Method for rapidly predicting hydrodynamic load of a vehicle under internal wave influence

Abstract

The application discloses a method for rapidly predicting hydrodynamic load of a sailing body under the influence of internal waves, and belongs to the technical field of hydrodynamic load prediction of the coupling of ocean internal waves and a sailing body. The method steps comprise the following steps: establishing a numerical model of the coupling of internal solitary waves and a sailing body; constructing a mapping relationship between time and spatial position, converting time-domain load data into spatial position load data; data preprocessing and division; constructing and training a polynomial chaos expansion model; calculating a prediction residual based on each load component predicted by the polynomial chaos expansion model; performing Gaussian process regression correction on the standardized prediction residual; superimposing the prediction value output by the polynomial chaos expansion model and the correction value to obtain a rapid prediction model of the hydrodynamic load of the sailing body; and load prediction and model verification. The application solves the technical problem of real-time prediction of the load of the sailing body under the influence of internal waves, and solves problems such as complicated steps, high time cost and poor timeliness.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of water dynamic load prediction of the coupling effect of internal waves and a sailing body, and particularly relates to a fast water dynamic load prediction method for a sailing body under the influence of internal waves. BACKGROUND

[0002] In a real marine environment, the uneven heating of seawater by solar radiation and the dynamic and thermodynamic effects of the atmosphere in different climate zones result in a regular macroscopic hierarchical structure of the temperature, salinity and density of seawater in the vertical direction in different sea areas, forming a stable and continuous density jump. When the density jump is disturbed by external forces, internal solitary waves (i.e. internal waves) are excited. Compared with the waves on the sea surface, internal solitary waves have a larger amplitude and a lower frequency. Moreover, internal solitary waves have huge energy, and they can maintain a fixed wave shape and wave speed during long-distance propagation in the ocean. In the induced flow field of internal solitary waves, sudden strong currents and significant convergence and divergence are caused, and thus strong shear occurs on the upper and lower sides of the density jump, which poses a serious threat to the safe navigation of a sailing body. According to existing data, more than 30% of submarine accidents are related to the action of internal solitary waves. In the ocean, internal solitary waves are a key factor that cannot be ignored in the safe navigation of a sailing body, and therefore it is crucial to accurately predict the water dynamic load of a sailing body under the action of internal waves.

[0003] At present, this field mainly relies on two types of methods, but both have serious limitations:

[0004] Computational fluid dynamics (CFD) method:

[0005] The wave-body coupling effect can be simulated by solving RANS / LES equations, and although the load data under a specific working condition can be obtained, 12-24 hours are needed for a single high-precision simulation;

[0006] When the working condition parameters such as wave amplitude and diving depth change, the grid and boundary conditions need to be repeatedly set, and the millisecond-level response requirement of real-time maneuvering decision cannot be met.

[0007] Data-driven proxy model:

[0008] The reduced order model (ROM) relies on the steady background flow field assumption and cannot handle the spatiotemporal coupling effect of the dynamic propagation of internal waves;

[0009] The black box model such as neural network needs a large amount of training samples, and the high-fidelity data of internal solitary wave-sailing body coupling are scarce, resulting in poor generalization ability and significant residual under small sample.

[0010] Therefore, a new method that combines physical mechanism and data-driven is urgently needed to break through the three technical barriers of real-time, dynamics and small sample accuracy, and to provide core support for the safety prevention and control of internal waves of a sailing body. SUMMARY

[0011] Technical problems to be solved:

[0012] In order to avoid the shortcomings of the prior art, the present application provides a kind of under the influence of internal wave fast prediction method of hydrodynamic load of navigation body, based on the regression technique of data-driven problem of polynomial chaos expansion (PCE), by introducing additional interpolation type Kriging (Kriging) for error correction of prediction model, to enhance the prediction ability of the model constructed, form the fast prediction model of the hydrodynamic load of navigation body under the influence of internal wave, reach the fast prediction of the hydrodynamic load of navigation body under the environment of ocean internal wave, provide guidance for the safe navigation of navigation body. Through the innovative combination of space-time data conversion and PCE-Kriging residual correction, the technical bottleneck of real-time prediction of navigation body load under internal wave environment is solved, such as tedious steps, high time cost, poor timeliness and the like.

[0013] The technical scheme of the present application is: a kind of under the influence of internal wave fast prediction method of hydrodynamic load of navigation body, comprising the following steps:

[0014] A numerical model of internal solitary wave-navigation body coupling is established, the wave amplitude of internal solitary wave and the depth of navigation body under different working conditions are set, the interaction of internal solitary wave-navigation body under different working conditions is simulated, and the time domain load data of the load component of navigation body is extracted;The load component includes horizontal force, vertical force and torque;

[0015] The mapping relationship between time and space position is constructed, and the time domain load data is converted into spatial position load data;The mapping relationship between time and space position is expressed as follows:

[0016]

[0017] In the formula, represents the initial distance of internal solitary wave trough from navigation body at 0 time, c represents the propagation speed of internal solitary wave, represents the relative distance of navigation body from internal solitary wave trough, t is the propagation time of internal solitary wave;

[0018] The wave amplitude of internal solitary wave, the depth data of navigation body under different working conditions and the spatial position load data are constructed as initial sample data set, the initial sample data set is standardized to obtain sample data set, and the sample data set is divided into training set and validation set;

[0019] A polynomial chaos expansion model is constructed and trained, the input is the wave amplitude of internal solitary wave, the depth of navigation body and the relative distance of navigation body from internal solitary wave trough, and the output is the predicted value of horizontal force, vertical force and torque of navigation body;

[0020] The prediction residual is calculated based on each load component predicted based on the polynomial chaos expansion model; the prediction residual is the difference between the true value and the predicted value of the horizontal force, vertical force and moment of the sailing body;

[0021] The standardized prediction residual is corrected by using the interpolation Kriging model for Gaussian process regression, so that the corrected value of the horizontal force, vertical force and moment of the sailing body is obtained.

