A method and system for ship performance evaluation and optimization based on neural network

Abstract

The present application provides a method and system for evaluating and optimizing ship performance based on a neural network. The method includes: a ship type image step, in which a cross-sectional view of a generated three-dimensional geometric ship type is intercepted, a plan view is generated using the cross-sectional lines in each view, and the enclosed areas are expressed in different colors to obtain a color cross-sectional view of the ship type; a neural network model training step, in which the hydrostatic force and resistance of the ship are labeled, normalized, and standardized, and the model is obtained by training the neural network with the color cross-sectional view of the ship type as input and the hydrostatic force and resistance of the ship as output; a model calculation step, in which the ship to be evaluated is converted into a new ship type diagram, and the diagram is input into the model, and the hydrostatic force and resistance of the ship are output; a judgment and optimization step, in which the ship hydrostatic force and resistance data output by the model are judged to meet the requirements and, if not, are converted into a variable in a weighted manner, and a genetic algorithm is used to optimize the variable with corresponding hydrostatic constraints, thereby optimizing the ship type.

Description

Technical Field

[0001] The present application relates to the field of ship performance, and in particular to a method and system for ship performance evaluation and optimization based on neural networks. Background Art

[0002] A ship's resistance characteristics are one of its most important performance indicators. Its superiority has a significant impact on energy conservation, emission reduction, cost reduction, and efficiency improvement, and has long been a focus of attention in the ship design community. The assessment and optimization of ship resistance are crucial for the shipbuilding and shipping industries. By reducing the ship's resistance in the water, fuel consumption and emissions can be reduced, which in turn helps reduce operating costs and environmental impact. Organizations such as the International Maritime Organization are continuously raising environmental standards for ships, and optimizing ship resistance is an effective means of helping shipowners meet these regulations. High-performance ship designs enhance the competitiveness of ship designers and shipowners in the market, attracting more customers and generating greater economic benefits. Ship optimization contributes to the shipping industry's response to global climate change and demonstrates the shipbuilding and shipping industries' commitment to social responsibility and environmental protection.

[0003] Current approaches to optimizing ship resistance typically rely on numerical simulation, combined with parametric modeling. After extensive sample calculations, optimization algorithms are employed to identify the optimal parameter combination and ultimately achieve the target ship form. Alternatively, surrogate models, such as response surface models and Kriging models, can be constructed by combining the ship form's parameters with a sample set of corresponding ship form resistance to find the optimal solution. These methods rely on a parametric representation of the ship form. However, the reality is that different ship forms have varying construction parameters. Consequently, past sample datasets cannot be applied to the analysis of new ships, resulting in a waste of data resources. Summary of the Invention

[0004] In order to solve the problem of data resource waste caused by the different construction parameters of different ship types in traditional methods and the inability of past sample data sets to be applied to the analysis of new ships, the present application provides a ship performance evaluation and optimization method and system based on neural networks. The ship performance evaluation model is trained using past sample data sets, which can quickly and efficiently evaluate the hydrostatic force, resistance and other performance of different types of new ships. Combined with the optimization method, it can easily realize ship type optimization with constraints, meet actual use needs, and optimize ship parameters more efficiently.

[0005] This application is achieved through the following technical solutions:

[0006] A method for evaluating and optimizing ship performance based on a neural network comprises the following steps:

[0007] Ship form image generation step: using 3D modeling technology to generate multiple different 3D geometric ship forms, intercepting multiple cross-sectional views of each of the 3D geometric ship forms, and using the cross-sectional lines in each of the cross-sectional views to generate a plan view, and expressing the areas enclosed by each of the cross-sectional lines in different colors to obtain color cross-sectional views of the ship form corresponding to each of the 3D geometric ship forms;

[0008] The neural network training model step includes: labeling the ship hydrostatic data and ship resistance data corresponding to each of the three-dimensional geometric ship forms; normalizing and standardizing the ship form color profile and the labeled data, and then using the ship form color profile as input and the ship hydrostatic data and ship resistance data as output to obtain a ship performance evaluation model through neural network training;

[0009] The model calculates hydrostatic force and resistance: the ship to be evaluated is converted into a corresponding new ship type image, and the new ship type image is input into the trained ship performance evaluation model. After calculation, the model outputs the corresponding ship hydrostatic force data and ship resistance data;

[0010] Judgment and optimization steps: The ship hydrostatic data and ship resistance data output by the model calculation are compared with their respective predetermined parameters to determine whether they meet the requirements. If the requirements are not met, the ship hydrostatic data and ship resistance data output by the model calculation are converted into a variable in a weighted manner, and the genetic algorithm is used to optimize the variable with corresponding hydrostatic constraints to obtain an optimized new ship type image; the optimized new ship type image is input into the ship performance evaluation model again for model calculation, and the output ship hydrostatic data and ship resistance data are compared and judged again. The judgment and optimization are continuously cycled until the requirements are met, and the optimized ship type is output to achieve resistance optimization under hydrostatic constraints.

[0011] Preferably, in the neural network training model step, the ship hydrostatic data involved in the annotation and neural network training of the ship performance evaluation model include displacement data and center of buoyancy position data; the model calculating hydrostatic force and resistance step outputs the corresponding displacement data, center of buoyancy position data and ship resistance data; the judgment and optimization step compares the displacement data, center of buoyancy position data and ship resistance data with their respective predetermined parameters to determine whether the requirements are met. If the requirements are not met, the displacement data, center of buoyancy position data and ship resistance data output by the model calculation are converted into a variable in a weighted manner, and a genetic algorithm is used to optimize the variable with respect to the loading capacity constraint conditions and the hull internal layout constraint conditions to obtain an optimized new ship type image; and the optimized new ship type image is inputted into the ship performance evaluation model again for model calculation, and the displacement data, center of buoyancy position data and ship resistance data are output for comparison and judgment again, and the judgment and optimization are continuously cycled until the requirements are met, and the optimized ship type is output to achieve resistance optimization under several hydrostatic constraint conditions.

