Ship resistance prediction method based on point cloud deep learning

Abstract

The invention discloses a ship resistance prediction method based on point cloud deep learning, and belongs to the technical field of ship resistance prediction, and the method comprises the steps: obtaining ship type value point coordinates and corresponding resistance values in a ship variable geometrical shape; acquiring coordinate values of the surface of the ship on the grid points; building a neural network model of ship point cloud and ship resistance, and training the neural network model by using coordinate values on grid points as input and corresponding ship resistance as output; and testing the prediction precision of the neural network model. According to the method, the ship performance can be rapidly evaluated by constructing the deep neural network model of the ship type point cloud and the ship resistance, the geometric value points of the ship are used as the input of the neural network model, the corresponding ship resistance is used as the output, direct mapping from geometric data to the resistance value is realized, and the prediction precision and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of ship resistance prediction, and in particular to a ship resistance prediction method based on point cloud deep learning. Background Art

[0002] Ship resistance prediction technology is primarily used in ship design and optimization, particularly in analyzing ship hydrodynamic performance. This technology aims to study the impact of changes in hull shape (i.e., ship form) on ship resistance. Ship resistance directly affects a ship's propulsion power and fuel consumption. Accurately predicting resistance can optimize hull line design, reducing resistance, thereby reducing energy consumption and improving economic efficiency, thus possessing significant engineering application value.

[0003] Traditional ship resistance prediction technologies are mainly divided into three categories: empirical formula method, model test method and numerical simulation method. The empirical formula method is based on historical test data and statistical analysis. It quickly estimates the ship resistance through simplified mathematical formulas. It is suitable for the preliminary design stage and has the advantage of high computational efficiency, but low accuracy. The model test method measures the ship resistance by testing a scaled model of the ship in a water tank, but this method is costly and time-consuming. The numerical simulation method is based on computational fluid dynamics (CFD) technology. It simulates the flow field and resistance around the ship through numerical solution. It has the advantages of high accuracy and detailed analysis of flow field characteristics, but it requires high computing resources and is time-consuming to calculate.

[0004] In recent years, with the rapid development of computer technology and the significant improvement of computing power, machine learning has become a hot field in scientific research and technological application. Machine learning methods can efficiently handle complex nonlinear problems in a data-driven manner, providing a new solution for ship resistance prediction. Neural networks can capture complex nonlinear relationships in data and have been applied to ship resistance prediction. The input of the neural network is the main parameters of the ship, and the output is the ship resistance. However, using the main parameters of the ship (such as length, width, draft, etc.) as the input of the resistance prediction model has certain defects. When the ship type value point is changed, additional steps are required to calculate the main parameters from the ship type point data. This process increases the computational cost of data processing. In addition, the main parameters of the ship are a simplified description of the hull geometry and cannot fully reflect the complex and local detailed geometric characteristics of the hull, thereby limiting the prediction accuracy. Summary of the Invention

[0005] The purpose of the present invention is to provide a ship resistance prediction method based on point cloud deep learning, which uses ship geometry value points as the input of the neural network model and the corresponding ship resistance as the output, realizing a direct mapping from geometric data to resistance value, and can efficiently and accurately predict ship resistance.

[0006] To achieve the above objectives, the present invention provides a method for predicting ship resistance based on point cloud deep learning, comprising the following steps:

[0007] S1. Obtain the coordinates of the ship type value points and the corresponding resistance values ​​under the ship's changing geometric shape, and construct a ship geometry data set;

[0008] S2. Obtain the coordinate values ​​of the ship surface at the grid points;

[0009] S3. Build a neural network model of the ship point cloud and ship resistance, and train the neural network model, using the coordinate values ​​on the grid points as input and the corresponding ship resistance as output;

[0010] S4. Test the prediction accuracy of the neural network model.

[0011] Preferably, the geometric shape change of the ship in step S1 includes deformation of the ship width, deformation of the bow and deformation of the stern.

