Intelligent forecasting method and system for ship control hydrodynamic derivative

Abstract

The invention belongs to the technical field of ship maneuvering performance prediction, and discloses a ship maneuvering hydrodynamic derivative intelligent forecasting method and system. The method comprises the following steps: compressing a plurality of extracted ship parameters such as a ship three-dimensional geometrical shape, a ship control movement speed and an angular speed into low-dimensional representation data by using a defined auto-encoder, and constructing a physical information neural network model based on the preprocessed compressed low-dimensional representation data; introducing an improved total loss function, and optimizing the physical information neural network model; based on the physical information neural network model with the improved total loss function introduced, a combined training strategy of an optimization algorithm is adopted for further optimization, an optimized physical information neural network model is obtained, and verification and iterative optimization are conducted on the optimized physical information neural network model PINN model. According to the method, high prediction precision can be kept under different working conditions, and good adaptability is shown in complex sea conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of ship maneuverability prediction, and in particular relates to a method and system for intelligent prediction of ship maneuverability hydrodynamic derivatives. Background Art

[0002] Ship maneuverability is of great significance to navigation safety, fuel economy and environmental impact. Accurately predicting the ship's maneuvering hydrodynamic derivatives is a key link in achieving efficient ship design and optimizing maneuverability. Traditional ship hydrodynamic derivative prediction methods mainly rely on model tests, empirical formulas and computational fluid dynamics (CFD) simulations. Although these methods have been widely used in the industry, they still have some inherent defects.

[0003] First, model tests are costly and time-consuming, and it is difficult to quickly adapt to the needs of different ship types and operating conditions. Second, empirical formulas are usually based on statistical data of specific ship types and operating conditions, and the prediction accuracy is insufficient for non-standard ship types or complex sea conditions. In addition, although CFD simulation can provide detailed flow field information, it consumes a lot of computing resources and has a long calculation cycle, making it difficult to be widely used in actual engineering.

[0004] In order to solve these problems, data-driven intelligent forecasting methods have begun to attract attention in recent years. In particular, deep learning technology has been gradually introduced into the field of ship hydrodynamics due to its powerful capabilities in pattern recognition and nonlinear function fitting. Deep learning models can quickly predict ship hydrodynamic derivatives with good prediction accuracy through training with a large amount of data. However, such models usually rely on a large amount of high-quality training data, and may have insufficient generalization capabilities when faced with new or unseen working conditions.

[0005] In order to improve the generalization ability of the model, the physical information neural network (PINN) introduces physical constraints, combines physical laws with data-driven methods, and avoids the limitations of traditional deep learning models that rely purely on data. By adding physical equation constraints to the loss function of the neural network, PINN makes the model conform to physical laws during the learning process and can flexibly adapt to complex nonlinear relationships. Therefore, the PINN method shows good results in dealing with problems involving complex physical processes, and is particularly suitable for the prediction of hydrodynamic derivatives of ship maneuvers.

[0006] Although PINN technology has significant advantages, its application in the field of ships still faces some challenges. For example, the processing efficiency of high-dimensional input data is a key issue. For ship maneuvering problems, the input data often contains a large number of ship parameters and navigation conditions, which brings a large computational burden to the training of the model. In addition, how to effectively combine physical information with data-driven models to improve the stability and prediction accuracy of the model is also a hot topic and difficulty in current research. Summary of the invention

[0007] In order to overcome the problems existing in the related technologies, the disclosed embodiments of the present invention provide a method and system for intelligent prediction of hydrodynamic derivatives of ship maneuvering. Specifically, it relates to an intelligent prediction method of hydrodynamic derivatives of ship maneuvering based on encoders and physical information neural networks, and specifically provides a prediction technology of hydrodynamic derivatives of ships under different maneuvering conditions, and belongs to the cross-technical field of combining machine learning and deep learning technology with ship hydrodynamics, with the purpose of improving the accuracy and efficiency of ship maneuvering performance prediction, and facilitating intelligent ship design and navigation safety.

