A ship type design method and system

Abstract

The application belongs to the field of ship and ocean engineering, and particularly discloses a ship type design method and system, which comprises the following steps: determining key design parameters based on the shape and performance requirements of a design ship; inputting the key design parameters into a trained deep learning model to obtain the transverse section area and the length between vertical lines at each station of the design ship, so as to form a transverse section area curve; performing Bezier curve fitting on the transverse section area curves of the design ship and each parent ship respectively, and reducing the control points on the transverse section area curve; for each control point, constructing a homogeneous linear equation based on the corresponding transverse section area on the transverse section area curve of the design ship and each parent ship, and solving to obtain the weight of each parent ship; and fusing the ship types of each parent ship according to the weight of each parent ship to obtain the ship type of the design ship. The application can ensure that the design ship is close to the requirements in performance, and greatly improves the ship type design efficiency.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of ship and ocean engineering, and more particularly, relates to a ship type design method and system. BACKGROUND

[0002] Hull line design is the core content of ship design, and its design quality directly affects the performance, economy and safety of the ship. With the rapid development of global marine economy, the demand for ship design is increasing, especially in the field of high-performance ships such as planing boats, yachts, high-speed passenger ships, etc.

[0003] Traditional ship type design methods mainly include experience design method, parent ship modification method, etc. The experience design method is the earliest and most basic method in ship type design. According to the specific requirements of the new ship, the designer analyzes and thinks, and has a certain understanding of the new ship line characteristics, and then draws the line drawing according to the basic principles and rules of line design and referring to the excellent line data of similar ship types. Although this method is simple and easy to implement, it is highly subjective and difficult to meet complex design requirements. The parent ship modification method inherits the excellent lines of the parent ship or series ship type. In the design process, the modification of the parent type is first based on the existing ship type data to determine the initial ship type, and then the initial ship type is repeatedly adjusted and calculated to make the ship performance meet the design requirements.

[0004] With the development of computer technology, numerical simulation methods based on CFD have been gradually introduced into the field of ship type design. By establishing a three-dimensional calculation model of the ship body, the fluid characteristics of the ship body are solved by using fluid mechanics equations, so as to optimize the ship body and perform ship type transformation to reduce resistance and improve efficiency. The advantage of this method is its accuracy and scientificity, but it has large amount of calculation, long time consumption, and high requirement for computing resources, which is difficult to be applied in large scale in actual design.

[0005] In recent years, optimization algorithms have been widely used in the field of ship type design. By parameterizing the shape of the ship body and combining with CFD simulation, the resistance or efficiency is established as the objective function, and the optimization algorithm is used to search for the optimal solution. This method can improve the design efficiency to a certain extent, but it still faces challenges in dealing with multi-objective optimization problems, and the convergence speed of the optimization process is slow. SUMMARY

[0006] In view of the above defects or improvement needs of the prior art, the present application provides a ship type design method and system, which aims to improve the efficiency of ship type design and better meet the performance requirements.

[0007] To achieve the above-mentioned purpose, according to one aspect of the present application, a ship type design method is provided, comprising the following steps:

[0008] determining key design parameters based on the shape and performance requirements of the designed ship;

[0009] By inputting key design parameters into a trained deep learning model, the cross-sectional area and length between verticals at each station of the design vessel are obtained, thus forming a cross-sectional area curve.

[0010] Bezier curve fitting is performed on the cross section area curves of the design ship and each parent ship to reduce the number of control points on the cross section area curves. For each control point, a homogeneous linear equation is constructed based on the corresponding cross section area on the cross section area curves of the design ship and each parent ship, and the weights of each parent ship are obtained by solving the equation.

[0011] Based on the weights of each parent ship type, the ship types of each parent ship type are merged to obtain the design ship type.

[0012] As a further preferred embodiment, the homogeneous linear equation is expressed as:

[0013]

[0014] in, n It refers to the number of mother ships; y , y i They are the design ship, the first i The ordinate of the control point on the cross-sectional area curve of the mother ship is the corresponding cross-sectional area. w i It is the first i The weight of a mother ship.

[0015] As a further optimization, before constructing the homogeneous linear equation, the lengths of all parent ships are adjusted to the same value through a uniform scaling process.

[0016] As a further optimization, the performance requirements of the designed ship are input into a large language model to obtain key design parameters.

