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the rapid prediction method for underwater vehicle flow noise as described in the first aspect.

[0016] The rapid prediction method for underwater vehicle flow noise proposed in this application has the following advantages over related technologies: 1. The rapid prediction method for underwater vehicle flow noise in this application first uses numerical simulation to obtain far-field radiated noise data of multi-configuration underwater vehicles, and then preprocesses it to obtain training and test datasets. This not only achieves the goal of concentrating numerical simulation computing resources on the core configuration to obtain high-precision data, but also provides a reliable data foundation for subsequent model training.

[0017] 2. After obtaining the training and test datasets, the hybrid prediction model, which includes a kernel ridge regression prediction sub-model and a radial basis function neural network prediction sub-model, is trained using these datasets. The prediction results of the two sub-models are weighted and summed based on the coefficient of determination, and the average deviation of the training dataset is subtracted to obtain the final prediction result, effectively ensuring prediction accuracy. Finally, only the configuration data of the underwater vehicle to be predicted needs to be input into the trained target hybrid prediction model to quickly obtain flow noise prediction results. Therefore, the computational resource consumption and time cost of single configuration flow noise prediction are significantly reduced, and the efficiency of large-scale underwater vehicle flow noise prediction is significantly improved. This enables the rapid provision of high-accuracy flow noise prediction results during the rapid design phase of underwater vehicles, solving the pain points of traditional numerical calculation methods. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a rapid prediction method for underwater vehicle flow noise in one embodiment of this application. Figure 2 This is a schematic diagram illustrating the use of a structured grid to divide the computational domain in one embodiment of this application; Figure 3 This is a schematic diagram of the hydrophone position and reference coordinate system corresponding to a preset field point in one embodiment of this application; Figure 4 This is a schematic diagram of the segmentation of an underwater vehicle in one embodiment of this application; Figure 5 This is a flowchart illustrating a rapid prediction method for underwater vehicle flow noise in another embodiment of this application. Figure 6 This is a schematic diagram showing the distribution of total sound level prediction results at six field points for various underwater vehicle configurations in one embodiment of this application; Figure 7 This is a schematic diagram of the error distribution of the total sound level prediction results at six field points for various underwater vehicle configurations in one embodiment of this application; Figure 8 This is a schematic diagram comparing the predicted and numerical calculation results of the sound pressure level spectrum at six points of a certain configuration of underwater vehicle in a test set in one embodiment of this application. Figure 9 This is a schematic diagram of the normalized root mean square error distribution of the sound pressure level spectrum of a certain configuration of underwater vehicle in a test set in one embodiment of this application. Figure 10 This is a schematic diagram of the structure of a rapid prediction device for underwater vehicle flow noise in one embodiment of this application. Detailed Implementation

[0020] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0021] In some embodiments, such as Figure 1 As shown, the present application provides a method for rapid prediction of underwater vehicle flow noise, which includes the following steps S101 to S103.

[0022] S101: Numerical simulation methods are used to obtain far-field radiated noise data of multi-configuration underwater vehicles, and the far-field radiated noise data is preprocessed to obtain training and test datasets.

[0023] When acquiring far-field radiated noise data for multi-configuration underwater vehicles using numerical simulation, a numerical simulation model can be constructed based on large eddy simulation (LES) in computational fluid dynamics (CFD) combined with acoustic analogy. This method efficiently and accurately captures the surface flow field fluctuation pressure and noise propagation characteristics of the underwater vehicle. Subsequently, numerical models of underwater vehicles with different configurations are designed, and the surface flow field fluctuation pressure distribution of the vehicle under each configuration is obtained through simulation calculations. Then, considering factors such as underwater medium attenuation and boundary reflection, and combined with the acoustic propagation model, multiple monitoring points are set on the far-field reference sphere to collect radiated noise sound pressure level data at each monitoring point.

[0024] It should be noted that preprocessing can include denoising and normalization. For example, outlier data points caused by numerical computation convergence issues or abnormal operating conditions can be removed using the 3σ principle. Then, moving average or Gaussian filtering is used to smooth the data and reduce random errors. Subsequently, Min-Max normalization or Z-score standardization is applied to the configuration parameters and noise sound pressure level data respectively to eliminate the interference of dimensional and numerical range differences on subsequent model training. After data preprocessing, the processed complete dataset is finally split into training and testing datasets according to a preset ratio using a random partitioning method. During the partitioning process, it is ensured that the two datasets maintain consistency in the distribution range of configuration parameters and the range of noise levels to avoid overfitting or distortion of prediction results due to data bias. The final result is a training and testing dataset with a well-structured, well-defined, and balanced distribution.