[0008] The technical solution is as follows: the intelligent prediction method of ship maneuvering hydrodynamic derivatives, using CFD simulation data and MMG model, by minimizing the loss function based on physical equations and data prediction loss, optimizing network parameters, and realizing accurate prediction of ship maneuvering hydrodynamic derivative values; specifically comprising the following steps: S1. Using the defined autoencoder, the extracted three-dimensional geometric shape of the ship, the ship maneuvering motion speed and angular velocity parameters are compressed into low-dimensional representation data and preprocessed; S2. Construct a physical information neural network model based on the preprocessed low-dimensional representation data; S3, introduce an improved total loss function to optimize the physical information neural network model; S4. Based on the physical information neural network model after the improved total loss function is introduced, the combined training strategy of the optimization algorithm is used for optimization to obtain the optimized physical information neural network model, which is suitable for complex manipulation motion prediction; S5. Verify and iteratively optimize the optimized physical information neural network model and apply it to the prediction analysis of hydrodynamic derivatives of ship maneuvering.

[0009] In step S1, the autoencoder includes an encoder and a decoder, which extracts 151 ship parameters according to the geometric shape of the ship; by training the autoencoder, the 151 ship parameters are compressed into a 20-dimensional low-dimensional representation; the 20-dimensional representation is combined with the ship speed as the input of the physical information neural network model; the original data sources include ship water tank test data and CFD numerical simulation data.

[0010] In step S2, a physical information neural network model is constructed, including: establishing a physical information neural network model based on the partial differential equation of the ship maneuvering motion, embedding the physical law into the structure of the network, and capturing the physical law of the ship maneuvering motion; The partial differential equation is: ; The above three equations represent Translational motion in the direction and rotation Rotation about the axis, subscript The meanings are: inertial fluid hydrodynamics, viscous fluid hydrodynamics, propeller fluid hydrodynamics and rudder fluid hydrodynamics. For the quality of the ship, is the ship body coordinate system The acceleration of the ship's speed in the axial direction, is the ship body coordinate system Axis ship speed, The ship's body coordinate system is The angular velocity of the ship about the axis, for Inertial fluid hydrodynamic force in the axial direction, for Viscous fluid hydrodynamic force in the axial direction, for The hydrodynamic force of the fluid acting on the rudder in the axial direction is for The hydrodynamic force of the fluid acting on the propeller in the axial direction is is the ship body coordinate system The acceleration of the ship's speed in the axial direction, for Inertial fluid hydrodynamic force in the axial direction, for Viscous fluid hydrodynamic force in the axial direction, for The hydrodynamic force of the fluid acting on the propeller in the axial direction is for The hydrodynamic force of the fluid acting on the rudder in the axial direction is The ship's body coordinate system is The acceleration of the ship's angular velocity about the axis, For around The hydrodynamic moment of inertial fluid in the axial direction, For around The viscous fluid hydrodynamic moment in the axial direction, For around The hydrodynamic torque of the propeller in the axial direction is: For around The hydrodynamic moment of the fluid acting on the rudder in the axial direction is For ships in The force in the axial direction, is the distance from the center of the ship to the center of gravity. The center of the coordinate system is located at the center of gravity of the ship. ; Hull pair The moment of inertia of the shaft and the additional moment of inertia; ; In the formula, is the resistance of the ship when sailing straight, is the position derivative, is the coupled derivative, is the rotation derivative; Respectively indicate the hull Additional mass in direction; All are linear hydrodynamic derivatives; are the higher-order derivatives of the lateral forces and moments; are the higher-order coupled derivatives of the lateral force and moment.

[0011] In step S3, the total loss function consists of two parts: data prediction loss and motion equation loss based on the partial differential equation of the ship maneuvering motion; The data prediction loss expression is: ; In the formula, For data prediction loss, is the number of the hydrodynamic derivative of the ship maneuvering, is a predicted maneuvering hydrodynamic derivative value under the current speed and angular velocity, is the true value of a maneuvering hydrodynamic derivative under the current velocity and angular velocity, The ratio of the change in the hydrodynamic force on the hull caused by changing a certain motion parameter to the motion parameter while keeping other parameters unchanged while keeping the other parameters unchanged. The loss expression of the motion equation based on the partial differential equation of ship maneuvering motion is: ; In the formula, is the motion equation loss, is the position derivative, is the rotation derivative; The total loss function expression is: ; In the formula, are the weight coefficients of the two parts of loss respectively.

