Intelligent fast prediction method for unsteady force of waterjet
By integrating the whole ship model with deep learning technology, a prediction model for turbulent wake parameters and unsteady forces was constructed, which solved the problems of slow prediction speed and insufficient accuracy of unsteady forces in waterjet propulsion, and achieved efficient and fast unsteady force prediction, supporting the low-noise optimization design of waterjet propulsion.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- RES INST 708 OF CHINA STATE SHIPBUILDING CORP
- Filing Date
- 2026-03-04
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies for predicting unsteady forces in waterjet propulsion systems suffer from high computational resource consumption, slow speed, and insufficient accuracy, making it difficult to meet the requirements for real-time and high precision.
By integrating turbulent wake calculation with deep learning technology using a full ship model, and combining support vector regression (SVR) and multilayer perceptron (MLP) techniques, a prediction model for turbulent wake parameters at the stern and an intelligent rapid forecasting model for unsteady forces are constructed, reducing computational resource requirements and improving forecasting speed and accuracy.
It achieves high-fidelity, millisecond-level prediction of unsteady forces in waterjet propulsion, improving the iterative efficiency and design level of low-noise optimization design for waterjet propulsion.
Smart Images
Figure CN122433201A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a shipbuilding and marine engineering technology, and in particular to an intelligent and rapid prediction method for unsteady forces in waterjet propulsion. Background Technology
[0002] Waterjet propulsion utilizes the momentum difference between incoming and outgoing water to generate thrust, propelling the platform forward. It boasts advantages such as high propulsion efficiency, excellent maneuverability, and low radiated noise. The device comprises a duct, rotor, and stator, making it the preferred propulsion system for underwater vehicles. However, its operating environment is complex, often facing unsteady conditions such as non-uniform incoming flow, maneuvering and variable speed, and oblique flow. This results in the rotor, stator, and other components experiencing strong unsteady fluid excitation forces. These unsteady forces are key factors contributing to propulsion noise, and accurate and rapid prediction of these unsteady forces is crucial for the stealth, safety, and reliability of underwater vehicles. Currently, the main prediction methods include empirical modeling, model testing, and computational fluid dynamics (CFD). Among these methods, empirical modeling, based on lifting line / lift surface theory or simplified formulas, offers fast computation but relies heavily on steady-state assumptions, making it difficult to accurately capture core unsteady physical phenomena such as rotor-stator interference, vortex shedding, and cavitation collapse, resulting in limited prediction accuracy. Model testing, conducted in cavitation tanks or circulating water tanks, yields high-confidence results but is extremely costly, time-consuming, and suffers from unavoidable scale and installation effects, making it unsuitable for real-time ship-based prediction. CFD methods, especially Delayed Separated Eddy Stirring (DDES) or Large Eddy Simulation (LES), provide detailed flow field and force information, but consume enormous computational resources, are highly dependent on grids and turbulence models, and single calculations can take several days, severely limiting their application in scheme optimization, fault diagnosis, and real-time control. Therefore, there is an urgent need in this field for a new method that can overcome the contradiction between "accuracy" and "efficiency" to achieve high-fidelity, millisecond-level prediction of unsteady forces in waterjet propulsion. Summary of the Invention
[0003] To address the problem that current CFD calculations for unsteady forces in waterjet propulsion systems are generally based on a full-ship + propulsion model, resulting in tens or even hundreds of millions of meshes and requiring consideration of numerous operating conditions, thus consuming enormous computational resources, this invention proposes an intelligent and rapid prediction method for unsteady forces in waterjet propulsion systems. This invention uses a full-ship model for turbulent wake calculations at the stern position and a stern + propulsion model for unsteady force calculations, fusing CFD calculations with deep learning techniques in two stages. The former is integrated with Support Vector Regression (SVR) to train a stern turbulent wake parameter prediction model based on finite-condition calculation results, providing inflow boundary conditions for the latter. The latter is integrated with Multilayer Perceptron (MLP) technology to form an intelligent and rapid prediction model and program for unsteady forces based on finite-condition calculation results, thereby achieving unsteady force prediction for waterjet propulsion systems with near-CFD accuracy and significantly faster speed.