[0017] As a further preferred option, the key design parameters include displacement, longitudinal position of the buoyancy center, rhombus coefficient, main engine power, and design speed.

[0018] As a further preferred embodiment, the training method for the deep learning model is as follows:

[0019] Obtain key design parameters and corresponding cross-sectional area curves for multiple parent ships; adjust the cross-sectional area at each station in the cross-sectional area curve within ±30% and match it with the key design parameters to expand the data volume and construct a training set; train the deep learning model using this training set to obtain a trained deep learning model.

[0020] As a further preferred embodiment, the deep learning model employs a multilayer perceptron (MLP) model.

[0021] As a further preferred, according to the weight of each mother ship, the ship type of each mother ship is fused, including:

[0022] According to the weight of each mother ship, the transverse section line, the type width, the type depth and the draft of each mother ship are fused, and the transverse section line, the type width, the type depth and the draft of the design ship are obtained correspondingly.

[0023] As a further preferred, the method further comprises the following steps:

[0024] The resistance performance of the design ship type is predicted using a machine learning method, and the effective power curve of the design ship is drawn, so as to realize the performance prediction and verification of the ship type;

[0025] If the performance does not meet the requirements, the ship type design is re-performed after adjustment until the performance requirements are met; the adjustment is performed in the following manner one and / or manner two:

[0026] Manner one: adjusting the number and type of mother ships;

[0027] Manner two: adjusting the key design parameters of the design ship.

[0028] According to another aspect of the present application, a ship type design system is provided, comprising a processor for executing the above ship type design method.

[0029] Overall, compared with the prior art, the above technical solutions conceived by the present application mainly have the following technical advantages:

[0030] 1. The present application can generate a corresponding ship type according to the ship shape and performance requirements; specifically, first, based on a deep learning model, the transverse section area curve of the design ship is obtained, then the ship type fusion technology is used to fit the transverse section area curve with a Bezier curve and optimize the control points, the control point equation set is solved to determine the fusion weight, and then the ship types of multiple mother ships are fused according to the corresponding weight to generate the design ship type, thereby realizing the optimal design result, ensuring that the design ship is close to the requirements in performance, and greatly improving the ship type design efficiency.

[0031] 2. The present application uses a large language model to analyze user input parameters and extract key ship design parameters; and based on a deep learning model and a large amount of transverse section data of mother ships, a large amount of training data is generated through data augmentation to train the model to generate a transverse section area curve; the present application combines a large language model, deep learning and ship type fusion technology to provide an efficient and intelligent solution for ship design. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 It is a flowchart of the ship type design method of the embodiment of the present application.

[0033] Figure 2A typical parent ship transverse section area curve and upper and lower limit schematic diagram of the embodiment of the present application.

[0034] Figure 3 A process schematic diagram for generating a transverse section area curve using a multi-layer perceptron (MLP) of the embodiment of the present application. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the present application clearer and more apparent, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0036] The ship type design method provided by the embodiment of the present application, as shown in Figure 1 includes the following steps:

[0037] (1) Key design parameter identification and extraction based on a large language model

[0038] In ship type design, the identification and extraction of key design parameters is the first step of the entire process. This process aims to accurately extract the key design parameters required for ship design from the user's input and ensure the accuracy and completeness of these parameters. Since ship design involves a large number of professional terms and complex parameter relationships, it is necessary to efficiently and accurately extract these parameters. The present application uses an external large language model to achieve key parameter identification and extraction. Specifically as follows:

[0039] First, user input analysis. Users can provide ship design requirements through natural language text input, i.e. design ship shape and performance requirements. The input can be free text or structured parameter list. For example:

[0040] Free text: "Design a cargo ship with a displacement of 5000 tons, main engine power of 3000 horsepower, and design speed requirement of 20 knots."

[0041] Structured input: "Cargo ship, displacement = 5000 tons, main engine power = 3000 horsepower, design speed = 20 knots."

[0042] Second, term identification and preprocessing. The field of ship design involves a large number of professional terms and specific parameter names, such as "displacement", "main engine power", "speed", etc. The model needs to be able to accurately identify these terms and match them with the corresponding parameter types. In order to achieve this, ship design related professional text data can be introduced during model training. These data help the model understand the contextual meaning of professional terms and improve its recognition ability in complex text.

