Model construction method and device based on ship type optimization parameter abstract feature representation

Abstract

The invention provides a ship type optimization parameter abstract feature representation-based model construction method and device, and relates to the technical field of three-dimensional modeling, and the method comprises the steps: carrying out the coding extraction of features of a ship type optimization parameter of a first dimension, and obtaining the abstract feature representation of a second dimension, and the second dimension is lower than the first dimension; taking the abstract feature representation as an optimization variable of a ship type optimization problem, and constructing a mathematical model of ship related performance based on the optimization variable; a mathematical model is adopted to replace a numerical simulation tool to participate in the ship form optimization process until an optimization result represented by abstract features is obtained; performing reverse mapping decoding on the optimization result to obtain an actual ship type optimization parameter; and according to the ship type optimization parameters, generating a corresponding optimized ship type through parameterization. According to the method, the modeling efficiency in the high-dimensional parameter space can be improved.

Description

Technical Field

[0001] This application relates to the technical field of three-dimensional modeling, and in particular, to a model construction method and device based on the abstract feature representation of ship form optimization parameters. Background Art

[0002] In the prior art, in the process of ship form optimization, an optimization process centered on numerical simulation is usually adopted, that is, directly using high-dimensional ship form optimization parameters as input variables, and calculating and evaluating the ship performance through computational fluid dynamics simulation, structural analysis or multi-objective evaluation models. In such methods, as the dimension of the optimization parameters continues to increase, the number of samples required to cover the parameter space increases exponentially, resulting in the need for thousands to tens of thousands of high-complexity simulation calculations in a single optimization process, with a huge cumulative simulation time cost, a long optimization cycle, and serious consumption of computing resources. Therefore, there are obvious technical bottlenecks in the modeling efficiency in the high-dimensional parameter space in the prior art, and it is urgent to introduce a technical path with dimensionality reduction ability to construct a low-dimensional but efficient performance prediction model, so as to break through the limitations of the traditional optimization system in large-scale design applications. Summary of the Invention

[0003] This application provides a model construction method and device based on the abstract feature representation of ship form optimization parameters, which can improve the modeling efficiency in the high-dimensional parameter space.

[0004] In the first aspect of this application, a model construction method based on the abstract feature representation of ship form optimization parameters is provided. The method includes:

[0005] Encoding and extracting features from the ship form optimization parameters of the first dimension to obtain an abstract feature representation of the second dimension, where the second dimension is lower than the first dimension;

[0006] Using the abstract feature representation as an optimization variable for the ship form optimization problem, and constructing a mathematical model of the ship-related performance based on the optimization variable;

[0007] Using the mathematical model to replace the numerical simulation tool and participating in the ship form optimization process until an optimization result represented by abstract features is obtained;

[0008] Inversely mapping and decoding the optimization result into actual ship form optimization parameters;

[0009] Generating a corresponding optimized ship form through parameterization according to the ship form optimization parameters.

[0010] Based on the above technical solutions, preferably, the encoding and extracting features from the ship form optimization parameters of the first dimension to obtain an abstract feature representation of the second dimension specifically includes:

[0011] Set the structure of the encoding network, and complete the encoding of the data samples in the first-dimensional parameter space through non-linear mapping and non-linear activation functions. The structural expression of the encoding network is as follows:

[0012] H = g(WX + b m )

[0013] where H is the abstract feature representation, H ∈ R^(k×n), R is the real number field, k is the second dimension, n is the number of samples, g(·) is the non-linear activation function, W is the weight matrix, and b m is the bias vector, X is the original input corresponding to the data samples in the first-dimensional parameter space, X ∈ R^(d×n), k is less than d, and d is the first dimension.

[0014] Based on the above technical solutions, preferably, after encoding and extracting features from the ship shape optimization parameters of the first dimension to obtain the abstract feature representation of the second dimension, the method further includes:

[0015] Introduce a loss function to measure the error between the abstract feature representation and the original input. The loss function includes the mean squared error loss function and the cross-entropy loss function. Among them, the expression of the mean squared error loss function is:

[0016]

[0017] The expression of the cross-entropy loss function is:

[0018]

[0019] where, is the error value, x i is the original input, is the abstract feature representation;

[0020] During the training process, by minimizing the loss function, optimize the weight matrix and the bias vector to make the obtained abstract feature representation close to the essential features of the ship shape optimization parameters of the first dimension in terms of information expression.

[0021] Based on the above technical solutions, preferably, the process of reverse mapping and decoding the optimization result into the actual ship shape optimization parameters specifically includes:

[0022] Complete the reverse mapping from the optimization result to the ship shape optimization parameters through linear mapping to obtain the actual ship shape optimization parameters;

[0023] Train the output weight matrix included in the linear mapping, and aim to minimize the difference between the optimization result and the ship shape optimization parameters to optimize the reverse mapping and decoding process.

[0024] Based on the above technical solutions, preferably, taking the abstract feature representation as the optimization variable of the ship form optimization problem, and based on the optimization variable, constructing a mathematical model of ship-related performance, specifically including:

[0025] Taking the abstract feature representation as the optimization variable, constructing a supervised learning model with the abstract feature representation as the input and the ship performance index as the output;

[0026] Completing the training of the supervised learning model by minimizing the prediction error to form a mathematical model that can be used for rapid performance evaluation.