[0004] The technical solution of this invention is as follows:
[0005] A smart and rapid prediction method for unsteady forces in waterjet propulsion includes: Generation of a prediction model for stern turbulent wake boundary conditions: First, a hydrodynamic calculation model of the underwater vehicle's hull is established and the hull parameters are transformed. Then, multi-condition LES simulations are conducted for hulls with different parameters, and a turbulent wake database is formed based on the simulation results. Using this turbulent wake database as a training set, deep learning training is performed in conjunction with SVR technology to construct a prediction model for stern turbulent wake parameters. This prediction model for stern turbulent wake parameters can quickly predict and obtain the turbulent wake parameters at the stern position based on given hull parameters and operating condition parameters, providing inflow boundary conditions for subsequent calculation of unsteady forces of the waterjet propulsion system. Establishment of an intelligent rapid prediction model for unsteady forces: A stern + propeller model is established, and stern line shape transformation, stern rudder geometry transformation, and propeller arrangement transformation are performed. Data from the turbulent wake database at the stern position is used as the inlet boundary, and batch DDES simulations are conducted. Based on the DDES simulation results, an unsteady force database is established. Using this unsteady force database as the training set, deep learning training is performed using MLP (Multilayer Perceptron) technology to construct an intelligent rapid prediction model for unsteady forces and form the corresponding program. This intelligent rapid prediction model and program predicts the unsteady forces of the waterjet propeller based on the given stern line shape, stern rudder geometry, and propeller arrangement.
[0006] Furthermore, the hydrodynamic calculation model for the hull adopts a teardrop-shaped rotating body with good drag performance; based on the governing equations, hull parameters can be transformed to establish hull models with different parameters; the teardrop-shaped rotating hull is divided into bow, midship, and stern sections, and the radii of rotation of each section are expressed according to the following governing equations:
[0007] In the formula, the fixed parameters include the ship's length. L , bow section length L 1. Midships section length L 2. Stern section length L 3, of which L= L 1+ L 2+ L 3; x Let [the coordinates] be the position coordinates along the length of the ship, with the center of the ship as the farthest point, and its range is [-]. L / 2, L / 2], the stern direction is negative, and the bow direction is positive; Let be the radius of rotation at different coordinate positions, where the subscript is the radius of rotation at different coordinate positions. i The numbers 1, 2, and 3 represent the bow section, midship section, and stern section, respectively. R The radius of slewing of the midship section is usually a constant. α 1,α 2, α 3, α 4 is the shape factor, which is also the main transformable parameter. α 1 and α 2 represents the bow linear shape factor. α 3 and α 4 represents the stern shape factor. By giving different shape factors, the hull shape can be quantitatively changed.
[0008] Furthermore, different parameter hull models were established as simulation objects. Within the speed range of 0kn to 40kn, different speed conditions were taken at intervals of 1kn, 2kn, or 5kn, and batch LES simulations were carried out. The interval values were flexibly adjusted according to the accuracy requirements of unsteady force prediction. The smaller the interval, the higher the accuracy.
[0009] Furthermore, at the stern inlet section, x = - L / 2+ L 3. Nodes are set at equal intervals, and the flow field information of all nodes is extracted based on the LES simulation results of the hull, constructing a turbulent wake database at the stern position; each data point in the turbulent wake database corresponds to a single node, specifically containing: hull shape factor. α 1, α 2, α 3, α 4; Ship speed V n ; Node location information, including X, Y, and Z components; Flow field velocity at the node, including VelocityX, VelocityY, and VelocityZ components; Turbulent intensity at the node, Turbulent_Intensity.
[0010] Furthermore, after the turbulent wake database at the stern position is constructed, deep learning training is performed using SVR technology with the hull shape factor, hull speed, and the position information of different nodes on the stern inlet section as input parameters, and the flow field velocity and turbulence intensity of different nodes on the stern inlet section as output parameters. The core idea of SVR technology is to find a function such that the flow field velocity and turbulence intensity at most nodes fall within a width of 2. Within the strip-shaped region; the SVR objective function is:
[0011] In the formula, These are the weight vectors for different nodes; For bias terms; These are positive and negative relaxation variables, used to improve the model's robustness to noise and outliers. This is a regularization term that controls the smoothness of the model. As a punishment factor, it controls the degree of tolerance; subscript j Representing different nodes; The SVR constraint is:
[0012]
[0013]
[0014] In the formula, These represent the true values of flow field velocity and turbulence intensity at different nodes; Here are the predicted values, where This is the transpose of the weight vector. It is a nonlinear function used to transform the original spatial samples Mapping to a higher-dimensional space; using the Lagrange method, it can be transformed into a dual problem for solution, ultimately yielding:
[0015] In the formula, n is the total number of nodes. For the Lagrange multipliers corresponding to the upper boundary constraints; For the Lagrange multipliers corresponding to the lower boundary constraints; Let be the Gaussian kernel function, defined as the inner product of two high-dimensional mappings, i.e. .