[0043] Finally, parameter identification and verification are performed. The user's text intent and specific parameter values are determined, and the parameter type is classified. The identified parameter values are classified into specific parameter types, such as "displacement", "main engine power", etc.

[0044] To ensure the consistency and calculability of the parameters, the model needs to standardize the extracted parameter values. For example: convert "tons" to cubic meters; convert "horsepower" to kilowatts; convert "knots" to meters per second.

[0045] Through the above steps, the key design parameters of the ship can be determined based on the shape and performance requirements of the designed ship, realizing the intelligent identification and extraction of the key design parameters of the ship, and providing reliable basic data support for subsequent ship design optimization and performance evaluation.

[0046] (2) Transformed and generated based on the cross-sectional area curve of the deep learning

[0047] Through the analysis of the ship design requirements, the following key design parameters are identified and extracted: ship type, displacement, main engine power, and design speed. These parameters are the basic information of ship design, which can effectively reflect the basic performance requirements and design goals of the ship.

[0048] The above key design parameters are used as input parameters to the trained deep learning model to generate the ship's cross-sectional area curve. The cross-sectional area curve is a curve with ship length as the horizontal coordinate and each station cross-sectional area as the vertical coordinate. The key is to obtain the cross-sectional area of each station of the ship, i.e. the vertical coordinate of each point on the curve. It is worth noting that the design waterline intersects the 10th station transverse line at the lowest point of the ship, so the cross-sectional area of the 10th station is 0 and does not need to be analyzed. In addition, the horizontal coordinate is the ship length, and the distance between each station is the 10th part of the vertical length. This means that in the model training process, in addition to predicting the cross-sectional area of each station, the vertical length of the ship also needs to be obtained.

[0049] Therefore, it can be determined that in the deep learning model training task, the input parameters are displacement, center of buoyancy longitudinal position, prismatic coefficient, main engine power, and design speed, and the output parameters are the cross-sectional area of each station and the vertical length, as shown in Table 1:

[0050]

[0051] In the design of ship type, the relationship between the area curve of each station of the ship and the key parameters is not a simple linear relationship, but involves complex nonlinear relationships, forming a typical multiple output regression problem. The embodiment adopts a multilayer perceptron (MLP) as a basic model architecture. The MLP can efficiently map a single input to a high-dimensional output space with its fully connected structure, can effectively capture the trend of change of each constraint parameter, and can map the input features to a high-dimensional space through a nonlinear activation function, thereby realizing the learning and prediction of complex patterns.

[0052] Specifically, as shown in Figure 3 , through the MLP model, different ship key design parameters are input, and the corresponding 0-10 station cross-sectional area can be obtained, and the design ship cross-sectional area curve can be drawn. The MLP is composed of three types of layers: input layer, hidden layer and output layer, wherein the input layer is responsible for receiving the original data features; the hidden layer is the core part of the model, which performs deep transformation on the features through a nonlinear activation function to extract high-order features; and the output layer generates the final prediction results.

[0053] Further, the training of the deep learning model needs to rely on a large amount of high-quality data support, so a complete structure and rich parameter parent ship shape database needs to be constructed as a training set. As many parent ships as possible can be selected as the original data; in the case of limited number of parent ships, in order to obtain sufficient data scale and improve the generalization ability of the model, data sample expansion can be performed through scaling adjustment, as shown in Figure 2 , specifically, the longitudinal coordinate value of each coordinate point of the different parent ship cross-sectional area curve is adjusted within ±30%, and different cross-sectional area curves can be obtained by randomly taking values within the obtained variable range, while the ship structure rationality can be maintained.

[0054] Then the key design parameters corresponding to different cross-sectional area curves need to be obtained, i.e. the displacement, the longitudinal position of the center of buoyancy, the prismatic coefficient, the main engine power and the design speed, and the effective power and other key design parameters are obtained by using CFD simulation, model test and empirical formula resistance prediction method. On this basis, the MLP model is trained to obtain the ship key design parameter-cross-sectional area curve model.

[0055] According to the design requirements, the key design parameters of the design ship are input into the trained model, and the corresponding 0-10 station cross-sectional area and the length between perpendiculars can be obtained, and the design ship cross-sectional area curve can be drawn.

