Convolutional neural network-based wake field forecasting method and equipment

Abstract

The invention discloses a wake field forecasting method and device based on a convolutional neural network, and relates to the technical field of wake field forecasting, and the method comprises the steps: obtaining a hull geometric file and fluid dynamic parameters; inputting the ship body geometric file and the fluid dynamic parameters into a pre-trained convolutional neural network model, and reconstructing a ship body corresponding to a ship body molded line by using a control grid with a preset size based on the ship body point cloud data to obtain reconstructed point cloud data; inputting the reconstructed point cloud data into a Gaussian mixture model corresponding to the hull molded line to obtain a point cloud feature matrix output by the Gaussian mixture model; and predicting and outputting a predicted wake flow field corresponding to the current ship type based on the point cloud feature matrix and the fluid dynamic parameters. The method is used for solving the problem that in the prior art, accuracy is poor when a proxy model is used for conducting wake field forecasting in the hull geometric configuration process, and the wake field forecasting accuracy and forecasting efficiency are improved.

Description

Technical Field

[0001] The present application relates to the technical field of wake field prediction, and in particular to a wake field prediction method and device based on a convolutional neural network. Background Art

[0002] In the optimization design of complex geometric configurations such as hulls, propellers and energy-saving appendages, and in the process of geometric uncertainty analysis, surrogate modeling methods are usually used to alleviate the problem of excessive consumption of computing resources caused by geometric space sampling and searching. Traditional surrogate models often rely on specific geometric parameterized modeling techniques and are not universally adaptable.

[0003] Moreover, the mapping relationship between the hull lines and the wake field is a typical image-to-image mapping problem, which belongs to the category of high-dimensional mapping. For this type of problem, the biggest difficulty in constructing a proxy model to represent its mapping relationship is that the number of samples in the numerical simulation data set is very limited, and the input and output dimensions of the model are high. There is a problem that the proxy model is difficult to expand to high dimensions, which ultimately leads to the problem of poor accuracy in using the proxy model to predict the wake field. Summary of the invention

[0004] In response to the above-mentioned problems and technical needs, the applicant has proposed a method and device for predicting the wake field based on a convolutional neural network, so as to solve the problem of poor accuracy in the prior art when using a proxy model to predict the wake field during the hull geometry configuration process, thereby improving the prediction accuracy and efficiency of the wake field.

[0005] The present application embodiment provides a method for predicting wake field based on a convolutional neural network, the method comprising:

[0006] Acquire a hull geometry file and fluid dynamics parameters, wherein the hull geometry file includes: hull point cloud data, the hull point cloud data is obtained based on the hull line, and the fluid dynamics parameters include: water flow velocity and physical properties of water corresponding to the ship when sailing;

[0007] The hull geometry file and the fluid dynamics parameters are input into a pre-trained convolutional neural network model, and the hull corresponding to the hull line is reconstructed by the convolutional neural network model based on the hull point cloud data using a control grid of a preset size to obtain reconstructed point cloud data; the reconstructed point cloud data is input into a Gaussian mixture model corresponding to the hull line to obtain a point cloud feature matrix output by the Gaussian mixture model; based on the point cloud feature matrix and the fluid dynamics parameters, a predicted wake field corresponding to the current ship type is predicted and output;

[0008] Wherein, the hull lines correspond to the ship type;

[0009] The convolutional neural network model is trained based on hull geometry file samples, fluid dynamics parameter samples and wake field samples.

[0010] According to a convolutional neural network-based wake field prediction method according to an embodiment of the present application, before inputting the reconstructed point cloud data into a Gaussian mixture model corresponding to the hull line, the method further includes:

[0011] The characteristics of the hull corresponding to the hull lines are analyzed, and based on the preset corresponding relationship between the characteristics of the hull and the Gaussian mixture model, a Gaussian mixture model corresponding to the characteristics of the hull is obtained.

[0012] According to a convolutional neural network-based wake field prediction method according to an embodiment of the present application, based on the hull point cloud data, a hull corresponding to the hull line is reconstructed using a control grid of a preset size to obtain reconstructed point cloud data, including:

[0013] Based on the preset curvature, resampling operations are performed on the control grid in the width direction and the height direction of the hull point cloud data to obtain a reconstruction quantity;

[0014] The three-dimensional hull model corresponding to the hull line is fitted based on the control grid of the reconstructed quantity to obtain the reconstructed point cloud data, wherein the three-dimensional hull model includes: a hull part, a bow part, a stern part and a tail shaft part, one part of the three-dimensional hull model corresponds to a curved surface, and different curves are in G1 continuity.