[0027] Based on the above technical solutions, preferably, using the mathematical model to replace the numerical simulation tool to participate in the ship form optimization process until an optimization result represented by abstract features is obtained, specifically including:

[0028] Using the mathematical model to replace the numerical simulation tool to participate in the optimization solution process. The optimizer uses the abstract feature representation as the input variable, continuously calls the objective function value based on the mathematical model, and completes multiple rounds of evaluation and gradient update of the target performance until an optimal abstract feature solution that meets the preset conditions is obtained to obtain the optimization result.

[0029] Based on the above technical solutions, preferably, according to the ship form optimization parameters, parametrically generating the corresponding optimized ship form, specifically including:

[0030] Obtaining a complete set of ship form optimization parameters;

[0031] Introducing a parametric modeling framework, taking the set of ship form optimization parameters as the input of the geometric generation function, and using interpolation technology to construct continuously differentiable hull outer surface in the longitudinal section, transverse section and waterline direction;

[0032] Synchronously establishing a two-way traceability relationship between parameters and geometric structures during the modeling process;

[0033] Adopting a local modular combination strategy, dividing the hull into a bow part, a main body and a stern part, respectively modeling and then splicing and uniformly controlling to generate a three-dimensional optimized ship form model.

[0034] In the second aspect of the present application, a model construction device based on the abstract feature representation of ship form optimization parameters is provided. The device includes an acquisition module, a processing module and an output module, wherein:

[0035] The acquisition module is used to encode and extract features from the ship form optimization parameters in the first dimension to obtain an abstract feature representation in the second dimension, wherein the second dimension is lower than the first dimension;

[0036] The processing module is configured to use the abstract feature representation as an optimization variable for the ship form optimization problem, and based on the optimization variable, construct a mathematical model of the ship-related performance;

[0037] The processing module is configured to use the mathematical model to replace the numerical simulation tool to participate in the ship form optimization process until an optimization result represented by abstract features is obtained;

[0038] The processing module is configured to inversely map and decode the optimization result into actual ship form optimization parameters;

[0039] The output module is configured to generate a corresponding optimized ship form through parameterization according to the ship form optimization parameters.

[0040] Based on the above technical solution, preferably, the processing module is configured to set the structure of the encoding network, and complete the encoding of the data samples in the first-dimensional parameter space through non-linear mapping and non-linear activation functions. The structural expression of the encoding network is as follows:

[0041] H = g(WX + b m )

[0042] where H is the abstract feature representation, k is the second dimension, n is the number of samples, g(·) is the non-linear activation function, W is the weight matrix, b m is the bias vector, X is the original input corresponding to the data samples in the first-dimensional parameter space, X ∈ R^(d×n), k is less than d, and d is the first dimension.

[0043] Based on the above technical solution, preferably, the processing module is configured to introduce a loss function to measure the error between the abstract feature representation and the original input. The loss function includes a mean squared error loss function and a cross-entropy loss function. Among them, the expression of the mean squared error loss function is:

[0044]

[0045] The expression of the cross-entropy loss function is:

[0046]

[0047] where, is the error value, x i is the original input, is the abstract feature representation;

[0048] The processing module is used to optimize the weight matrix and the bias vector by minimizing the loss function during the training process, so that the obtained abstract feature representation is close to the essential features of the ship type optimization parameters in the first dimension in terms of information expression.

[0049] Based on the above technical solutions, preferably, the processing module is used to complete the inverse mapping from the optimization result to the ship type optimization parameters through linear mapping to obtain the actual ship type optimization parameters;

[0050] The processing module is used to optimize the process of inverse mapping decoding by training the output weight matrix included in the linear mapping with the goal of minimizing the difference between the optimization result and the ship type optimization parameters.

[0051] Based on the above technical solutions, preferably, the processing module is used to construct a supervised learning model with the abstract feature representation as the optimization variable, where the input is the abstract feature representation and the output is the ship performance index;

[0052] The processing module is used to complete the training of the supervised learning model by minimizing the prediction error to form a mathematical model that can be used for rapid performance evaluation.

[0053] Based on the above technical solutions, preferably, the processing module is used to use the mathematical model to replace the numerical simulation tool to participate in the optimization solution process. The optimizer takes the abstract feature representation as the input variable, continuously calls the objective function value based on the mathematical model, and completes multiple rounds of evaluation and gradient update of the target performance until an optimal abstract feature solution that meets the preset conditions is obtained to get the optimization result.

[0054] Based on the above technical solutions, preferably, the acquisition module is used to obtain a complete set of ship type optimization parameters;

[0055] The processing module is used to introduce a parametric modeling framework, take the set of ship type optimization parameters as the input of the geometric generation function, and use interpolation technology to construct a continuously differentiable hull outer surface in the longitudinal section, transverse section and waterline direction;

[0056] The processing module is used to establish a two-way traceability relationship between parameters and geometric structures synchronously during the modeling process;

[0057] The processing module is used to adopt a local modular combination strategy to divide the hull into a bow part, a main body and a stern part, model them separately and then splice and control them uniformly to generate a three-dimensional optimized ship type model.

[0058] In a third aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions. Both the user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory, so that the electronic device executes the method described in any one of the above.

[0059] In a fourth aspect of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores instructions, and when the instructions are executed, the method described in any one of the above is executed.

[0060] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0061] 1. In the present application, a non-linear coding mechanism is introduced in the high-dimensional ship form optimization parameter space to extract low-dimensional abstract feature representations, and a supervised learning mathematical model is constructed with these abstract features as optimization variables to replace numerical simulation tools in the performance prediction and optimization process, significantly reducing the input variable dimension and modeling complexity. Compared with the traditional method of directly modeling in the original high-dimensional space, the solution realizes a reduction in the required number of modeling samples, an acceleration in the model training convergence speed, and an improvement in the prediction calculation efficiency while maintaining the integrity of the information structure, thus significantly improving the modeling efficiency in the high-dimensional parameter space.