[0022] The prediction value output by the polynomial chaos expansion model is superimposed with the corrected value obtained by the interpolation Kriging model, so that the rapid prediction model of the hydrodynamic load of the sailing body is obtained; and the final load prediction of the horizontal force, vertical force and moment is output.

[0023]

[0024]

[0025]

[0026] In the formula, F h-final The predicted value of the horizontal force of the sailing body is represented by Fx, F v-final The predicted value of the vertical force of the sailing body is represented by Fz, F M-final The predicted value of the moment of the sailing body is represented by M, a The wave amplitude of the internal solitary wave is represented by a, d The diving depth of the sailing body is represented by h, The relative distance of the sailing body from the trough of the internal solitary wave is represented by x; The horizontal force prediction value output by the polynomial chaos expansion model is represented by Fx, The vertical force prediction value output by the polynomial chaos expansion model is represented by Fz, The moment prediction value output by the polynomial chaos expansion model is represented by M; The correction value of the horizontal force is represented by Fx, The correction value of the vertical force is represented by Fz, The correction value of the moment is represented by M.

[0027] The performance of the rapid prediction model of the hydrodynamic load of the sailing body is evaluated by the root mean square error and the correlation coefficient, and the rapid prediction of the hydrodynamic load of the sailing body under the influence of internal waves is completed.

[0028] A further technical solution of the present application is that the numerical model of the internal solitary wave-sailing body coupling effect is obtained by using the Reynolds-averaged Navier-Stokes equation RANS or large eddy simulation LES, and the initial flow field of the internal solitary wave is obtained by solving the DJL internal solitary wave theoretical equation through the DJLES open source software package.

[0029] A further technical solution of the present application is that the polynomial chaos expansion model expression is as follows:

[0030]

[0031]

[0032]

[0033] In the formula, represents a predicted value of a horizontal force on the vehicle, represents a predicted value of a vertical force on the vehicle, represents a predicted value of a moment on the vehicle; is an index set corresponding to the polynomial, defines the structure of the polynomial basis function, and represents the structure and order of all possible polynomial terms in the polynomial expansion; represents the order of each variable; represents an orthogonal polynomial basis function based on . is a polynomial expansion coefficient, and represents the weight of the polynomial term.

[0034] A further technical solution of the present application is that the expression of the prediction residual is as follows:

[0035]

[0036]

[0037]

[0038] In the formula, represents a true value of a horizontal force on the vehicle, represents a true value of a vertical force on the vehicle, represents a true value of a moment on the vehicle; represents a predicted residual value of a horizontal force on the vehicle, represents a predicted residual value of a vertical force on the vehicle, represents a predicted residual value of a moment on the vehicle;

[0039] The prediction residual is normalized as follows:

[0040]

[0041]

[0042]

[0043] In the formula, , , respectively represent the mean of the residual error corresponding to the horizontal force, vertical force and moment, , , respectively represent the standard deviation of the residual error corresponding to the horizontal force, vertical force and moment, , , respectively represent the normalized predicted residual error value corresponding to the horizontal force, vertical force and moment.

[0044] A further technical solution of the present application is that the interpolation type Kriging method model corrects the expression of the predicted residual error by Gaussian process regression as follows:

[0045]

[0046] In the formula, represents the mean of the predicted residual error of all samples; represents the covariance matrix, and the element is , wherein, represents the noise variance hyperparameter, represents the Kronecker symbol; represents the similarity of and calculated by the kernel function; represents the input variable of the new sample to be predicted at present, which is used for model prediction; represents the input value of the i-th sample in the training set, which is used for model training and kernel function calculation; i represents the input value of the i-th sample in the training set; represents the input value of the i-th sample in the training set; j respectively correspond to , , , ; respectively correspond to , , , n represents the total number of samples in the training set;

[0047] The kernel function adopts a composite form as shown below:

[0048]

[0049] In the formula, is the signal variance, which is used to control the output amplitude of the kernel function; represents the length scale hyperparameter for controlling the amplitude of the internal solitary wave directional smoothness, represents the length scale hyperparameter for controlling the potential depth of the vehicle, dThe length-scale hyperparameter of directional smoothness Indicates the relative distance between the control vehicle and the trough of an isolated wave. The length-scale hyperparameter of directional smoothness; Indicates the first i , j The amplitude of isolated waves within a sample. Indicates the first i , j The diving depth of a sample navigation body 、 Indicates the first i , j The relative distance between each sample vessel and the isolated wave trough.