[0025] S102: The hybrid prediction model is trained using the training dataset and the test dataset to obtain the target hybrid prediction model. The hybrid prediction model includes a kernel ridge regression prediction sub-model and a radial basis function neural network prediction sub-model. The prediction result of the hybrid prediction model is obtained by weighted summation of the prediction results of the two prediction sub-models based on the coefficient of determination, and then subtracting the average deviation of the training dataset.

[0026] For the kernel ridge regression prediction sub-model, it should be noted that kernel ridge regression (KRR) is a nonlinear regression method that combines ridge regression regularization techniques with the kernel trick. It is particularly suitable for modeling datasets with complex nonlinear relationships between inputs and outputs. Its basic idea is to use a kernel function to map the original low-dimensional input data to a high-dimensional feature space, perform linear regression analysis in this high-dimensional space, and simultaneously suppress overfitting through an L2 regularization term.

[0027] Linear regression assumes input features ( For the sample size, (Input dimension) and output ( There is a linear relationship (for the output dimension):

[0028] In the formula, This is the weight matrix. This is the error term. Linear regression solves this by minimizing the squared loss. :

[0029]

[0030] In the formula, This is the square of the Frobenius norm of the error matrix, used to measure the overall error magnitude. To find the square of the Frobenius norm of a given matrix. However, when multicollinearity exists in the input features, Approaching the singular, leading to The solution is unstable. Ridge regression addresses this issue by introducing L2 regularization, with the objective function being:

[0031]

[0032] In the formula, This is the regularization parameter, which controls the strength of regularization and prevents the model from overfitting. Let be the square of the Frobenius norm of the weight matrix W, which is equal to the sum of the squares of all weight values ​​and is used to measure the complexity of the weight matrix. The closed-form solution of ridge regression is:

[0033]

[0034] in, The identity matrix is ​​multiplied by the regularization parameter and added. Within the matrix, ensure that the matrix is ​​invertible. When the relationship between input and output is nonlinear (the nonlinear scattering relationship between UUV sound pressure at different frequencies and field parameters), the linear model cannot fit the data. In this case, a nonlinear mapping can be used to map the input to a high-dimensional feature space, where a linear model can be used for fitting.

[0035]

[0036] In the formula, Let x be a function that maps the original feature x to a high-dimensional (or even infinite-dimensional) feature space. In the radial basis functions, Mapping the original feature x to a Gaussian function space centered at each training sample, directly computes... Its complexity is extremely high in high-dimensional spaces. Kernel tricks utilize kernel functions:

[0037]

[0038] In the formula, For original spatial features, Given a high-dimensional mapping space, the above equation means using the inner product of two vectors in the high-dimensional mapping space as the value of the kernel function. This avoids explicit mapping and directly calculates the inner product in the high-dimensional space to reduce computational complexity.

[0039] Kernel ridge regression performs ridge regression in a high-dimensional feature space, with the objective function being... Using the Representer Theorem, the weights W in high-dimensional space can be represented as a linear combination of the training sample mappings:

[0040] In the formula, As the combined coefficient matrix, substituting it into the objective function and simplifying, the objective function of kernel ridge regression becomes:

[0041] In the formula, For the kernel matrix, elements .

[0042] For the Radial Basis Function (RBF) neural network prediction sub-model, the RBF neural network is a three-layer feedforward network consisting of an input layer, hidden layers, and an output layer. The input layer only transmits data and does not transform the input information. The number of nodes in the hidden layer depends on the needs, and its neuron kernel function is the radial basis function, commonly the Gaussian kernel function. This function can transform low-dimensional pattern input data into a high-dimensional space, making linearly inseparable problems in low-dimensional space linearly separable in high-dimensional space. Specifically, it takes the following form:

[0043]

[0044] in, Indicates the input point. As the "center", The distance function, defined as the "width," maps the low-dimensional input to a high-dimensional space, achieving a "linearized fit" to the nonlinear curve. The distance function expression is:

[0045]

[0046] This represents the square of the coordinate difference between the input point and the center point in the k-th dimension. The value of the input point in the k-th dimension. Let be the value of the j-th center point in the k-th dimension. The action function of the output layer neurons is a linear function, which is a linear weighted sum of the information output by the hidden layer neurons, and is the output of the entire neural network. The output layer is usually a linear combination of the outputs of the hidden layers. For a node in the output layer, its output is:

[0047]

[0048] in, It is the number of hidden layer nodes. It represents the weights from the i-th node in the hidden layer to the j-th node in the output layer. This is the bias. It refers to the value of the kernel function.

[0049] It is understandable that, for the kernel ridge regression prediction sub-model, a suitable kernel function (such as Gaussian kernel or polynomial kernel) and regularization coefficient are selected, and the input features of the training dataset and the corresponding far-field radiative noise labels are used for fitting training. The kernel function is used to map the low-dimensional input space to the high-dimensional feature space to solve the nonlinear fitting problem. At the same time, the regularization term is used to suppress the model overfitting and improve its fitting stability to high-dimensional sparse data.