[0016] Furthermore, after training with SVR deep learning, a prediction model for stern turbulent wake parameters is constructed. Based on this prediction model, given the hull shape factor, hull speed, and the position information of different nodes on the stern inlet section, the flow field velocity and turbulence intensity at different nodes on the stern inlet section can be quickly obtained.
[0017] Furthermore, since the data in the turbulent wake database at the stern position is used as the inlet boundary, the unsteady force numerical DDES simulation does not require wake calculation. Therefore, the stern + propeller model is used instead of the whole ship + propeller model. The stern + propeller model includes the stern section of the hull, rudder, duct, stator, and rotor; the stern lines are defined by a shape factor. α 3 and α 4. The tail rudder geometry is transformed into a cross-shaped tail rudder structure and an X-shaped tail rudder structure. The propeller arrangement is transformed by different numbers of stator blades, different numbers of rotor blades, and different stator-rotor spacing.
[0018] Furthermore, using the established stern + propeller model as the simulation object, batch DDES simulations were conducted under different speed conditions. Based on the simulation results, an unsteady force database was established. The individual data entries in the unsteady force database consist of: hull shape factor. α 3 and α 4; Ship speed V n Tail rudder geometry, cruciform or X-shaped tail rudder structure; propeller arrangement, number of stator blades. n rotor Different numbers of rotor blades n stator Different rotor-stator spacing D ; unsteady force F unsteady After the unsteady force database is constructed, deep learning training is performed using MLP technology with hull shape factor, hull speed, rudder geometry, and propeller arrangement as input parameters and unsteady forces as output parameters. The mathematical model is as follows:
[0019] In the formula, For the first l Net input to layer neurons; To connect the ( l -1) layer neurons to the first l The weights of layer neurons; For the ( l -1) Activation output of neurons in layer 1; For the first l Bias of layer neurons; It is an activation function, using the ReLU activation function, in the form of: The training process includes parameter initialization, forward propagation, loss calculation, backpropagation, parameter update, and regularization.
[0020] Furthermore, after training with MLP deep learning, an intelligent rapid prediction model for unsteady forces is constructed, and a corresponding program is written. Based on this intelligent rapid prediction model and program for unsteady forces, the unsteady forces of waterjet propulsion can be quickly predicted and obtained given the hull shape factor, hull speed, rudder geometry, and propeller arrangement.
[0021] Optionally, the intelligent rapid prediction method for unsteady forces in waterjet propulsion is designed for underwater vehicles and underwater waterjet propulsion. When applying it to other types of underwater surface vessels and propulsion systems, it is only necessary to redefine the input and output parameters of the turbulent wake boundary condition prediction model and the unsteady force intelligent rapid prediction model.
[0022] The beneficial effects of this invention are as follows: The method proposed in this invention has significant advantages in predicting velocity compared with the commonly used unsteady force CFD calculation method. It helps to accelerate the iteration efficiency of the scheme in the low-noise optimization design of waterjet propulsion and improve the low-noise design level of waterjet propulsion. Attached Figure Description
[0023] Figure 1 This is a diagram showing the main steps of the intelligent rapid prediction method for unsteady forces in waterjet propulsion systems according to the present invention. Figure 2 This is a model diagram of the ship's hull for the present invention; Figure 3 This is a schematic diagram of the input and output of the prediction model for turbulent wake parameters at the stern of the ship according to the present invention; Figure 4 This is a model diagram of the stern and propeller of the present invention; Figure 5 This is a schematic diagram of the input and output of the intelligent rapid prediction of unsteady forces according to the present invention. Detailed Implementation
[0024] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.
[0025] This invention relates to an intelligent and rapid prediction method for unsteady forces in waterjet propulsion, comprising seven main steps, such as... Figure 1 Its core components include two parts: the first part is the generation of a predictive model for turbulent wake parameters at the stern position, which incorporates support vector regression (SVR) technology; the second part is the establishment of an intelligent rapid prediction method for unsteady forces, which incorporates multilayer perceptron (MLP) technology.