[0050] A further technical solution of the present invention is: the method for standardizing the initial sample dataset is to use StandardScaler to perform z-score standardization on the input variables, the input variables including the amplitude of the inner solitary wave, the depth of the vehicle, and the relative distance between the vehicle and the trough of the inner solitary wave.

[0051] A further technical solution of the present invention is: the evaluation method of the rapid prediction model of hydrodynamic loads of the vehicle body includes,

[0052] Calculate the root mean square error (RMSE):

[0053]

[0054] Calculate the correlation coefficient R:

[0055]

[0056] In the formula, M It is the total number of samples in the validation set. y i This represents the true response value of each load component. This represents the response predicted by the rapid prediction model of hydrodynamic loads on the vehicle body; , ;

[0057] Verification criteria: When the correlation coefficient R ≥ 0.9 and RMSE ≤ 3% of the mean of the actual value, the predicted value and the actual value have a high degree of linearity and have high-quality predictive ability.

[0058] A rapid prediction system for hydrodynamic loads on a vehicle under the influence of internal waves includes:

[0059] Data acquisition module: a numerical model of internal solitary wave-vehicle coupling is established to simulate the interaction between internal solitary wave and vehicle under different working conditions; the time-domain load data of the horizontal force, vertical force and moment of the vehicle are extracted;

[0060] Space-time conversion module: connecting the data acquisition module, the time-domain load data is converted into spatial position load data;

[0061] Preprocessing module: connecting the space-time conversion module, the initial sample data is standardized;

[0062] Hybrid agent model module: connecting the preprocessing module, including constructing a polynomial chaos expansion model and an interpolation Kriging model, superimposing the prediction value output by the polynomial chaos expansion model and the correction value obtained by the interpolation Kriging model to obtain a rapid prediction model of the hydrodynamic load of the vehicle; output the final load prediction of the horizontal force, vertical force and moment;

[0063] Verification module: connecting the hybrid agent model module, determining the validity of the rapid prediction model of the hydrodynamic load of the vehicle.

[0064] An electronic device comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to realize the steps of the rapid prediction method of the hydrodynamic load of the vehicle under the influence of internal waves.

[0065] A computer-readable storage medium storing a computer program, wherein the computer program is executed to realize the steps of the rapid prediction method of the hydrodynamic load of the vehicle under the influence of internal waves.

[0066] Advantages

[0067] The present application has the advantages that: by constructing a PCE-Kriging hybrid agent model, the load prediction time is shortened from hours to milliseconds (only 111.8 milliseconds in the actual measurement case). After loading the model, only the amplitude of the internal solitary wave, the depth of the vehicle and the relative distance of the vehicle from the trough of the internal solitary wave are input, and the hydrodynamic load (horizontal force, vertical force and moment) at any position can be output in real time, providing real-time data support for the safe operation of the vehicle. The specific advantages are as follows:

[0068] 1. Based on the physical characteristics of the constant propagation speed of internal solitary waves, the space-time conversion formula is proposed to convert the time-domain load data into a spatial position function, and the dynamic internal solitary wave-vehicle coupling problem is converted into a static spatial position problem, which significantly improves the adaptability of the model to dynamic working conditions.

[0069] 2. The present application aims at PCE prediction error, adopts Gaussian process regression of composite kernel function, adaptively learns the local nonlinear characteristics of the amplitude of internal solitary wave, the depth of the underwater body and the relative distance of the underwater body from the internal solitary wave trough, and significantly reduces the residual error. The correlation coefficient R of the predicted value and the true value of the verification set is greater than or equal to 0.9, and the root mean square error RMSE is less than or equal to 0.3%.

[0070] 3. The present application outputs horizontal force, vertical force and torque data in real time, providing a theoretical basis for the underwater body to avoid internal wave danger area or adjust attitude. It can predict the sudden strong flow and shear effect induced by internal wave, and avoid accidents caused by "dropping deep" out of control. BRIEF DESCRIPTION OF DRAWINGS

[0071] Figure 1 The flowchart of the method for quickly predicting the hydrodynamic load of the underwater body in the embodiment of the present application is shown in the figure;

[0072] Figure 2 The schematic diagram of the numerical model of the internal solitary wave-underwater body coupling in the embodiment of the present application is shown in the figure;

[0073] Figure 3 The correlation between the predicted hydrodynamic force and the real hydrodynamic force of the quick prediction model of the hydrodynamic load of the underwater body in the embodiment of the present application is shown in the figure;

[0074] Figure 4 The comparison between the predicted value and the true value of the verification set in the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0075] The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present application, but cannot be understood as a limitation of the present application.

[0076] After the underwater body encounters internal solitary wave, the hydrodynamic load of its vertical surface will change sharply, the motion attitude of the underwater body will change sharply under the action of internal solitary wave, and the motion stability will be poor, and finally a serious "dropping deep" phenomenon will occur due to loss of control, and when the depth of the underwater body exceeds the limit depth of its submersion, the boat will be destroyed and the people will be killed. In order to avoid such accidents, at present, the field of underwater body hydrodynamic load under internal wave environment mainly relies on the following two methods, but both have significant defects:

[0077] The first method is computational fluid dynamics (CFD) method (such as CN201910154630.7, CN202410050749.0):

[0078] Although the wave-body coupling effect can be simulated by combining numerical wave making with RANS equations to obtain load data under specific working conditions, a single high-precision simulation requires several hours, and the grid and boundary conditions need to be repeatedly set when the working conditions change, which cannot meet the timeliness requirements of real-time operation decisions.