[0056] (3) Cross-sectional line and main dimension parameter fusion and generation based on ship type fusion technology

[0057] Bézier curves are spline curves based on Bernstein polynomials. They possess strong approximation capabilities and excellent convergence in the interval [a, b], and can accurately describe complex geometric shapes. An n-order Bézier curve consists of n+1 Bernstein polynomials and is controlled by n+1 control points. To achieve accurate fitting of the cross-sectional area curve, the control points of the Bézier curve need to be optimized. The shape of the Bézier curve is closely related to the position of its control points. By adjusting the position of the intermediate control points, precise control of the curve shape can be achieved while maintaining the smoothness of the curve.

[0058] The `fmincon` function from the MATLAB Optimization Toolbox was used. This function is a constrained nonlinear optimization tool based on the Sequential Quadratic Programming (SQP) algorithm, which can effectively handle nonlinear constraints in the control point optimization process. In the optimization process, the initial control point positions are first set, and then the coordinates of the control points are adjusted through iterative calculations until the convergence condition is met. A fourth-order Bézier curve is used to fit the cross-sectional area curve, which is fitted by three intermediate control points and two endpoints. An optimization algorithm is then used to optimize the control points of the Bézier curve to obtain the best fit; thus, the number of control points on the cross-sectional area curve can be reduced to three. The smoothness of the fitted curve is also observed.

[0059] Following the above method, the control points on the cross-sectional area curves of the design ship and each parent ship are reduced to 3; through uniform scale scaling, the lengths of each parent ship are adjusted to the same design value to ensure consistency during parameter fusion.

[0060] Then, based on the cross-sectional area curve under the design length, a homogeneous linear equation system is established to achieve the fusion calculation of parameters for each parent ship type, as detailed below:

[0061] For each control point, construct the following homogeneous linear equation:

[0062]

[0063] In the formula, n It refers to the number of mother ships; y i It is the first i The ordinate of the control points on the cross-sectional area curve of the mother ship. w i It is the first i The weight of each mother ship y It is the ordinate of the control point on the same abscissa curve of the designed ship's cross section.

[0064] After the 4th order Bezier curve is used to fit the transverse area curve, the control points in the middle of the transverse area curve are reduced to 3, and then 3 homogeneous linear equations are constructed to form a homogeneous linear equation set, and the transverse lines of 3 or more than 3 mother ships are fused to solve the homogeneous linear equation set to obtain the weight of each mother ship w i It should be noted that when more than 3 control points are selected (i.e. more than 3), a non-unique solution is obtained, which can be filtered according to the solution or uniformly fused, and the optimal solution is selected by performance prediction.

[0065] Based on the weight, the transverse lines of each mother ship are fused to obtain the transverse line of the design ship. The coordinates of the midpoint of the fused transverse line of the design ship are as follows:

[0066]

[0067] In the formula, w i is the fusion weight of the first i mother ship; x i is the transverse line coordinate of the mother ship, x is the transverse line coordinate fused and generated; y i is the transverse line coordinate of the mother ship, y is the transverse line coordinate fused and generated.

[0068] The main dimension parameters of the design ship are obtained, including the length between perpendiculars, the width, the depth and the draft. The length between perpendiculars is obtained by a deep learning model, the width, the depth and the draft are fused based on the weight according to the similar method of the transverse line fusion. Unlike the transverse line which has two dimensions of horizontal and vertical coordinates, the width, the depth and the draft only have one dimension, so the parameters of the mother ship can be directly weighted to obtain the parameters.

[0069] (4) Design ship performance verification and optimization

[0070] The machine learning resistance prediction method is used to predict the resistance, main engine power and speed of the generated ship type, and to verify whether it meets the user's requirements. If it does not meet the requirements, the design parameters need to be adjusted for a new round of iterative design optimization.

[0071] There are two adjustment methods: 1) adjusting the number and type of selected mother ships, changing the fusion weight, and re-fusing; 2) adjusting the key design parameters of the ship to change the generated transverse area curve and the transverse line of the design ship.

[0072] The present application provides a ship type design system, which comprises a processor for executing the ship type design method as described above.