[0012] Preferably, in the neural network training model step, the ship hydrostatic data involved in the annotation and neural network training of the ship performance evaluation model include displacement data, buoyancy center position data, drift center position data and stability data; the model calculates hydrostatic force and resistance step outputs the corresponding displacement data, buoyancy center position data, drift center position data, stability data and ship resistance data; the judgment and optimization step compares the displacement data, buoyancy center position data, drift center position data, stability data and ship resistance data with their respective predetermined parameters to judge whether the requirements are met. If the requirements are not met, the displacement data calculated and output by the model is compared. The center of buoyancy position data, drift center position data, stability data and ship resistance data are converted into a variable in a weighted manner, and the genetic algorithm is used to optimize the variable with respect to the loading capacity constraints, the internal layout constraints of the hull and the navigation safety constraints to obtain an optimized new ship type picture; and the optimized new ship type picture is input into the ship performance evaluation model again for model calculation, and the displacement data, center of buoyancy position data, drift center position data, stability data and ship resistance data are output for comparison and judgment again, and the judgment and optimization are continuously circulated until the requirements are met, and the optimized ship type is output to achieve resistance optimization under several hydrostatic constraints.

[0013] Preferably, in the ship form imaging step, the color cross-section images of different three-dimensional geometric ship forms have the same size and image format, and the color cross-section images adopt a red, green, and blue (RGB) color mode.

[0014] Preferably, in the ship form image generation step, the method of intercepting a plurality of cross-sectional views of each of the three-dimensional geometric ship forms and generating a plan view using cross-sectional lines in each of the cross-sectional views comprises:

[0015] Take the designated station as the dividing position, and cut the cross section perpendicular to the line connecting the bow and stern to the waterline of the ship as the cross section;

[0016] The cross-sectional lines in each captured cross-sectional view are used to generate a plan view with the left side as the stern direction and the right side as the bow direction.

[0017] Preferably, in the neural network model training step, marking the ship hydrostatic data and ship resistance data corresponding to each of the three-dimensional geometric ship forms includes:

[0018] Determine the hydrostatic data of the ship corresponding to each of the three-dimensional geometric ship forms using professional software or a self-programming method;

[0019] Determining ship resistance data corresponding to each of the three-dimensional geometric ship forms using experiments, computational fluid dynamics methods, or potential flow methods;

[0020] The ship hydrostatic data and ship resistance data corresponding to each of the three-dimensional geometric ship types are marked.

[0021] Preferably, in the judgment and optimization steps, in addition to using genetic algorithms, simulated annealing algorithms and / or particle swarm algorithms are also used to optimize the corresponding hydrostatic constraints of the converted variables, and the optimized ship type is evaluated and screened in combination with the ship performance evaluation model.

[0022] A ship performance evaluation and optimization system based on neural network includes an image generation module, a neural network training module, a ship hydrostatic data and ship resistance data evaluation module, and a ship type optimization module connected in sequence; wherein,

[0023] The image generation module is used to generate a plurality of different three-dimensional geometric ship shapes using three-dimensional modeling technology, intercept a plurality of cross-sectional views of each of the three-dimensional geometric ship shapes, and generate a plan view using the cross-sectional lines in each of the cross-sectional views, and express the areas enclosed by each of the cross-sectional lines in different colors to obtain a color cross-sectional view of the ship shape corresponding to each of the three-dimensional geometric ship shapes;

[0024] The neural network training module is used to label the ship hydrostatic data and ship resistance data corresponding to each of the three-dimensional geometric ship forms; then the ship form color cross-section image and the labeled data are normalized and standardized, and then the ship performance evaluation model is obtained through neural network training using the ship form color cross-section image as input and the ship hydrostatic data and ship resistance data as output;

[0025] The ship hydrostatic data and ship resistance data evaluation module is used to convert the ship to be evaluated into a corresponding new ship type image, and input the new ship type image into a trained ship performance evaluation model. After the model calculates, it outputs the corresponding ship hydrostatic data and ship resistance data;

[0026] The ship form optimization module is used to compare the ship hydrostatic data and ship resistance data output by the model calculation with their respective predetermined parameters to determine whether the requirements are met. If the requirements are not met, the ship hydrostatic data and ship resistance data output by the model calculation are converted into a variable in a weighted manner, and a genetic algorithm is used to optimize the variable with corresponding hydrostatic constraints to obtain an optimized new ship form image; and the optimized new ship form image is input into the ship performance evaluation model again for model calculation, and the output ship hydrostatic data and ship resistance data are compared and judged again, and the judgment and optimization are continuously cycled until the requirements are met, and the optimized ship form is output to achieve resistance optimization under hydrostatic constraints.

[0027] Preferably, the ship hydrostatic data involved in the neural network training module and the neural network training of the ship performance evaluation model include displacement data and center of buoyancy position data; the ship hydrostatic data and ship resistance data evaluation module outputs the corresponding displacement data, center of buoyancy position data and ship resistance data; the ship form optimization module compares the displacement data, center of buoyancy position data and ship resistance data with their respective predetermined parameters to determine whether the requirements are met. If the requirements are not met, the displacement data, center of buoyancy position data and ship resistance data output by the model calculation are converted into a variable in a weighted manner, and a genetic algorithm is used to optimize the variable with respect to the loading capacity constraint conditions and the hull internal layout constraint conditions to obtain an optimized new ship form image; and the optimized new ship form image is inputted into the ship performance evaluation model again for model calculation, and the displacement data, center of buoyancy position data and ship resistance data are output for comparison and judgment again, and the judgment and optimization are continuously cycled until the requirements are met, and the optimized ship form is output to achieve resistance optimization under several hydrostatic constraint conditions.

[0028] Preferably, in the ship type optimization module, in addition to the genetic algorithm, the simulated annealing algorithm and / or the particle swarm algorithm are also used to optimize the corresponding hydrostatic constraints of the converted variables, and the optimized ship type is evaluated and screened in combination with the ship performance evaluation model.