[0050] For the radial basis function neural network prediction sub-model, the number of hidden layer nodes, the center and width parameters of the radial basis function are optimized and determined. The network weights are iteratively updated based on the training dataset. By utilizing the local approximation property of the radial basis function, the model's ability to fit complex nonlinear mapping relationships is enhanced.

[0051] Kernel Ridge Regression (KR) models, by incorporating regularization coefficients into the loss function, force the model weights to be moderate, effectively suppressing overfitting to noise and outliers in the training data, ultimately resulting in a smoother and more stable model overall. The value of the radial basis function (RBF) neural network kernel function decays exponentially with increasing distance between two points; for a test point, only training samples very close to it significantly influence the prediction. This characteristic makes it better at capturing local variations, high-frequency details, and sharp peaks. Calculating the coefficient of determination (R²) for each sub-model on the training dataset is crucial. Since the coefficient of determination quantifies the goodness of fit between the sub-model's predicted values ​​and the true values, a coefficient closer to 1 indicates a better fit. Using this coefficient as a weighting factor, the prediction results of the two sub-models are weighted and summed. Simultaneously, the mean deviation between the predicted and true values ​​of all samples in the training dataset is calculated to obtain the average deviation of the training dataset. Subtracting this average deviation from the preliminary prediction result obtained from the weighted summation corrects the model's systematic error, effectively reducing the overall prediction bias.

[0052] The test dataset was then input into the initially constructed hybrid prediction model. By evaluating the model's prediction accuracy on the test dataset, the hybrid prediction model was iteratively optimized in reverse until its prediction accuracy on the test dataset met a preset threshold. Finally, a target hybrid prediction model was obtained that balances the stability of the kernel ridge regression prediction sub-model with the nonlinear fitting capability of the radial basis function neural network prediction sub-model. Forecasting using this target hybrid prediction model effectively ensured both prediction accuracy and efficiency.

[0053] S103: Input the configuration data of the underwater vehicle to be predicted into the target hybrid prediction model to obtain the flow noise prediction result of the underwater vehicle to be predicted.

[0054] In the application, the configuration data of the underwater vehicle to be predicted is first preprocessed in a manner identical to that of the training dataset to ensure that the input data format fully matches the input requirements of the target fusion prediction model. Then, the preprocessed configuration feature parameters are synchronously input into the kernel ridge regression prediction sub-model and the radial basis function neural network prediction sub-model included in the target fusion prediction model. Each sub-model independently outputs its corresponding preliminary flow noise prediction result based on the mapping relationship learned during the training phase. Next, the predetermined determination coefficient weights of the two sub-models, calculated based on the training dataset and already fixed in the target fusion prediction model, are called. Following a preset weighted summation rule, the prediction result of the kernel ridge regression prediction sub-model is multiplied by its corresponding determination coefficient weight, and then added to the prediction result of the radial basis function neural network prediction sub-model multiplied by its corresponding determination coefficient weight to obtain the weighted fused prediction value. Finally, the average deviation calculated during the model training phase is subtracted from this weighted fused prediction value to obtain the deviation-corrected prediction value.

[0055] Finally, the predicted values ​​after bias correction are denormalized or destandardized to restore them to the true far-field radiated noise sound pressure level. At the same time, full-band noise spectrum data or noise peak data of characteristic frequency bands can be output according to actual needs, and finally the flow noise prediction data of the underwater vehicle to be predicted is obtained.

[0056] The aforementioned rapid prediction method for underwater vehicle flow noise first employs numerical simulation to acquire far-field radiated noise data of multi-configuration underwater vehicles. After preprocessing, training and testing datasets are obtained. This approach concentrates numerical simulation resources on the core configuration to obtain high-precision data and provides a reliable data foundation for subsequent model training. After acquiring the training and testing datasets, a hybrid prediction model, including a kernel ridge regression prediction sub-model and a radial basis function neural network prediction sub-model, is trained using these datasets. The prediction results of the two sub-models are weighted and summed based on the coefficient of determination, and the average deviation of the training dataset is subtracted to obtain the final prediction result, effectively ensuring prediction accuracy. Finally, only the configuration data of the underwater vehicle to be predicted needs to be input into the trained target hybrid prediction model to quickly obtain flow noise prediction results. Therefore, this significantly reduces the computational resource consumption and time cost of single-configuration flow noise prediction, significantly improving the efficiency of large-scale underwater vehicle flow noise prediction. It can rapidly provide high-accuracy flow noise prediction results during the rapid design phase of underwater vehicles, solving the pain points of traditional numerical calculation methods.