[0015] According to a convolutional neural network-based wake field prediction method according to an embodiment of the present application, fitting a three-dimensional hull model corresponding to the hull line based on the control grid of the reconstructed number to obtain the reconstructed point cloud data includes:

[0016] Obtaining reconstructed point cloud data based on the first calculation formula;

[0017] Wherein, the first calculation formula includes:

[0018]

[0019] Among them, S(u,v) represents the reconstructed point cloud data, P i,j represents the control grid, which includes the i direction and the j direction, ω i,j represents the weight factor of the corresponding control grid, u represents the node in the i direction, the corresponding total number of nodes is r+1, p represents the number of spline curves in the i direction, v represents the node in the j direction, the corresponding total number of nodes is S+1, q represents the number of spline curves in the j direction, N i,p (u) represents the non-rational B-spline basis function on the knot vector U; N j,q(v) represents the non-rational B-spline basis function on the node vector V, n represents the number of reconstructions of the control grid in the i direction, and m represents the number of reconstructions of the control grid in the j direction;

[0020] in,

[0021] According to a method for predicting wake fields based on a convolutional neural network in one embodiment of the present application, the reconstructed point cloud data is input into a Gaussian mixture model corresponding to the hull line, and a point cloud feature matrix output by the Gaussian mixture model is obtained, including:

[0022] The reconstructed point cloud data is input into the Gaussian mixture model, and the likelihood function of each control grid is calculated by the Gaussian mixture model; the obtained likelihood function is subjected to regularized gradient statistic summation processing to obtain the point cloud feature matrix.

[0023] According to a convolutional neural network-based wake field prediction method according to an embodiment of the present application, the likelihood function of each control grid is calculated; the obtained likelihood function is subjected to regularized gradient statistics summation processing to obtain the point cloud feature matrix, including:

[0024] Calculating the likelihood function of each control network based on the second calculation formula;

[0025] Wherein, the second calculation formula includes:

[0026]

[0027] Among them, u λ (p) represents the likelihood function corresponding to the control grid, λ represents the parameter set of the Gaussian mixture model with K components, λ={(w k ,μ k ,σ k ),k=1,...,K},w k represents the weight of the kth component, σ k represents the variance of the kth component, μ k represents the expectation of the kth component; p represents the coordinate information of each coordinate point in the coordinate system created based on the reconstructed point cloud data, and there is a preset corresponding relationship between the coordinate point and the control grid;

[0028] in,

[0029] The obtained likelihood function is calculated by the third calculation formula and the regularized gradient statistics are summed;

[0030] Among them, the third calculation formula includes:

[0031]

[0032] in, represents the sum of gradient statistics, T represents the number of coordinate points, L λ represents the preset information matrix, represents the preset evaluation function, p t represents the tth coordinate point;

[0033] w k Transformed to α k , obtained by the fourth calculation formula;

[0034] Wherein, the fourth calculation formula includes:

[0035]

[0036] The final point cloud feature matrix is ​​obtained through the fifth calculation formula;

[0037] Among them, the fifth calculation formula includes:

[0038]

[0039] in, Represents the point cloud feature matrix.

[0040] According to a method for predicting wake fields based on a convolutional neural network in one embodiment of the present application, the training process of the convolutional neural network model includes:

[0041] A training sample set is obtained, wherein the training sample set includes: hull point cloud data samples corresponding to a plurality of ship types, and wake field samples and fluid dynamics parameter samples corresponding to the hull point cloud data samples of each ship type.

[0042] Input the training sample set into a convolutional neural network model, reconstruct the hull corresponding to the hull line through the convolutional neural network model based on the hull point cloud data samples using a control grid of a preset size to obtain a reconstructed point cloud data sample; input the reconstructed point cloud data sample into a Gaussian mixture model corresponding to the hull line to obtain a point cloud feature matrix sample output by the Gaussian mixture model; predict and output a predicted wake field sample corresponding to the current ship type based on the point cloud feature matrix sample and the fluid dynamics parameter sample;

[0043] The similarity between the wake field sample and the predicted wake field sample is compared, and the model parameters of the convolutional neural network model are optimized based on the comparison result until the similarity reaches a preset similarity, and it is determined that the training of the convolutional neural network model is completed.

