Ship propeller-rudder system disturbing force modeling method based on data driving

Abstract

The invention discloses a ship propeller-rudder system disturbing force modeling method based on data driving, and relates to the technical field of ship maneuverability, and the method comprises the steps: building a function relation between propeller thrust and rudder normal force based on motion state parameters of a ship; the method comprises the following steps: classifying motion characteristics of a ship in a rotation process, dividing a sampling interval by using a curvature adaptive method, and determining a working condition matrix by using an orthogonal sampling method; propeller thrust and rudder normal force under different parameters are tested through a revolving arm simulation test, and a paddle-rudder system disturbing force database is established; neural network modeling and training of a database are carried out, specific parameters of a propeller thrust and rudder normal force function are generated, a propeller-rudder system disturbing force model is formed, ship propeller-rudder system disturbing force forecasting under data driving is achieved, the database and the neural network can be applied to characterize the non-linear and coupling characteristics of propeller-rudder disturbing force, and the disturbing force forecasting accuracy is improved. And the ship maneuverability motion forecasting precision is obviously improved.

Description

Technical Field

[0001] This application relates to the field of ship maneuvering motion, and in particular to a data-driven modeling method for the disturbance forces of a ship's propeller and rudder system. Background Technology

[0002] The dynamic disturbance characteristics of a ship's propeller and rudder system are a key element of its maneuverability. During navigation, the thrust generated by the rotating propeller provides forward propulsion, while the rudder, by adjusting the angle of attack, generates the normal force to create a steering control torque. By coordinating these two control variables—propeller speed and rudder deflection—the ship can achieve precise control of its course and reliable execution of maneuvering commands.

[0003] The unsteady wake field generated by a ship during actual navigation exhibits significant spatiotemporal variations. This complex flow environment leads to strong interference effects between the propeller and rudder, consequently affecting the thrust and steering torque provided by the propeller-rudder system. Traditional analytical methods often rely on simplified empirical formulas or ideal flow assumptions to establish mathematical models to characterize this complex coupled interference. However, when ships face severe sea states or emergency maneuvering conditions, traditional methods cannot accurately characterize the coupled interference effects under these circumstances, making it difficult to achieve optimal matching of the ship's propeller-rudder system. This, in turn, affects the ship's maneuvering response performance and may even jeopardize navigational safety.

[0004] Therefore, accurately characterizing the coupling interference between the propeller and the rudder under various operating conditions is an important issue that the industry urgently needs to address. Summary of the Invention

[0005] To address the aforementioned problems and technical requirements, the applicant proposes a data-driven modeling method for the disturbance forces of ship propeller and rudder systems. This method aims to solve the problem that existing technologies cannot accurately characterize the nonlinear and strong coupling effects between propellers and rudders in complex flow field environments, and to achieve rapid and accurate prediction of the disturbance forces of ship propeller and rudder systems.

[0006] This application provides a data-driven method for modeling the disturbance forces of a ship's propeller and rudder system, the method comprising:

[0007] Construct functional relationships between propeller thrust and rudder normal force and motion state parameters, where propeller thrust and rudder normal force constitute the ship's propeller-rudder system disturbance force;

[0008] Extract the characteristic curves of the ship motion data corresponding to the motion state parameters under typical turning maneuvering motions, and determine the sampling area based on the characteristic curves;

[0009] Data samples are obtained from the sampling area, and the working condition matrix of the slewing arm simulation test is designed using the data samples to obtain the test thrust and test normal force corresponding to the data samples.

[0010] The data samples, the test thrust, the test normal force, and the functional relationship are input into a preset neural network model for training to obtain a specific expression of the functional relationship, so as to predict the propeller thrust and rudder normal force through the trained neural network model.

[0011] According to the data-driven modeling method for ship propeller and rudder system disturbance forces provided in the embodiments of this application, characteristic curves of ship motion data corresponding to the motion state parameters under typical yaw maneuvering motions are extracted, including:

[0012] Obtain the distribution of three-dimensional scatter points of ship motion data under typical turning and maneuvering motions;

[0013] Based on the aforementioned distribution, the following feature extraction operation is performed on each three-dimensional scatter point:

[0014] A neighborhood with a preset radius is set at the three-dimensional scatter point; the centroid and covariance matrix of the neighborhood are calculated based on the data points in the neighborhood; the eigenvector of the neighborhood is calculated based on the covariance matrix; and two feature points corresponding to the three-dimensional scatter point are generated based on the centroid, the preset radius, and the eigenvector.

[0015] The feature curve is obtained by connecting the feature points of each three-dimensional scatter point using B-splines.

[0016] According to the data-driven modeling method for ship propeller and rudder system interference provided in this application, the feature curve is obtained by connecting the feature points of each three-dimensional scatter point using B-splines, including:

[0017] Based on the feature points of each three-dimensional scattered point connected by B-spline, a three-dimensional feature curve is obtained;

[0018] The three-dimensional feature curves are reduced in dimension to obtain a first two-dimensional feature curve showing the change of slewing speed and rudder angle with lateral speed, and a second two-dimensional feature curve showing the change of slewing speed and longitudinal speed with lateral speed.