[0079] Although the existing improvement scheme (such as the "intervention module" of CN202410050749.0) attempts to compensate for the density jump error, it still cannot break through the CFD calculation efficiency bottleneck and lacks dynamic working condition adaptability.

[0080] The second method is a data-driven proxy model (such as CN202310113949.1 and CN202211474451.X):

[0081] The reduced order model (ROM) compresses the flow field dimension through singular value decomposition (such as CN202310113949.1), but is only suitable for steady-state background flow fields and cannot handle the spatio-temporal coupling effect of internal wave dynamic propagation.

[0082] The neural network model (such as the transfer learning of CN202211474451.X) relies on a large number of training samples, but high-fidelity data of internal wave-vehicle coupling is scarce, resulting in poor model generalization ability.

[0083] In summary, the existing technology has the following problems:

[0084] Timeliness defect: CFD method has high calculation cost and cannot realize millisecond-level real-time prediction;

[0085] Dynamic coupling is missing: existing proxy models do not consider the spatio-temporal mapping relationship between the propagation speed of internal solitary waves c The relative distance between the internal solitary wave and the wave trough of the vehicle ;

[0086] Small sample precision: traditional machine learning methods (such as pure PCE or neural networks) have significant residuals under limited data.

[0087] To break through the above technical bottlenecks, the present application proposes an internal wave load rapid prediction method that combines physical mechanism driving and data driving, based on the propagation speed of internal solitary waves cThe constant characteristics of the internal wave are established, the mapping function of the time domain load to the spatial position is established, the dynamic coupling problem is converted into a spatial position function, a polynomial chaos expansion model (PCE) is used to construct a basic load prediction model, a Kriging Gaussian process regression is introduced to accurately correct the prediction residual, and the accuracy degradation problem under small samples is solved, and the relative distance of the internal solitary wave trough from the internal solitary wave is taken as the input, and the horizontal force, vertical force and torque of the navigation body are synchronously output, so that the real-time prediction of any working condition is supported. The method shortens the prediction time from hours to milliseconds under the premise of ensuring the engineering accuracy, and provides core technical support for the safe operation of the navigation body. Through the advance prediction of the hydrodynamic load of the navigation body, the theoretical guidance for the effective operation of the navigation body is provided, and the safety accidents when encountering internal solitary waves are reduced. The specific technical scheme is as follows:

[0088] The application provides a kind of internal wave influence under the fast prediction method of navigation body hydrodynamic load, comprising the following steps:

[0089] A numerical model of internal solitary wave-navigation body coupling is established, the amplitude of internal solitary wave and the depth of navigation body under different working conditions are set, the interaction of internal solitary wave-navigation body under different working conditions is simulated, and the time domain load data of the load component of navigation body is extracted;The load component includes horizontal force, vertical force and torque;

[0090] The mapping relationship between time and space position is constructed, and the time domain load data is converted into spatial position load data;The mapping relationship between time and space position is expressed as follows:

[0091]

[0092] In the formula, represents the initial distance of internal solitary wave trough from navigation body at 0 time, c represents the propagation speed of internal solitary wave, represents the relative distance of navigation body from internal solitary wave trough, t is the propagation time of internal solitary wave;

[0093] The spatial position load data and the amplitude of internal solitary wave and the depth of navigation body under different working conditions are constructed as initial sample data set, the initial sample data set is standardized to obtain sample data set, and the sample data set is divided into training set and validation set;

[0094] A polynomial chaos expansion model is constructed and trained, the amplitude of internal solitary wave, the depth of navigation body and the relative distance of navigation body from internal solitary wave trough are input, and the prediction values of horizontal force, vertical force and torque of navigation body are output;

[0095] The prediction residual is calculated based on each load component predicted by the polynomial chaos expansion model, and the prediction residual is the difference between the true value and the predicted value of the horizontal force, vertical force and moment of the sailing body;

[0096] The standardized prediction residual is corrected by the Gaussian process regression of the interpolation Kriging model to obtain the corrected value of the horizontal force, vertical force and moment of the sailing body;

[0097] The prediction value output by the polynomial chaos expansion model is superimposed with the corrected value obtained by the interpolation Kriging model to obtain a rapid prediction model of the hydrodynamic load of the sailing body; and the horizontal force, vertical force and moment of the final load prediction are output.

[0098]

[0099]

[0100]

[0101] In the formula, F h-final represents the predicted value of the horizontal force acting on the sailing body, F v-final represents the predicted value of the vertical force acting on the sailing body, F M-final represents the predicted value of the moment acting on the sailing body, a represents the wave amplitude of the internal solitary wave, d is the diving depth of the sailing body, is the relative distance of the sailing body from the trough of the internal solitary wave; represents the horizontal force prediction value output by the polynomial chaos expansion model, represents the vertical force prediction value output by the polynomial chaos expansion model, represents the moment prediction value output by the polynomial chaos expansion model; represents the correction value of the horizontal force, represents the correction value of the vertical force, represents the correction value of the moment;

[0102] The performance of the rapid prediction model of the hydrodynamic load of the sailing body is evaluated by the root mean square error and the correlation coefficient, and the rapid prediction of the hydrodynamic load of the sailing body under the influence of internal waves is completed.