[0029] The beneficial effects of this application are:

[0030] The present application provides a method for evaluating and optimizing ship performance based on a neural network. The method uses 3D modeling technology to generate a plurality of different 3D geometric ship forms, intercepts a plurality of cross-sectional views of each of the 3D geometric ship forms, and uses the cross-sectional lines in each of the cross-sectional views to generate a plan view. In this way, a non-parametric expression method of the ship form is established with the cross-sectional views of the ship form, and a universal representation method of the ship form is constructed as sample data, which solves the problem of data resource waste caused by the different construction parameters of different ship forms in the traditional method and the inability of past sample data sets to be applied to the analysis of new ships; the area surrounded by each of the cross-sectional lines is expressed in different colors, and the area of ​​each of the 3D geometric ship forms is obtained. The ship type color profiles corresponding to the ship types are respectively obtained. In this way, the color blocks are divided by conventional cross-section lines to generate pictures. The pictures are obtained simply and efficiently, and the color changes of different ship types are clearly distinguished, which is conducive to deep learning and feature extraction; the ship hydrostatic data and ship resistance data corresponding to each of the three-dimensional geometric ship types are marked, and then the ship type color profiles and the marked data are normalized and standardized. Then, the ship type color profiles are used as input, and the ship hydrostatic data and ship resistance data are used as output. After the neural network training, a ship performance evaluation model is obtained. In this way, the ship type color profile is used as the input feature data, and the picture does not need to be scaled in the two directions of width and height. Or the scaling is small, which maintains the relative relationship between the various parts of the ship type characteristics and improves the training speed and accuracy of the model; the ship to be evaluated is converted into a corresponding new ship type picture, and the new ship type picture is input into the trained ship performance evaluation model. After the model is calculated, the corresponding ship hydrostatic data and ship resistance data are output. In this way, the ship performance evaluation model is obtained by training with the past sample data set, and then the hydrostatic, resistance and other performances of different types of new ships can be evaluated quickly and efficiently; the ship hydrostatic data and ship resistance data calculated and output by the model are compared with their respective predetermined parameters to determine whether the requirements are met. If the requirements are not met, the model calculation output is The hydrostatic data and resistance data of the ship are converted into a variable in a weighted manner, and the genetic algorithm is used to optimize the variable with the corresponding hydrostatic constraint conditions to obtain the optimized new ship type picture, and the optimized new ship type picture is input into the ship performance evaluation model again for model calculation, and the hydrostatic data and resistance data of the ship are output for comparison and judgment again, and the judgment and optimization are continuously circulated until the requirements are met, and the optimized ship type is output to achieve resistance optimization under hydrostatic constraint conditions. In this way, on the basis of the model, combined with the optimization method, it is convenient to optimize the ship type under given constraints, meet actual use needs, and significantly improve the efficiency of ship parameter optimization. After adopting the neural network-based ship performance evaluation and optimization method of this application, the ship performance evaluation model can be obtained by training with past sample data sets and applied to the analysis of new ships, thereby quickly and efficiently evaluating the hydrostatic, resistance and other performances of different types of new ships.Combined with the optimization method, it is easy to realize the optimization of ship types with constraints, meet the actual use requirements, and optimize the ship parameters more efficiently.

[0031] The ship performance evaluation and optimization method based on neural networks in this application can also be called a method for ship hydrostatic evaluation and resistance optimization based on deep learning. A convolutional neural network agent model, a ship performance evaluation model, is constructed to express the relationship between ship characteristics and performance. It can not only obtain the ship resistance, but also the hydrostatic data of the corresponding ship type, providing a feasible method for ship type optimization under restricted conditions, and realizing the optimization of resistance performance while meeting the hydrostatic requirements.

[0032] This application uses a neural network-deep learning method to establish a proxy model (ship performance evaluation model) for ship types and ship hydrostatic forces and resistance. This proxy model can not only evaluate the resistance of different ship types, but also evaluate hydrostatic forces as needed, such as displacement volume, center of buoyancy, center of drift, stability, etc. That is, ship hydrostatic force data can further involve displacement data, center of buoyancy position data, or further involve displacement data, center of buoyancy position data, center of drift position data and stability data. Compared with traditional solutions, it provides richer output content and can more comprehensively and comprehensively characterize the various characteristics of the ship, thereby ensuring that the trained ship performance evaluation model can achieve more accurate ship performance evaluation. In actual engineering, for ship performance optimization, there are ship hydrostatic force requirements, such as general displacement (displacement volume), which is related to loading capacity; center of buoyancy position is related to the internal layout of the hull; stability is related to navigation safety, etc. It is necessary to achieve optimal resistance while meeting certain requirements (constraints). In the past, the calculation of ship hydrostatic force required complex processes or professional software, and could not be carried out simultaneously with resistance evaluation. The required work was complicated and required the cooperation of multiple people. Since the hydrostatic force and resistance of a ship are both related to the ship's lines, deep learning is used to unify the evaluation of different performance characteristics with the line images, simplifying the workflow and making it more suitable for actual engineering use.

[0033] Based on the surrogate model (ship performance evaluation model), combined with optimization methods, it is easy to optimize ship types with constraints to meet actual usage requirements. Constraints refer to hydrostatic limits, such as displacement that cannot be less than a fixed value, otherwise the ship cannot carry sufficient cargo. Because ship types are represented by images, they can be easily and diversely obtained. The surrogate model combined with optimization methods can quickly evaluate a large number of options. Optimization can be based on manual experience or optimization algorithms such as genetic algorithms, simulated annealing algorithms, and particle swarm optimization algorithms. Using these optimization algorithms in combination with the surrogate model allows for the rapid evaluation and screening of a large number of excellent ship types in a short period of time, improving work efficiency.

[0034] In this application, the color profiles of different three-dimensional geometric ship types have different color distributions, and the color changes are clearly distinguishable. In this way, different ship types will have obvious distinguishing features, which is conducive to the extraction of features for deep learning, thereby making the training speed of the ship performance evaluation model faster and the convergence better.

[0035] This application divides the color blocks by conventional cross-section station lines, which is friendly to ship designers. Pictures can be obtained from the type value table, reducing extra work.