[0019] According to the data-driven modeling method for ship propeller and rudder system interference forces provided in the embodiments of this application, the typical turning maneuver includes: a rudder turning phase and a turning phase;

[0020] The distribution of three-dimensional scatter points for ship motion data under typical turning maneuvers is obtained, including:

[0021] Obtain the first distribution of the ship's lateral velocity and turning speed as a function of the rudder angle during the turning phase, and the second distribution of the ship's lateral velocity and turning speed as a function of the longitudinal velocity during the turning phase.

[0022] According to the data-driven modeling method for ship propeller and rudder system disturbance forces provided in this application embodiment, data samples are obtained from the sampling region, including:

[0023] The sampling area is uniformly divided into multiple sub-regions, and the curvature of the feature curves in each sub-region changes sequentially.

[0024] If the curvature change is determined to be greater than a preset value, the sub-region is further divided to obtain multiple sub-intervals;

[0025] In the divided sub-regions and sub-intervals, the data samples are obtained by sampling using orthogonal methods and / or by randomly selecting points.

[0026] According to the data-driven modeling method for disturbance forces in a ship propeller and rudder system provided in this application, the working condition matrix for a slewing arm simulation test is designed using the data samples to obtain the test thrust and test normal force corresponding to the data samples, including:

[0027] Based on the aforementioned working condition matrix, a slewing arm simulation test was conducted to test the test thrust and test normal force corresponding to different parameters.

[0028] According to the data-driven modeling method for ship propeller and rudder system disturbance forces provided in the embodiments of this application, the functional relationship includes:

[0029] f T =T u u+T v v+T r r+T δ δ+T uv uv+T ur ur+T uδ uδ+T vr vr+T vδ vδ+T rδ rδ;

[0030] Among them, f T Let u represent the propeller thrust function, v represent the lateral velocity, r represent the angular velocity of the turn, δ represent the rudder angle, and T represent the propeller thrust function. u T represents the linear hydrodynamic derivative of propeller thrust with respect to longitudinal velocity. v T represents the linear hydrodynamic derivative of propeller thrust with respect to lateral velocity. r T represents the linear hydrodynamic derivative of propeller thrust with respect to the bow angular velocity. δT represents the linear hydrodynamic derivative of propeller thrust with respect to rudder angle. uv The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on the longitudinal and lateral velocities. ur The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on the longitudinal velocity and the bow angular velocity. uδ The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on longitudinal velocity and rudder angle. vr The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on the lateral velocity and the bow angular velocity. vδ The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on lateral velocity and rudder angle. rδ The linear hydrodynamic derivative represents the coupled effect between propeller thrust and rudder angular velocity and rudder angle.

[0031] f R =R u u+R v v+R r r+R δ δ+R uv uv+R ur ur+R uδ uδ+R vr vr+R vδ vδ+R rδ rδ;

[0032] Among them, f R R represents the rudder normal force function. u R represents the linear hydrodynamic derivative of the rudder normal force with respect to the longitudinal velocity. v R represents the linear hydrodynamic derivative of the rudder normal force with respect to the lateral velocity. r R represents the linear hydrodynamic derivative of the rudder normal force with respect to the bow angular velocity. δ R represents the linear hydrodynamic derivative of the rudder normal force with respect to the rudder angle. uv R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the longitudinal and lateral velocities. ur R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the longitudinal velocity and the bow angular velocity. uδ R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the longitudinal velocity and rudder angle. vr R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the lateral velocity and the bow angular velocity. vδ R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the lateral velocity and rudder angle. rδ The linear hydrodynamic derivative represents the coupling effect between the rudder normal force and the rudder angular velocity and rudder angle.

[0033] According to the data-driven modeling method for disturbance forces of a ship propeller and rudder system provided in this application embodiment, after inputting the data sample, the test thrust, the test normal force, and the functional relationship into a preset neural network model for training, and obtaining a specific expression of the functional relationship, the method further includes:

[0034] Physical experiments were conducted on some or all of the data samples to verify the physical thrust and physical normal force.

[0035] The neural network model is optimized based on the physical thrust and the physical normal force.

[0036] According to the data-driven modeling method for disturbance forces of a ship propeller and rudder system provided in the embodiments of this application, the physical test includes a slewing arm simulation test.

[0037] According to the data-driven modeling method for disturbance forces of a ship propeller and rudder system provided in the embodiments of this application, the rudder turning phase includes the process of the rudder angle changing uniformly from 0° to 35°, and the turning phase includes the process of the rudder angle being maintained at 35°.

[0038] The data-driven modeling method for ship propeller and rudder system disturbance forces provided in this application constructs functional relationships between propeller thrust and rudder normal force and motion state parameters, respectively. The propeller thrust and rudder normal force are obtained by decomposing the disturbance forces of the ship propeller and rudder system. This application pre-creates the correspondence between the ship's motion state parameters and disturbance forces during navigation; extracts characteristic curves of ship motion data corresponding to motion state parameters under typical slewing maneuvers, and determines sampling areas based on these characteristic curves; obtains data samples from the sampling areas, and uses these data samples to design the working condition matrix for the slewing arm simulation test, obtaining the test thrust and test normal force corresponding to the data samples. This application analyzes multiple ships... The characteristics of the motion data provide an effective data foundation for the sample data used in subsequent simulation experiments and model training, enabling the model to learn the details of the flow field generated by ship motion and more accurately obtain the interference between the propeller and the rudder. The data samples, test thrust, test normal force and functional relationship are input into the preset neural network model for training to obtain the specific expression of the functional relationship. The trained neural network model is then used to predict the propeller thrust and rudder normal force. The nonlinearity and coupling characteristics of the propeller and rudder interference force are analyzed using the sample data and the neural network model, thereby improving the prediction accuracy of ship maneuvering motion under various working conditions. Finally, the neural network model is used to quickly and accurately predict the interference force. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart illustrating the data-driven modeling method for disturbance forces in a ship propeller and rudder system provided in an embodiment of this application.