[0062] 2. In the present application, a loss function is introduced to quantitatively evaluate the information reconstruction error between the abstract feature representation and the original input, and the weight matrix and bias vector in the coding network are optimized with the goal of minimizing this error, so as to ensure that the extracted abstract feature representation maximally retains the essential structural information of the original ship form optimization parameters while achieving dimension compression. This mechanism effectively improves the representation ability and reversibility of the abstract features, provides a high-fidelity feature basis for subsequent construction of a performance prediction model and accurate decoding of the optimization solution, and thus enhances the stability, generalization ability, and reconstruction accuracy of the entire modeling system.

[0063] 3. In the present application, a reverse decoding mechanism based on linear mapping is constructed to accurately restore the optimization result of the low-dimensional abstract feature representation to the high-dimensional ship form optimization parameters, and the output weight matrix is trained to minimize the difference between the decoding result and the true ship form optimization parameters, achieving an optimal approximation of the decoding process. It ensures the reducibility and design feasibility of the optimization result, effectively improves the accuracy and stability of the mapping from the feature space to the engineering parameter space, constructs a reliable bridge from the optimization calculation to the generation of actual ship form parameters, and thus guarantees the direct application ability of the model optimization result in actual ship form design. Description of the Drawings

[0064] Figure 1It is a schematic flowchart of a model construction method based on the abstract feature representation of ship form optimization parameters disclosed in an embodiment of the present application;

[0065] Figure 2 It is a schematic diagram of the extraction and reconstruction process of the abstract feature representation of ship form optimization parameters disclosed in an embodiment of the present application.

[0066] Figure 3 It is a schematic module diagram of a model construction device based on the abstract feature representation of ship form optimization parameters disclosed in an embodiment of the present application;

[0067] Figure 4 It is a schematic structural diagram of an electronic device disclosed in an embodiment of the present application.

[0068] Explanation of reference numerals: 301, acquisition module; 302, processing module; 303, output module; 401, processor; 402, communication bus; 403, user interface; 404, network interface; 405, memory. Detailed implementation manners

[0069] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.

[0070] In the description of the embodiments of the present application, words such as "for example" or "for illustration" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "for example" or "for illustration" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of words such as "for example" or "for illustration" is intended to present relevant concepts in a specific manner.

[0071] In the description of the embodiments of the present application, the meaning of the term "a plurality" refers to two or more. For example, a plurality of systems refers to two or more systems, and a plurality of screen terminals refers to two or more screen terminals. In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly indicating the technical features indicated. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. The terms "include", "comprise", "have" and their variants all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0072] This embodiment discloses a model construction method based on the abstract feature representation of ship form optimization parameters. Referring to Figure 1 , it includes the following steps S110 - S150:

[0073] S110, encode and extract features from the ship shape optimization parameters in the first dimension to obtain the abstract feature representation in the second dimension.

[0074] A model construction method based on the abstract feature representation of ship shape optimization parameters disclosed in the embodiments of the present application is applied to a server. The server includes, but is not limited to, electronic devices such as mobile phones, tablet computers, wearable devices, and PCs (Personal Computers), or can also be a background server running a model construction method based on the abstract feature representation of ship shape optimization parameters. The server can be implemented by an independent server or a server cluster composed of multiple servers.

[0075] First, set the structure form of the encoding network, and clarify that the input is the original input sample matrix X ∈ R d×n , where R is the real number field, d is the dimension of the ship shape optimization parameters in the first dimension, and n is the number of samples. The set encoding network consists of a weight matrix W ∈ R k×d and a bias vector b m ∈ R k×1 to perform a linear transformation and a non-linear activation process on each input sample, and output the abstract feature representation H ∈ R k×n , where k < d is the dimension of the abstract feature.

[0076] The non-linear mapping process adopts the following expression:

[0077] H = g(WX + b m )

[0078] where g(·) is a non-linear activation function, and common function forms include the Sigmoid function, the Tanh function, or the ReLU function, etc., which are used to introduce non-linear capabilities to enhance the fitting ability of the model to complex structures. In specific implementation, first randomly initialize the weight matrix W and the bias vector b m , then perform matrix multiplication WX on the input sample X one by one, add the bias term b m , and then apply the activation function g(·) to perform element-wise non-linear transformation on the result, thereby generating the corresponding abstract feature representation H.

[0079] The goal of this encoding network is to compress high-dimensional inputs into low-dimensional representations while minimizing information loss, preserving the main variation trends and key feature structures among the input parameters, which serves as the input basis for subsequent optimization modeling and predictive analysis. During the training process, through joint optimization with the decoding network and the reconstruction error loss function, it is ensured that the extracted abstract feature representations have good representativeness and reversibility, can effectively reduce the dimensional complexity of the ship form optimization parameter space while maintaining the original structural information, thus achieving the dual goals of parameter compression and improved optimization efficiency.

[0080] Further, first, after encoding and extracting the first-dimensional ship form optimization parameters and obtaining the abstract feature representation H = g(WX + b m ), a corresponding decoding network is constructed to achieve the reconstruction of the original input, that is, mapping the abstract feature representation H back to the high-dimensional parameter space through a symmetric structure to generate an approximate reconstruction value for measuring information loss.

