A rapid prediction method, program, equipment, and storage medium for ship motion flow field in waves based on a combination of POD and DMD.

Abstract

The present invention belongs to the technical field of ship flow field analysis and prediction, and specifically relates to a method, program, device and storage medium for rapid prediction of the flow field of a ship moving in waves based on the combination of POD and DMD. The present invention obtains the initial moment reconstructed flow field of any sample ship structure based on the POD method, obtains the linear transfer matrix of any sample ship structure based on the DMD method, and combines with the proxy model. When the ship to be predicted is sailing in the monitored sea area, the initial moment flow field data vector and the POD basis coefficient of the transfer matrix of the corresponding ship structure are obtained according to the proxy model, and then the initial moment flow field data vector and transfer matrix are obtained, so as to realize the rapid prediction of the flow field in waves for any ship structure at any time.

Description

Technical Field

[0001] The present invention belongs to the technical field of ship flow field analysis and prediction, and specifically relates to a method, program, device and storage medium for quickly predicting the flow field of a ship moving in waves based on a combination of POD and DMD. Background Art

[0002] As research into the flow fields of ships moving in waves deepens, the corresponding numerical simulation computational costs are also increasing. Specifically, the computational memory and time consumed by high-precision numerical simulations are growing exponentially. As a result, rapid flow field prediction through data analysis of limited flow field simulation results has become a research hotspot.

[0003] However, most numerical flow simulation results are high-dimensional data. Although the information contained is very detailed, the computational cost of high-dimensional data is high, and the high dimensionality of the data itself makes subsequent processing complex and difficult. Therefore, it is necessary to use "dimensionality reduction" methods to extract important features from these high-dimensional data and achieve a mapping from high-dimensional feature space to low-dimensional feature space. Commonly used dimensionality reduction methods in current research include Proper Orthogonal Decomposition (POD) and Dynamic Mode Decomposition (DMD).

[0004] After searching, it was found that there is a flow field reconstruction method based on intrinsic orthogonal decomposition and deep learning fusion (publication number CN116029205A, publication date 20230428). The invention first trains a neural network model and inputs the sensor observation data into the trained neural network model to obtain a neural network predicted flow field; then uses the POD method to extract the POD mode from the flow field data and solves to obtain the POD modal coefficient; finally, based on the solved modal coefficient, the POD mode is linearly combined to output the flow field reconstruction result. However, the invention only processes the data observed by the sensor at a certain moment by the POD method and uses the neural network to predict the flow field, and cannot achieve the prediction of the flow field at any time under different working conditions; when the amount of flow field data is large, directly using a reduced-order model such as POD for dimensionality reduction may cause memory overflow; when using a deep learning model such as a neural network for dimensionality reduction, the relative error level has a large randomness and is difficult to control in advance. Therefore, the invention has limitations in practical application, that is, its universality is poor. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for quickly predicting the flow field of a ship moving in waves based on the combination of POD and DMD, which can realize the rapid prediction of unknown flow fields.

[0006] A method for quickly predicting the flow field of a ship moving in waves based on a combination of POD and DMD includes the following steps:

[0007] Step 1: Obtain N types of ship structures, obtain the flow field data of each ship structure at M typical moments in the monitoring sea area through numerical simulation methods, and establish the correlation matrix X of the typical moment information of the flow field; the elements in the matrix are represents the flow field data vector of the i-th ship type structure at the j-th typical moment in the monitoring sea area;

[0008] Step 2: For each ship structure, obtain the corresponding initial flow field data vector Establish the flow field data matrix at the initial moment

[0009] Step 3: Use the POD method to reduce the dimension and reconstruct the matrix Y to obtain the initial flow field data vector corresponding to each ship structure

[0010] Step 4: For each ship structure, use the DMD method to obtain the transfer matrix A of the flow field data at different times i , and then establish the total transfer matrix vector B=[A1,A2,...,A N ];

[0011] Step 5: Use the POD method to reduce the dimension and reconstruct the matrix B to obtain the transfer matrix A corresponding to each ship structure i_rec ;