[0041] Figure 2 This is a schematic diagram of the motion state parameters provided in the embodiments of this application;

[0042] Figure 3 This is a schematic diagram of the motion changes corresponding to the steering phase provided in the embodiments of this application;

[0043] Figure 4 This is a schematic diagram illustrating the motion changes corresponding to the rotation stage provided in the embodiments of this application;

[0044] Figure 5 This is a schematic diagram of the data scatter distribution during the steering phase provided in an embodiment of this application;

[0045] Figure 6 This is a schematic diagram of the data scatter distribution during the rotation stage provided in an embodiment of this application;

[0046] Figure 7 This is a schematic diagram of the two-dimensional feature curves of the steering stage provided in the embodiments of this application;

[0047] Figure 8 This is a schematic diagram of the two-dimensional feature curve of the rotation stage provided in the embodiments of this application;

[0048] Figure 9 This is a schematic diagram of the spatial distribution of characteristic curve samples during the steering phase provided in an embodiment of this application;

[0049] Figure 10 This is a schematic diagram of the spatial distribution of characteristic curve samples in the rotation stage provided in the embodiments of this application;

[0050] Figure 11 This is a schematic diagram of a ship slewing arm test provided in an embodiment of this application;

[0051] Figure 12 This is a schematic diagram of the prediction results of the ship's motion trajectory under a 35° rudder angle provided in the embodiments of this application. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0053] This application provides a data-driven method for modeling the disturbance forces of a ship's propeller and rudder system. This method can be applied to smart terminals and servers. This application uses the application of this method in a server as an example for illustration, and some other descriptions in the embodiments are illustrative and not intended to limit the scope of protection of this application, and will not be described in detail thereafter. The specific implementation of the method is as follows... Figure 1 As shown:

[0054] Step 101: Construct the functional relationships between propeller thrust and rudder normal force and motion state parameters, respectively.

[0055] Among them, the propeller thrust and rudder normal force are obtained by decomposing the interference force of the ship's propeller and rudder system, that is, the ship's propeller and rudder system interference force is composed of the propeller thrust and rudder normal force.

[0056] The motion parameters include: the ship's lateral speed, longitudinal speed, turning angular velocity, and rudder angle.

[0057] Step 102: Extract the characteristic curves of the ship motion data corresponding to the motion state parameters under typical turning maneuvering motion, and determine the sampling area based on the characteristic curves.

[0058] Typical turning maneuvers include the steering phase and the turning phase.

[0059] Step 103: Obtain data samples from the sampling area and use the data samples to design the working condition matrix for the slewing arm simulation test, and obtain the test thrust and test normal force corresponding to the data samples.

[0060] Step 104: Input the data samples, test thrust, test normal force and functional relationship into the preset neural network model for training to obtain the specific expression of the functional relationship, so as to predict the propeller thrust and rudder normal force through the trained neural network model.

[0061] The data-driven modeling method for ship propeller and rudder system disturbance forces provided in this application constructs functional relationships between propeller thrust and rudder normal force and motion state parameters, respectively. The propeller thrust and rudder normal force are obtained by decomposing the disturbance forces of the ship propeller and rudder system. This application pre-creates the correspondence between the ship's motion state parameters and disturbance forces during navigation; extracts characteristic curves of ship motion data corresponding to motion state parameters under typical slewing maneuvers, and determines sampling areas based on these characteristic curves; obtains data samples from the sampling areas, and uses these data samples to design the working condition matrix for the slewing arm simulation test, obtaining the test thrust and test normal force corresponding to the data samples. This application analyzes multiple ships... The characteristics of the motion data provide an effective data foundation for the sample data used in subsequent simulation experiments and model training, enabling the model to learn the details of the flow field generated by ship motion and more accurately obtain the interference between the propeller and the rudder. The data samples, test thrust, test normal force and functional relationship are input into the preset neural network model for training to obtain the specific expression of the functional relationship. The trained neural network model is then used to predict the propeller thrust and rudder normal force. The nonlinearity and coupling characteristics of the propeller and rudder interference force are analyzed using the sample data and the neural network model, thereby improving the prediction accuracy of ship maneuvering motion under various working conditions. Finally, the neural network model is used to quickly and accurately predict the interference force.