[0081] To quantitatively evaluate the ability of the abstract feature representation to retain the original input, a loss function is introduced for calculating the reconstruction error of each group of samples. In actual training, one of the following two typical loss functions can be selected according to the data type and distribution characteristics: One is the mean squared error loss function, which is applicable to continuous variable inputs, and its expression is:

[0082]

[0083] where, x i represents the i-th original input sample, is its corresponding reconstructed sample, and this loss function measures the reconstruction accuracy in the Euclidean sense. The other is the cross-entropy loss function, which is applicable to scenarios where the input data has a probability distribution or discrete binary features, and its expression is:

[0084]

[0085] This loss function measures the deviation degree between two probability distributions from the perspective of information entropy. During the training process, minimizing the above loss function is used as the optimization goal, and the weight matrix W and bias vector b in the encoding network are continuously updated through the backpropagation algorithm m , thereby optimizing the parameter structure of the network. In each round of iteration, based on the reconstructed value output by the current network, the loss function is recalculated, and the gradient is corrected and updated until the loss function converges to the minimum value range, indicating that the abstract feature representation can fully express the essential feature information of the original input parameters.

[0086] Finally, the encoding network structure optimized by minimizing the loss function can stably output a low-dimensional abstract feature representation H, and ensure that the abstract feature has high fidelity to the first-dimensional ship form optimization parameters during the mapping reduction process, providing a usable data basis and mathematical guarantee for subsequent optimization modeling and reverse decoding based on the abstract feature.

[0087] S120. Use the abstract feature representation as an optimization variable for the ship form optimization problem, and based on the optimization variable, construct a mathematical model of ship-related performance.

[0088] In a possible implementation, use the abstract feature representation as an optimization variable for the ship form optimization problem, and based on the optimization variable, construct a mathematical model of ship-related performance, specifically including: using the abstract feature representation as an optimization variable to construct a supervised learning model with the abstract feature representation as the input and the ship performance index as the output; training the supervised learning model by minimizing the prediction error to form a mathematical model that can be used for rapid performance evaluation.

[0089] Specifically, first, the abstract feature representation needs to be used as a new optimization variable for performance modeling. This step uses the abstract feature representation H∈R output by the encoding network k×n as the input variable of the supervised learning model, and at the same time uses the ship performance index Y∈R of the corresponding sample m×n as the target output of the supervised learning model, where k is the abstract feature dimension, m is the performance index dimension, and n is the number of samples. Through this input-output pairing method, a supervised learning data set with the abstract feature representation as the input and the ship performance index as the output is constructed to establish the mapping relationship between the abstract feature space and the performance response space.

[0090] Subsequently, construct a supervised learning model to learn this mapping relationship. The model structure can select prediction models with non-linear expression capabilities such as multi-layer perceptron, support vector regression, radial basis neural network, or extreme learning machine. Using the input feature H as the independent variable and the performance index Y as the dependent variable, train the model to approximate the function f:R k →R m . In model construction, technical parameters such as network topology structure, activation function, initialization strategy, and regularization means need to be set to make the model have good generalization ability and adapt to the non-linear structural characteristics implied by the abstract features.

[0091] In the model training stage, optimize the model parameters by minimizing the error function between the predicted output and the true performance index. The loss function often selects the mean square error form, and its expression is:

[0092]

[0093] where θ is the parameter set of the supervised learning model, f(H i; θ) is the prediction result of the model for the i-th input sample, Y i is the corresponding target performance metric. Through optimization algorithms such as backpropagation or the least squares method, the parameters are iteratively adjusted to converge the loss function to the minimum value, ultimately forming a mathematical model that can quickly respond to input abstract features and output high-precision performance prediction results.

[0094] After training, the supervised learning model has the ability to predict ship performance in the low-dimensional abstract feature space and can be directly embedded into the ship form optimization process as an objective function or a constraint condition for quick call, significantly replacing the high-cost operations of traditional numerical simulation tools in performance evaluation, thereby achieving a magnitude improvement in optimization efficiency.

[0095] S130, use the mathematical model to replace the numerical simulation tool and participate in the ship form optimization process until an optimization result represented by abstract features is obtained.

[0096] In a possible implementation, using the mathematical model to replace the numerical simulation tool and participate in the ship form optimization process until an optimization result represented by abstract features is obtained, specifically including: the mathematical model replaces the numerical simulation tool and participates in the optimization solution process. The optimizer uses the abstract feature representation as the input variable, continuously calls the objective function value based on the mathematical model, completes multiple rounds of evaluation and gradient update of the target performance until an optimal abstract feature solution that meets the preset conditions is obtained, and the optimization result is obtained.

[0097] Specifically, first, set the optimizer structure, select an optimization algorithm suitable for global search in the continuous space, such as the particle swarm optimization algorithm, genetic algorithm, differential evolution algorithm, or gradient-enhanced optimization algorithm, and use the abstract feature representation H ∈ R k as the input variable of the optimizer, where k is the dimension of the abstract feature, and this feature representation is output by the encoding network and has a compressed global information structure.

[0098] In the optimizer initialization stage, generate a set of initial sample points of abstract features as the optimization starting point. Subsequently, based on the aforementioned trained supervised learning model, input each sample point H i into the mathematical model f(H i ) to predict the corresponding ship performance metric, and define the objective function value according to the optimization goal. If the goal is to minimize the total ship resistance or maximize the propulsion efficiency, the objective function can be set as:

[0099]

[0100] where f(H) is the performance metric output by the mathematical model, The optimization objective is mapped to the indicator. In each optimization iteration, the optimizer performs operations such as sample selection, update, crossover, or gradient correction based on the objective function value of the current abstract feature variable to generate a new generation of abstract feature candidate solutions.