[0012] Step 6: Based on the N types of ship structure data obtained and the initial flow field data vector corresponding to each ship structure and transfer matrix A i_rec , establish a proxy model, and construct the mapping relationship between the ship structure data and the flow field data vector at its initial moment and the POD basis coefficient of the transfer matrix;

[0013] Step 7: When the ship to be forecasted is sailing in the monitored sea area, the initial flow field data vector of the corresponding ship structure and the POD basis coefficient of the transfer matrix are obtained according to the proxy model, and then the initial flow field data vector is obtained and the transfer matrix A rec , according to the formula Obtain rapid forecast results of the ship's flow field in the monitored sea area; Denotes the transfer matrix A rec The kth power of k represents the sailing time of the ship in the monitored sea area.

[0014] Furthermore, in step 4, if the transfer matrix A iIs a low-dimensional matrix, then directly establish the total transfer matrix vector B=[A1,A2,...,A N ]; If the transfer matrix A i Is a high-dimensional matrix, then first matrix A i Perform compression processing to obtain the compressed matrix D i , then establish the total transfer matrix vector B=[D1,D2,...,D N ]; among them, A i Whether it is a low-dimensional matrix or a high-dimensional matrix is ​​determined by whether memory overflow occurs when the rapid forecast system processes data.

[0015] Furthermore, the matrix A i Perform compression processing to obtain the compressed matrix D i The specific method is:

[0016] Step 4.1: For the M-order square matrix A i , given cutoff values ​​θ1, θ2, 0<θ1≤1, 0<θ2≤1;

[0017] Step 4.2: Through the Matrix The unit orthogonal vectors corresponding to the first l1 largest eigenvalues ​​of are arranged into L columns, where l1 is calculated according to the truncation formula. Get, λ j N L The monotonically decreasing eigenvalue of ;

[0018] Step 4.3: Compress the given matrix A using a left linear transformation L with orthogonal columns i And get its compressed version B i =L T A i ;

[0019] Step 4.4: Through the Matrix The unit orthogonal vectors corresponding to the first l2 largest eigenvalues ​​of are arranged into R in columns, where l2 can be calculated according to the truncation formula Get, μ j N R The monotonically decreasing eigenvalue of ;

[0020] Step 4.5: Compress the given matrix B using a right linear transformation R with orthogonal columns i And get its compressed version D i =B i R.

[0021] Furthermore, the method of reducing the dimension and reconstructing the matrix using the POD method in step 3 is the same as that in step 5. Specifically, step 3 is:

[0022] Step 3.1: Note Pick Define the matrix Pair Matrix Perform singular value decomposition:

[0023]

[0024] Where U is The orthogonal eigenvector matrix, that is, the POD basis vector matrix, U=(u1,u2,…,u r ); D is the singular value matrix corresponding to U, D=diag(σ1,σ2,…,σ r ) and σ1>σ2>…>σ r ; V is an orthogonal matrix, V T V = E;

[0025] Step 3.2: Given a cutoff value α1, 0<α1≤1, according to the cutoff formula To determine the truncation position α;

[0026] Step 3.3: Calculate POD basis coefficients Calculate the initial flow field data vector corresponding to each ship structure

[0027] Furthermore, in step 4, the DMD method is used to obtain the transfer matrix A of the flow field data at different times. i The specific method is: for the ship structure C i , select the flow field data from the first moment to the (M-1)th typical moment, recorded as Select the flow field data from the second moment to the Mth typical moment and record it as and The transfer matrix A i Expressed as The transfer matrix A is obtained by solving the generalized inverse of the matrix i , express The generalized inverse of .

[0028] A computer device / equipment / system comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-mentioned method for rapid prediction of the flow field of a ship moving in waves based on the combination of POD and DMD.

[0029] A computer-readable storage medium stores a computer program / instruction, which, when executed by a processor, implements the steps of the method for rapidly predicting the flow field of a ship moving in waves based on a combination of POD and DMD.

[0030] A computer program product includes a computer program / instruction, which, when executed by a processor, implements the steps of the method for quickly predicting the flow field of a ship moving in waves based on a combination of POD and DMD.