[0062] In one specific embodiment, the functional relationship is shown in formulas (1) and (2):

[0063] f T =T u u+T v v+T r r+T δ δ+T uv uv+T ur ur+T uδ uδ+T vr vr+T vδ vδ+T rδ rδ……………………(1)

[0064] Among them, f T Let represent the propeller thrust function, u represent the longitudinal velocity (the longitudinal velocity of the ship along the bow and stern direction), v represent the lateral velocity (the drift velocity of the ship along the lateral direction), r represent the turning angular velocity (the turning velocity of the ship about its vertical axis at its center of gravity, also known as the swivel velocity), δ represent the rudder angle (the rudder angle of the ship, i.e., the deflection angle of the rudder relative to the mid-longitudinal section of the ship), and T represent the propeller thrust function. u T represents the linear hydrodynamic derivative of propeller thrust with respect to longitudinal velocity. v T represents the linear hydrodynamic derivative of propeller thrust with respect to lateral velocity. rT represents the linear hydrodynamic derivative of propeller thrust with respect to the bow angular velocity. δ T represents the linear hydrodynamic derivative of propeller thrust with respect to rudder angle. uv The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on the longitudinal and lateral velocities. ur The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on the longitudinal velocity and the bow angular velocity. uδ The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on longitudinal velocity and rudder angle. vr The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on the lateral velocity and the bow angular velocity. vδ The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on lateral velocity and rudder angle. rδ The linear hydrodynamic derivative represents the coupled effect of propeller thrust on rudder angular velocity and rudder angle.

[0065] f R =R u u+R v v+R r r+R δ δ+R uv uv+R ur ur+R uδ uδ+R vr vr+R vδ vδ+R rδ rδ…………………(2)

[0066] Among them, f R R represents the rudder normal force function. u R represents the linear hydrodynamic derivative of the rudder normal force with respect to the longitudinal velocity. v R represents the linear hydrodynamic derivative of the rudder normal force with respect to the lateral velocity. r R represents the linear hydrodynamic derivative of the rudder normal force with respect to the bow angular velocity. δ R represents the linear hydrodynamic derivative of the rudder normal force with respect to the rudder angle. uv R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the longitudinal and lateral velocities. ur R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the longitudinal velocity and the bow angular velocity. uδ R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the longitudinal velocity and rudder angle. vr R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the lateral velocity and the bow angular velocity. vδ R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the lateral velocity and rudder angle. rδ The linear hydrodynamic derivative represents the coupling effect between the rudder normal force and the rudder angular velocity and rudder angle.

[0067] In formulas (1) and (2), u is equivalent to v is equivalent to r is equivalent to δ is equivalent to Therefore, it is interpreted as consistent when explaining its physical meaning.

[0068] For a schematic diagram of the motion state parameters, please refer to [link / reference needed]. Figure 2 .

[0069] Specifically, the mutual interference of these four motion parameters significantly affects the hydrodynamic characteristics of the ship. The coupling of longitudinal velocity and bow speed determines the propeller inlet conditions, the coupling of lateral velocity and rudder angle changes the actual effective angle of attack of the rudder, and the bow speed affects the local flow field distribution in the propeller-rudder region. In addition, the interference effect between the propeller wake and the rudder body is also affected by the combined influence of these motion parameters, ultimately manifesting as dynamic changes in propeller thrust and rudder normal force.

[0070] Specifically, the disturbance force of the ship's propeller and rudder system is decomposed into propeller thrust and rudder normal force. Dimensionless functions of propeller thrust and rudder normal force are established using motion state parameters, as shown in formulas (3) and (4):

[0071]

[0072] Where U0 represents the ship's design speed, which is a constant; δ0 represents the ship's maximum rudder angle of 35°; and L represents the ship's characteristic length, which is a constant.

[0073] U0 and δ0 are used for rigid-free quantization of velocity and rudder angle.

[0074] In one specific embodiment, typical slewing maneuvers include a steering phase and a slewing phase.

[0075] The specific implementation of obtaining the distribution of three-dimensional scatter points of ship motion data under typical turning maneuvers includes:

[0076] Obtain the first distribution of the ship's lateral velocity and turning speed as a function of the rudder angle during the turning phase, and the second distribution of the ship's lateral velocity and turning speed as a function of the longitudinal velocity during the turning phase.

[0077] Specifically, ship operation data is acquired during the rudder turning and yaw phases. Yaw motion at a 35° rudder angle represents an extreme maneuvering condition, highlighting the coupled nonlinear characteristics of the propeller-rudder system.

[0078] The turning phase is the process of the rudder angle changing uniformly from 0° to 35°. During this phase, the longitudinal velocity of the hull remains basically stable at its initial value, and the changes in lateral velocity and bow turning speed are both less than 10%, exhibiting quasi-static motion characteristics. This phase can be analyzed in detail for its impact on the propeller-rudder interference force.

[0079] The turning phase is the process of maintaining the rudder angle at 35°, during which all ship motion parameters change significantly. This phase allows for a focused study of the quantitative relationship between ship motion parameters and disturbance forces under constant rudder angle conditions.

[0080] Among them, through Figure 3 The diagram illustrates the motion changes during the steering phase. Figure 4 The motion changes during the rotation phase are illustrated.

[0081] In one specific embodiment, the specific implementation of extracting the feature curves of ship motion data corresponding to motion state parameters under typical turning maneuvering motions includes:

[0082] The distribution of three-dimensional scatter points in ship motion data under typical turning and maneuvering motions is obtained. Based on the distribution, the following feature extraction operations are performed on each three-dimensional scatter point: a neighborhood with a preset radius is set at the scatter point; the centroid and covariance matrix of the neighborhood are calculated based on the data points in the neighborhood; the eigenvector of the neighborhood is calculated based on the covariance matrix; two feature points corresponding to the scatter point are generated based on the centroid, preset radius, and eigenvector. The feature points of each scatter point are connected using B-splines to obtain the feature curve.

[0083] Specifically, the characteristics of ship motion data are analyzed, and the changes in the lateral velocity and turning speed of multiple ships with the rudder angle during the turning phase and the changes in the lateral velocity and turning speed of multiple ships with the longitudinal direction during the turning phase are analyzed to obtain the distribution of motion data scatter points.