[0026] The detailed implementation process is as follows: Generation of a prediction model for turbulent wake boundary conditions at the stern: First, a hydrodynamic calculation model of the underwater vehicle's hull is established and its parameters are transformed. Then, multi-condition LES simulations are conducted for hulls with different parameters, and a turbulent wake database is formed based on the simulation results. Using this database as a training set, deep learning training is performed using SVR technology to construct a prediction model for turbulent wake parameters at the stern. This model can quickly predict and obtain the turbulent wake parameters at the stern position based on given hull parameters and operating condition parameters, providing inflow boundary conditions for subsequent calculations of unsteady forces in the waterjet propulsion system.
[0027] The hydrodynamic calculation model for the hull uses a teardrop-shaped rotating body with good drag performance, such as... Figure 2 .
[0028] Based on the governing equations, hull parameters can be transformed, and hull models with different parameters can be established. The teardrop-shaped hull of revolution is divided into bow, midship, and stern sections, and the turning radius of each section is expressed by the following governing equations:
[0029] In the formula, the fixed parameters include the ship's length. L , bow section length L 1. Midships section length L 2. Stern section length L 3, of which L= L 1+ L 2+ L 3; x Let [the coordinates] be the position coordinates along the length of the ship, with the center of the ship as the farthest point, and its range is [-]. L / 2, L / 2], the stern direction is negative, and the bow direction is positive; Let be the radius of rotation at different coordinate positions, where the subscript is the radius of rotation at different coordinate positions. i The numbers 1, 2, and 3 represent the bow section, midship section, and stern section, respectively. R The radius of slewing of the midship section is usually a constant. α 1, α 2, α 3, α 4 is the shape factor, which is also the main transformable parameter. α 1 and α 2 represents the bow linear shape factor. α 3 and α 4 represents the stern shape factor. By giving different shape factors, the hull shape can be quantitatively changed.
[0030] Using the aforementioned ship models with different parameters as simulation objects, different speed conditions were simulated within the speed range of 0 knots to 40 knots, with intervals of 1 knot, 2 knots, or 5 knots, and batch LES simulations were conducted. The interval values can be flexibly adjusted according to the accuracy requirements of unsteady force prediction; the smaller the interval, the higher the accuracy.
[0031] At the stern inlet section (x=- L / 2+ L 3) Nodes are set at equal intervals, and the flow field information of all nodes is extracted based on the LES simulation results of the hull, constructing a turbulent wake database at the stern. Each data point in the turbulent wake database corresponds to a single node, specifically containing: hull shape factor. α 1, α 2, α 3, α 4; Ship speed V n; Node location information, including X, Y, and Z components; Flow field velocity at the node, including VelocityX, VelocityY, and VelocityZ components; Turbulent intensity at the node, Turbulent_Intensity.
[0032] After the turbulent wake database at the stern position is constructed, deep learning training is performed using SVR technology with the hull shape factor, hull speed, and the position information of different nodes on the stern inlet section as input parameters, and the flow field velocity and turbulence intensity of different nodes on the stern inlet section as output parameters.
[0033] The core idea of SVR technology is to find a function such that the flow field velocity and turbulence intensity at most nodes fall within a width of 2. Within the strip-shaped region. The SVR objective function is:
[0034] In the formula, These are the weight vectors for different nodes; For bias terms; These are positive and negative relaxation variables, used to improve the model's robustness to noise and outliers. This is a regularization term that controls the smoothness of the model. As a punishment factor, it controls the degree of tolerance; subscript j Represents different nodes.
[0035] The SVR constraint is:
[0036]
[0037]
[0038] In the formula, These represent the true values of flow field velocity and turbulence intensity at different nodes; Here are the predicted values, where This is the transpose of the weight vector. It is a nonlinear function used to transform the original spatial samples Mapping to a higher-dimensional space. Using the Lagrange method, this can be transformed into a dual problem for solution, ultimately yielding:
[0039] In the formula, n is the total number of nodes. For the Lagrange multipliers corresponding to the upper boundary constraints; For the Lagrange multipliers corresponding to the lower boundary constraints; Let be the Gaussian kernel function, defined as the inner product of two high-dimensional mappings, i.e. .