[0031] The beneficial effects of the present invention are:

[0032] The present invention obtains the initial moment reconstructed flow field of any sample ship structure based on the POD method, obtains the linear transfer matrix of any sample ship structure based on the DMD method, and realizes the rapid prediction of the flow field of any ship structure in waves at any time in combination with the proxy model. The present invention takes into account the rapid and accurate prediction of the wave field around the ship at each typical moment in a wave encounter cycle when different ship geometries make periodic motions in waves; adopts matrix compression means to solve the problems such as memory overflow caused by directly using a reduced-order model for dimensionality reduction when the amount of numerical simulation data of the flow field (the number of grid points) is large; by reasonably setting the truncation error parameter, the energy contained in the entire smooth database can be grasped, and the overall relative error level of the prediction can be effectively controlled, solving the problem that the relative error level is highly random and difficult to control in advance when using deep learning models such as neural networks for dimensionality reduction. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Flowchart of the present invention.

[0034] Figure 2 This is a wave field diagram of a sample trimaran at a typical moment in an embodiment of the present invention.

[0035] Figure 3 This is a comparison diagram of the errors between the CFD numerical simulation flow field and the POD reconstructed flow field of a sample ship type at the initial moment in an embodiment of the present invention.

[0036] Figure 4 Graph showing the variation of θ1 and NMSE with the value of l1 in an embodiment of the present invention.

[0037] Figure 5 Graph showing the CFD numerical simulation results of the wave field at a typical moment for a new trimaran in an embodiment of the present invention.

[0038] Figure 6 This is a diagram showing the rapid prediction results of the wave field for a new trimaran at a typical time in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described below with reference to the accompanying drawings.

[0040] This invention belongs to the field of ship flow field analysis and prediction, and discloses a method and system for rapidly predicting the flow field of a ship moving in waves, based on a combination of POD and DMD. The present invention uses the POD method to obtain the initial moment reconstructed flow field for any sample ship structure, and the DMD method to obtain the linear transfer matrix for any sample ship structure. Combining these two methods with a proxy model enables rapid prediction of the flow field in waves for any ship structure at any time.

[0041] A method for quickly predicting the flow field of a ship moving in waves based on a combination of POD and DMD includes the following steps:

[0042] Step 1: Obtain N types of ship structures, obtain the flow field data of each ship structure at M typical moments in the monitoring sea area through numerical simulation methods, and establish the correlation matrix X of the typical moment information of the flow field; the elements in the matrix are represents the flow field data vector of the i-th ship type structure at the j-th typical moment in the monitoring sea area, i = 1, 2, ..., N, j = 1, 2, ..., M;

[0043] Step 2: For each ship structure, obtain the corresponding initial flow field data vector Establish the flow field data matrix at the initial moment The POD method is used to reduce the dimension and reconstruct the matrix Y to obtain the initial flow field data vector corresponding to each ship structure.

[0044] Step 2.1: Remember Pick Define the matrix Pair Matrix Perform singular value decomposition:

[0045]

[0046] Where U is The orthogonal eigenvector matrix, that is, the POD basis vector matrix, U=(u1,u2,…,u r ); D is the singular value matrix corresponding to U, D=diag(σ1,σ2,…,σ r ) and σ1>σ2>…>σ r ; V is an orthogonal matrix, V T V = E;

[0047] Step 2.2: Given a cutoff value α1, 0<α1≤1, according to the cutoff formula To determine the truncation position α;

[0048] Step 2.3: Calculate POD basis coefficients Calculate the initial flow field data vector corresponding to each ship structure

[0049] Step 3: For each ship structure, use the DMD method to obtain the transfer matrix A of the flow field data at different times i ; For ship structure C i , select the flow field data from the first moment to the (M-1)th typical moment, recorded as Select the flow field data from the second moment to the Mth typical moment and record it as and The transfer matrix A i Expressed as The transfer matrix A is obtained by solving the generalized inverse of the matrix i , express The generalized inverse of