[0084] Among them, through Figure 5 The distribution of data points during the steering phase is illustrated. Figure 6 A schematic diagram of the scatter distribution of data points during the rotation phase is provided.

[0085] Specifically, based on three-dimensional scatter data, principal component analysis is applied to extract the characteristic curves of the three-dimensional scatter data.

[0086] For example, a preset radius of 0.02 is used as an illustration. This preset radius was obtained through data analysis and can cover 90% of the data points within this radius range.

[0087] The centroid is calculated using the first calculation formula, as shown in formula (5):

[0088]

[0089] Where, μ i N represents the center of mass. i Let x represent the neighborhood of the i-th 3D scatter point. j This represents the j-th data point in the neighborhood.

[0090]

[0091] Among them, C i Let represent the covariance matrix.

[0092] C i v i =λv i …………………………………(7)

[0093] Among them, v i Let λ represent the eigenvector and λ represent the eigenvalue.

[0094]

[0095] in, and These are two feature points generated along the eigenvector direction of the centroid.

[0096] In one specific embodiment, the specific implementation of obtaining the feature curve based on the feature points of each three-dimensional scatter point connected by B-splines includes:

[0097] Based on the feature points of each three-dimensional scattered point connected by B-splines, a three-dimensional feature curve is obtained; the three-dimensional feature curve is then subjected to dimensionality reduction processing to obtain the first two-dimensional feature curve of the rotation speed and rudder angle changing with the lateral velocity, and the second two-dimensional feature curve of the rotation speed and longitudinal velocity changing with the lateral velocity.

[0098] Specifically, for the rudder turning phase, a first two-dimensional characteristic curve is obtained showing the changes in slewing speed and rudder angle with lateral velocity; for the turning phase, a second two-dimensional characteristic curve is obtained showing the changes in slewing speed and longitudinal velocity with lateral velocity. Among these, through... Figure 7 The two-dimensional characteristic curves of the steering phase are illustrated. Figure 8 A schematic diagram of the two-dimensional characteristic curves during the rotation phase is provided.

[0099] In one specific embodiment, the specific implementation of obtaining data samples from the sampling region includes:

[0100] The sampling area is uniformly divided into multiple sub-regions, and the curvature change of the feature curve in each sub-region is measured sequentially. If the curvature change is greater than a preset value, the sub-region is divided again to obtain multiple sub-intervals. Data samples are obtained in the sub-regions and sub-intervals by orthogonal sampling and / or random point selection.

[0101] Specifically, the implementation of determining the sampling region within the neighborhood of the characteristic curve includes:

[0102] For the four characteristic curves, a strip sampling area is formed by extending 0.02 units on both sides of each characteristic curve as the center line, and all data samples are selected within this area.

[0103] Among them, through Figure 9 The spatial distribution of characteristic curve samples during the steering phase is illustrated. Figure 10 A schematic diagram of the spatial distribution of characteristic curve samples during the rotation phase is provided.

[0104] Among them, the four characteristic curves include: a two-dimensional characteristic curve showing the change of slewing speed with lateral speed, a two-dimensional characteristic curve showing the change of rudder angle with lateral speed, a two-dimensional characteristic curve showing the change of slewing speed with lateral speed, and a two-dimensional characteristic curve showing the change of longitudinal speed with lateral speed.

[0105] Specifically, the following is a detailed explanation of data sampling, using two-dimensional characteristic curves of slewing speed versus lateral velocity and rudder angle versus lateral velocity as examples:

[0106] For the two-dimensional characteristic curve of rotational speed as a function of lateral speed:

[0107] Divide the curve into 16 equal-width intervals (sub-regions) along the x-axis (v / U0 direction), and adaptively divide the rL / U0 curve along the y-axis, with the division range being the range between the minimum and maximum values ​​of rL / U0.

[0108] First, a basic division of 16 equal parts is performed, and the curvature change of the characteristic curve within each interval is calculated sequentially. When a curvature change exceeding 50% is detected within an interval, that interval is further subdivided into two equal-width sub-intervals. Then, orthogonal sampling is performed for each subdivided region, with the x-coordinate of the k-th sampling point being (v / U0). k The y-coordinate is (rL / U0). k This curvature-adaptive partitioning strategy ensures denser sampling points in high-curvature regions.

[0109] For the two-dimensional characteristic curve of the rudder angle as a function of lateral velocity:

[0110] The same method was applied to adaptively divide the curve along the y-axis, and random points were selected within each division. This resulted in a series of test condition combinations for the steering phase: a fixed longitudinal velocity of U0 and a lateral velocity of (v / U0). k The rotational speed is (rL / U0). k The rudder angle is (δ / δ0).k .

[0111] Specifically, similar procedures were used in advance to obtain the two-dimensional characteristic curves of rotational speed versus lateral speed and longitudinal speed versus lateral speed test samples.

[0112] First, the x-axis (v / U0 direction) is divided into 16 equal parts. Then, the u / U0 curve and rL / U0 curve are adaptively divided using curvature. Test samples are selected within each division interval using an orthogonal method. Finally, a series of test condition combinations for the steering phase are obtained: longitudinal velocity is (u / U0). k The lateral velocity is (v / U0). k The rotational speed is (rL / U0). k The rudder angle is fixed at δ0.