[0101] This optimization process continuously calls upon mathematical models to complete multiple rounds of objective function evaluation. Compared to the traditional method of using numerical simulation tools to gradually evaluate each ship design, the model's computational overhead is significantly reduced, allowing thousands or even tens of thousands of performance evaluations to be completed in a relatively short period of time. It also supports multi-objective collaborative optimization or solving constrained optimization problems. After each round of iteration, the optimizer determines the convergence of the current optimal objective function value and the amplitude of variable fluctuations. The iteration process terminates when convergence conditions are met, such as when the objective function change is less than a preset threshold, the maximum number of iterations is reached, or when fluctuations in the solution space stabilize.

[0102] The resulting abstract feature representation is the optimal abstract feature solution, representing the optimal performance solution achievable under the current mathematical model's predictions. This abstract feature solution is then passed to the subsequent decoding module as the optimization result, where it is converted into the actual ship optimization parameters. This entire process, through the complete replacement of simulation functions with mathematical models, constructs an efficient optimization framework using abstract features as input and a supervised learning model as its computational core. This approach achieves a solution path that balances precision and efficiency for complex multidimensional ship optimization problems in a low-dimensional feature space.

[0103] S140, reverse mapping and decoding the optimization results into actual ship type optimization parameters.

[0104] In one possible implementation, the optimization results are reverse mapped and decoded into actual ship type optimization parameters, specifically including: completing the reverse mapping from the optimization results to the ship type optimization parameters through linear mapping to obtain the actual ship type optimization parameters; and optimizing the reverse mapping decoding process by training the output weight matrix contained in the linear mapping with the goal of minimizing the difference between the optimization results and the ship type optimization parameters.

[0105] Specifically, first, the decoding structure is set as a set of linear mapping relationships, that is, the final optimization result is set as the abstract feature representation matrix H∈R k×n The goal is to restore the original ship type optimization parameter matrix Where k is the abstract feature dimension, d is the dimension of the first dimension ship type optimization parameter, and n is the number of samples. The linear mapping expression is defined as:

[0106]

[0107] Among them, β∈R k×dis the output weight matrix, which controls the mapping direction and amplitude from the abstract features to the high-dimensional parameters. To ensure that this linear mapping has a high reconstruction accuracy, it is necessary to train and optimize the output weight matrix β. The training objective is to make the reconstructed parameter as close as possible to the original ship form optimization parameter X, and the least square error is introduced as the form of the loss function:

[0108]

[0109] By solving the minimum value of this loss function, the optimal solution of the output weight matrix is obtained. When the matrix H is invertible or can be generalized inversed, the Moore-Penrose generalized inverse method is used to calculate the weight matrix, and the solution is:

[0110]

[0111] where is the generalized inverse of the abstract feature representation matrix, and the singular value decomposition method can be used to solve it stably. To further enhance the generalization ability and decoding robustness of the reverse mapping, the loss function in the form of the extreme learning machine autoencoder regularization is introduced:

[0112]

[0113] where C is the regularization coefficient, which is used to balance the optimization objective between the norm of the output weight and the reconstruction error. In the actual solution, according to the relationship between the number of samples and the feature dimension, the following closed-form solutions are respectively adopted:

[0114]

[0115] Once this weight matrix is obtained, it can be used to decode any optimized abstract feature representation, and the corresponding actual ship form optimization parameter is calculated through to ensure that the decoding result has engineering feasibility in terms of structural consistency and parameter accuracy. The finally output actual ship form optimization parameter can be used as the input of the ship form parametric modeling module to drive the subsequent hull 3D modeling and performance verification, and realize the effective transition from the solution in the feature space to the solution in the physical design space. This implementation process has the characteristics of simple structure, stable convergence and high computational efficiency, and is the key technical path to realize the reversibility and accuracy control of the decoding process.

[0116] S150. According to the ship form optimization parameter, the corresponding optimized ship form is generated parametrically.

[0117] In a possible implementation, according to the ship form optimization parameters, the corresponding optimized ship form is generated parametrically, specifically including: obtaining a complete set of ship form optimization parameters; introducing a parametric modeling framework, taking the set of ship form optimization parameters as the input of the geometric generation function, and using interpolation techniques to construct a continuously differentiable hull outer surface in the longitudinal section, transverse section and waterline direction; establishing a two-way traceability relationship between parameters and geometric structures synchronously during the modeling process; adopting a local modular combination strategy, dividing the hull into a bow part, a main body and a stern part, modeling them separately and then splicing and uniformly controlling to generate a three-dimensional optimized ship form model.

[0118] Specifically, first, after completing the abstract feature decoding, a complete set of ship form optimization parameters is obtained. This parameter set is a numerical vector with a dimension of d, containing the core variables that control the hull outer shape, such as main dimension parameters, section control point coordinates, waterline curvature coefficients, volume coefficients, bilge line positions, bow and stern shape parameters, etc. This parameter set serves as the input basis for subsequent hull geometric modeling, has clear structural semantics and geometric constraint meanings, and is the basic data carrier for realizing the transition from the optimization space to the three-dimensional ship form space.

[0119] Subsequently, a parametric modeling framework is introduced to achieve the conversion from optimization parameters to geometric surfaces. Specifically, geometric generation functions are used to perform interpolation modeling on the longitudinal section, transverse section and waterline direction. The optimization parameters are used as function inputs to control the node positions, curvature changes and boundary shapes, generating a set of surfaces required in the three-dimensional hull structure. Commonly used interpolation techniques include B-spline curves, NURBS surfaces or Bezier methods, combined with C1 or C2 continuity constraints to achieve smooth transitions of the surfaces, ensuring that the hull surface is continuously differentiable and meets the requirements of streamline and manufacturability, thereby establishing a parametrically driven spatial configuration modeling mechanism.