[0044] According to a convolutional neural network-based wake field prediction method according to an embodiment of the present application, after obtaining the hull geometry file, the method further includes:

[0045] The first seven modal parameters corresponding to the hull geometry file are extracted based on experimental modal analysis.

[0046] According to a convolutional neural network-based wake field prediction method according to an embodiment of the present application, before obtaining the hull geometry file and the fluid dynamics parameters, the method further includes:

[0047] A three-dimensional Cartesian network corresponding to the flow field region to be analyzed is obtained, and a Gaussian mixture model is obtained based on the three-dimensional Cartesian network.

[0048] An embodiment of the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the accompanying flow field prediction method based on a convolutional neural network as described above are implemented.

[0049] The convolutional neural network-based wake field prediction method and device provided in the embodiment of the present application, by inputting the acquired hull geometry file and fluid dynamic parameters into a pre-trained convolutional neural network model, reconstructing the hull corresponding to the hull line based on the hull point cloud data through the convolutional neural network model using a control grid of a preset size to obtain reconstructed point cloud data; inputting the reconstructed point cloud data into a Gaussian mixture model corresponding to the hull line to obtain a point cloud feature matrix output by the Gaussian mixture model; based on the point cloud feature matrix and the fluid dynamic parameters, predicting and outputting the predicted wake field corresponding to the current ship type, the present application uses a convolutional neural network model to obtain the hull geometry configuration and the corresponding fluid dynamic parameters, and uses a Gaussian mixture model to extract high-dimensional input features, predicting the wake field corresponding to the high-dimensional input features and the fluid dynamic parameters, effectively improving the accuracy and efficiency of the wake field prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 It is a flow chart of a method for predicting a wake field based on a convolutional neural network provided in an embodiment of the present application;

[0052] Figure 2 It is a structural schematic diagram of a wake field prediction device based on a convolutional neural network provided in an embodiment of the present application;

[0053] Figure 3 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] The embodiment of the present application provides a method for predicting the wake field based on a convolutional neural network. The method can be applied to a smart terminal, a server, or a ship's controller. The present application uses the method applied to a ship's controller as an example for illustration, and some other descriptions in the embodiment are for illustrative purposes only and are not used to limit the scope of protection of the present application, and will not be described one by one later. The specific implementation of the method is as follows Figure 1 As shown:

[0056] Step 101, obtaining hull geometry files and fluid dynamics parameters.

[0057] The hull geometry file includes: hull point cloud data, which is obtained based on the hull lines; and the fluid dynamics parameters include: the water flow velocity and the physical properties of water corresponding to the ship when sailing.

[0058] Step 102: input the hull geometry file and fluid dynamics parameters into a pre-trained convolutional neural network model; reconstruct the hull corresponding to the hull line using a control grid of a preset size based on the hull point cloud data through the convolutional neural network model to obtain reconstructed point cloud data; input the reconstructed point cloud data into a Gaussian mixture model corresponding to the hull line to obtain a point cloud feature matrix output by the Gaussian mixture model; and predict and output the predicted wake field corresponding to the current ship type based on the point cloud feature matrix and the fluid dynamics parameters.

[0059] Among them, the hull lines correspond to the ship type.

[0060] Among them, the convolutional neural network model is trained based on hull geometry file samples, fluid dynamics parameter samples and wake field samples.

[0061] Among them, the physical properties of water include: water density and water viscosity, etc.

[0062] Specifically, the process of obtaining the point cloud feature matrix of the reconstructed point cloud data through the Gaussian mixture model is equivalent to obtaining the FV expression of the reconstructed point cloud data.

[0063] Specifically, the hull point cloud data can indicate information such as the length, width, draft, bow shape, stern shape, propeller type, size and position, etc. of the hull.

[0064] The convolutional neural network-based wake field prediction method provided in the embodiment of the present application is characterized in that the acquired hull geometry file and fluid dynamic parameters are input into a pre-trained convolutional neural network model, and the hull corresponding to the hull line is reconstructed by the convolutional neural network model based on the hull point cloud data using a control grid of a preset size to obtain reconstructed point cloud data; the reconstructed point cloud data is input into a Gaussian mixture model corresponding to the hull line to obtain a point cloud feature matrix output by the Gaussian mixture model; based on the point cloud feature matrix and the fluid dynamic parameters, the predicted wake field corresponding to the current ship type is predicted and output. The present application utilizes a convolutional neural network model to obtain the hull geometry configuration and the fluid dynamic parameters corresponding thereto, and utilizes a Gaussian mixture model to extract high-dimensional input features, and predicts the wake field corresponding to the high-dimensional input features and the fluid dynamic parameters, thereby effectively improving the accuracy and efficiency of the wake field prediction.