[0103] In one embodiment, the numerical model of the internal solitary wave-sailing body coupling is obtained by using the Reynolds-averaged Navier-Stokes equation (RANS) or large eddy simulation (LES), and the initial flow field of the internal solitary wave is obtained by solving the DJL internal solitary wave theory equation by using the DJLES open source software package.

[0104] A further technical solution of the present invention is: the expression of the polynomial chaotic expansion model is as follows:

[0105]

[0106]

[0107]

[0108] In the formula, This represents the predicted horizontal force acting on the vehicle. This represents the predicted vertical force acting on the vehicle. This indicates the predicted value of the torque acting on the vehicle. For the index set corresponding to the polynomial, define the structure of the polynomial basis functions, and represent the structure and order of all possible polynomial terms in the polynomial expansion; Indicates the order of each variable; Indicates based on orthogonal polynomial basis functions; These are the polynomial expansion coefficients, representing the weights of the polynomial terms.

[0109] In one embodiment, the expression for the predicted residual is as follows:

[0110]

[0111]

[0112]

[0113] In the formula, This represents the true value of the horizontal force acting on the vehicle. This represents the true value of the vertical force acting on the vehicle. This represents the actual value of the torque acting on the vehicle. This represents the predicted residual value of the horizontal force acting on the vehicle. This represents the predicted residual value of the vertical force acting on the vehicle. This represents the predicted residual value of the torque acting on the vehicle.

[0114] The predicted residuals are standardized:

[0115]

[0116]

[0117]

[0118] in, , , respectively represent the mean of the prediction residual of the horizontal force, vertical force, and moment, respectively, , , respectively represent the standard deviation of the prediction residual of the horizontal force, vertical force, and moment, respectively, , , respectively represent the normalized prediction residual value of the horizontal force, vertical force, and moment, respectively.

[0119] In one embodiment, the interpolation-type Kriging model corrects the prediction residual by a Gaussian process regression expression as follows:

[0120]

[0121] In the formula, represents the mean of the prediction residual of all samples; represents the covariance matrix, and the element is , wherein, represents the noise variance hyperparameter, represents the Kronecker symbol; represents the similarity of and calculated by the kernel function; represents the input variable of the new sample to be predicted, which is used for model prediction; represents the input value of the i th sample in the training set, which is used for model training and kernel function calculation; represents the input value of the j th sample in the training set; respectively correspond to , , in each load component; respectively correspond to , , , n represents the total number of samples in the training set;

[0122] The kernel function adopts a composite form as shown below:

[0123]

[0124] In the formula, is the signal variance, which is used to control the output amplitude of the kernel function; represents the length scale hyperparameter that controls the amplitude of the internal solitary wave in the direction smoothness, represents the length scale hyperparameter that controls the amplitude of the internal solitary wave in the d direction smoothness, relative distance of the vehicle from the trough of the internal solitary wave length scale hyperparameter of directional smoothness amplitude of the internal solitary wave i , j amplitude of the internal solitary wave submerged depth of the vehicle i , j submerged depth of the vehicle 、 relative distance of the vehicle from the trough of the internal solitary wave i , j relative distance of the vehicle from the trough of the internal solitary wave

[0125] In one embodiment, the method for standardizing the initial sample data set is to use StandardScaler to perform z-score standardization on the input variables, including the amplitude of the internal solitary wave, the submerged depth of the vehicle, and the relative distance of the vehicle from the trough of the internal solitary wave.

[0126] In one embodiment, the evaluation method of the vehicle hydrodynamic load fast prediction model comprises

[0127] Calculate the root mean square error (RMSE):

[0128]

[0129] Calculate the correlation coefficient R:

[0130]

[0131] wherein, M is the total number of samples in the validation set, y i is the true response value of each load component, is the response predicted by the vehicle hydrodynamic load fast prediction model; , ;

[0132] Determine the validation standard: when the correlation coefficient R is greater than or equal to 0.9 and the RMSE is less than or equal to 3% of the mean value of the actual value, the linear trend of the predicted value and the actual value is highly consistent, and the prediction has high quality.

[0133] The present application also provides a vehicle hydrodynamic load fast prediction system under the influence of internal waves, comprising:

[0134] Data acquisition module: establish a numerical model of internal solitary wave-vehicle coupling interaction, simulate the interaction between internal solitary waves and vehicles under different working conditions; extract time-domain load data of horizontal force, vertical force and moment on the vehicle;

[0135] Space-time conversion module: connected to the data acquisition module, converts time domain load data into spatial position load data;

[0136] Preprocessing module: connected to the space-time conversion module, standardizes the initial sample data;

[0137] Hybrid proxy model module; connected to the preprocessing module, including constructing a polynomial chaos expansion model and an interpolation-type Kriging model, superimposing the prediction value output by the polynomial chaos expansion model and the correction value obtained by the interpolation-type Kriging model to obtain a rapid prediction model of the water dynamic load of the navigation body; output the horizontal force, vertical force and torque of the final load prediction;

[0138] Verification module: connected to the hybrid proxy model module, determines the validity of the rapid prediction model of the water dynamic load of the navigation body.