[0050] Step 4: If the transfer matrix A i Is a low-dimensional matrix, then establish the total transfer matrix vector B = [A1, A2, ..., A N ]; If the transfer matrix A i Is a high-dimensional matrix, then first matrix A i Perform compression processing to obtain the compressed matrix D i , then establish the total transfer matrix vector B=[D1,D2,...,D N ]; among them, A i Whether it is a low-dimensional matrix or a high-dimensional matrix is ​​determined by whether memory overflow occurs when the rapid forecast system processes data;

[0051] The GLRAM method was used to i Perform compression processing. The specific method is:

[0052] Step 4.1: For the M-order square matrix A i , given cutoff values ​​θ1, θ2, 0<θ1≤1, 0<θ2≤1;

[0053] Step 4.2: Through the Matrix The unit orthogonal vectors corresponding to the first l1 largest eigenvalues ​​of are arranged into L columns, where l1 is calculated according to the truncation formula. Get, λ j N L The monotonically decreasing eigenvalue of ;

[0054] Step 4.3: Compress the given matrix A using a left linear transformation L with orthogonal columns i And get its compressed version B i =L T A i;

[0055] Step 4.4: Through the Matrix The unit orthogonal vectors corresponding to the first l2 largest eigenvalues ​​of are arranged into R in columns, where l2 can be calculated according to the truncation formula Get, μ j N R The monotonically decreasing eigenvalue of ;

[0056] Step 4.5: Compress the given matrix B using a right linear transformation R with orthogonal columns i And get its compressed version D i =B i R;

[0057] Step 5: Use the POD method to reduce the dimension and reconstruct the matrix B to obtain the transfer matrix A corresponding to each ship structure i_rec ;

[0058] Step 6: Based on the N types of ship structure data obtained and the initial flow field data vector corresponding to each ship structure and transfer matrix A i_rec , establish a proxy model, and construct the mapping relationship between the ship structure data and the flow field data vector at its initial moment and the POD basis coefficient of the transfer matrix;

[0059] Step 7: When the ship to be forecasted is sailing in the monitored sea area, the initial flow field data vector of the corresponding ship structure and the POD basis coefficient of the transfer matrix are obtained according to the proxy model, and then the initial flow field data vector is obtained and the transfer matrix A rec , according to the formula Obtain rapid forecast results of the ship's flow field in the monitored sea area; Denotes the transfer matrix A rec The kth power of k represents the sailing time of the ship in the monitored sea area.

[0060] Example 1:

[0061] This embodiment provides a method and system for quickly predicting the flow field of a ship moving in waves based on the combination of POD and DMD. Figure 1 As shown, the method of the present invention mainly comprises the following steps:

[0062] S1, for different sample ship structures {C i A certain structure C in}(i=1,2,…,N) i For example, we select the flow field data at M typical moments and establish the correlation matrix X about the typical moment information of the flow field;

[0063] S2, based on the initial flow field data corresponding to different ship structures, the Proper Orthogonal Decomposition (POD) method is used for dimensionality reduction and reconstruction;

[0064] S3, based on the flow field data corresponding to different ship structures, for each ship structure, the dynamic mode decomposition (DMD) method is used to obtain the transfer matrix A of the flow field data at different times;

[0065] S4, based on the transfer matrix A obtained from different ship structures, is reconstructed using the POD method;

[0066] S5, based on the results obtained by the above POD method, establish a proxy model that maps any ship structure to the flow field at the initial moment and transfers the matrix basis vector coefficients;

[0067] S6, the initial flow field obtained based on the above proxy model and the transfer matrix A rec , which can realize the rapid prediction of the flow field in waves for any ship structure at any time.

[0068] In step S3, the transfer matrix A is obtained based on the DMD method. i , the specific method is as follows:

[0069] For a certain ship structure C i For example, the flow field data from the first moment to the (M-1)th typical moment are selected and recorded as The flow field data from the second moment to the Mth typical moment is recorded as

[0070] Since DMD is a method that approximates a nonlinear dynamic system as a linear dynamic system, that is, it assumes that there is a certain linear change relationship between the data sampled at different times, and when the ship makes periodic motion in the waves, the flow field at different times also has an approximately linear change law.