[0120] During the modeling process, a two-way traceability relationship between parameters and geometric structures is established synchronously. By recording the geometric influence areas and control variable paths corresponding to each parameter dimension, a parameter-to-geometric mapping table and a geometric-to-parameter reverse lookup channel are constructed to support subsequent sensitivity analysis, variable regulation and rapid correction operations. This traceability mechanism can achieve transparent control of the structure, facilitate accurately locking the affected structure areas in performance feedback, optimization iteration or local modification scenarios, and improve the regulation efficiency and result interpretability of ship form modeling.

[0121] Finally, to enhance the modeling flexibility and structural control accuracy, a local modular combination strategy is introduced. The hull is divided into structural units such as the bow, the main body, and the stern, and local parameter control models are constructed respectively. Each module independently performs parameter analysis, geometric construction, and surface splicing, and continuity constraint conditions are imposed at the boundaries to ensure the smooth integration of the overall structure. By uniformly controlling the connection points and transition curvatures between the modules, the modeling freedom and local optimization ability can be improved while ensuring the integrity of the modeling. Finally, the three-dimensional optimized ship form model assembled by each module can be output in the standard CAD format for numerical verification, rapid prototyping, or engineering deployment, realizing the efficient conversion from parameter results to physical ship forms.

[0122] Referring to Figure 2 , the figure shows a schematic diagram of the extraction and reconstruction process of the abstract feature representation of ship form optimization parameters disclosed in the embodiment of the present application. The process includes three functional regions: the original input data input region, the feature extraction layer by layer region, the feature reconstruction layer by layer region, and the output reconstruction region. The leftmost is the input of the original data X ∈ R d×n , representing a ship form optimization parameter matrix with dimension d, which contains a total of n samples. This input is passed through the input layer to the multi-layer feature extraction structure on the right to form an encoding network.

[0123] The feature extraction layer by layer region consists of multiple non-linear hidden layers. Starting from the first layer, the number of nodes in each layer gradually decreases, and finally converges to the minimum abstract feature representation with dimension p. This compression process is driven by a series of weight matrices and non-linear activation functions, and the feature representations from the original high-dimensional space to the low-dimensional abstract space are extracted layer by layer. The intermediate layers marked as K1, K2, and K3 from left to right represent the multi-level compression process of the abstract features, and the output is the core abstract feature representation H ∈ R p×n , where p < d.

[0124] The feature reconstruction layer by layer region is a decoding network, whose structure is symmetric to the encoding network, and the number of nodes gradually increases, simulating the inverse mapping process from the low-dimensional abstract feature space to the high-dimensional parameter space. This reconstruction process depends on a set of independent weight matrices and activation functions, and its goal is to recover an approximate reconstruction value as close as possible to the original parameters from the abstract feature representation. Finally, the output reconstruction region on the right outputs the reconstructed data with dimension d for comparison with the original input X.

[0125] The entire network minimizes the input X and the output during the training phase Taking the loss function between them as the objective, the mean squared error loss function or the cross-entropy loss function is often adopted. All encoding weight matrices and decoding weight matrices are optimized through the backpropagation algorithm, so as to ensure that the abstract feature representation has both information compression ability and good information fidelity. This structure is used in ship form optimization to perform dimensionality reduction encoding on ship form optimization parameters and support subsequent inverse mapping decoding, and is the core neural network framework for realizing the integrated process of feature extraction, approximate modeling, optimization decoding and 3D modeling.

[0126] This embodiment also discloses a model construction device based on the abstract feature representation of ship form optimization parameters. Referring to Figure 3 , it includes an acquisition module 301, a processing module 302 and an output module 303, where:

[0127] The acquisition module 301 is configured to encode and extract features from the ship form optimization parameters in the first dimension to obtain an abstract feature representation in the second dimension, where the second dimension is lower than the first dimension.

[0128] The processing module 302 is configured to use the abstract feature representation as an optimization variable for the ship form optimization problem, and construct a mathematical model of the ship-related performance based on the optimization variable.

[0129] The processing module 302 is configured to use the mathematical model to replace the numerical simulation tool to participate in the ship form optimization process until an optimization result represented by the abstract feature is obtained.

[0130] The processing module 302 is configured to inverse map and decode the optimization result into actual ship form optimization parameters.

[0131] The output module 303 is configured to generate a corresponding optimized ship form through parameterization according to the ship form optimization parameters.

[0132] In a possible implementation manner, the processing module 302 is configured to set the structure of the encoding network, and complete the encoding of the data samples in the first-dimensional parameter space through non-linear mapping and non-linear activation functions. The structural expression of the encoding network is as follows:

[0133] H = g(WX + b m )

[0134] where H is the abstract feature representation, k is the second dimension, n is the number of samples, g(·) is the non-linear activation function, W is the weight matrix, b m is the bias vector, X is the original input corresponding to the data sample in the first-dimensional parameter space, X ∈ R^(d×n), k is less than d, and d is the first dimension.

[0135] In a possible implementation, the processing module 302 is configured to introduce a loss function to measure the error between the abstract feature representation and the original input. The loss function includes a mean squared error loss function and a cross-entropy loss function. The expression of the mean squared error loss function is as follows:

[0136]

[0137] The expression of the cross-entropy loss function is as follows:

[0138]

[0139] Where is the error value, x i is the original input, is the abstract feature representation.