[0012] In step S4, based on the physical information neural network model after the introduction of the improved total loss function, a combination strategy of the Adam and LBFGS optimization algorithms is adopted in the network training process of PINN; the Adam optimizer is used for convergence; the LBFGS optimizer is used for fine-tuning in the later stage of training; in the PINN training process, when the training step reaches 20,000 steps or the loss value decreases by less than 10% compared with the previous step, the Adam method is switched to the LBFGS method for fine-tuning.

[0013] In step S5, the optimized physical information neural network model is verified and iteratively optimized, including: By separating the training set from the test set, the physical information neural network is optimized using the training set. After the training is completed, the trained neural network is tested using the test set to evaluate its accuracy and effectiveness in predicting the hydrodynamic derivatives of ship maneuvering.

[0014] Another object of the present invention is to provide a ship maneuvering hydrodynamic derivative intelligent prediction system, which implements the ship maneuvering hydrodynamic derivative intelligent prediction method, and the system comprises: Construct an autoencoder for data preprocessing module, use the defined autoencoder to compress the extracted three-dimensional geometric shape of the ship, the ship maneuvering motion speed and angular velocity parameters into low-dimensional representation data, and perform preprocessing; A physical information neural network model building module is used to build a physical information neural network model based on the preprocessed low-dimensional representation data; The total loss function introduction module is used to introduce the improved total loss function and optimize the physical information neural network model; The combined training strategy module of the optimization algorithm is used to optimize the physical information neural network model after the introduction of the improved total loss function, and adopt the combined training strategy of the optimization algorithm to obtain the optimized physical information neural network model to adapt to the complex control motion prediction; The model verification and optimization module is used to verify and iteratively optimize the optimized physical information neural network model, and is applied to the prediction and analysis of hydrodynamic derivatives of ship maneuvering.

[0015] Furthermore, the ship maneuvering hydrodynamic derivative intelligent prediction system is installed on a computer device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, and the processor implements the functions of the above-mentioned ship maneuvering hydrodynamic derivative intelligent prediction system when executing the computer program.

[0016] Furthermore, the ship maneuvering hydrodynamic derivative intelligent prediction system is installed on a computer-readable storage medium, and the computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the functions of the above-mentioned ship maneuvering hydrodynamic derivative intelligent prediction system can be realized.

[0017] Combining all the above technical solutions, the beneficial effects of the present invention are as follows: The present invention proposes an intelligent prediction method for ship maneuvering hydrodynamic derivatives based on encoders and physical information neural networks. By introducing an encoder module into the model, high-dimensional input data is reduced in dimensionality to reduce computational complexity, and the hydrodynamic derivatives are predicted in combination with the PINN model, which not only improves computational efficiency but also ensures the accuracy of physical constraints. This method can maintain a high prediction accuracy under different working conditions and show good adaptability in complex sea conditions. This intelligent prediction method based on encoders and PINN will provide new technical means for the prediction and optimization of ship maneuvering performance, help the development of intelligent ships, and promote the intelligent and efficient process of ship design and operation.

[0018] The application of the present invention can reduce the cost of ship maneuverability prediction, improve computational efficiency, and improve prediction accuracy. The expected commercial value is reflected in the market demand for intelligent ship design and maneuverability analysis, which can help shipyards and ship research institutes reduce experimental and numerical simulation costs. In addition, the technology can be widely used in maneuverability prediction and safety analysis in the maritime field, and has potential market promotion value. At present, there is a lack of intelligent prediction methods for ship maneuverability hydrodynamic derivatives that combine physical information in the world. Traditional methods are difficult to cope with maneuverability predictions for complex sea conditions or non-standard ship types. The present invention innovatively combines physical equation constraints with neural networks, which can provide stable and high-precision predictions, filling the gap in intelligent prediction technology in complex environments. The present invention solves the problem of intelligent prediction of ship hydrodynamic derivatives. Traditional CFD-based numerical simulation calculations have long calculation cycles, high costs, and are difficult to apply in real time. The present method not only accelerates the prediction process, but also ensures the physical rationality of the prediction through dimensionality reduction of the autoencoder and PINN integration, solving the demand for efficient and accurate maneuverability performance prediction in the maritime field. The present invention overcomes the limitation of the traditional deep learning method that the existing hydrodynamic derivative intelligent prediction relies on big data training, and introduces physical equations as constraints. By combining physical information with data-driven models, the bias problem of deep learning models predicting failure under insufficient data or unseen working conditions is avoided, thereby improving the generalization ability and applicability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings herein are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description, serve to explain the principles of the present disclosure; Figure 1 It is a flow chart of a method for intelligent prediction of ship maneuvering hydrodynamic derivatives provided by an embodiment of the present invention; Figure 2 is a network model test loss curve diagram provided by an embodiment of the present invention; Figure 3Schematic diagram of an intelligent prediction system for ship maneuvering hydrodynamic derivatives provided by an embodiment of the present invention; In the figure: 1. Module for building an autoencoder for data preprocessing; 2. Module for building a physical information neural network model; 3. Module for introducing the total loss function; 4. Module for combining training strategies for optimization algorithms; 5. Module for model verification and optimization. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific implementation disclosed below.