[0057] In some embodiments, the far-field radiated noise data of a multi-configuration underwater vehicle is obtained using a numerical simulation method, including: setting the entire computational domain as a cylinder, setting the incoming and side surfaces of the computational domain as velocity inlets, and the outgoing surface as a pressure outlet; meshing the computational domain using a structured mesh; and based on the meshing results, calculating the instantaneous flow field and radiated noise of the multi-configuration underwater vehicle in the computational domain using large eddy simulation and acoustic analogy methods to obtain the far-field radiated noise data of the multi-configuration underwater vehicle.

[0058] In the application, the geometric parameters of the cylindrical computational domain are determined based on the maximum external dimensions of the underwater vehicle and the requirements for far-field noise monitoring. The radial range of the cylinder needs to cover the surface of the vehicle to the far-field noise monitoring sphere, and the axial length needs to meet the requirements of sufficient inflow development and sufficient wake diffusion, thereby avoiding the interference of boundary effects on the flow field and noise calculation results. Subsequently, the inflow surface of the computational domain and the side of the cylinder are set as the velocity inlet boundary. A uniform inflow velocity matching the actual navigation conditions of the underwater vehicle is input, and the medium physical property parameters corresponding to the navigation depth are set. The outflow surface of the computational domain is set as the pressure outlet boundary, and a static pressure value consistent with the underwater environment is applied to ensure that there is no backflow at the flow field outlet and that the calculation process is stable and converges.

[0059] like Figure 2 As shown, a structured grid is then used to mesh the computational domain. This allows for mesh refinement in critical areas with drastic flow field gradient changes, such as the vehicle surface, appendage connections, and wake region. Boundary layer meshes are deployed to accurately capture the distribution of wall shear stress and pulsating pressure. For the far-field region, a uniform gradient transition meshing strategy is adopted to control the computational load while ensuring the accuracy of noise propagation path calculations. Mesh independence verification is also conducted. By comparing the calculation results of key parameters such as the average pressure coefficient of the vehicle surface and the wake vortex shedding frequency under different mesh densities, the optimal mesh scheme that balances computational accuracy and efficiency is determined.

[0060] Finally, based on the mesh generation results, the Large Eddy Simulation (LES) method was used to solve the three-dimensional incompressible turbulent Navier-Stokes equations. Large-scale and small-scale vortices in the flow field were separated using a filtering function. The dissipation effect of small-scale vortices was modeled using a subgrid model, accurately capturing the generation, development, shedding, and dissipation processes of vortex structures in the instantaneous flow field around multi-configuration underwater vehicles. The time history data of fluctuating pressure on the surface and near-field region of the vehicle were obtained. Then, combined with acoustic analogy methods (such as the FW-H equations), the fluctuating pressure obtained from the flow field calculation was used as the acoustic equivalent source term and substituted into the acoustic control equations considering the acoustic propagation characteristics of the underwater medium. The time history data of radiated noise sound pressure at each measuring point on the far-field monitoring sphere were obtained through integral calculation. The noise spectrum characteristics were obtained through Fourier transform. Finally, the far-field radiated noise data of different configuration underwater vehicles under corresponding operating conditions were summarized, providing basic data support for the subsequent construction of training and testing datasets.

[0061] It should be noted that large eddy simulation divides physical quantities into filtered values ​​in space. and subgrid values ,Right now:

[0062] It should also be noted that in the relevant formulas below, the subscripts i, j, and k refer to the x, y, and z directions, respectively. Different subscripts mean that the physical meaning of the symbols remains the same, but the directions they refer to are different. The "-" above the symbols represents the average value. Inserting the turbulent kinetic energy transport equation into the incompressible Navier-Stokes equations can also simplify it to the same form as the unsteady Reynolds-averaged Navier-Stokes (URANS) model:

[0063]

[0064] Where t is time, Where μ is the fluid density and μ is the fluid viscosity. For small-scale stress, For fluid pressure, The average velocity in the x-direction is... The average velocity in the y-direction is... , These represent the independent variables along the x and y coordinate axes. This paper adopts the WALE subgrid-scale stress model, whose expression is:

[0065]

[0066] The strain tensor represents large-scale motion. The symbol for Kronecker. The small-scale eddy viscosity coefficient is expressed as follows:

[0067] in, It is a constant. For unit scale, It is related to the vorticity tensor and strain rate, and the specific relationship is as follows:

[0068] in, For the isotropic contraction term, the vorticity tensor is given by the equation. The expression is:

[0069] Hydrodynamic noise and aerodynamic noise both belong to fluid dynamics noise. The theoretical methods and tools for calculating hydrodynamic noise of aircraft in this invention are basically derived from a hybrid approach of computational fluid dynamics (CFD) and computational hydroacoustics (CHA). The hybrid approach can meet the accuracy requirements while having relatively low computational resource and time costs, and is therefore widely used in engineering. The integral and variational forms of the Lighthill acoustic analogy in the hybrid approach are introduced below.