[0139] The application also provides an electronic device comprising a memory, a processor and a computer program stored in the memory, wherein the processor implements the steps of the method for rapidly predicting the water dynamic load of a navigation body under the influence of internal waves when executing the computer program.

[0140] The application also provides a computer-readable storage medium storing a computer program, wherein the computer program implements the steps of the method for rapidly predicting the water dynamic load of a navigation body under the influence of internal waves when executed.

[0141] In one embodiment, the water dynamic load prediction when the standard SUBOFF model interacts with internal solitary waves is taken as an example:

[0142] Step 1: Numerical modeling of internal solitary wave-navigation body coupling;

[0143] A SUBOFF submarine model is used to establish a numerical water tank with a size of 260m*20m*30m;

[0144] Density jump layer setting: upper layer density (thickness ), lower layer density (thickness ).

[0145] The SUBOFF model is placed at a position 180m away from the entrance of the water tank.

[0146] Based on the internal wave theory, an internal wave is generated for the numerical water tank model, and monitoring points for horizontal force, vertical force and torque are set on the numerical model to record the time history curve of the water dynamic load of the navigation body when encountering an internal solitary wave.

[0147] Internal solitary wave generation: solve the DJL equation through the DJLES open source software, input the wave amplitude parameter to generate an initial wave field (see Figure 2 ).

[0148] By changing the wave amplitude of internal solitary wave, the submerged depth of the vehicle, the interaction between the vehicle and the internal solitary wave under different working conditions is simulated, and the time history curve of the hydrodynamic load on the vehicle is extracted. In this implementation case, 7 wave amplitudes in the range of 2.4m-7.2m and 7 submerged depths in the range of -8m-20m are set, wherein the submerged depth is defined as the distance between the vehicle and the horizontal plane, and the coordinate of the horizontal plane is 0. The submerged depth below the water surface is negative. A total of 49 different working conditions are simulated in this implementation case.

[0149] Step 2: Time-space data conversion;

[0150] Mapping relationship: convert time-domain load data to spatial position load data;

[0151] Based on The time-dependent hydrodynamic load is converted to spatial position-dependent data by converting the extracted time items to spatial positions. In this case, the initial distance between the vehicle and the trough of the internal solitary wave is Each wave amplitude corresponds to a wave speed, and the wave speeds corresponding to the wave amplitudes set in this case are shown in the following table.

[0152]

[0153] Step 3: Data preprocessing and division;

[0154] The converted load component data and the set wave amplitude of the internal solitary wave and the submerged depth of the vehicle are subjected to z-score standardization processing by StandardScaler.

[0155] After data preprocessing, the data is divided. The sample size of each load component corresponding to the horizontal force, vertical force and moment is 5042. The samples and corresponding hydrodynamic loads are divided into a training set (80%: 4033 samples) and a validation set (20%: 1009 samples).

[0156] Step 4: Construction of PCE-Kriging hybrid surrogate model;

[0157] A data-driven polynomial chaos expansion (PCE) model is constructed, with the wave amplitude of the internal solitary wave, the submerged depth of the vehicle and the relative distance between the vehicle and the trough of the internal solitary wave as input variables, and the predicted values of the horizontal force, vertical force and moment on the vehicle as output. In the polynomial expansion section, the polynomial chaos expansion model uses a 3rd order polynomial expansion scheme, and generates the corresponding orthogonal polynomial basis functions based on the defined input variable distribution by means of the generate_expansion function of the chaospy library.

[0158] The coefficients of the polynomial expansion are determined in the training process based on the least square regression method by fitting the training data, so as to obtain the PCE prediction model. The standardized data are predicted by the constructed PCE model, and the preliminary force prediction value is obtained.

[0159] According to each load component predicted by the polynomial chaos expansion model, the prediction residual (i.e. the difference between the true value and the predicted value of the horizontal force, vertical force and moment of the navigation body) is calculated. The Kriging model is fitted, the Gaussian Process Regressor (GPR) model is used to correct the prediction residual, and the corrected value of the horizontal force, vertical force and moment of the navigation body is obtained.

[0160] The prediction value output by the polynomial chaos expansion model is superimposed with the corrected value obtained by the interpolation Kriging method, so as to obtain the rapid prediction model of the hydrodynamic load of the navigation body; and the final load prediction of the horizontal force, vertical force and moment is output.

[0161] Step 5: load prediction and model verification

[0162] Based on the rapid prediction model of the hydrodynamic load of the navigation body, the new input sample (the amplitude of the internal solitary wave, the depth of the navigation body and the relative distance of the navigation body from the trough of the internal solitary wave) is predicted, and the final predicted horizontal force, vertical force and moment are obtained by summation.