[0113] In one specific embodiment, the specific implementation of acquiring data samples from the sampling area and using the data samples for simulation experiments includes:

[0114] Data samples are obtained from the sampling area, and the working condition matrix for the slewing arm simulation test is designed using the data samples. Based on the working condition matrix, the slewing arm simulation test is carried out to test the test thrust and test normal force corresponding to different parameters.

[0115] Specifically, characteristic curves of ship motion data during typical slewing maneuvering motions are extracted, and the range of sample data is determined. This range of sample data is used to design the working condition matrix for the slewing arm simulation test.

[0116] Specifically, the working condition matrix constructed based on data samples takes into account the uniform coverage of the parameter space, and also captures the key change areas of the feature curves through curvature adaptive partitioning, providing an effective data foundation for the training of neural network models.

[0117] Specifically, a slewing arm simulation test was conducted for the operating condition matrix to measure the propeller thrust and rudder normal force under different parameters, and a database of disturbance forces of the ship's propeller-rudder system was established.

[0118] Select the hull type and propeller / rudder model for the test, and determine the basic parameters such as the characteristic length L of the hull and the design speed U0, as well as the diameter D of the propeller and the profile area A of the rudder. R Isoplasmic type parameters.

[0119] First, a high-speed self-propulsion simulation test of the ship was conducted. During the simulation, the ship started from a stationary state, and the propeller speed was dynamically adjusted through a PID control algorithm to drive the ship to reach the target speed, thereby obtaining the stable propeller speed at the target speed. The PID control algorithm is shown in formulas (10), (11), and (12):

[0120] e=uU……………………(10)

[0121]

[0122] n new =n old +Δn…………………(12)

[0123] Where e represents the speed deviation, u represents the actual speed of the ship, U represents the target speed of the ship, Δn represents the adjustment amount of the propeller speed, P and I represent the PID control parameters, t represents time, and n old This represents the propeller speed before adjustment, n. new This indicates the adjusted propeller speed.

[0124] For example, the proportional system control parameter P is set to 0.2, and the integral coefficient I is set to 2.

[0125] Then, simulation calculations were carried out for the ship's slewing arm test (slewing arm simulation test). The test principle is as follows: Figure 11 As shown, with the ship's center of gravity as the towing point, the towing ship rotates at a constant speed (slewing speed), while maintaining the towing ship's linear velocity (towing speed) at a constant value. The constant radius R of rotational motion is calculated from the rotational velocity and the linear velocity, i.e.

[0126] Based on the defined working condition matrix, the four motion parameters for each slewing arm simulation test are controlled, namely longitudinal speed... Lateral velocity Rotation speed Rudder angle k represents the k-th sample condition in the test matrix, and the turning radius value in each test is adjusted by these four motion parameters. Furthermore, the propeller speed remains constant during the simulation, and its value is determined by the corresponding speed. The results of the self-propelled simulation test will determine the outcome.

[0127] Finally, the propeller thrust and rudder normal force measured in each slewing arm simulation test are summarized, and dimensionless processing is performed using formulas (13) and (14) to obtain a database of disturbance forces in the ship's propeller-rudder system. This database consists of 6 parameters: T p and F n .

[0128]

[0129] Where ρ represents the density of water, n represents the propeller speed, D represents the propeller diameter, and T... pF represents propeller thrust (test thrust). N Indicates the rudder normal force (test normal force), A R The rudder's outline area is represented by δ, U represents the target speed, and δ represents the rudder angle.

[0130] Specifically, a multi-level perceptual neural network model is built and trained based on a database to achieve rapid prediction of propeller thrust and rudder normal force.

[0131] First, the established polynomial functions of propeller thrust and rudder normal force are transformed into a weighted sum form, as shown in formula (15):

[0132] y(x)=c0+c1x1+c2x2+c3x3+c4x4+c5x1x2+c6x1x3+c7x1x4+

[0133] c8x2x3+c9x2x4+c 10 x3x4……………(15)

[0134] Where y represents the function output of propeller thrust or rudder normal force, and x1, x2, x3, and x4 correspond to dimensionless ship motion parameters, respectively. and c0 to c 10 These are coefficients.

[0135] Design a deep neural network model structure with three hidden layers. The network input layer receives four ship motion parameters, and the output layer simultaneously predicts two target variables: propeller thrust and rudder normal force.

[0136] The Sigmoid activation function is used in the network, and a batch normalization layer is introduced to improve training stability. The mathematical expression of the Sigmoid activation function is given by formula (16):

[0137]

[0138] Then, the complete dataset of disturbance forces (including propeller thrust and rudder normal force) was divided into a training set (80%) and a validation set (20%) in an 8:2 ratio, and stratified sampling was used to ensure that the two sets of data have similar distribution characteristics in the parameter space.

[0139] An iterative strategy with a fixed 500 training epochs is adopted, with each epoch fully traversing the training dataset. The optimizer uses the Adam algorithm with adaptive learning rate characteristics, initially set to 0.01, and implements a dynamic learning rate decay strategy: after every 50 training epochs, the learning rate decays exponentially at a rate of 0.9. Simultaneously, the first-order moment estimation coefficients of the Adam algorithm are set to 0.9, the second-order moment estimation coefficients to 0.999, and le...-7 Numerical stability terms are used to ensure the stability of the calculation process.