[0012] Preferably, step S2 specifically includes: setting the three-dimensional coordinate origin (0,0,0) at the intersection of the ship baseline and the stern perpendicular, obtaining the coordinate value of the ship grid point through the known ship point cloud coordinates, that is, obtaining the coordinate value of the ship grid point when x i =X m *i / N x , i=1,2,…,N x , and z i =w*j / N z ,j=1,2,…N z When the coordinate y corresponding to the ship surface i Value, where y i The value is obtained by interpolating the nearest coordinates of the known ship type points, y i >0, w represents the draft value, X m Indicates the maximum coordinate of the ship's length, x i 、y i and z i Indicates the coordinate value of the ship grid point, 1 / N x Indicates the spacing value of the grid points along the x-axis, 1 / N z Indicates the spacing value of the grid points along the z-axis.

[0013] Preferably, in step S3, the coordinate values ​​of the grid points are used as input, and the corresponding ship resistance F is used as output. The input vector is represented by y in =(y1,y2,…,y N ), N=N x *N z .

[0014] Preferably, in step S3, the neural network model uses the mean square error of the target resistance value and the predicted resistance value as a loss function, and uses an ADAM (Adaptive Moment Estimation) optimizer to optimize the weight parameters of the neural network model.

[0015] Therefore, the present invention adopts the above-mentioned ship resistance prediction method based on point cloud deep learning. By constructing a neural network model of ship point cloud and ship resistance, it can quickly evaluate ship performance, avoid time-consuming problems of methods such as experiments and numerical simulations, and make full use of the ability of deep learning to capture complex nonlinear mapping relationships, thereby improving the accuracy and efficiency of resistance prediction.

[0016] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0018] Figure 2 This is a schematic diagram of a ship grid point according to an embodiment of the present invention;

[0019] Figure 3 A schematic structural diagram of a KCS ship model according to an embodiment of the present invention;

[0020] Figure 4 A schematic diagram of a deep neural network according to an embodiment of the present invention;

[0021] Figure 5 This is a comparison chart of the predicted and actual values ​​of ship resistance values ​​according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0023] Example

[0024] like Figure 1 As shown, the present invention provides a ship resistance prediction method based on point cloud deep learning, the steps comprising:

[0025] S1. Based on the ship point cloud, the coordinates of the ship's profile points and the corresponding resistance values ​​under the ship's varying geometric shapes (breadth deformation, bow deformation, and stern deformation) are obtained to construct a ship geometry dataset. A ship point cloud is a collection of discrete spatial points on the ship's surface, recording their geometric position information.

[0026] S2. Get the coordinate values ​​of the ship surface at the grid points. Assume that the origin of the three-dimensional coordinates (0,0,0) is located at the intersection of the ship baseline and the stern perpendicular. The coordinate values ​​of the ship grid points are obtained by using the known ship point cloud coordinates, that is, when x i =X m *i / N x , i=1,2,…,N x , and z i =w*j / N z ,j=1,2,…N z When the coordinate y corresponding to the ship surface i Value, where y i The value is obtained by interpolating the nearest coordinates of the known ship type points, y i >0, w represents the draft value, X m Indicates the maximum coordinate of the ship's length, x i 、y i and z i Indicates the coordinate value of the ship grid point, 1 / N x Indicates the spacing value of the grid points along the x-axis, 1 / N z Indicates the spacing value of the grid points along the z-axis. The schematic diagram of the ship grid points is as follows Figure 2 shown.

[0027] S3. Build a back propagation (BP) artificial neural network model of ship point cloud and ship resistance, and train the neural network model. The schematic diagram of the deep neural network is as follows: Figure 3 As shown in Figure 2, a deep neural network is an artificial neural network with multiple hidden layers. It consists of multiple layers of neurons (nodes), including input layers, hidden layers, and output layers. Each neuron receives input signals through weight parameters and generates outputs through activation functions.

[0028] The dataset is divided into a training set and a test set. The training set is used to build a neural network model, using the coordinate values ​​on the grid points as input and the corresponding ship resistance as output. Specifically, assuming that the ship's draft is w, the three-dimensional coordinate origin (0,0,0) is located at the intersection of the ship's baseline and the stern perpendicular, and the maximum coordinate of the ship's length is X m Then, the input features of the neural network are: x i =X m *i / N x , i=1,2,…,Nx , and z i =w*j / N z ,j=1,2,…N z When the coordinate y corresponding to the ship surface i Value(y i >0), that is, the input vector is y in =(y1,y2,…,y N ),N=N x *N z The output dimension of the neural network is 1, which is the predicted value F of the ship's resistance.