[0073] In addition, the interactive page of the ship type design system is divided into two areas, namely a "design parameter input area" and a "design interaction area". The user first edits and sends the design requirements in the "performance input area", including the ship type, displacement, longitudinal position of the buoyancy center, prismatic coefficient, main engine power and design speed. The system will respond in the "design interaction area": extract and confirm the key design requirements, display the number of parent ships of this type and preview the transverse section lines, obtain the transverse section area and perpendicular length of each station of the design ship through the ship key design parameter-transverse section area curve MLP model according to the existing parent ships and design requirements, and further draw the transverse section area curve of the design ship. Then the user can view the transverse section lines of different parent ships, and finally select and input the numbers of three or more fusion parent ships in the "design parameter input area". The system will continue to respond in the "design interaction area": calculate the fusion weights of different ships according to the selected fusion parent ships, and perform fusion and generation of the transverse section lines of the design ship. The transverse section lines can be exported in dwg format, and the main dimensions of the design ship are also generated through the same fusion method. Then, through the machine learning resistance prediction method, the effective power and design speed of the design ship are predicted, and the effective power curve of the design ship is drawn. All data can be exported in xlsx format.

[0074] The ship type design method and system provided by the present application can generate corresponding ship transverse section line curves according to the key ship design parameters input by the user. First, the large language model is used to analyze the user input parameters and extract the key ship design parameters. Second, based on the deep learning model and a large amount of transverse section data of parent ships, a large amount of training data is generated through data augmentation, and the model is trained to generate the transverse section area curve. Then, the ship type fusion technology is used. First, the transverse section area curve is fitted with a fourth-order Bezier curve and the control points are optimized to determine the fusion weights. Then, the transverse lines of multiple parent ships are fused according to the corresponding weights to generate the transverse line of the design ship. Finally, the machine learning resistance prediction method is used to predict and optimize the performance of the generated ship type to ensure that the performance of the design ship meets the user's requirements. The present application combines large language models, deep learning and ship type fusion technology to provide an efficient and intelligent solution for ship design.

[0075] Those skilled in the art will readily understand that the above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method of designing a ship form, characterized by, The method comprises the following steps: determining key design parameters based on the shape and performance requirements of the design ship; inputting the key design parameters into the trained deep learning model to obtain the transverse area and vertical length at each station of the design ship, thereby forming a transverse area curve; performing Bezier curve fitting on the transverse area curves of the design ship and each parent ship to reduce the control points on the transverse area curves; adjusting the lengths of the parent ships to the same value through unified scale processing; for each control point, constructing a homogeneous linear equation based on the corresponding transverse areas on the transverse area curves of the design ship and each parent ship, and solving to obtain the weight of each parent ship; the homogeneous linear equation is expressed as: wherein, n is the number of parent ships; y , y i is the longitudinal coordinate of the control point on the curve of the cross-sectional area of the design ship, the first i parent ship, respectively, i.e. the corresponding cross-sectional area; w i is the weight of the first i parent ship; fusing the ship types of the parent ships according to the weights to obtain the design ship type.

2. The ship design method of claim 1, wherein inputting the performance requirements of the design ship into the large language model to obtain the key design parameters.

3. The ship design method of claim 1, wherein The key design parameters include displacement, longitudinal position of the center of buoyancy, prismatic coefficient, main engine power, and design speed.

4. The ship design method of claim 1, wherein The training method of the deep learning model is: obtaining the key design parameters and corresponding transverse area curves of a plurality of parent ships; adjusting the transverse areas at each station in the transverse area curves within ±30% of the amplitude, and corresponding to the key design parameters, thereby expanding the data volume and constructing a training set; training the deep learning model through the training set to obtain the trained deep learning model.

5. The ship design method of claim 1, wherein, The deep learning model adopts a multi-layer perception model MLP.

6. The ship design method of claim 1, wherein, Fusing the ship types of the parent ships according to the weights, comprising: Fusing the transverse lines, widths, depths, and draughts of the parent ships according to the weights to obtain the transverse lines, widths, depths, and draughts of the design ship.

7. A hull form design method according to any one of claims 1 to 6, wherein, The method further comprises the following steps: using a machine learning method to predict the resistance performance of the design ship type, drawing an effective power curve of the design ship, and realizing performance prediction and verification of the ship type; if the performance does not meet the requirements, adjusting and redesigning the ship type until the performance requirements are met; the adjustment is performed in the following ways one and / or two: way one: adjusting the number and type of parent ships; way two: adjusting the key design parameters of the design ship.

8. A ship design system characterized by comprising: The method comprises a processor configured to perform the ship type design method according to any one of claims 1-7.

Images