[0065] In a specific embodiment, based on the hull point cloud data, a hull corresponding to the hull line is reconstructed using a control grid of a preset size, and a specific implementation of obtaining the reconstructed point cloud data includes:

[0066] Based on the preset curvature and the hull point cloud data, the control grid is resampled in the width and height directions of the hull point cloud data to obtain the reconstructed quantity; the control grid based on the reconstructed quantity is fitted with a three-dimensional hull model corresponding to the hull line to obtain the reconstructed point cloud data.

[0067] The three-dimensional hull model includes: a hull part, a bow part, a stern part and a tail shaft part. One part of the three-dimensional hull model corresponds to a curved surface, and different curves are in G1 continuity.

[0068] The size of the control grid can be set by the user according to their actual needs, and this application does not impose any restrictions. For example, a 60*40 control grid (or control point) is used for fitting operations. The unit user of the control network can set it according to their actual needs, and this application does not impose any restrictions.

[0069] Specifically, the process is to perform NURBS fitting on the hull point cloud data, use a 60*40 control grid to approximate the entire hull, resample based on the preset curvature, and obtain the reconstructed point cloud data of the reconstructed quantity.

[0070] Specifically, the point cloud reconstruction process is to fit the hull through NURBS surfaces. Since the topology of the hull is relatively complex, four curves are generally required for fitting, including: the hull part, the bow part, the stern part and the tail shaft part. Different surfaces need to satisfy G1 continuity, so it is necessary to align the NURBS nodes at the junction of the surfaces and adjust the position of the control grid to meet G1 continuity.

[0071] Specifically, the reconstructed point cloud data performs resampling operations on the width direction and height direction of the hull point cloud data according to the control grid, and generally uniform sampling along the arc length can be used.

[0072] Among them, the reconstruction number of different control networks will affect the reconstruction accuracy, which can be adjusted appropriately according to the size of the hull.

[0073] In a specific embodiment, the specific implementation of fitting the three-dimensional hull model corresponding to the hull line based on the control grid of the reconstructed quantity to obtain the reconstructed point cloud data includes:

[0074] The reconstructed point cloud data is obtained through the first calculation formula.

[0075] The first calculation formula is shown in formula (1):

[0076]

[0077] Among them, S(u,v) represents the reconstructed point cloud data, P i,j represents the control grid, which includes the i direction and the j direction, ω i,j represents the weight factor of the corresponding control grid, u represents the node in the i direction, the corresponding total number of nodes is r+1, p represents the number of spline curves in the i direction, v represents the node in the j direction, the corresponding total number of nodes is S+1, q represents the number of spline curves in the j direction, N i,p (u) represents the non-rational B-spline basis function on the knot vector U; N j,q (v) represents the non-rational B-spline basis function on the node vector V, n represents the number of reconstructions of the control grid in the i direction, and m represents the number of reconstructions of the control grid in the j direction.

[0078] in,

[0079] Specifically, if the ship type is obtained at the very beginning, the ship type can be approximated by NURBS curves through the least squares method, and after local smoothing control at the junction of the bow and stern, the reconstructed point cloud data can be obtained.

[0080] In a specific embodiment, before the reconstructed point cloud data is input into the Gaussian mixture model corresponding to the hull lines, the characteristics of the hull corresponding to the hull lines are analyzed, and based on the preset correspondence between the characteristics of the hull and the Gaussian mixture model, a Gaussian mixture model corresponding to the characteristics of the hull is obtained.

[0081] Specifically, the hull lines intersect the hull through three sets of mutually perpendicular planes, and the three sets of curves (waterline, transverse section line, and longitudinal section line) obtained represent the shape of the hull in all directions. The projections of these curves on the corresponding planes, such as the transverse section view, longitudinal section view, and waterline plane view, together constitute the hull line diagram, which is used to describe the three-dimensional shape of the hull. In the hull line diagram, the transverse section view can show the shape and contour of the hull at different positions (i.e., different cross-sections), including the shape changes of the bow, the middle of the ship, and the stern. The longitudinal section view shows the shape and contour of the hull along the centerline plane and its parallel planes, reflecting the longitudinal characteristics of the hull. The characteristics of the hull are obtained based on the above information.