[0070] Lighthill, focusing on jet aircraft turbulence noise, restructured the Navier-Stokes equations, rewriting them as a wave equation with a generalized source term:

[0071] In the formula The reference speed of sound in the fluid. For the Lighthill stress tensor:

[0072] In the above formula, the first term is the Reynolds stress, the second term represents the entropy fluctuation contribution, which can be ignored for isentropic flow, and the third term... It is the viscous stress tensor. and For pressure and fluid density under turbulent disturbance, and The pressure and density are those under undisturbed conditions. For flow rate, Let be the Dirichlet function. In isentropic fluids with high Reynolds numbers and low Mach numbers, the Lighthill tensor can be approximated as: .

[0073] Combining the mass conservation equation and the momentum equation, the wave equation can be written as:

[0074] Before applying the Lighthill acoustic analogy, it is necessary to assume that sound and fluid are not coupled, meaning the influence of sound on the flow field is negligible. Since the Lighthill stress tensor and density fluctuations are independent, this makes the equations true wave equations. The Lighthill equations only consider turbulent quadrupole noise under free flow. Then, Curle considered the effect of stationary solid boundaries on noise in the flow field and, using Kirchhoff integration, derived the Curle equations considering the influence of solid walls. Ffowcs-William and Hawkings introduced the Heaviside generalized function to extend Curle to the problem of sound generation by moving objects in a flow field, obtaining the FW-H equations:

[0075]

[0076] The first term on the right side of the equation is the spatial turbulence source term in the Lighthill equation theory, corresponding to quadrupole noise; the second term is the dipole noise source term caused by pressure fluctuations between the flow field and the solid wall; and the third term is the monopole noise source term caused by the solid wall velocity. For the Heaviside function; surface =0 represents the surface of the computational model.

[0077] In some embodiments, based on the mesh generation results, the instantaneous flow field and radiated noise of the multi-configuration underwater vehicle in the computational domain are calculated using large eddy simulation (LES) and acoustic analogy methods to obtain the far-field radiated noise data of the multi-configuration underwater vehicle. This includes: using the meshed cylindrical computational domain as the simulation space, performing unsteady flow field calculations using LES with a first time step until the flow field reaches a stable state; after the flow field stabilizes, switching to a second time step to calculate the instantaneous flow field data; using the instantaneous flow field data as the input source, solving for the radiated noise of the multi-configuration underwater vehicle at multiple preset field points using acoustic analogy methods; and collecting the total sound level and sound pressure level spectrum data of each configuration underwater vehicle at each preset field point, summarizing them to obtain the far-field radiated noise data of the multi-configuration underwater vehicle. The first time step is larger than the second time step, and the LES method uses a LES turbulence model combined with a sub-grid scale model as the computational model, with a bounded-central difference scheme used for separation.

[0078] For example, the inflow surface and lateral surfaces of the computational domain can be 20m from the bow, and the outflow surface can be 60m from the bow, eliminating the influence of the boundaries on the numerical simulation results. Figure 2As shown, a structured mesh is used for mesh generation. To ensure a smooth mesh transition, a three-layer volumetric mesh transition control is employed, with the mesh size increasing sequentially from the innermost to the outermost refinement layers (refine1-refine3), for example, 0.1m, 0.4m, and 0.8m respectively. Due to the involvement of multiple submarine hull shapes, to ensure the universality of the mesh scheme, the first layer of refinement mesh is set as a cylinder with a length of 26m and a radius of 1.6m, making it adaptable to all submarine models within the design parameter range and ensuring the uniformity of the mesh scheme. Furthermore, for example, during the calculation, a larger time step of 0.01s can be used initially to calculate the flow field, and after the calculation stabilizes, a smaller time step of 0.0005s can be used to calculate the radiated noise.

[0079] In this embodiment, a cylindrical computational domain after meshing is used as the simulation space. A large eddy simulation turbulence model is configured, paired with a sub-mesh-scale model as the core computational model. A bounded-central difference scheme is employed as the separation scheme to ensure computational accuracy and numerical stability. In the unsteady flow field calculation process, a relatively large first time step is used for iterative calculations to quickly propel the flow field towards a stable state. The criteria for determining the stable state of the flow field are that the fluctuation amplitude of key parameters such as the average pressure coefficient of the vehicle surface, the vortex distribution characteristics of the wake region, and the intensity of near-field velocity fluctuations are below a preset threshold over multiple consecutive iteration cycles. Once the flow field reaches a stable state, a smaller second time step is used to continue the calculation. This second time step must match the sub-mesh-scale model's requirement for capturing small-scale vortex fluctuations, ensuring accurate acquisition of high-precision instantaneous flow field data such as the time history of the vehicle surface fluctuation pressure and the instantaneous distribution of near-field velocity and vortex.