[0071] Therefore, the transfer matrix in the DMD method can be used to describe the flow field of the ship in the waves. i Ship-type structure C i The wave field transfer matrix (i.e., Koopman operator matrix) is: and It can be represented by the transfer matrix A i Expressed as And the transfer matrix A can be obtained by solving the generalized inverse of the matrix i ,Right now in for The generalized inverse of .

[0072] In step S4, if the transfer matrix A is a low-dimensional matrix (low dimension and high dimension are determined by whether memory overflow occurs when the rapid prediction system processes data), the matrix can be directly vectorized, and the matrix composed of the vectorized transfer matrices corresponding to different ship structures is subjected to POD dimensionality reduction and reconstruction processing to obtain the transfer matrix A corresponding to any sample ship structure. rec .

[0073] In step S4, if the transfer matrix A i It is a high-dimensional matrix (low-dimensional and high-dimensional are determined by whether memory overflow occurs when the rapid forecast system processes data). The matrix needs to be compressed first using the Generalized Low Rank Approximation Method (GLRAM). The specific method is as follows:

[0074] For a given M-order square matrix A i (i=1,2,…,N), first pass the matrix The unit orthogonal vectors corresponding to the first l1 largest eigenvalues ​​of are arranged into L columns, where l1 can be calculated according to the truncation formula Get, λ i (i=1,2,…,M) is N L After that, the given matrix A is compressed using the left linear transformation L with orthogonal columns i And get its compressed version B i =L T A i .

[0075] Secondly, through the matrix The unit orthogonal vectors corresponding to the first l2 largest eigenvalues ​​of are arranged into R in columns, where l2 can be calculated according to the truncation formula Get, μ i (i=1,2,…,M) is N R Then, the given matrix B is compressed using the right linear transformation R with orthogonal columns i And get its compressed version D i =B i R.

[0076] For the above truncation formula, given θ1,θ2, the normalized mean square error can be quickly estimated The upper limit of NMSE is 1-θ1θ2, where

[0077] Finally, you can get A i The final compressed version D i =LT A i R.

[0078] The compression matrix D corresponding to the different sample ship structures obtained after GLRAM processing i The matrix formed after vectorization is processed by POD dimension reduction and reconstruction, and then Restore and obtain the transfer matrix A corresponding to any ship structure rec .

[0079] The specific implementation process of the Proper Orthogonal Decomposition (POD) method mentioned in step S2 and step S4 is as follows:

[0080] Let X=[x1,x2,…,x M ], where x i is an N-dimensional column vector.

[0081] remember Pick Define the matrix right Perform singular value decomposition

[0082]

[0083] Where U is The orthogonal eigenvector matrix, V is an orthogonal matrix, V T V=E, and U=(u1,u2,…,u r ) is the POD basis vector matrix, D is the singular value matrix corresponding to U, that is, D=diag(σ1,σ2,…,σ r ) and σ1>σ2>…>σ r .

[0084] Afterwards, since the basis vector u j The energy contained can be represented by the corresponding eigenvalue. Under the premise of a given cutoff value α0, the cutoff formula can be To determine the truncation position α, and use the proxy model to obtain the corresponding base coefficient a ij , after the final reconstruction It can be expressed as

[0085] In step S5, different surrogate models (such as Kriging surrogate model and response surface surrogate model) are established to construct the mapping relationship between the POD basis coefficients of the flow field and transfer matrix at the initial moment for any ship structure, thereby realizing the initial moment flow field. and the transfer matrix A rec to obtain.

[0086] The present invention consists of a POD calculation module for dimensionality reduction and reconstruction of the initial flow field matrix and the transfer matrix, a DMD calculation module for solving the transfer matrix corresponding to different ship structures and performing flow field prediction at any time, a GLRAM compression module for processing the high-dimensional transfer matrix, a proxy model mapping module for establishing the mapping between different ship structures and the basis vector coefficients in the POD dimensionality reduction and the transfer matrix, and an error analysis module for calculating the error to determine the truncation position.