[0082] The correspondence between the characteristics of the hull and the Gaussian mixture model is preset, and the Gaussian mixture model is obtained based on the correspondence.

[0083] Specifically, different Gaussian mixture models correspond to different likelihood functions.

[0084] This application selects different Gaussian mixture models through the characteristics of different hulls, which can more accurately obtain the point cloud feature matrix of reconstructed point cloud data, providing an effective data basis for subsequent wake field prediction.

[0085] In a specific embodiment, the reconstructed point cloud data is input into a Gaussian mixture model corresponding to the hull line, and the specific implementation of obtaining the point cloud feature matrix output by the Gaussian mixture model includes:

[0086] The reconstructed point cloud data is input into the Gaussian mixture model, and the likelihood function of each control grid is calculated through the Gaussian mixture model; the obtained likelihood function is summed by regularizing the gradient statistics to obtain the point cloud feature matrix.

[0087] The regularized gradient statistic summing process includes: performing regularization on the loss function of the Gaussian mixture model, calculating the gradient on the regularized loss function, and finally summing the obtained gradients.

[0088] In a specific embodiment, the likelihood function of each control grid is calculated; the obtained likelihood function is subjected to regularized gradient statistic summation processing, and the specific implementation of obtaining the point cloud feature moment includes:

[0089] The likelihood function of each control network is calculated by the second calculation formula.

[0090] The second calculation formula is shown in formula (2):

[0091]

[0092] Among them, u λ (p) represents the likelihood function corresponding to the control grid, λ represents the parameter set of the Gaussian mixture model with K components, λ={(wk ,μ k ,σ k ),k=1,...,K},w k represents the weight of the kth component, σ k represents the variance of the kth component, μ k represents the expectation of the kth component; p represents the coordinate information of each coordinate point in the coordinate system created based on the reconstructed point cloud data. There is a preset correspondence between the coordinate point and the control grid. The user can set the correspondence according to their actual needs. It can be one-to-one, many-to-one or one-to-many.

[0093] in,

[0094] The third calculation formula is used to calculate the likelihood function and perform regularized gradient statistics summation processing.

[0095] The third calculation formula is shown in formula (3):

[0096]

[0097] in, represents the sum of gradient statistics, T represents the number of coordinate points, L λ represents the preset information matrix, represents the preset evaluation function, p t Represents the tth coordinate point.

[0098] Specifically, in order to ensure that the likelihood function has a valid distribution and simplify the gradient calculation, w k Transformed to α k , obtained by the fourth calculation formula, the relationship between the two is shown in formula (4):

[0099]

[0100] The final point cloud feature matrix is ​​obtained through the fifth calculation formula.

[0101] The fifth calculation formula is shown in formula (5):

[0102]

[0103] in, Represents the point cloud feature matrix.

[0104] Specifically, this process is equivalent to FV expression. Using a Gaussian mixture model with K components, T coordinate points with three-directional scalars (i.e., 3T variables) can be expressed using a Fisher Vector with 7K elements, where each element is a continuous function of 3T variables. When T<7K3, the system of equations that determine the positions of unknown point clouds by known Fisher components is hyperstatic, so its only solution is the original point cloud coordinate point. When the solution to the equation is unique, the FV expression does not lose any information and is equivalent to the original point cloud. Finally, a four-dimensional matrix is ​​obtained.

[0105] In a specific embodiment, the training process of the convolutional neural network model includes:

[0106] A training sample set is obtained, wherein the training sample set includes: hull point cloud data samples corresponding to a plurality of ship types, and wake field samples and fluid dynamics parameter samples corresponding to the hull point cloud data samples of each ship type.

[0107] Input the training sample set into a convolutional neural network model, reconstruct the hull corresponding to the hull line through the convolutional neural network model based on the hull point cloud data samples using a control grid of a preset size to obtain a reconstructed point cloud data sample; input the reconstructed point cloud data sample into a Gaussian mixture model corresponding to the hull line to obtain a point cloud feature matrix sample output by the Gaussian mixture model; predict and output a predicted wake field sample corresponding to the current ship type based on the point cloud feature matrix sample and the fluid dynamics parameter sample;

[0108] The similarity between the wake field sample and the predicted wake field sample is compared, and the model parameters of the convolutional neural network model are optimized based on the comparison result until the similarity reaches a preset similarity, and it is determined that the training of the convolutional neural network model is completed.