[0036] The present application also relates to a neural network-based ship performance evaluation and optimization system, which corresponds to the above-mentioned neural network-based ship performance evaluation and optimization method, and can be understood as a system that implements the above-mentioned neural network-based ship performance evaluation and optimization method, including an image generation module, a neural network training module, a ship hydrostatic data and ship resistance data evaluation module, and a ship type optimization module. The image generation module can use three-dimensional modeling technology to generate a variety of three-dimensional geometric ship types, and intercept multiple cross-sectional views. By extracting the cross-sectional lines in each cross-sectional view, a non-parametric ship type expression method is constructed. That is, a universal ship type representation method. As sample data, it effectively solves the problem of data resource waste caused by the different construction parameters of different ship types in traditional methods, and the past sample data sets cannot be applied to the analysis of new ships. The module further fills the areas surrounded by cross-sections with different colors to generate color cross-sections corresponding to each three-dimensional geometric ship type. This image generation method based on the conventional cross-section line to divide the color blocks is not only simple and efficient, but also has significant color distinctions between ship types, which is conducive to the efficient feature extraction of deep learning algorithms; the neural network training module can mark the ship hydrostatic data and ship corresponding to each three-dimensional geometric ship type. Resistance data, and then normalize and standardize the ship color profile and annotated data, and then use the ship color profile as input, ship hydrostatic data and ship resistance data as output, and obtain the ship performance evaluation model through neural network training. In this process, the input color profile maintains or approximately maintains the original proportion in width and height dimensions, effectively retaining the relative relationship of ship characteristics, thereby significantly improving the speed and accuracy of model training; the ship hydrostatic data and ship resistance data evaluation module can convert the ship to be evaluated into the corresponding new ship type picture, and input it into the trained ship performance evaluation model. After the model calculation, the corresponding ship hydrostatic data and ship resistance data are output. The ship performance evaluation model is obtained by training with past sample data sets, which can quickly output the hydrostatic, resistance and other data of the new ship type to achieve efficient and accurate performance evaluation; the ship type optimization module uses genetic algorithms to weightedly integrate the ship hydrostatic data and resistance data into a single variable, and optimize the variable under the condition of meeting the hydrostatic constraints, so as to generate a better new ship type image and minimize the resistance under the hydrostatic constraints, which is convenient for ship type optimization under given restrictions, meets actual use needs, and significantly improves the efficiency of ship parameter optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flow chart of the neural network-based ship performance evaluation and optimization method for this application;

[0038] Figure 2 RGB color image converted from the cross-section of the ship lines of this application;

[0039] Figure 3 This is the framework diagram of the neural network ship performance evaluation model for this application;

[0040] Figure 4 Comparison diagram of ship waveforms before and after optimization for this application;

[0041] Figure 5 This is a structural block diagram of the neural network-based ship performance evaluation and optimization system for this application. DETAILED DESCRIPTION

[0042] This application discloses a method for ship performance evaluation and optimization based on a neural network, which aims to solve the problem of data resource waste caused by the inability of past sample data sets to be applied to the analysis of new ships, and to achieve ship type optimization with restricted conditions. The method can be executed by a processor or an electronic device with processing capabilities.

[0043] Current ship resistance optimization relies mainly on numerical simulation technology, which, combined with parametric modeling, calculates a large number of samples and applies optimization algorithms to explore the optimal parameter configuration, thereby determining the best ship design. Another strategy is to construct a sample set of ship parameters and corresponding ship resistance to construct a proxy model, such as a response surface model or a Kriging model. The proxy model is used to find the optimal solution. However, these methods are all based on the parametric expression of the ship type, but the actual situation is that the construction parameters of different ship types are not the same, which means that past sample data sets cannot be applied to the analysis of new ships, resulting in a waste of data resources. In addition, traditional methods often ignore the construction of proxy models between ship parameters and ship hydrostatic forces, so that the evaluation of hydrostatic forces still needs to rely on professional software, involving complex information exchange and the generation of ship geometry.

[0044] In order to overcome these limitations, the present application proposes an innovative method, namely, a non-parametric expression of the ship form is established based on the cross-sectional view of the ship form, and a proxy model between the ship characteristics and performance is constructed, especially using deep learning technology to evaluate ship performance and optimize ship resistance. Specifically, a convolutional neural network proxy model is designed, which can accurately capture and express the complex relationship between the ship characteristics and its performance. This model can not only predict the resistance of the ship form, but also provide the hydrostatic value of the corresponding ship form at the same time. This method provides a practical way for ship form optimization under restricted conditions (ship form optimization under given hydrostatic constraints), so that while meeting the hydrostatic requirements, the resistance performance can be optimized. In other words, the present application designs a general, non-parametric ship form expression method, and constructs a proxy model based on this. The proxy model is used to express the ship form and the ship's hydrostatic force and resistance. Combined with the optimization algorithm, the ship form optimization that meets the hydrostatic constraints can be achieved. The non-parametric representation of ship forms is achieved through images. These images serve as input for deep learning, and the corresponding hydrostatic force and resistance are used as output labels. A neural network model, such as VGG16 or a custom convolutional neural network, is trained to generate a multi-output deep learning model of hydrostatic force and resistance based on the images. New ship images are fed into the model to obtain hydrostatic force and resistance results. By using geometric modeling, image editing, and custom programming, images of the same format can be obtained, allowing the deep learning model to calculate hydrostatic force and resistance data. Optimization algorithms, such as genetic algorithms, are then applied to control image changes and identify ship forms that meet the requirements.

[0045] like Figure 1 As shown in the figure, the dotted box mainly includes the deep learning process and the ship form optimization process. The first is the establishment of a deep learning model (ship performance evaluation model) between the ship form sample image and the hydrostatic force and resistance of the ship form; the second is the ship form optimization process constructed using this deep learning model (ship performance evaluation model, also known as performance evaluation agent model). Specifically, the method includes the following steps:

[0046] 1. Ship Form Graphics Step: Utilizing 3D modeling technology, multiple different 3D geometric ship forms are generated. Multiple cross-sectional views of each 3D geometric ship form are captured. A plan view is generated using the cross-sectional lines within each cross-sectional view. The areas enclosed by each cross-sectional line are represented using different colors, resulting in a color cross-sectional view corresponding to each 3D geometric ship form. This step involves both sample ship form generation and ship form representation.

[0047] Preferably, the embodiment of the present application can use three-dimensional modeling software or professional ship design software to generate a certain number of three-dimensional geometric ship models, and the methods for obtaining ship models are diverse and convenient.