[0140] Finally, through systematic training, the parameters c0 to c in the form of weighted sums in each network layer are obtained. 10 By combining the Sigmoid activation function, a rapid prediction model for the disturbance forces of the ship's propeller-rudder system was finally formed. This model can quickly and accurately predict propeller thrust and rudder normal force based on the input ship motion parameters, providing an important computational tool for ship maneuvering performance analysis and control system design.

[0141] In one specific embodiment, data samples, test thrust, test normal force, and functional relationship are input into a preset neural network model for training. After obtaining a specific expression of the functional relationship, physical experiments are conducted on some or all of the data samples to verify the physical thrust and physical normal force. The neural network model is then optimized based on the physical thrust and physical normal force.

[0142] Specifically, physical tests can be conducted under typical operating conditions to verify the simulation model, and the results of the tests on the ship's hydrodynamics and propeller interference forces can be compared and compared to adjust and optimize the convergence and accuracy of the simulation model.

[0143] This application establishes a direct functional relationship between propeller thrust and rudder normal force based on four motion state parameters of the ship: longitudinal speed, lateral speed, bow turning angular velocity, and rudder angle. It classifies the motion characteristics of the ship during turning, uses a curvature adaptive method to divide the sampling interval, and applies an orthogonal sampling method to determine the working condition matrix. Through slewing arm simulation experiments, it tests the propeller thrust and rudder normal force under different parameters, establishing a propeller-rudder system disturbance force database. Finally, it performs neural network modeling and training on the database to generate specific parameters for the propeller thrust and rudder normal force functions, forming a propeller-rudder system disturbance force model. By analyzing the nonlinear and coupling characteristics of the propeller-rudder disturbance force (propeller thrust and rudder normal force) through the database and neural network model, it constructs a propeller-rudder disturbance force model (neural network model) with higher accuracy and better generalization ability, thereby significantly improving the accuracy of ship maneuverability prediction. Specific effects include:

[0144] 1. A direct functional relationship between propeller-rudder disturbance force and ship motion state parameters (longitudinal velocity, lateral velocity, bow turning angular velocity, rudder angle) was established, breaking through the limitation of traditional modeling that requires the introduction of multiple intermediate parameters (such as advance coefficient, rudder effectiveness coefficient, etc.). By enriching the database and using artificial neural networks, end-to-end modeling from the original motion parameters to the disturbance force was realized, simplifying the model structure and improving computational efficiency.

[0145] 2. Based on the motion characteristics of the ship during the turning process, the motion process is divided into two types: the steering phase and the turning phase. Test conditions with engineering significance are selected for each phase to avoid the non-physical and invalid nature of the sample conditions. Furthermore, an orthogonal sampling method with curvature adaptation is applied to select test conditions, effectively overcoming the subjectivity and randomness caused by manual or random partitioning, especially in the region of large curvature change of the characteristic curve, ensuring the spatial coverage of the samples and improving the generalization ability of the model.

[0146] 3. By applying simulation methods to establish a propeller-rudder interference force database, the testing time and cost can be reduced compared with traditional physical experiments, achieving full coverage of test conditions, and providing details of the flow field generated by ship motion, which can reveal the interference mechanism between the propeller and the rudder.

[0147] 4. By leveraging a rich database and multi-layered perceptual neural networks, it is possible to deeply capture the nonlinear and strongly coupled characteristics of the disturbance forces in the propeller-rudder system, breaking through the theoretical bottlenecks of traditional models, such as... Figure 12 As shown, taking large rudder angle turning maneuvering motion as an example, the application of the novel propeller-rudder coupling model (neural network model, propeller-rudder interference force model) in this application can improve the prediction accuracy of ship maneuvering performance under complex working conditions.

[0148] Finally, it should be noted that the above descriptions are merely preferred embodiments of this application, and this application is not limited to the above embodiments. It is understood that other improvements and variations directly derived or conceived by those skilled in the art without departing from the spirit and concept of this application should be considered to be included within the protection scope of this application.

Claims

1. A data-driven modeling method for disturbance forces in a ship's propeller and rudder system, characterized in that, The method includes: Construct functional relationships between propeller thrust and rudder normal force and motion state parameters, where propeller thrust and rudder normal force constitute the ship's propeller-rudder system disturbance force; Extract the characteristic curves of the ship motion data corresponding to the motion state parameters under typical turning maneuvering motions, and determine the sampling area based on the characteristic curves; Data samples are obtained from the sampling area, and the working condition matrix of the slewing arm simulation test is designed using the data samples to obtain the test thrust and test normal force corresponding to the data samples. The data samples, the test thrust, the test normal force, and the functional relationship are input into a preset neural network model for training to obtain a specific expression of the functional relationship, so as to predict the propeller thrust and rudder normal force through the trained neural network model.

2. The data-driven modeling method for disturbance forces in a ship propeller and rudder system according to claim 1, characterized in that, The characteristic curves of the ship motion data corresponding to the motion state parameters under typical turning maneuvering motions are extracted, including: Obtain the distribution of three-dimensional scatter points of ship motion data under typical turning and maneuvering motions; Based on the aforementioned distribution, the following feature extraction operation is performed on each three-dimensional scatter point: A neighborhood with a preset radius is set at the three-dimensional scatter point; the centroid and covariance matrix of the neighborhood are calculated based on the data points in the neighborhood; the eigenvector of the neighborhood is calculated based on the covariance matrix; and two feature points corresponding to the three-dimensional scatter point are generated based on the centroid, the preset radius, and the eigenvector. The feature curve is obtained by connecting the feature points of each three-dimensional scatter point using B-splines.