[0087] In a specific embodiment of the present invention, step S1: 60 trimaran types are selected as sample types, and the wave fields around the 60 trimaran types are calculated by CFD numerical simulation (wherein the main parameters are: the incident wave is a first-order Stokes wave, the amplitude is 0.02 meters, the wavelength is 4 meters, the waterline length of the ship body is 4 meters, and the ship speed is 1.88 meters per second). Figure 2 As shown, 9 typical moments are selected and the flow field information matrix at the initial moment is established.

[0088] In the step S2, a cutoff value α0=0.95 is given, and the POD dimension reduction is used to obtain the initial moment reconstructed flow field of any sample ship structure, such as Figure 3 shown.

[0089] In step S3, for each set of flow field data in the 60 sets of ship types, the corresponding transfer matrix is ​​obtained using DMD and recorded as A i (i=1,2,…,60), since after vectorization (the corresponding matrix dimension is 19950 2 ×60) in the POD processing process (need to process a dimension of 19950 2 ×19950 2 The server memory overflow problem occurs in the matrix of , so it is necessary to use GLRAM to compress the matrix. For this embodiment, the truncation value θ1 = θ2 = 0.99999 is given, as shown in Figure 4 As shown, through the error analysis module, the values ​​of l1 and l2 can be calculated to be 420, and the normalized mean square error NMSE = 6×10 -6 Then, the compression matrix D corresponding to the different sample ship structures obtained after GLRAM processing is obtained. i The matrix formed after vectorization is processed by POD dimension reduction and reconstruction, and then Restore and obtain the transfer matrix A corresponding to any ship structure rec .

[0090] In the steps S4 and S5, the cutoff value α0=0.95 is given, and the transfer matrix A after the GLRAM processing is recThe POD dimension reduction process is performed, and by establishing a Kriging proxy model, the mapping relationship between the POD basis coefficients of any ship structure to the flow field and transfer matrix at the initial moment can be obtained.

[0091] Finally, in S6, the initial flow field information obtained based on the above proxy model is and the transfer matrix A rec , which can realize the rapid prediction of the flow field in waves for any ship structure at any time.

[0092] Select a new ship type and under known regular wave conditions, the CFD numerical simulation results are as follows: Figure 5 As shown in the figure, the wave field forecast results obtained based on this rapid forecast system are as follows Figure 6 shown.

[0093] from Figure 5 and Figure 6 It can be seen that the wave field predicted by this system for a trimaran at a typical moment within a wave cycle is consistent with the evolution trend of the CFD simulation results, with a small overall error. Therefore, the computational cost of numerical simulation can be greatly reduced, achieving highly efficient and high-precision prediction of the flow field of a trimaran in waves under any layout.

[0094] The present invention takes into account the rapid and accurate prediction of the wave field around the ship at each typical moment within a wave encounter cycle when different ship geometries (including ship shape and sheet layout) make periodic motion in the waves; adopts matrix compression means to solve the problems such as memory overflow caused by directly using the reduced-order model for dimensionality reduction when the amount of flow field numerical simulation data (the number of grid points) is large; by reasonably setting the truncation error parameter, the energy contained in the entire flow database can be grasped, and the overall relative error level of the forecast can be effectively controlled, which solves the problem that the relative error level has a large randomness when using deep learning models such as neural networks for dimensionality reduction and is difficult to control in advance.

[0095] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A rapid prediction method for the flow field of a ship moving in waves based on the combination of POD and DMD, characterized in that: The following steps are involved: Step 1: Obtain N types of ship structures, obtain the flow field data of each ship structure at M typical times in the monitored sea area through numerical simulation methods, and establish the correlation matrix X about the typical time information of the flow field; the elements in the matrix are It represents the flow field data vector of the i-th ship type structure in the monitoring sea area at the j-th typical moment; Step 2: For each ship type structure, obtain the corresponding initial moment flow field data vector Establish the flow field data matrix about the initial moment Step 3: Use the POD method to reduce the dimension and reconstruct the matrix Y to obtain the initial flow field data vector corresponding to each ship structure Step 4: For each ship type structure, use the DMD method to obtain the transfer matrix A of the flow field data at different times i , and then establish the total transfer matrix vector B = [A1, A2, ..., A N ]; Step 5: Use the POD method to reduce the dimension and reconstruct the matrix B to obtain the transfer matrix A corresponding to each ship type structure i_rec ; Step 6: Based on the N types of ship structure data obtained and the initial flow field data vector corresponding to each ship structure And the transfer matrix A i_rec , establish a proxy model, and construct the mapping relationship between the ship structure data and the POD basis coefficients of the flow field data vector and transfer matrix at its initial moment; Step 7: When the forecasted ship is sailing in the monitored sea area, the initial flow field data vector of the corresponding ship structure and the POD basis coefficient of the transfer matrix are obtained according to the proxy model, and then the initial flow field data vector is obtained and the transfer matrix A rec , according to the formula Obtain rapid forecast results of the flow field of ships in the monitored sea area; among them, Denotes the transfer matrix A rec The kth power, k represents the sailing time of the ship in the monitored sea area.