[0048] The embodiment of the present application is based on a three-dimensional geometric ship type, and divides the hull of the three-dimensional geometric ship type into a preset number of stations (in the process of ship design and construction, in order to facilitate construction and management, the hull is usually divided into several longitudinal segments, each segment is called a station) through a series of sectional planes perpendicular to the longitudinal axis of the hull (the so-called longitudinal axis of the hull refers to the direction pointing to the bow, which determines the front and rear direction of the ship, that is, the heading of the ship) and perpendicular to the bottom plane. In the specific operation, we use the designated station as the segmentation position, with the stern on the left and the bow on the right, and intercept in the height direction at the draft position, and intercept the section perpendicular to the line connecting the bow and the stern as the cross-sectional view. Based on the cross-sectional line formed by the intersection of these sectional planes and the hull of the three-dimensional geometric ship type, a two-dimensional plane view that integrates all sectional plane information is constructed. As Figure 2 As shown, in this two-dimensional plan view, the area enclosed by each two adjacent transverse lines is defined as a color block and colored using the RGB (Red, Green, Blue) color scheme to ensure distinct color differences between adjacent blocks. It is noteworthy that different ship models are distinguished by the unique color distribution. Using conventional transverse lines to divide the color blocks is user-friendly for ship designers, as images can be obtained from the model value table, reducing unnecessary work. Color profiles of different three-dimensional ship geometries have distinct color distributions, and the color variations are clearly distinguishable. This creates distinct characteristics for each ship type, facilitating feature extraction through deep learning, leading to faster training and better convergence of ship performance evaluation models. To ensure consistency, the segmentation method for each ship is consistent, and the generated color images are standardized in terms of format, size, coverage, and other parameters. Another advantage of using transverse sections is that during deep learning, the images do not need to be scaled in width or height, or the scaling is minimal, preserving the relative relationships between the various components of the ship type. A non-parametric representation of ship forms has been established using cross-sectional views, and a universal representation method for ship forms has been constructed as sample data. This method can comprehensively characterize the various performance characteristics of ships, resolving the data resource waste caused by traditional methods whereby past sample datasets cannot be applied to the analysis of new ships due to the varying construction parameters of different ship forms. Experienced ship designers can analyze resistance performance based on cross-sectional characteristics. Deep learning based on cross-sectional features can provide ship design with a differentiation capability similar to human experience, assisting designers in optimizing resistance performance using experience or optimization algorithms, thereby improving the efficiency of ship optimization.

[0049] Preferably, the preset number of stations may be 8 stations, 10 stations, 12 stations, etc.

[0050] Second, the neural network model training step involves labeling the hydrostatic and resistance data corresponding to each of the three-dimensional geometric ship forms. The ship form color profiles and the labeled data are then normalized and standardized. A ship performance evaluation model is then generated through neural network training using the ship form color profiles as input and the hydrostatic and resistance data as output. This step involves labeling data and deep learning training.

[0051] Preferably, the embodiments of the present application can use professional software (such as MAXSURF or MATLAB, etc.) or self-programming methods to determine the hydrostatic data of different ships (including displacement volume, center of buoyancy position, etc.).

[0052] Preferably, the embodiment of the present application can use experiments, computational fluid dynamics (CFD) methods or potential flow methods to determine the ship resistance data (including resistance value or resistance coefficient).

[0053] The embodiment of the present application marks the ship hydrostatic data and ship resistance data corresponding to each three-dimensional geometric ship type, which are used to verify the results output during the model training process, thereby ensuring the diversity and accuracy of the output.

[0054] like Figure 3 As shown, the embodiment of the present application can use a deep learning framework such as Pytorch to build a convolutional neural network, and can use a convolutional neural network such as VGG16 (Visual Geometry Group-16). Among them, the neural network includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. VGG16 has 16 layers of depth (13 convolutional layers and 3 fully connected layers). It provides a strong image feature extraction capability by stacking small convolution kernels (3x3), which is particularly suitable for image classification tasks.

[0055] This embodiment utilizes deep learning technology to construct a proxy model for ship types and their hydrostatic and resistance characteristics. This model not only accurately predicts the resistance performance of various ship types but also comprehensively evaluates multiple hydrostatic parameters, including displacement volume, center of buoyancy, drift center, and stability. Compared to traditional methods, it provides more detailed output information, comprehensively reflecting various ship performance indicators. In actual engineering applications, ship performance optimization requires comprehensive consideration of hydrostatic requirements. For example, displacement volume is directly related to loading capacity, center of buoyancy position affects the internal layout of the hull, and stability is the cornerstone of navigation safety. All of these must be achieved while ensuring optimal resistance. In the past, hydrostatic calculations involved cumbersome processes or relied on specialized software, and were performed separately from resistance assessments, resulting in a heavy workload and requiring teamwork. Given that both ship hydrostatic and resistance are directly affected by the ship's lines, this solution integrates line images with multiple performance assessments through deep learning, significantly simplifying the workflow and better meeting the needs of actual engineering applications.

[0056] 3. Model calculation of hydrostatic force and resistance step: The ship to be evaluated is converted into a corresponding new ship type image, and the new ship type image is input into the trained ship performance evaluation model. After calculation, the model outputs the corresponding ship hydrostatic force data and ship resistance data.

[0057] Preferably, in order to evaluate the performance of a new ship, a pictorial representation of its lines is required, which can be achieved in a variety of ways: using 3D modeling software to process the curved surface of the ship to generate a picture; or using a programming language such as Python to write a program to draw a 2D picture; or using image processing software to directly obtain it, etc.

[0058] 4. Judgment and optimization steps: The ship hydrostatic data and ship resistance data output by the model calculation are compared with their respective predetermined parameters to determine whether they meet the requirements. If the requirements are not met, the ship hydrostatic data and ship resistance data output by the model calculation are converted into a variable in a weighted manner, and the genetic algorithm (optimization algorithm) is used to optimize the variable with corresponding hydrostatic constraints to obtain an optimized new ship type image; and the optimized new ship type image is input into the ship performance evaluation model again for model calculation, and the output ship hydrostatic data and ship resistance data are compared and judged again, and the judgment and optimization are continuously cycled until the requirements are met, and the optimized ship type is output to achieve resistance optimization under hydrostatic constraints.