[0040] After training with SVR deep learning, a prediction model for turbulent wake parameters at the stern of a ship is constructed. Based on this prediction model, given the ship's shape factor, ship speed, and the location information of different nodes at the stern inlet section, the flow field velocity and turbulence intensity at different nodes on the stern inlet section can be quickly obtained, such as... Figure 3 .
[0041] Establishment of an Intelligent Rapid Prediction Model for Unsteady Forces: A stern + propeller model was established, and stern alignment, rudder geometry, and propeller arrangement were transformed. Data from a turbulent wake database at the stern location was used as the inlet boundary, and batch DDES simulations were conducted. Based on the DDES simulation results, an unsteady force database was established. Using this unsteady force database as the training set, deep learning training was performed using MLP (Multilayer Perceptron) technology to construct an intelligent rapid prediction model for unsteady forces and generate the corresponding program. This intelligent rapid prediction model and program can predict the unsteady forces of waterjet propellers with high fidelity and millisecond-level accuracy based on given stern alignment, rudder geometry, and propeller arrangement.
[0042] Since the data in the turbulent wake database at the stern position is used as the inlet boundary, the unsteady force numerical DDES simulation involved in the above process does not need to perform wake calculation. Therefore, the stern + propeller model can be used to replace the whole ship + propeller model, reducing the overall mesh size and saving computational resources.
[0043] The stern + propeller model includes the stern section of the hull, rudder, duct, stator, and rotor, such as... Figure 4 The stern line shape is determined by the shape factor. α 3 and α 4. The tail rudder geometry is transformed into a cross-shaped tail rudder structure and an X-shaped tail rudder structure. The propeller arrangement is transformed by different numbers of stator blades, different numbers of rotor blades, and different stator-rotor spacing.
[0044] Using the established stern + propeller model as the simulation object, batch DDES simulations were conducted under different speed conditions. An unsteady force database was then established based on the simulation results. Each data entry in the unsteady force database contains: hull shape factor. α 3 and α 4; Ship speed V n Tail rudder geometry, cruciform or X-shaped tail rudder structure; propeller arrangement, number of stator blades. n rotor Different numbers of rotor blades n stator Different rotor-stator spacing D ; unsteady forceF unsteady After the unsteady force database is constructed, deep learning training is performed using MLP technology, with hull shape factor, hull speed, rudder geometry, and propeller arrangement as input parameters and unsteady forces as output parameters. The mathematical model is as follows:
[0045] In the formula, For the first l Net input to layer neurons; To connect the ( l -1) layer neurons to the first l The weights of layer neurons; For the ( l -1) Activation output of neurons in layer 1; For the first l Bias of layer neurons; It is an activation function, using the ReLU activation function, in the form of: The training process includes parameter initialization, forward propagation, loss calculation, back propagation, parameter update, and regularization.
[0046] After training with MLP deep learning, an intelligent rapid prediction model for unsteady forces was constructed, and the corresponding program was written in Python. Based on this intelligent rapid prediction model and program for unsteady forces, given the ship's hull shape factor, hull speed, rudder geometry, and propeller arrangement, the unsteady forces of the waterjet propulsion system can be quickly predicted, such as... Figure 5 .
[0047] The present invention provides an intelligent and rapid prediction method for unsteady forces in waterjet propulsion, which can be implemented in the low-noise optimization design of waterjet propulsion. Through the establishment of a hull hydrodynamic calculation model, LES numerical calculation and turbulence wake database construction, stern turbulence wake parameter prediction model construction, stern + propulsion hydrodynamic calculation model establishment, DDES numerical calculation and unsteady force database construction, multilayer perceptron (MLP) training, and the construction of an intelligent and rapid prediction model and program for unsteady forces, high-fidelity, millisecond-level prediction of unsteady forces is achieved, supporting the waterjet propulsion's compliance with noise performance requirements.
[0048] The aforementioned intelligent rapid prediction method for unsteady forces in waterjet propulsion is primarily designed for underwater vehicles and underwater waterjet propulsion. When applying it to other types of underwater and surface vessels and propulsion systems, it is only necessary to redefine the input and output parameters of the turbulent wake boundary condition prediction model and the unsteady force intelligent rapid prediction model.