[0080] Subsequently, using instantaneous flow field data as the acoustic equivalent input source, a coupled computational model of the flow field and sound field was constructed using acoustic analogy methods (such as the FW-H equation). The pulsating pressure on the surface of the underwater vehicle was substituted into the acoustic control equation as the sound radiation source term to solve for the radiated noise of the multi-configuration underwater vehicle at multiple preset far-field monitoring points. The preset points need to be arranged on the far-field monitoring sphere according to the principle of uniform distribution, covering different azimuth and pitch angles to comprehensively obtain the noise radiation characteristics. Finally, for each configuration of the underwater vehicle, the total noise level data and sound pressure level spectrum data of each preset point were collected, and the noise data of all points were sorted and summarized according to the configuration. Finally, the far-field radiated noise data of the multi-configuration underwater vehicle can be obtained.

[0081] In some embodiments, the number of preset field points is six. The reference coordinate system is set at the bow of the underwater vehicle, and the coordinate positions of the six preset field points are (-100,0,0), (100,0,0), (0,-100,0), (0,100,0), (0,0,-100), and (0,0,100). It should be noted that hydrophones can be deployed at the preset field points to collect radiated noise data at the corresponding field points. Figure 3 As shown, Figure 3 This is a schematic diagram showing the position of the hydrophone and the reference coordinate system corresponding to the preset field point.

[0082] It is understandable that the reference coordinate system is set at the bow of the underwater vehicle, and six preset monitoring points are arranged in the far field region with coordinate positions of (-100,0,0), (100,0,0), (0,-100,0), (0,100,0), (0,0,-100), and (0,0,100). These six points correspond to six directions: directly in front of the bow, directly behind the tail, directly to the left, directly to the right, directly below the bottom, and directly above the top of the vehicle. This can fully cover the key directions of the vehicle's far field noise radiation, thereby accurately solving the radiated noise of the multi-configuration underwater vehicle at the six preset points.

[0083] In some embodiments, before acquiring far-field radiated noise data of the multi-configuration underwater vehicle using numerical simulation methods, the method further includes: acquiring design parameter types and corresponding parameter ranges; and constructing underwater vehicle models of multiple configurations based on the design parameter types and corresponding parameter ranges. The design parameter types include overall length, overall width, inlet section length, outlet section length, longitudinal profile coefficient in the inlet section, longitudinal profile coefficient in the transition section, and longitudinal profile coefficient in the outlet section.

[0084] For example, the dataset can be established by designing 55 underwater vehicle configurations using 7 design parameters, and then calculating the total sound level and sound pressure level spectrum of the flow noise at six field points for each of the 55 underwater vehicle configurations at a speed of 4 km / h using numerical simulation methods based on large eddy simulation and acoustic analogy theory (such as the FW-H equation). This data serves as the dataset for this method. Specific design parameters and their design ranges are shown in Table 1, and the segmented schematic diagram of the underwater vehicle can be seen as follows. Figure 4 As shown.

[0085] Table 1 Design parameters and their design range

[0086] In some embodiments, training the hybrid prediction model using a training dataset and a test dataset to obtain a target hybrid prediction model includes: training the kernel ridge regression prediction sub-model and the radial basis function neural network prediction sub-model in the hybrid prediction model using the training dataset, and verifying the prediction accuracy of the hybrid prediction model using the test dataset; if the prediction accuracy reaches a preset target, outputting the target hybrid prediction model; if the prediction accuracy does not reach the preset target, adjusting the accuracy-related parameters in the hybrid prediction model and repeating the training steps. The accuracy-related parameters include regularization parameters, kernel width, and the number of hidden layer nodes.

[0087] For example, the adjustable parameter is the regularization parameter. nuclear width Given the number of hidden layer nodes h, it can be initially set to 20.

[0088] It is understandable that training the hybrid prediction model using training and test datasets involves first decomposing the training dataset into an input feature set. Based on this input-output pair, targeted training is then conducted on the kernel ridge regression prediction sub-model and the radial basis function neural network prediction sub-model within the hybrid prediction model. After completing the first round of training, the test dataset is input into the initially trained hybrid prediction model. The prediction accuracy of the hybrid prediction model is verified by calculating the quantitative accuracy index of the model's prediction results against the actual noise labels on the test dataset. For example, the preset target can be set as follows: the error in total sound level prediction accuracy compared to numerical calculation is within 5 dB, and the normalized root mean square error of sound pressure level spectrum prediction is within 0.66 (which can be adjusted according to actual engineering needs).