[0163]

[0164]

[0165]

[0166] In the formula, F h-final represents the predicted value of the horizontal force of the navigation body, F v-final represents the predicted value of the vertical force of the navigation body, F M-final represents the predicted value of the moment of the navigation body, a is the amplitude of the internal solitary wave, d is the depth of the navigation body, is the relative distance of the navigation body from the trough of the internal solitary wave. represents the horizontal force prediction value output by the polynomial chaos expansion model, represents the vertical force prediction value output by the polynomial chaos expansion model, represents the moment prediction value output by the polynomial chaos expansion model; represents the correction value of the horizontal force, represents the correction value of the vertical force, a correction value of the moment; RBF represents a radial basis function kernel, and length_scale controls the "influence range" or smoothness of the kernel function.

[0167] After obtaining the final hydrodynamic load, the performance of the constructed fast prediction model of the hydrodynamic load of the vehicle is evaluated by calculating the root mean square error (RMSE) and the correlation coefficient (R) on the validation set.

[0168] From the prediction results of the present example, it can be concluded that the fast prediction method proposed by the present application can accurately reflect the hydrodynamic load on any position of the vehicle under different working conditions. In the present example, the final root mean square error (RMSE) of the horizontal force, vertical force and moment hydrodynamic load prediction model is 0.032, 0.235 and 0.112 respectively, and the correlation (R) between the predicted value and the actual value is 0.99981, 0.99967 and 0.99927 respectively. If the numerical model used in the present example and the same number of grids are used to predict the hydrodynamic force of the vehicle under the action of internal solitary waves by the CFD numerical simulation method, each example needs to spend 21.5 hours, but based on the fast prediction method proposed in the present example, the hydrodynamic prediction of the vehicle only needs 111.8 milliseconds from loading data to completion.

[0169] Compared with the traditional CFD method, the prediction model proposed by the present application greatly reduces the evaluation time of the hydrodynamic load, and when the internal wave parameters and other working conditions change, it does not need to be repeatedly set complicatedly, but by inputting the wave amplitude of the internal solitary wave, the depth of the vehicle and the relative distance of the vehicle from the trough of the internal solitary wave, the force information of the vehicle at any spatial position can be output in real time, so as to realize the fast and real-time prediction of the hydrodynamic load of the vehicle under the influence of internal waves.

[0170] Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments without departing from the principles and purposes of the present application within the scope of the present application.

Claims

1. A method for fast prediction of hydrodynamic loads of a vehicle under internal wave influence, characterized in that The method comprises the following steps: A numerical model of internal solitary wave-vehicle interaction is established, the wave amplitude of the internal solitary wave and the submerged depth of the vehicle under different working conditions are set, the internal solitary wave-vehicle interaction under different working conditions is simulated, and time-domain load data of load components borne by the vehicle are extracted; The load components include horizontal force, vertical force and moment; A mapping relationship between time and spatial position is constructed to convert the time-domain load data into spatial position load data; the mapping relationship between time and spatial position is expressed as follows: wherein denotes the initial distance of the wave trough of the internal solitary wave from the vehicle at time t = 0, c denotes the propagation speed of the internal solitary wave, denotes the relative distance of the vehicle from the wave trough of the internal solitary wave, t is the propagation time of the internal solitary wave; The spatial position load data, the wave amplitude of the internal solitary wave and the submerged depth of the vehicle under different working conditions are constructed into an initial sample data set, the initial sample data set is subjected to standardization processing to obtain a sample data set, and the sample data set is divided into a training set and a verification set; A polynomial chaos expansion model is constructed and trained, the input of the polynomial chaos expansion model is the wave amplitude of the internal solitary wave, the submerged depth of the vehicle and the relative distance between the vehicle and the wave trough of the internal solitary wave, and the output of the polynomial chaos expansion model is the predicted values of the horizontal force, the vertical force and the moment borne by the vehicle; Based on the load components predicted by the polynomial chaos expansion model, a prediction residual is calculated; the prediction residual is the difference between the true values and the predicted values of the horizontal force, the vertical force and the moment borne by the vehicle; An interpolation Kriging model is used to correct the standardized prediction residual by Gaussian process regression to obtain the corrected values of the horizontal force, the vertical force and the moment borne by the vehicle; The predicted values output by the polynomial chaos expansion model and the corrected values obtained by the interpolation Kriging model are superimposed to obtain a vehicle hydrodynamic load rapid prediction model; the final load prediction of the horizontal force, the vertical force and the moment is output: wherein, F h-final represents a predicted value of the horizontal force on the vehicle, F v-final represents a predicted value of the vertical force on the vehicle, F M-final represents a predicted value of the moment on the vehicle, a represents the amplitude of the internal solitary wave, d is the depth of the vehicle, is the relative distance of the vehicle from the trough of the internal solitary wave; represents the predicted value of the horizontal force output by the polynomial chaos expansion model, represents the predicted value of the vertical force output by the polynomial chaos expansion model, represents the predicted value of the moment output by the polynomial chaos expansion model; represents the correction value of the horizontal force, represents the correction value of the vertical force, represents the correction value of the moment; The performance of the vehicle hydrodynamic load rapid prediction model is evaluated by the root mean square error and the correlation coefficient to complete the rapid prediction of the vehicle hydrodynamic load under the influence of internal waves.