2. According to claim 1, a method for rapid prediction of the flow field of a ship moving in waves based on the combination of POD and DMD, characterized in that: In step 4, if the transfer matrix A i is a low-dimensional matrix, then directly establish the total transfer matrix vector B = [A1, A2, ..., A N ]; if the transfer matrix A i Is a high-dimensional matrix, then first matrix A i Perform compression processing to obtain the compressed matrix D i , and then establish the total transfer matrix vector B = [D1, D2, ..., D N ]; among them, A i Whether it is a low-dimensional matrix or a high-dimensional matrix is ​​determined by whether memory overflow occurs when the rapid forecasting system processes data.

3. A method for rapid prediction of the flow field of a ship moving in waves based on the combination of POD and DMD according to claim 2, characterized in that: The matrix A i Perform compression processing to obtain the compressed matrix D i The specific method is: Step 4.1: For the M-order square matrix A i , given cutoff values ​​θ1, θ2, 0<θ1≤1, 0<θ2≤1; Step 4.2: Through the matrix The unit orthogonal vectors corresponding to the first l1 largest eigenvalues ​​of are arranged in columns to form L, where l1 is calculated according to the truncation formula Get, λ j N L The monotonically decreasing eigenvalue of ; Step 4.3: Compress the given matrix A using a left linear transformation L with orthogonal columns i And get its compressed version B i =L T A i ; Step 4.4: Through the matrix The unit orthogonal vectors corresponding to the first l2 largest eigenvalues ​​of are arranged in columns to form R, where l2 can be calculated according to the truncation formula Get, μ j N R The monotonically decreasing eigenvalue of ; Step 4.5: Compress the given matrix B using a right linear transformation R with orthogonal columns i And get its compressed version D i =B i R.

4. According to claim 1, a method for rapid prediction of the flow field of a ship moving in waves based on the combination of POD and DMD, characterized in that: The method of reducing the dimension and reconstructing the matrix using the POD method in step 3 is the same as that in step 5. Specifically, step 3 is as follows: Step 3.1: Remember Pick Defining the Matrix Pair Matrix Perform singular value decomposition: Where U is The orthogonal eigenvector matrix, that is, the POD basis vector matrix, U = (u1, u2, ..., u r ), D is the singular value matrix corresponding to U, D = diag(σ1,σ2,…,σ r ) and σ1>σ2>…>σ r ; V is an orthogonal matrix, V T V = E; Step 3.2: Given a cutoff value α1, 0<α1≤1, according to the cutoff formula To determine the truncation position α; Step 3.3: Calculate POD basis coefficients Calculate the flow field data vector corresponding to the initial moment of each ship structure 5. According to claim 1, a method for rapid prediction of the flow field of a ship moving in waves based on the combination of POD and DMD, characterized in that: In step 4, the DMD method is used to obtain the transfer matrix A of the flow field data at different times. i The specific method is as follows: For the ship structure C i , select the flow field data from the 1st moment to the (M-1)th typical moment, recorded as Select the flow field data from the second moment to the Mth typical moment and record it as and The transfer matrix A i Expressed as The transfer matrix A is obtained by solving the generalized inverse of the matrix i , express The generalized inverse of .

6. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 5.

7. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 5.

8. A computer system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 5.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images