[0089] If the predicted accuracy obtained from the verification meets the preset target, it indicates that the model has sufficient generalization ability and prediction accuracy, and this hybrid prediction model is directly output as the target hybrid prediction model. If the predicted accuracy does not meet the preset target, the accuracy-related parameters in the hybrid prediction model are adjusted accordingly. The regularization parameter is used to optimize the overfitting problem of the kernel ridge regression sub-model, the kernel width is used to adjust the mapping range of the kernel function of the kernel ridge regression sub-model, and the number of hidden layer nodes is used to optimize the nonlinear fitting ability of the radial basis function neural network sub-model. Parameter adjustment can be performed using efficient optimization methods such as grid search and Bayesian optimization to determine new parameter combinations, or it can be adjusted manually. Then, the two sub-models are retrained based on the training dataset using the adjusted accuracy-related parameters, and the predicted accuracy is verified again using the test dataset. This iterative process of training-verification-parameter adjustment is repeated until the predicted accuracy of the hybrid prediction model meets the preset target, ultimately obtaining the target hybrid prediction model with the best performance.

[0090] Based on the above embodiments, the process of the rapid prediction method for underwater vehicle flow noise can be as follows: Figure 5As shown in the figure. Based on this process, the total sound level and sound pressure level spectra of the six fields of the Type 55 underwater vehicle (i.e., the Type 55 configuration underwater vehicle designed in the aforementioned example) are predicted. The distribution of the predicted total sound level of the Type 55 underwater vehicle at the six fields is shown in the figure. Figure 6 As shown in the diagram, the error distribution of the total sound level prediction results at six field points for various underwater vehicle configurations is illustrated in the figure below. Figure 7 As shown. Furthermore, Figure 8 and Figure 9 The results show the sound pressure level spectrum and its normalized root mean square error distribution at six points on a specific underwater vehicle configuration in the test set. Through the above experiments, it is verified that, compared with traditional numerical calculation methods, this method reduces the flow noise prediction efficiency of underwater vehicle configurations within the design parameter range from 2-3 days (using a 128-core CPU, Large Eddy Simulation method, 10 million grids, and a computation time step of 1e-5 seconds) to less than 5 minutes after model training. The overall sound level prediction accuracy is within 5 dB of numerical calculation, and the normalized root mean square error of the sound pressure level spectrum prediction is around 0.66, enabling accurate prediction of the overall trend of flow noise sound pressure level spectrum changes in a short time.

[0091] In some embodiments, please refer to Figure 10 This application provides a rapid prediction device 100 for underwater vehicle flow noise, including: an acquisition module 101, a training module 102 and a prediction module 103.

[0092] The acquisition module 101 is used to acquire far-field radiated noise data of multi-configuration underwater vehicles using numerical simulation methods, and to preprocess the far-field radiated noise data to obtain training datasets and test datasets.

[0093] The training module 102 is used to train the hybrid prediction model using the training dataset and the test dataset to obtain the target hybrid prediction model. The hybrid prediction model includes a kernel ridge regression prediction sub-model and a radial basis function neural network prediction sub-model. The prediction result of the hybrid prediction model is obtained by weighting and summing the prediction results of the two prediction sub-models based on the coefficient of determination and subtracting the average deviation of the training dataset.

[0094] The forecast module 103 is used to input the configuration data of the underwater vehicle to be forecasted into the target hybrid prediction model to obtain the flow noise forecast result of the underwater vehicle to be forecasted.

[0095] It should be noted that the underwater vehicle flow noise rapid prediction device 100 provided in this application embodiment and the underwater vehicle flow noise rapid prediction method provided in this application embodiment are based on the same inventive concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned underwater vehicle flow noise rapid prediction method, and the repeated parts will not be described again.

[0096] In some embodiments, an electronic device provided in this application includes a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the above-described method for rapid prediction of underwater vehicle flow noise.

[0097] Specifically, the processor may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0098] Memory can be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, memory can include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, instruments, or propagation media. Specific examples of memory include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and also random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0099] This application also provides a non-transitory computer storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for rapid prediction of underwater vehicle flow noise. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method as described in the embodiments of this application.

[0100] According to embodiments of this application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.

[0101] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0102] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for rapid prediction of underwater vehicle flow noise, characterized in that, include: Numerical simulation was used to obtain far-field radiated noise data of a multi-configuration underwater vehicle, and the far-field radiated noise data was preprocessed to obtain training datasets and test datasets. The hybrid prediction model is trained using the training dataset and the test dataset to obtain the target hybrid prediction model. The hybrid prediction model includes a kernel ridge regression prediction sub-model and a radial basis function neural network prediction sub-model. The prediction result of the hybrid prediction model is obtained by weighting and summing the prediction results of the two prediction sub-models based on the coefficient of determination, and then subtracting the average deviation of the training dataset. The configuration data of the underwater vehicle to be predicted is input into the target hybrid prediction model to obtain the flow noise prediction result of the underwater vehicle to be predicted.