2. The method of claim 1, wherein: The numerical model of internal solitary wave-vehicle interaction is obtained by using Reynolds-averaged Navier-Stokes equation (RANS) or large eddy simulation (LES), and the initial flow field of the internal solitary wave is obtained by solving the DJL internal solitary wave theoretical equation by using DJLES open source software package.

3. The method of claim 1, wherein: The polynomial chaos expansion model is expressed as follows: wherein, represents the predicted value of the horizontal force on the vehicle, represents the predicted value of the vertical force on the vehicle, represents the predicted value of the moment on the vehicle; is a set of indices corresponding to the polynomial, defining the structure of the polynomial basis function, representing the structure and order of all possible polynomial terms in the polynomial expansion; represents the order of each variable; represents the orthogonal polynomial basis function based on ; and is a polynomial expansion coefficient, representing the weight of the polynomial term.

4. The method of claim 3, wherein: The prediction residual is expressed as follows: In the formula, represents a real value of the horizontal force on the vehicle, represents a real value of the vertical force on the vehicle, represents a real value of the moment on the vehicle; represents a predicted residual value of the horizontal force on the vehicle, represents a predicted residual value of the vertical force on the vehicle, represents a predicted residual value of the moment on the vehicle; The prediction residual is subjected to standardization processing: wherein, , , respectively represent the mean of the horizontal force, vertical force, moment corresponding residual error, , , respectively represent the standard deviation of the horizontal force, vertical force, moment corresponding residual error, , , respectively represent the normalized predicted residual value corresponding to the horizontal force, vertical force, moment.

5. The method of claim 4, wherein: The expression for correcting the standardized prediction residual by Gaussian process regression using the interpolation Kriging model is as follows: In the formula, represents the average value of the prediction residual of all samples; represents the covariance matrix, and the element is , wherein, represents a noise variance hyperparameter, represents the Kronecker symbol; represents the of the kernel function calculation; similarity with represents the input variable of the new sample to be predicted currently, for model prediction; represents the input value of the i sample in the training set, for model training and kernel function calculation; represents the input value of the j sample in the training set; corresponds to , , respectively in each load component; corresponds to , , respectively in each load component, n represents the total number of samples in the training set; The kernel function adopts a composite form as follows: where, is the signal variance, used to control the output amplitude of the kernel; denotes the wave amplitude of the internal solitary wave is the length scale hyperparameter of the directional smoothness, denotes the submerged depth of the vehicle d is the length scale hyperparameter of the directional smoothness, denotes the relative distance of the vehicle from the trough of the internal solitary wave is the length scale hyperparameter of the directional smoothness; denotes the wave amplitude of the internal solitary wave i , j denotes the wave amplitude of the internal solitary wave , i , j denotes the submerged depth of the vehicle 、 denotes the relative distance of the vehicle from the trough of the internal solitary wave i , j denotes the relative distance of the vehicle from the trough of the internal solitary wave 6. The method of claim 1, wherein: The method for standardizing the initial sample data set is to perform z-score standardization processing on the input variables by using StandardScaler, and the input variables include the wave amplitude of the internal solitary wave, the submerged depth of the vehicle and the relative distance between the vehicle and the wave trough of the internal solitary wave.

7. The method of claim 1, wherein: The evaluation method of the vehicle hydrodynamic load rapid prediction model comprises, The root mean square error (RMSE) is calculated as follows: The correlation coefficient (R) is calculated as follows: In the formula, M denotes the total number of samples in the validation set, y i denotes the true response value of each load component, denotes the response predicted by the fast prediction model of the water dynamic load of the vehicle; , ; The judgment standard is that when the correlation coefficient R is greater than or equal to 0.9 and the RMSE is less than or equal to 3% of the actual value mean, the linear trend of the predicted value and the actual value is highly consistent, and the prediction has high quality.

8. A system for rapidly predicting the hydrodynamic loads of a vehicle under the influence of internal waves, for implementing the method for rapidly predicting the hydrodynamic loads of a vehicle under the influence of internal waves according to any one of claims 1 to 7, characterized in that, The method comprises the following steps: Data acquisition module: a numerical model of the interaction between internal solitary wave and vehicle is established to simulate the interaction between internal solitary wave and vehicle under different conditions; the time-domain load data of the horizontal force, vertical force and moment of the vehicle are extracted; Space-time conversion module: connected to the data acquisition module, the time-domain load data is converted into spatial position load data; Preprocessing module: connected to the space-time conversion module, the initial sample data is standardized; Mixed proxy model module: connected to the preprocessing module, including the construction of polynomial chaos expansion model and interpolation kriging model, the predicted value of the polynomial chaos expansion model is superimposed with the correction value obtained by the interpolation kriging model to obtain a fast prediction model of the hydrodynamic load of the vehicle; the final load prediction of the horizontal force, vertical force and moment is output; Verification module: connected to the mixed proxy model module, the effectiveness of the fast prediction model of the hydrodynamic load of the vehicle is determined.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory, wherein: The processor executes the computer program to realize the steps of the fast prediction method of the hydrodynamic load of the vehicle under the influence of internal wave according to any one of claims 1-7.

10. A computer readable storage medium storing a computer program, characterized in that: The computer program is executed to realize the steps of the fast prediction method of the hydrodynamic load of the vehicle under the influence of internal wave according to any one of claims 1-7.

Images