[0140] The processing module 302 is configured to optimize the weight matrix and the bias vector by minimizing the loss function during the training process, so that the obtained abstract feature representation is close to the essential features of the ship form optimization parameters in the first dimension in terms of information expression.

[0141] In a possible implementation, the processing module 302 is configured to perform an inverse mapping from the optimization result to the ship form optimization parameters through a linear mapping to obtain the actual ship form optimization parameters.

[0142] The processing module 302 is configured to optimize the inverse mapping decoding process by training the output weight matrix included in the linear mapping with the goal of minimizing the difference between the optimization result and the ship form optimization parameters.

[0143] In a possible implementation, the processing module 302 is configured to construct a supervised learning model with the abstract feature representation as the optimization variable and the ship performance index as the output, where the input is the abstract feature representation.

[0144] The processing module 302 is configured to complete the training of the supervised learning model by minimizing the prediction error to form a mathematical model that can be used for rapid performance evaluation.

[0145] In a possible implementation, the processing module 302 is configured to use the mathematical model to replace the numerical simulation tool to participate in the optimization solution process. The optimizer uses the abstract feature representation as the input variable, continuously calls the objective function value based on the mathematical model, and completes multiple rounds of evaluation and gradient update of the target performance until an optimal abstract feature solution that meets the preset conditions is obtained to get the optimization result.

[0146] In a possible implementation, the acquisition module 301 is configured to obtain a complete set of ship form optimization parameters.

[0147] The processing module 302 is configured to introduce a parametric modeling framework, use the set of ship form optimization parameters as the input of the geometric generation function, and adopt an interpolation technique to construct a continuously differentiable hull outer surface in the longitudinal section, transverse section, and waterline directions.

[0148] The processing module 302 is configured to synchronously establish a two-way traceability relationship between parameters and geometric structures during the modeling process.

[0149] The processing module 302 is configured to adopt a local modular combination strategy, divide the hull into a bow part, a main body, and a stern part, model them separately and then splice and uniformly control them to generate a three-dimensional optimized ship form model.

[0150] It should be noted that when the device provided in the above embodiments realizes its functions, only the division of the above functional modules is used for illustration. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be seen in the method embodiments, which will not be elaborated here.

[0151] This embodiment also discloses an electronic device. Referring to Figure 4 , the electronic device may include: at least one processor 401, at least one communication bus 402, a user interface 403, a network interface 404, and at least one memory 405.

[0152] Among them, the communication bus 402 is used to realize the connection and communication between these components.

[0153] Among them, the user interface 403 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 403 may further include a standard wired interface and a wireless interface.

[0154] Among them, the network interface 404 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0155] Among them, the processor 401 may include one or more processing cores. The processor 401 connects various parts within the entire server through various interfaces and lines. By running or executing instructions, programs, code sets, or instruction sets stored in the memory 405, and by calling the data stored in the memory 405, it performs various functions of the server and processes data. Optionally, the processor 401 may be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). The processor 401 may integrate a combination of one or several of a central processing unit (CPU), a graphics processing unit (GPU), and a modem, etc. Among them, the CPU mainly processes the operating system, user interface, application programs, etc.; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 401 and may be implemented separately through a single chip.

[0156] Among them, the memory 405 may include random access memory (RAM) and may also include read-only memory. Optionally, the memory includes a non-transitory computer-readable storage medium. The memory 405 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 405 may include a program storage area and a data storage area. Among them, the program storage area can store instructions for implementing the operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area can store the data involved in the above-mentioned various method embodiments. Optionally, the memory 405 may also be at least one storage device located far from the aforementioned processor 401. The memory 405, as a computer storage medium, may include an operating system, a network communication module, a user interface 403 module, and an application program for a model construction method based on the abstract features of ship type optimization parameters.

[0157] In Figure 4In the electronic device shown, the user interface 403 is mainly used to provide an interface for the user to input and obtain the data input by the user; while the processor 401 can be used to call an application program stored in the memory 405, which is a model construction method based on the abstract features of ship-shaped optimization parameters. When executed by one or more processors 401, the electronic device is enabled to execute the method of one or more of the above embodiments.

[0158] It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0159] In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0160] In several embodiments provided by this application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some service interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical or other form.

[0161] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0162] In addition, in each embodiment of this application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0163] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on such understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory 405 and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of various embodiments of this application. And the aforementioned memory 405 includes: various media such as USB flash drives, mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0164] This application also discloses a computer-readable storage medium, and the computer-readable storage medium stores instructions. When executed by one or more processors 401, it causes the electronic device to execute the method as described in one or more of the above embodiments.

[0165] The above are only exemplary embodiments of the present disclosure and should not be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made in accordance with the teachings of the present disclosure still fall within the scope covered by the present disclosure. After considering the specification and the disclosure of the practical truth, those skilled in the art will easily think of other implementation schemes of the present disclosure. This application aims to cover any variations, uses, or adaptive changes of the present disclosure, and these variations, uses, or adaptive changes follow the general principles of the present disclosure and include the common general knowledge or conventional technical means in the technical field not recorded in the present disclosure. The specification and the embodiments are only regarded as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A model construction method based on the abstract feature representation of ship form optimization parameters, characterized in that, The method includes: Encoding and extracting features from the ship form optimization parameters in the first dimension to obtain an abstract feature representation in the second dimension, where the second dimension is lower than the first dimension; Taking the abstract feature representation as an optimization variable for the ship form optimization problem, and based on the optimization variable, constructing a mathematical model of ship-related performance; Using the mathematical model to replace the numerical simulation tool and participating in the ship form optimization process until an optimization result represented by abstract features is obtained; Inverse mapping and decoding the optimization result into actual ship form optimization parameters; According to the ship form optimization parameters, parametrically generating a corresponding optimized ship form.