2. The rapid prediction method for underwater vehicle flow noise as described in claim 1, characterized in that, The method of obtaining far-field radiated noise data of multi-configuration underwater vehicles using numerical simulation includes: The computational domain is set as a cylinder, and the inflow surface and side surface of the computational domain are set as velocity inlets, and the outflow surface is set as pressure outlets; The computational domain is divided into grids using a structured grid. Based on the mesh generation results, the instantaneous flow field and radiated noise of the multi-configuration underwater vehicle in the computational domain are calculated using large eddy simulation and acoustic analogy methods, and the far-field radiated noise data of the multi-configuration underwater vehicle are obtained.

3. The rapid prediction method for underwater vehicle flow noise as described in claim 2, characterized in that, Based on the mesh generation results, the instantaneous flow field and radiated noise of the multi-configuration underwater vehicle in the computational domain are calculated using large eddy simulation and acoustic analogy methods, yielding far-field radiated noise data of the multi-configuration underwater vehicle, including: Using a cylindrical computational domain after meshing as the simulation space, the Large Eddy Simulation (LES) method is first used to calculate the unsteady flow field at a first time step until the flow field reaches a steady state. After the flow field stabilizes, the calculation is switched to a second time step to calculate the instantaneous flow field data. The first time step is longer than the second time step. The LES method uses a LES turbulence model combined with a subgrid-scale model as the calculation model, and the separation scheme adopts a bounded-central difference scheme. Using the instantaneous flow field data as the input source, the acoustic analogy method is used to solve the radiated noise of the multi-configuration underwater vehicle at multiple preset field points; The total sound level and sound pressure level spectrum data of each underwater vehicle configuration were collected at each preset field point, and the far-field radiated noise data of the multi-configuration underwater vehicle were obtained by summarizing them.

4. The rapid prediction method for underwater vehicle flow noise as described in claim 2, characterized in that, The structured mesh is controlled by a three-layer volumetric mesh transition, with the size of the volumetric mesh increasing sequentially from the inside out.

5. The rapid prediction method for underwater vehicle flow noise as described in claim 3, characterized in that, The number of preset field points is six; The reference coordinate system is set at the bow of the underwater vehicle, and the coordinate positions of the six preset field points are (-100,0,0), (100,0,0), (0,-100,0), (0,100,0), (0,0,-100), and (0,0,100).

6. The rapid prediction method for underwater vehicle flow noise as described in claim 1, characterized in that, Before using numerical simulation to obtain far-field radiated noise data of a multi-configuration underwater vehicle, the method further includes: Obtain the design parameter types and corresponding parameter ranges, wherein the design parameter types include total length, total width, inlet section length, outlet section length, longitudinal section coefficient in the inlet section, longitudinal section coefficient in the transition section, and longitudinal section coefficient in the outlet section; Multiple underwater vehicle models with different configurations are constructed based on the design parameter types and corresponding parameter ranges.

7. The rapid prediction method for underwater vehicle flow noise as described in claim 1, characterized in that, The step of training the hybrid prediction model using the training dataset and the test dataset to obtain the target hybrid prediction model includes: The kernel ridge regression prediction sub-model and radial basis function neural network prediction sub-model in the hybrid prediction model are trained using the training dataset, and the prediction accuracy of the hybrid prediction model is verified using the test dataset. If the prediction accuracy reaches the preset target, output the target hybrid prediction model; If the prediction accuracy does not reach the preset target, the accuracy correlation parameters in the hybrid prediction model are adjusted and the training is repeated. The accuracy correlation parameters include regularization parameters, kernel width, and the number of hidden layer nodes.

8. A rapid prediction device for underwater vehicle flow noise, characterized in that, include: The acquisition module is used to acquire far-field radiated noise data of a multi-configuration underwater vehicle using numerical simulation methods, and to preprocess the far-field radiated noise data to obtain a training dataset and a test dataset. The training module is used to train the hybrid prediction model using the training dataset and the test dataset to obtain the target hybrid prediction model. The hybrid prediction model includes a kernel ridge regression prediction sub-model and a radial basis function neural network prediction sub-model. The prediction result of the hybrid prediction model is obtained by weighting and summing the prediction results of the two prediction sub-models based on the coefficient of determination, and then subtracting the average deviation of the training dataset. The forecasting module is used to input the configuration data of the underwater vehicle to be predicted into the target hybrid prediction model to obtain the flow noise forecasting result of the underwater vehicle to be predicted.

9. An electronic device, characterized in that, It includes a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the rapid prediction method for underwater vehicle flow noise as described in any one of claims 1 to 7.

10. A non-transitory computer storage medium, characterized in that, It stores a computer program, wherein the computer program, when executed by a processor, implements the method for rapid prediction of underwater vehicle flow noise as described in any one of claims 1 to 7.

Images