[0109] In a specific embodiment, after the hull geometry file is obtained, the first seven modal parameters corresponding to the hull geometry file are extracted based on experimental modal analysis.

[0110] Specifically, the first seven-order modal parameters and fluid dynamic parameters are input into a convolutional neural network model, and the hull corresponding to the hull line is reconstructed by the convolutional neural network model based on the first seven-order modal parameters using a control grid of a preset size to obtain reconstructed point cloud data; the reconstructed point cloud data is input into a Gaussian mixture model corresponding to the hull line to obtain a point cloud feature matrix output by the Gaussian mixture model; and the predicted wake field corresponding to the current ship type is predicted and output based on the point cloud feature matrix and the fluid dynamic parameters.

[0111] Among them, modal analysis is a technique used to determine the dynamic characteristics of a system, which can identify the modal parameters of the system, including modal frequency, modal damping ratio, modal shape, etc. These modal parameters can be used to describe the response characteristics of the system under specific excitation.

[0112] Specifically, the first 7 order modal parameters can be ship type-dependent, that is, their main characteristics are more sensitive to changes in hull geometry, so the present application adopts the first 7 order modal parameters.

[0113] In a specific embodiment, after obtaining the training sample set, the first 7-order modal parameter samples corresponding to the hull point cloud data samples are extracted based on experimental modal analysis.

[0114] Specifically, the wake field samples and fluid dynamics parameter samples of the first seven-order modal parameter samples are input into a convolutional neural network model, and the hull corresponding to the hull line is reconstructed by the convolutional neural network model based on the seven-order modal parameter samples using a control grid of a preset size to obtain reconstructed point cloud data samples; the reconstructed point cloud data samples are input into a Gaussian mixture model corresponding to the hull line to obtain point cloud feature matrix samples output by the Gaussian mixture model; and the predicted wake field samples corresponding to the current ship type are predicted and output based on the point cloud feature matrix samples and the fluid dynamics parameter samples.

[0115] In a specific embodiment, before obtaining the hull geometry file and the fluid dynamics parameters, a three-dimensional Cartesian network corresponding to the flow field area to be analyzed is obtained, and a Gaussian mixture model is obtained based on the three-dimensional Cartesian network.

[0116] Specifically, the three-dimensional Cartesian network is composed of three-dimensional grid units, and the center point of each Gaussian mixture model is located at the center of the three-dimensional Cartesian network. The FV expression generated by this Gaussian mixture model can retain the structure of the point cloud dataset, that is, the range of Fisher components that can be significantly affected by a point appearing at a specific three-dimensional spatial position is limited.

[0117] This application combines NURBS surface reconstruction, FV hull point cloud recognition, deep convolutional neural network and flow field POD decomposition method (corresponding to experimental modal analysis) to establish a convolutional neural network model from geometric point cloud to flow field distribution at specified position (corresponding to fluid dynamics parameters, i.e. fluid dynamics parameters at specified position, different positions correspond to different fluid dynamics parameters). The hull value points are reduced in dimension through FV expression, and the flow field reconstruction on the specified interface is realized through a two-dimensional deconvolutional neural network, realizing the rapid prediction capability of complex geometric flow field based on point cloud input.

[0118] After training the wake field dataset of this application, the obtained model can more accurately predict the wake field of a given ship point cloud. The prediction accuracy is fully verified on both the validation set and the test set, proving that the deep proxy model (convolutional neural network model) constructed by this method can well capture the relationship between complex geometric configurations and their flow field characteristics. In the encoding-decoding framework, a deep proxy model from geometric point cloud to typical flow field is designed based on a deep convolutional neural network, which better solves the problem of the proxy model's dependence on the geometric parameterized modeling method.

[0119] The deep proxy model developed in this paper can effectively capture the main flow characteristics of the ship's wake field, and can more accurately predict the direction and magnitude of the lateral velocity at the propeller disk. It shows a relatively stable prediction ability for the wake fields of different ship types, different vortex core positions and sizes, and can capture the main characteristics of the flow field. This proves that the prediction method constructed in this paper can better capture the mapping relationship between the complex geometry of the ship type and the flow field characteristics.