2. The model construction method based on the abstract feature representation of ship type optimization parameters according to claim 1, wherein, The encoding and extracting features from the ship form optimization parameters in the first dimension to obtain an abstract feature representation in the second dimension specifically includes: Setting the structure of the encoding network, and completing the encoding of data samples in the parameter space of the first dimension through non-linear mapping and non-linear activation functions. The structural expression of the encoding network is as follows: H = g(WX + b m ) Among them, H is the abstract feature representation, H ∈ R^(k×n), R is the real number field, k is the second dimension, n is the number of samples, g(·) is a non-linear activation function, W is a weight matrix, and b m is a bias vector, X is the original input corresponding to the data sample in the first-dimensional parameter space, X ∈ R^(d×n), k is less than d, and d is the first dimension.

3. The model construction method based on the abstract feature representation of ship type optimization parameters according to claim 2, characterized in that, After encoding and extracting features from the ship form optimization parameters in the first dimension to obtain an abstract feature representation in the second dimension, the method further includes: Introducing a loss function to measure the error between the abstract feature representation and the original input. The loss function includes a mean squared error loss function and a cross-entropy loss function. Among them, the expression of the mean squared error loss function is: The expression of the cross-entropy loss function is: wherein, is the error value, and x i is the original input, is the abstract feature representation; During the training process, by minimizing the loss function, optimizing the weight matrix and the bias vector, so that the obtained abstract feature representation is close to the essential features of the ship form optimization parameters in the first dimension in terms of information expression.

4. A model construction method based on abstract feature representation of ship type optimization parameters according to claim 1, characterized in that The inverse mapping and decoding the optimization result into actual ship form optimization parameters specifically includes: Completing the inverse mapping from the optimization result to the ship form optimization parameters through linear mapping to obtain the actual ship form optimization parameters; By training the output weight matrix included in the linear mapping, with the goal of minimizing the difference between the optimization result and the ship form optimization parameters, optimizing the inverse mapping and decoding process.

5. The model construction method based on the abstract feature representation of ship type optimization parameters according to claim 1, characterized in that Taking the abstract feature representation as an optimization variable for the ship form optimization problem, and based on the optimization variable, constructing a mathematical model of ship-related performance specifically includes: Taking the abstract feature representation as the optimization variable, and constructing a supervised learning model with the abstract feature representation as the input and the ship performance index as the output; Completing the training of the supervised learning model by minimizing the prediction error to form a mathematical model that can be used for rapid performance evaluation.

6. The model construction method based on the abstract feature representation of ship type optimization parameters according to claim 1, wherein, Using the mathematical model to replace the numerical simulation tool and participating in the ship form optimization process until an optimization result represented by abstract features is obtained specifically includes: Using the mathematical model to replace the numerical simulation tool and participating in the optimization solution process. The optimizer takes the abstract feature representation as the input variable, and continuously calls the objective function value based on the mathematical model to complete multiple rounds of evaluation and gradient update of the target performance until an optimal abstract feature solution that meets the preset conditions is obtained, and the optimization result is obtained.

7. A method for constructing a model based on the abstract feature representation of ship type optimization parameters according to claim 1, characterized in that, According to the ship form optimization parameters, parametrically generating a corresponding optimized ship form specifically includes: Obtaining a complete set of ship form optimization parameters; Introduce a parametric modeling framework, take the set of ship form optimization parameters as the input of the geometric generation function, and use interpolation technology to construct a continuously differentiable hull surface in the longitudinal section, transverse section and waterline direction; Establish a two-way traceability relationship between parameters and geometric structures synchronously during the modeling process; Adopt a local modular combination strategy, divide the hull into the bow, the main body and the stern, model them separately and then splice and control them uniformly to generate a three-dimensional optimized ship form model.

8. A model construction device based on the abstract feature representation of ship type optimization parameters, characterized in that, The device is used to execute a model construction method based on the abstract feature representation of ship form optimization parameters as described in any one of claims 1-7. The device includes an acquisition module (301), a processing module (302) and an output module (303), wherein: The acquisition module (301) is used to encode and extract features from the ship form optimization parameters in the first dimension to obtain an abstract feature representation in the second dimension, where the second dimension is lower than the first dimension; The processing module (302) is used to use the abstract feature representation as an optimization variable for the ship form optimization problem, and based on the optimization variable, construct a mathematical model of the ship-related performance; The processing module (302) is used to use the mathematical model to replace the numerical simulation tool and participate in the ship form optimization process until an optimization result represented by abstract features is obtained; The processing module (302) is used to inversely map and decode the optimization result into actual ship form optimization parameters; The output module (302) is used to generate a corresponding optimized ship form by parameterization according to the ship form optimization parameters.

9. An electronic device, characterized in that, It includes a processor (401), a communication bus (402), a user interface (403), a network interface (404) and a memory (405). The memory (405) is used to store instructions. The user interface (403) and the network interface (404) are both used to communicate with other devices. The communication bus (402) is used to realize the connection and communication between components in the electronic device. The processor (401) is used to execute the instructions stored in the memory (405) so that the electronic device executes the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions, and when the instructions are executed, the method as described in any one of claims 1-7 is executed.

Images