[0049] The key technology lies in the design of the entire unsteady force rapid prediction process, including the establishment of turbulent wake database and unsteady force database, the deep integration of CFD calculation and deep learning technology, the construction of stern turbulent wake boundary condition prediction model and unsteady force intelligent rapid prediction model, etc., to ultimately achieve the goal of intelligent rapid prediction of unsteady forces of waterjet propulsion.
[0050] The above-described embodiments are merely one implementation of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention should be determined by the appended claims.
Claims
1. A smart and rapid prediction method for unsteady forces in waterjet propulsion, characterized in that, include: Generation of prediction model for turbulent wake boundary conditions at the stern of a ship: First, a hydrodynamic calculation model of the underwater vehicle's hull is established and the hull parameters are transformed. Then, LES simulations under different working conditions are carried out for hulls with different parameters, and a turbulent wake database is formed based on the simulation results. Using this turbulent wake database as a training set, deep learning training was conducted using SVR technology to construct a prediction model for turbulent wake parameters at the stern of a ship. This prediction model can quickly predict and obtain the turbulent wake parameters at the stern position based on given hull parameters and operating parameters, providing inflow boundary conditions for subsequent calculation of unsteady forces of waterjet propulsion. Establishment of intelligent rapid prediction model for unsteady forces: Establish a stern + propeller model, perform stern line shape transformation, rudder geometry transformation and propeller arrangement transformation, call data from the turbulence wake database at the stern position as the inlet boundary, conduct batch DDES simulation, and combine the DDES simulation results to establish an unsteady force database; Using this unsteady force database as the training set, deep learning training is performed using MLP (Multilayer Perceptron) technology to construct an intelligent rapid prediction model for unsteady forces and form a corresponding program. This intelligent rapid prediction model and program predicts the unsteady forces of waterjet propulsion based on the given stern shape, rudder geometry, and propeller arrangement.
2. The intelligent rapid prediction method for unsteady forces in waterjet propulsion systems according to claim 1, characterized in that, The hydrodynamic calculation model of the hull adopts a teardrop-shaped rotating body with good drag performance; based on the governing equations, the hull parameters can be transformed to establish hull models with different parameters; the teardrop-shaped rotating hull is divided into bow, midship, and stern sections, and the turning radius of each section is expressed according to the following governing equations: In the formula, the fixed parameters include the ship's length. L , bow section length L 1. Midships section length L 2. Stern section length L 3, of which L= L 1+ L 2+ L 3; x Let [the coordinates] be the position coordinates along the length of the ship, with the center of the ship as the farthest point, and its range is [-]. L / 2, L / 2], the stern direction is negative, and the bow direction is positive; Let be the radius of rotation at different coordinate positions, where the subscript is the radius of rotation at different coordinate positions. i The numbers 1, 2, and 3 represent the bow section, midship section, and stern section, respectively. R The radius of slewing of the midship section is usually a constant. α 1, α 2, α 3, α 4 is the shape factor, which is also the main transformable parameter. α 1 and α 2 represents the bow linear shape factor. α 3 and α 4 represents the stern shape factor. By giving different shape factors, the hull shape can be quantitatively changed.
3. The intelligent rapid prediction method for unsteady forces in waterjet propulsion systems according to claim 2, characterized in that, Different ship models with different parameters were used as simulation objects. Within the speed range of 0kn to 40kn, different speed conditions were taken at intervals of 1kn, 2kn or 5kn, and batch LES simulations were carried out. The interval value was flexibly adjusted according to the accuracy requirements of unsteady force prediction. The smaller the interval, the higher the accuracy.
4. The intelligent rapid prediction method for unsteady forces in waterjet propulsion systems according to claim 2, characterized in that, At the stern inlet section, x = - L / 2+ L 3. Nodes are set at equal intervals, and the flow field information of all nodes is extracted based on the LES simulation results of the hull, constructing a turbulent wake database at the stern position; each data point in the turbulent wake database corresponds to a single node, specifically containing: hull shape factor. α 1, α 2, α 3, α 4; Ship speed V n ; Node location information, including X, Y, and Z components; Flow field velocity at the node, including VelocityX, VelocityY, and VelocityZ components; Turbulent intensity at the node, Turbulent_Intensity.