[0120] The embodiment of the present application also provides a wake field prediction device based on a convolutional neural network. The specific implementation of the device can refer to the description of the wake field prediction method based on a convolutional neural network, and the repeated parts will not be repeated. Figure 2 As shown, the device comprises:

[0121] The acquisition module 201 is used to acquire a hull geometry file and fluid dynamics parameters, wherein the hull geometry file includes: hull point cloud data, the hull point cloud data is obtained based on the hull line, and the fluid dynamics parameters include: the water flow velocity and the physical properties of water corresponding to the ship when sailing;

[0122] The prediction module 202 is used to input the hull geometry file and the fluid dynamics parameters into the pre-trained convolutional neural network model, reconstruct the hull corresponding to the hull line based on the hull point cloud data by using the control grid of a preset size through the convolutional neural network model to obtain the reconstructed point cloud data; input the reconstructed point cloud data into the Gaussian mixture model corresponding to the hull line to obtain the point cloud feature matrix output by the Gaussian mixture model; predict and output the predicted wake field corresponding to the current ship type based on the point cloud feature matrix and the fluid dynamics parameters;

[0123] Among them, the hull lines correspond to the ship type;

[0124] Among them, the convolutional neural network model is trained based on hull geometry file samples, fluid dynamics parameter samples and wake field samples.

[0125] Figure 3 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 3As shown, the electronic device may include: a processor 301, a communication interface 302, a memory 303 and a communication bus 304, wherein the processor 301, the communication interface 302 and the memory 303 communicate with each other through the communication bus 304. The processor 301 may call the logic instructions in the memory 303 to execute the convolutional neural network-based wake field prediction method.

[0126] In addition, the logic instructions in the above-mentioned memory 303 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0127] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the convolutional neural network-based flow field prediction method provided by the above-mentioned methods.

[0128] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the convolutional neural network-based flow field prediction method provided in the above-mentioned embodiments.

[0129] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0130] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0131] Finally, it should be noted that the above is only the preferred implementation of the present application, and the present application is not limited to the above embodiments. It is understood that other improvements and changes directly derived or associated by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the scope of protection of the present application.

Claims

1. A method for predicting wake fields based on convolutional neural networks, characterized in that: The method comprises: Acquire a hull geometry file and fluid dynamics parameters, wherein the hull geometry file includes: hull point cloud data, the hull point cloud data is obtained based on the hull line, and the fluid dynamics parameters include: water flow velocity and physical properties of water corresponding to the ship when sailing; The hull geometry file and the fluid dynamics parameters are input into a pre-trained convolutional neural network model, and the hull corresponding to the hull line is reconstructed by the convolutional neural network model based on the hull point cloud data using a control grid of a preset size to obtain reconstructed point cloud data; the reconstructed point cloud data is input into a Gaussian mixture model corresponding to the hull line to obtain a point cloud feature matrix output by the Gaussian mixture model; based on the point cloud feature matrix and the fluid dynamics parameters, a predicted wake field corresponding to the current ship type is predicted and output; Wherein, the hull lines correspond to the ship type; The convolutional neural network model is trained based on hull geometry file samples, fluid dynamics parameter samples and wake field samples.

2. The method for predicting wake field based on convolutional neural network according to claim 1, characterized in that: Before inputting the reconstructed point cloud data into the Gaussian mixture model corresponding to the hull line, the method further includes: The characteristics of the hull corresponding to the hull lines are analyzed, and based on the preset corresponding relationship between the characteristics of the hull and the Gaussian mixture model, a Gaussian mixture model corresponding to the characteristics of the hull is obtained.

3. The method for predicting wake field based on convolutional neural network according to claim 1 or 2, characterized in that: Reconstructing the hull corresponding to the hull line using a control grid of a preset size based on the hull point cloud data to obtain reconstructed point cloud data, including: Based on the preset curvature, resampling operations are performed on the control grid in the width direction and the height direction of the hull point cloud data to obtain a reconstruction quantity; The three-dimensional hull model corresponding to the hull line is fitted based on the control grid of the reconstructed quantity to obtain the reconstructed point cloud data, wherein the three-dimensional hull model includes: a hull part, a bow part, a stern part and a tail shaft part, one part of the three-dimensional hull model corresponds to a curved surface, and different curves are in G1 continuity.