[0059] This step uses an optimization method with constraints, combined with the hydrostatic force and resistance of the ship calculated by the neural network agent model (ship performance evaluation model), to finally obtain a ship image that meets the corresponding hydrostatic force requirements.

[0060] Preferably, the optimization algorithm of this step can further use simulated annealing algorithm and / or particle swarm algorithm to optimize the corresponding hydrostatic constraints of the converted variables, and combine the ship performance evaluation model to evaluate and screen the optimized ship type.

[0061] The embodiment of the present application relies on a model framework and combines it with an optimization method to conveniently implement ship type optimization with constraints, ensuring that the ship meets actual use requirements such as the minimum hydrostatic limit. The hydrostatic limit can be that the displacement cannot be less than a fixed value, otherwise the ship cannot carry enough cargo. The ship type is presented in the form of pictures, which is convenient for diversified collection and processing. By combining the proxy model with optimization technology, many design solutions can be quickly evaluated. The optimization process can draw on manual experience, and can also use advanced algorithms such as simulated annealing and particle swarm. This solution can efficiently evaluate and screen the optimal ship type in a short period of time, significantly improving work efficiency.

[0062] Preferably, embodiments of the present application can convert the optimized ship image into a 3D geometric surface using different methods, depending on the method used to generate the new ship image. If the new ship image is derived from a 3D geometry, the corresponding 3D geometry can be directly used. If the new ship image is drawn from a 2D image, the 3D ship form must be reconstructed in 3D geometric modeling software based on the corresponding image.

[0063] For example, a cruise ship is used as an example to optimize the ship's resistance. The initial main dimensions of the cruise ship are shown in the table below:

[0064]

[0065] The optimization objective was to achieve a resistance of 22 kn at the design draft, while also meeting the hydrostatic constraints of a 0.5% variation in displacement volume and center of buoyancy. Using 3D modeling software, 100 different 3D ship geometries were modeled, using the ship's initial parameters as the basis data. Randomly adjusting any of these parameters allowed the modeling of 100 different 3D ship geometries. For each 3D ship geometries, a corresponding color cross-section image was generated. The displacement volume and center of buoyancy of each 3D ship geometries were calculated using Rhino software, and the resistance value of each 3D ship geometries was calculated using CFD methods. After normalizing and standardizing each color cross-section image and its corresponding displacement volume, center of buoyancy, and resistance values, the ship performance evaluation model was trained to convergence (using a VGG16 convolutional neural network) using each color cross-section image as input and outputting three data points: displacement volume, center of buoyancy, and resistance. The trained network implemented a fitting function, recognizing the ship image (color cross-section image) and providing the corresponding displacement volume, center of buoyancy, and resistance. Verified by the test set, the calculation error of the displacement volume and the center of buoyancy position is within ±0.5%, and the calculation error of the resistance value is within ±2%.

[0066] The displacement, center of buoyancy and resistance that need to be met are converted into a variable in a weighted manner, and a genetic algorithm is used to optimize this variable. The specific process is as follows: 200 different three-dimensional geometric ship types are randomly generated as the population size of a generation by parametric modeling, and a color profile corresponding to each three-dimensional geometric ship type is generated. The ship performance evaluation model then calculates the displacement, center of buoyancy and resistance corresponding to the color profile, converts the three outputs into one data, and the genetic algorithm is used to evaluate and generate the parameter values ​​for the next generation of control ship types, and repeats the above process. After 20 generations of screening, the ship type parameters reached a stable state, and a combination of ship type parameters that met the requirements was obtained. At this point, the displacement volume is reduced by 0.03% compared to the prototype, and the center of buoyancy is shifted back by 0.14% compared to the prototype. Numerical simulation verifies that the effective power of the optimized ship type is reduced by 2.6%, and the waveform is significantly improved (such as Figure 4 As shown in the figure, the upper part is the prototype waveform, and the lower part is the optimized waveform), which realizes the resistance optimization under the hydrostatic constraint conditions.

[0067] The above embodiment is based on ship hydrostatic data including displacement data and center of buoyancy data, but this embodiment is not intended to be limiting. Ship hydrostatic data may also include other content, in which case the specific model training, model calculation output, constraint setting, and ship form optimization will all change accordingly.

[0068] For example, in the neural network training model step, the ship hydrostatic data involved in the annotation and neural network training of the ship performance evaluation model include displacement data, center of buoyancy position data, drift center position data and stability data; the model calculates the hydrostatic force and resistance step outputs the corresponding displacement data, center of buoyancy position data, drift center position data, stability data and ship resistance data; the judgment and optimization step compares the displacement data, center of buoyancy position data, drift center position data, stability data and ship resistance data with their respective predetermined parameters to determine whether the requirements are met. If the requirements are not met, the displacement data, center of buoyancy position data, drift center position data, stability data and ship resistance data calculated by the model are compared. The displacement data, buoyancy center position data, stability data and ship resistance data are converted into a variable in a weighted manner, and the genetic algorithm is used to optimize the variable with respect to the loading capacity constraints, the internal layout constraints of the hull and the navigation safety constraints to obtain an optimized new ship type image; the optimized new ship type image is input into the ship performance evaluation model again for model calculation, and the output displacement data, buoyancy center position data, buoyancy center position data, stability data and ship resistance data are compared and judged with their respective predetermined parameters again, and the judgment and optimization are continuously circulated until the requirements are met, and the optimized ship type is output to achieve resistance optimization under several hydrostatic constraints.

[0069] Based on the same inventive concept, one or more embodiments of this specification also provide a neural network-based ship performance evaluation and optimization system. Since the principles of the problems solved by the neural network-based ship performance evaluation and optimization system are similar to those of the aforementioned neural network-based ship performance evaluation and optimization method, the implementation of the neural network-based ship performance evaluation and optimization system can refer to the aforementioned implementation of the neural network-based ship performance evaluation and optimization method, and the repeated parts will not be repeated.