[0029] The loss function of the neural network model is the mean square error between the target resistance value and the predicted resistance value, and the ADAM (Adaptive Moment Estimation) optimizer is used to minimize the loss function to optimize the weight parameters of the neural network.

[0030] S4. Test the prediction accuracy of the neural network model. Use the test set to predict the resistance values ​​of different ship types using the trained neural network model to evaluate its prediction accuracy.

[0031] The KCS ship model is taken as an example to verify the method of the present invention.

[0032] The model of the KCS ship is shown in the figure below. Figure 4 Its main parameters are shown in Table 1.

[0033] Table 1 Main parameters of KCS ship

[0034] mother ship KRISO Container Ship(KCS) Scale ratio 1:31.599 Length between perpendicular lines Lpp(m) 7.2786 Waterline length Lwl(m) 7.3570 Waterline width Bwl(m) 1.0190 Moulded depth D(m) 0.6013 Design draft T(m) 0.3418 <![CDATA[Displacement volume Displayment (m 3 )]]> 1.6490 <![CDATA[Wetted surface area of rudder Sw / orudder (m 2 )]]> 9.5441 Block coefficient CB 0.651 Medium cross-section coefficient CM 0.985 Center of buoyancy position LCB (%), midship forward fwd+ -1.48

[0035] The calculation software is used to obtain the coordinates of the ship's type value points and the corresponding resistance when the KCS ship changes its geometric shape (ship width deformation, bow deformation, stern deformation). A total of 631 sets of data are obtained, 90% of the data are used as training sets, and the remaining 10% of the data are used as test sets, and N is set. x =60, N z = 10. The results of the trained neural network predicting resistance on the test set are as follows Figure 5 As shown in the figure, the average relative error of the predicted resistance is 0.171%. The results show that the neural network model has a very high accuracy in predicting ship resistance.

[0036] Therefore, the present invention adopts the above-mentioned ship resistance prediction method based on point cloud deep learning. By constructing a deep neural network model of ship point cloud and ship resistance, the ship performance can be quickly evaluated. The ship geometry value points are used as the input of the neural network model, and the corresponding ship resistance is used as the output, realizing a direct mapping from geometric data to resistance value, thereby improving the prediction accuracy and efficiency.

[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for predicting ship resistance based on point cloud deep learning, characterized in that the steps include: S1. Obtain the coordinates of the ship type value points and the corresponding resistance values ​​under the ship's changing geometric shape, and construct a ship geometry data set; S2. Obtain the coordinate values ​​of the ship surface at the grid points; S3. Build a neural network model of the ship point cloud and ship resistance, and train the neural network model, using the coordinate values ​​on the grid points as input and the corresponding ship resistance as output; S4. Test the prediction accuracy of the neural network model.

2. The ship resistance prediction method based on point cloud deep learning according to claim 1 is characterized in that: The geometric shape change of the ship in step S1 includes deformation of the ship width, deformation of the bow and deformation of the stern.

3. The ship resistance prediction method based on point cloud deep learning according to claim 1 is characterized in that: Step S2 specifically includes: assuming that the origin of the three-dimensional coordinates (0,0,0) is located at the intersection of the ship baseline and the stern perpendicular, obtaining the coordinate values ​​of the ship grid points through the known ship point cloud coordinates, when x i =X m *i / N x , i=1,2,…,N x , and z i =w*j / N z ,j=1,2,…N z When , get the coordinate y corresponding to the ship surface i Value, where y i The value is obtained by interpolating the nearest coordinates of the known ship type points, y i >0, w represents the draft value, X m Indicates the maximum coordinate of the ship's length, x i 、y i and z i Indicates the coordinate value of the ship grid point, 1 / N x Indicates the spacing value of the grid points along the x-axis, 1 / N z Indicates the spacing value of the grid points along the z-axis.

4. The ship resistance prediction method based on point cloud deep learning according to claim 1, characterized in that: In step S3, the coordinate value of the grid point is used as input and the corresponding ship resistance F is used as output. The input vector is represented by y in =(y1,y2,…,y N ), N=N x *N z .

5. The ship resistance prediction method based on point cloud deep learning according to claim 1, characterized in that: In step S3, the neural network model uses the mean square error of the target resistance value and the predicted resistance value as the loss function, and uses the ADAM optimizer to optimize the weight parameters of the neural network model.

Images