3. The data-driven modeling method for disturbance forces in a ship propeller and rudder system according to claim 2, characterized in that, The feature curve is obtained by connecting the feature points of each three-dimensional scatter point using B-splines, including: Based on the feature points of each three-dimensional scattered point connected by B-spline, a three-dimensional feature curve is obtained; The three-dimensional feature curves are reduced in dimension to obtain a first two-dimensional feature curve showing the change of slewing speed and rudder angle with lateral speed, and a second two-dimensional feature curve showing the change of slewing speed and longitudinal speed with lateral speed.

4. The data-driven modeling method for disturbance forces in a ship propeller and rudder system according to claim 2, characterized in that, The typical gyratory maneuvers include: a steering phase and a gyratory phase; The distribution of three-dimensional scatter points for ship motion data under typical turning maneuvers is obtained, including: Obtain the first distribution of the ship's lateral velocity and turning speed as a function of the rudder angle during the turning phase, and the second distribution of the ship's lateral velocity and turning speed as a function of the longitudinal velocity during the turning phase.

5. The data-driven modeling method for disturbance forces in a ship propeller and rudder system according to any one of claims 1-4, characterized in that, Obtaining data samples from the sampling area includes: The sampling area is uniformly divided into multiple sub-regions, and the curvature of the feature curves in each sub-region changes sequentially. If the curvature change is determined to be greater than a preset value, the sub-region is further divided to obtain multiple sub-intervals; In the divided sub-regions and sub-intervals, the data samples are obtained by sampling using orthogonal methods and / or by randomly selecting points.

6. The data-driven modeling method for disturbance forces of a ship propeller and rudder system according to any one of claims 1-4, characterized in that, Using the data samples, a working condition matrix for the slewing boom simulation test is designed to obtain the test thrust and test normal force corresponding to the data samples, including: Based on the aforementioned working condition matrix, a slewing arm simulation test was conducted to test the test thrust and test normal force corresponding to different parameters.

7. The data-driven modeling method for disturbance forces in a ship propeller and rudder system according to any one of claims 1-4, characterized in that, The functional relationship includes: f T =T u u+T v v+T r r+T δ δ+T uv uv+T ur ur+T uδ uδ+T vr vr+T vδ vδ+T rδ rδ; Among them, f T Let u represent the propeller thrust function, v represent the lateral velocity, r represent the angular velocity of the turn, δ represent the rudder angle, and T represent the propeller thrust function. u T represents the linear hydrodynamic derivative of propeller thrust with respect to longitudinal velocity. v T represents the linear hydrodynamic derivative of propeller thrust with respect to lateral velocity. r T represents the linear hydrodynamic derivative of propeller thrust with respect to the bow angular velocity. δ T represents the linear hydrodynamic derivative of propeller thrust with respect to rudder angle. uv The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on the longitudinal and lateral velocities. ur The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on the longitudinal velocity and the bow angular velocity. uδ The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on longitudinal velocity and rudder angle. vr The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on the lateral velocity and the bow angular velocity. vδ The linear hydrodynamic derivative, T, represents the coupled effect of propeller thrust on lateral velocity and rudder angle. rδ The linear hydrodynamic derivative represents the coupled effect between propeller thrust and rudder angular velocity and rudder angle. f R =R u u+R v v+R r r+R δ δ+R uv uv+R ur ur+R uδ uδ+R vr vr+R vδ vδ+R rδ rδ; Among them, f R R represents the rudder normal force function. u R represents the linear hydrodynamic derivative of the rudder normal force with respect to the longitudinal velocity. v R represents the linear hydrodynamic derivative of the rudder normal force with respect to the lateral velocity. r R represents the linear hydrodynamic derivative of the rudder normal force with respect to the bow angular velocity. δ R represents the linear hydrodynamic derivative of the rudder normal force with respect to the rudder angle. uv R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the longitudinal and lateral velocities. ur R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the longitudinal velocity and the bow angular velocity. uδ R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the longitudinal velocity and rudder angle. vr R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the lateral velocity and the bow angular velocity. vδ R represents the linear hydrodynamic derivative of the coupling effect between the rudder normal force and the lateral velocity and rudder angle. rδ The linear hydrodynamic derivative represents the coupling effect between the rudder normal force and the rudder angular velocity and rudder angle.

8. The data-driven modeling method for disturbance forces of a ship propeller and rudder system according to any one of claims 1-4, characterized in that, After inputting the data sample, the test thrust, the test normal force, and the functional relationship into a preset neural network model for training to obtain a specific expression of the functional relationship, the process further includes: Physical experiments were conducted on some or all of the data samples to verify the physical thrust and physical normal force. The neural network model is optimized based on the physical thrust and the physical normal force.

9. The data-driven modeling method for disturbance forces in a ship propeller and rudder system according to claim 8, characterized in that, The physical tests include a rotary arm simulation test.

10. The data-driven modeling method for disturbance forces in a ship propeller and rudder system according to claim 4, characterized in that, The rudder turning phase includes the process of the rudder angle changing uniformly from 0° to 35°, and the turning phase includes the process of maintaining the rudder angle at 35°.

Images