[0070] Figure 5 This is a block diagram of the ship performance evaluation and optimization system based on neural network provided in one or more embodiments of this specification. Figure 5 As shown, the neural network-based ship performance evaluation and optimization system includes: an image generation module 101, a neural network training module 102, a ship hydrostatic data and ship resistance data evaluation module 103 and a ship type optimization module 104.

[0071] in,

[0072] The image generation module 101 is used to generate multiple different three-dimensional geometric ship shapes using three-dimensional modeling technology, intercept multiple cross-sectional views of each three-dimensional geometric ship shape, and use the cross-sectional lines in each cross-sectional view to generate a plan view, and express the area enclosed by each cross-sectional line with different colors to obtain a color cross-sectional view of the ship shape corresponding to each three-dimensional geometric ship shape.

[0073] The neural network training module 102 is used to label the ship hydrostatic data and ship resistance data corresponding to each three-dimensional geometric ship type; then the ship type color profile and the labeled data are normalized and standardized, and then the ship type color profile is used as input and the ship hydrostatic data and ship resistance data are used as output to obtain a ship performance evaluation model through neural network training.

[0074] The ship hydrostatic data and ship resistance data evaluation module 103 is used to convert the ship to be evaluated into a corresponding new ship type image, and input the new ship type image into the trained ship performance evaluation model. After the model calculation, the corresponding ship hydrostatic data and ship resistance data are output.

[0075] The ship type optimization module 104 is used to compare the ship hydrostatic data and ship resistance data output by the model calculation with their respective predetermined parameters to determine whether the requirements are met. If the requirements are not met, the ship hydrostatic data and ship resistance data output by the model calculation are converted into a variable in a weighted manner, and a genetic algorithm is used to optimize the variable with corresponding hydrostatic constraints to obtain an optimized new ship type image; the optimized new ship type image is input into the ship performance evaluation model again for model calculation, and the output ship hydrostatic data and ship resistance data are compared and judged again. The judgment and optimization are continuously cycled until the requirements are met, and the optimized ship type is output to achieve resistance optimization under hydrostatic constraints.

[0076] Furthermore, the ship hydrostatic data involved in the neural network training module and the neural network training of the ship performance evaluation model include displacement data and center of buoyancy position data; the ship hydrostatic data and ship resistance data evaluation module outputs the corresponding displacement data, center of buoyancy position data and ship resistance data; the ship type optimization module compares the displacement data, center of buoyancy position data and ship resistance data with their respective predetermined parameters to determine whether the requirements are met. If the requirements are not met, the displacement data, center of buoyancy position data and ship resistance data output by the model calculation are converted into a variable in a weighted manner, and a genetic algorithm is used to optimize the variable with respect to the loading capacity constraint conditions and the hull internal layout constraint conditions to obtain an optimized new ship type image; and the optimized new ship type image is input into the ship performance evaluation model again for model calculation, and the output displacement data, center of buoyancy position data and ship resistance data are compared and judged with their respective predetermined parameters again, and the judgment and optimization are continuously cycled until the requirements are met, and the optimized ship type is output to achieve resistance optimization under several hydrostatic constraint conditions.

[0077] Furthermore, in the ship type optimization module, in addition to the genetic algorithm, the simulated annealing algorithm and / or the particle swarm algorithm are also used to optimize the corresponding hydrostatic constraints of the converted variables, and the optimized ship type is evaluated and screened in combination with the ship performance evaluation model.

[0078] It should be noted that the specific embodiments described above can enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although this specification has described the present invention in detail with reference to the drawings and embodiments, those skilled in the art should understand that the present invention can still be modified or replaced with equivalents. In short, all technical solutions and improvements that do not depart from the spirit and scope of the present invention should be included in the scope of protection of the patent for the present invention.

Claims

1. A ship performance evaluation and optimization method based on neural network, characterized in that: The steps include: Ship form image generation step: using 3D modeling technology to generate multiple different 3D geometric ship forms, intercepting multiple cross-sectional views of each of the 3D geometric ship forms, and using the cross-sectional lines in each of the cross-sectional views to generate a plan view, and expressing the areas enclosed by each of the cross-sectional lines in different colors to obtain color cross-sectional views of the ship form corresponding to each of the 3D geometric ship forms; The neural network training model step includes: labeling the ship hydrostatic data and ship resistance data corresponding to each of the three-dimensional geometric ship forms; normalizing and standardizing the ship form color profile and the labeled data, and then using the ship form color profile as input and the ship hydrostatic data and ship resistance data as output to obtain a ship performance evaluation model through neural network training; The model calculates hydrostatic force and resistance: the ship to be evaluated is converted into a corresponding new ship type image, and the new ship type image is input into the trained ship performance evaluation model. After calculation, the model outputs the corresponding ship hydrostatic force data and ship resistance data; Judgment and optimization steps: The ship hydrostatic data and ship resistance data output by the model calculation are compared with their respective predetermined parameters to determine whether they meet the requirements. If the requirements are not met, the ship hydrostatic data and ship resistance data output by the model calculation are converted into a variable in a weighted manner, and the genetic algorithm is used to optimize the variable with corresponding hydrostatic constraints to obtain an optimized new ship type image; the optimized new ship type image is input into the ship performance evaluation model again for model calculation, and the output ship hydrostatic data and ship resistance data are compared and judged again. The judgment and optimization are continuously cycled until the requirements are met, and the optimized ship type is output to achieve resistance optimization under hydrostatic constraints.

2. The method according to claim 1, characterized in that In the neural network training model step, the ship hydrostatic data involved in the annotation and neural network training of the ship performance evaluation model include displacement data and center of buoyancy position data; the model calculates hydrostatic force and resistance step outputs corresponding displacement data, center of buoyancy position data and ship resistance data; the judgment and optimization step compares the displacement data, center of buoyancy position data and ship resistance data with their respective predetermined parameters to determine whether the requirements are met. If the requirements are not met, the displacement data, center of buoyancy position data and ship resistance data output by the model calculation are converted into a variable in a weighted manner, and a genetic algorithm is used to optimize the variable for loading capacity constraints and hull internal layout constraints to obtain an optimized new ship type image; the optimized new ship type image is again input into the ship performance evaluation model for model calculation, and the output displacement data, center of buoyancy position data and ship resistance data are compared and judged again, and the judgment and optimization are continuously cycled until the requirements are met, and the optimized ship type is output to achieve resistance optimization under several hydrostatic constraints.