5. The intelligent rapid prediction method for unsteady forces in waterjet propulsion according to claim 1, characterized in that, After the turbulent wake database at the stern position is constructed, deep learning training is carried out using SVR technology with the hull shape factor, hull speed and the position information of different nodes on the stern inlet section as input parameters, and the flow field velocity and turbulence intensity of different nodes on the stern inlet section as output parameters. The core idea of SVR technology is to find a function such that the flow field velocity and turbulence intensity at most nodes fall within a width of 2. Within the strip-shaped region; the SVR objective function is: In the formula, These are the weight vectors for different nodes; For bias terms; These are positive and negative relaxation variables, used to improve the model's robustness to noise and outliers. This is a regularization term that controls the smoothness of the model. As a punishment factor, it controls the degree of tolerance; subscript j Representing different nodes; The SVR constraint is: In the formula, These represent the true values of flow field velocity and turbulence intensity at different nodes; Here are the predicted values, where This is the transpose of the weight vector. It is a nonlinear function used to transform the original spatial samples Mapping to a higher-dimensional space; using the Lagrange method, it can be transformed into a dual problem for solution, ultimately yielding: In the formula, n is the total number of nodes. For the Lagrange multipliers corresponding to the upper boundary constraints; For the Lagrange multipliers corresponding to the lower boundary constraints; Let be the Gaussian kernel function, defined as the inner product of two high-dimensional mappings, i.e. .
6. The intelligent rapid prediction method for unsteady forces in waterjet propulsion according to claim 1, characterized in that, After training with SVR deep learning, a prediction model for stern turbulent wake parameters is constructed. Based on this prediction model for stern turbulence wake parameters, the flow field velocity and turbulence intensity at different nodes on the stern inlet section can be quickly obtained after giving the hull shape factor, hull speed, and the location information of different nodes on the stern inlet section.
7. The intelligent rapid prediction method for unsteady forces in waterjet propulsion according to claim 1, characterized in that, Since the data in the turbulent wake database at the stern position is used as the inlet boundary, the unsteady force numerical DDES simulation does not require wake calculation. Therefore, the stern + propeller model is used instead of the whole ship + propeller model. The stern + propeller model includes the stern section of the hull, rudder, duct, stator, and rotor; the stern lines are defined by a shape factor. α 3 and α 4. The tail rudder geometry is transformed into a cross-shaped tail rudder structure and an X-shaped tail rudder structure. The propeller arrangement is transformed by different numbers of stator blades, different numbers of rotor blades, and different stator-rotor spacing.
8. The intelligent rapid prediction method for unsteady forces in waterjet propulsion according to claim 1, characterized in that, Using the established stern + propeller model as the simulation object, batch DDES simulations were conducted under different speed conditions. An unsteady force database was established based on the simulation results. Each data entry in the unsteady force database contains: hull shape factor. α 3 and α 4; Ship speed V n Tail rudder geometry, cruciform or X-shaped tail rudder structure; propeller arrangement, number of stator blades. n rotor Different numbers of rotor blades n stator Different rotor-stator spacing D ; unsteady force F unsteady After the unsteady force database is built, deep learning training is carried out using MLP technology with hull shape factor, hull speed, rudder geometry, and propeller arrangement as input parameters and unsteady forces as output parameters. The mathematical model is as follows: In the formula, For the first l Net input to layer neurons; To connect the ( l -1) layer neurons to the first l The weights of layer neurons; For the ( l -1) Activation output of neurons in layer 1; For the first l Bias of layer neurons; It is an activation function, using the ReLU activation function, in the form of: ; The training process includes parameter initialization, forward propagation, loss calculation, backpropagation, parameter update, and regularization.
9. The intelligent rapid prediction method for unsteady forces in waterjet propulsion according to claim 1, characterized in that, After training with MLP deep learning, an intelligent rapid prediction model for unsteady forces is constructed, and a corresponding program is written. Based on this intelligent rapid prediction model and program for unsteady forces, the unsteady forces of waterjet propulsion can be quickly predicted and obtained given the hull shape factor, hull speed, rudder geometry, and propeller arrangement.
10. The intelligent rapid prediction method for unsteady forces in waterjet propulsion according to claim 1, characterized in that, A smart and rapid prediction method for unsteady forces in waterjet propulsion is proposed for underwater vehicles and underwater waterjet propulsion. When applied to other types of underwater and surface vessels and propulsion systems, it is only necessary to redefine the input and output parameters of the turbulent wake boundary condition prediction model and the unsteady force intelligent rapid prediction model.