4. The method for predicting wake field based on convolutional neural network according to claim 3, characterized in that: Fitting the three-dimensional hull model corresponding to the hull line based on the control grid of the reconstructed number to obtain the reconstructed point cloud data includes: Obtaining reconstructed point cloud data based on the first calculation formula; Wherein, the first calculation formula includes: Among them, S(u,v) represents the reconstructed point cloud data, P i,j represents the control grid, which includes the i direction and the j direction, ω i,j represents the weight factor of the corresponding control grid, u represents the node in the i direction, the corresponding total number of nodes is r+1, p represents the number of spline curves in the i direction, v represents the node in the j direction, the corresponding total number of nodes is S+1, q represents the number of spline curves in the j direction, N i,p (u) represents the non-rational B-spline basis function on the knot vector U; N j,q (v) represents the non-rational B-spline basis function on the node vector V, n represents the number of reconstructions of the control grid in the i direction, and m represents the number of reconstructions of the control grid in the j direction; in, 5. The method for predicting wake field based on convolutional neural network according to claim 1 or 2, characterized in that: Inputting the reconstructed point cloud data into a Gaussian mixture model corresponding to the hull line to obtain a point cloud feature matrix output by the Gaussian mixture model, including: The reconstructed point cloud data is input into the Gaussian mixture model, and the likelihood function of each control grid is calculated by the Gaussian mixture model; the obtained likelihood function is subjected to regularized gradient statistic summation processing to obtain the point cloud feature matrix.

6. The method for predicting wake field based on convolutional neural network according to claim 5, characterized in that: Calculate the likelihood function of each control grid; The obtained likelihood function is subjected to regularized gradient statistic summation processing to obtain the point cloud feature matrix, including: Calculating the likelihood function of each control network based on the second calculation formula; Wherein, the second calculation formula includes: Among them, u λ (p) represents the likelihood function corresponding to the control grid, λ represents the parameter set of the Gaussian mixture model with K components, λ={(w k ,μ k ,σ k ),k=1,...,K},w k represents the weight of the kth component, σ k represents the variance of the kth component, μ k represents the expectation of the kth component; p represents the coordinate information of each coordinate point in the coordinate system created based on the reconstructed point cloud data, and there is a preset corresponding relationship between the coordinate point and the control grid; in, The obtained likelihood function is calculated by the third calculation formula and the regularized gradient statistics are summed up; The third calculation formula includes: in, represents the sum of gradient statistics, T represents the number of coordinate points, L λ represents the preset information matrix, represents the preset evaluation function, p t Represents the tth coordinate point; w k Transformed to α k , obtained by the fourth calculation formula; Wherein, the fourth calculation formula includes: The final point cloud feature matrix is ​​obtained through the fifth calculation formula; Among them, the fifth calculation formula includes: in, Represents the point cloud feature matrix.

7. The method for predicting wake field based on convolutional neural network according to claim 1 or 2, characterized in that: The training process of the convolutional neural network model includes: A training sample set is obtained, wherein the training sample set includes: hull point cloud data samples corresponding to a plurality of ship types, and wake field samples and fluid dynamics parameter samples corresponding to the hull point cloud data samples of each ship type. Input the training sample set into a convolutional neural network model, reconstruct the hull corresponding to the hull line through the convolutional neural network model based on the hull point cloud data samples using a control grid of a preset size to obtain a reconstructed point cloud data sample; input the reconstructed point cloud data sample into a Gaussian mixture model corresponding to the hull line to obtain a point cloud feature matrix sample output by the Gaussian mixture model; predict and output a predicted wake field sample corresponding to the current ship type based on the point cloud feature matrix sample and the fluid dynamics parameter sample; The similarity between the wake field sample and the predicted wake field sample is compared, and the model parameters of the convolutional neural network model are optimized based on the comparison result until the similarity reaches a preset similarity, and it is determined that the training of the convolutional neural network model is completed.

8. The method for predicting wake field based on convolutional neural network according to claim 1 or 2, characterized in that: After obtaining the hull geometry file, it also includes: The first seven modal parameters corresponding to the hull geometry file are extracted based on experimental modal analysis.

9. The method for predicting wake field based on convolutional neural network according to claim 1 or 2, characterized in that: Before obtaining the hull geometry file and fluid dynamics parameters, it also includes: A three-dimensional Cartesian network corresponding to the flow field region to be analyzed is obtained, and a Gaussian mixture model is obtained based on the three-dimensional Cartesian network.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method for predicting the accompanying flow field based on a convolutional neural network as described in any one of claims 1 to 9 are implemented.

Images