3. The method according to claim 1, characterized in that In the neural network training model step, the ship hydrostatic data involved in the annotation and neural network training of the ship performance evaluation model include displacement data, buoyancy center position data, drift center position data and stability data; the model calculates hydrostatic force and resistance step outputs corresponding displacement data, buoyancy center position data, drift center position data, stability data and ship resistance data; the judgment and optimization step compares the displacement data, buoyancy center position data, drift center position data, stability data and ship resistance data with their respective predetermined parameters to judge whether the requirements are met. If the requirements are not met, the displacement data, buoyancy center position data, drift center position data, stability data and ship resistance data calculated by the model are compared. The center position data, center of buoyancy position data, stability data and ship resistance data are converted into a variable in a weighted manner. The genetic algorithm is used to optimize the variable under the loading capacity constraints, hull internal layout constraints and navigation safety constraints to obtain an optimized new ship type image. The optimized new ship type image is input into the ship performance evaluation model again for model calculation, and the displacement data, center of buoyancy position data, center of buoyancy position data, stability data and ship resistance data are output for comparison and judgment again. The judgment and optimization are continuously cycled until the requirements are met, and the optimized ship type is output to achieve resistance optimization under several hydrostatic constraints.

4. The method according to any one of claims 1 to 3, characterized in that In the ship type image generation step, the color cross-section images of different three-dimensional geometric ship types have the same size and image format, and the color cross-section images adopt a red, green, and blue (RGB) color mode.

5. The method according to any one of claims 1 to 3, characterized in that In the ship form image generation step, a method of intercepting a plurality of cross-sectional views of each of the three-dimensional geometric ship forms and generating a plan view using cross-sectional lines in each of the cross-sectional views comprises: Take the designated station as the dividing position, and cut the cross section perpendicular to the line connecting the bow and stern to the waterline of the ship as the cross section; The cross-sectional lines in each captured cross-sectional view are used to generate a plan view with the left side as the stern direction and the right side as the bow direction.

6. The method according to any one of claims 1 to 3, characterized in that In the neural network model training step, marking the ship hydrostatic data and ship resistance data corresponding to each of the three-dimensional geometric ship forms includes: Determine the hydrostatic data of the ship corresponding to each of the three-dimensional geometric ship forms using professional software or a self-programming method; Determining ship resistance data corresponding to each of the three-dimensional geometric ship forms using experiments, computational fluid dynamics methods, or potential flow methods; The ship hydrostatic data and ship resistance data corresponding to each of the three-dimensional geometric ship types are marked.

7. The method according to any one of claims 1 to 3, characterized in that In the judgment and optimization steps, in addition to the genetic algorithm, the simulated annealing algorithm and / or the particle swarm algorithm are also used to optimize the corresponding hydrostatic constraints of the converted variables, and the ship performance evaluation model is combined to evaluate and screen the optimized ship type.

8. A ship performance evaluation and optimization system based on neural network, characterized in that: It includes a picture generation module, a neural network training module, a ship hydrostatic data and ship resistance data evaluation module, and a ship type optimization module connected in sequence; among them, The image generation module is used to generate a plurality of different three-dimensional geometric ship shapes using three-dimensional modeling technology, intercept a plurality of cross-sectional views of each of the three-dimensional geometric ship shapes, and generate a plan view using the cross-sectional lines in each of the cross-sectional views, and express the areas enclosed by each of the cross-sectional lines in different colors to obtain a color cross-sectional view of the ship shape corresponding to each of the three-dimensional geometric ship shapes; The neural network training module is used to label the ship hydrostatic data and ship resistance data corresponding to each of the three-dimensional geometric ship forms; then the ship form color cross-section image and the labeled data are normalized and standardized, and then the ship performance evaluation model is obtained through neural network training using the ship form color cross-section image as input and the ship hydrostatic data and ship resistance data as output; The ship hydrostatic data and ship resistance data evaluation module is used to convert the ship to be evaluated into a corresponding new ship type image, and input the new ship type image into a trained ship performance evaluation model. After the model calculates, it outputs the corresponding ship hydrostatic data and ship resistance data; The ship form optimization module is used to compare the ship hydrostatic data and ship resistance data output by the model calculation with their respective predetermined parameters to determine whether the requirements are met. If the requirements are not met, the ship hydrostatic data and ship resistance data output by the model calculation are converted into a variable in a weighted manner, and a genetic algorithm is used to optimize the variable with corresponding hydrostatic constraints to obtain an optimized new ship form image; and the optimized new ship form image is input into the ship performance evaluation model again for model calculation, and the output ship hydrostatic data and ship resistance data are compared and judged again, and the judgment and optimization are continuously cycled until the requirements are met, and the optimized ship form is output to achieve resistance optimization under hydrostatic constraints.

9. The system according to claim 8, characterized in that The ship hydrostatic data involved in the neural network training module and the neural network training of the ship performance evaluation model include displacement data and center of buoyancy position data; the ship hydrostatic data and ship resistance data evaluation module outputs the corresponding displacement data, center of buoyancy position data and ship resistance data; the ship form optimization module compares the displacement data, center of buoyancy position data and ship resistance data with their respective predetermined parameters to determine whether the requirements are met. If the requirements are not met, the displacement data, center of buoyancy position data and ship resistance data output by the model calculation are converted into a variable in a weighted manner, and a genetic algorithm is used to optimize the variable for loading capacity constraints and hull internal layout constraints to obtain an optimized new ship form image; the optimized new ship form image is re-input into the ship performance evaluation model for model calculation, and the output displacement data, center of buoyancy position data and ship resistance data are compared and judged again, and the judgment and optimization are continuously cycled until the requirements are met, and the optimized ship form is output to achieve resistance optimization under several hydrostatic constraints.

10. The system according to claim 8 or 9, characterized in that In the ship type optimization module, in addition to the genetic algorithm, the simulated annealing algorithm and / or the particle swarm algorithm are also used to optimize the corresponding hydrostatic constraints of the converted variables, and the optimized ship type is evaluated and screened in combination with the ship performance evaluation model.

Images