[0021] The system innovation of the present invention lies in that the present invention uses an autoencoder to reduce the dimensionality of multi-dimensional parameters such as the three-dimensional geometry, maneuvering speed and angular velocity of the ship to reduce the computational complexity, and introduces a physical information neural network (PINN) model to embed the physical equations as constraints into the loss function to achieve an organic combination of data-driven and physical constraints. Furthermore, an improved total loss function and optimization algorithm combination strategy are adopted to improve the prediction accuracy and generalization ability of the model under complex sea conditions. Experimental results show that this method can not only maintain high-precision hydrodynamic derivative prediction under different maneuvering conditions, but also show good physical consistency and adaptability, providing effective technical support for intelligent ship design and maneuverability performance analysis.

[0022] Embodiment 1, as Figure 1 As shown, the intelligent prediction method for ship maneuvering hydrodynamic derivatives provided by the embodiment of the present invention comprehensively applies CFD simulation data and MMG model, minimizes the loss function based on physical equations and data prediction loss, optimizes network parameters, and realizes accurate prediction of ship maneuvering hydrodynamic derivative values.

[0023] like Figure 1 As shown, the specific steps include: S1. Using the defined autoencoder, the extracted three-dimensional geometric shape of the ship, the ship maneuvering motion speed and angular velocity parameters are compressed into low-dimensional representation data and preprocessed; S2. Construct a physical information neural network model based on the preprocessed low-dimensional representation data; S3, introduce an improved total loss function to optimize the physical information neural network model; S4. Based on the physical information neural network model after the improved total loss function is introduced, the combined training strategy of the optimization algorithm is used for optimization to obtain the optimized physical information neural network model, which is suitable for complex manipulation motion prediction; S5. Verify and iteratively optimize the optimized physical information neural network model and apply it to the prediction analysis of hydrodynamic derivatives of ship maneuvering.

[0024] Exemplarily, in step S1, it specifically includes: constructing an autoencoder for data preprocessing; first defining the structure of the autoencoder to achieve dimensionality reduction processing of input parameters such as the three-dimensional geometric shape of the ship, the ship's maneuvering motion speed and angular velocity; extracting 151 ship parameters based on the ship's geometric shape; compressing these 151 ship parameters to a 20-dimensional low-dimensional representation by training the autoencoder to reduce computational complexity and improve network stability; then, combining the 20-dimensional representation with the ship's speed as the input of the physical information neural network model PINN; the original data sources include ship water tank test data and CFD numerical simulation data.

[0025] Exemplarily, in step S2, constructing a physical information neural network model based on the low-dimensional representation data compressed after preprocessing includes: establishing a physical information neural network model PINN based on the partial differential equations of ship maneuvering motion (such as the MMG model); the physical information neural network model PINN embeds physical laws into the structure of the network so that it can capture the physical laws of ship maneuvering motion; by introducing inequality constraints of physical equations, the inequality constraints of physical equations are incorporated into the loss function of the physical information neural network model PINN, so that the prediction results of the physical information neural network model PINN are consistent with the actual physical background.

[0026] Exemplarily, in step S3, an improved total loss function is introduced to optimize the prediction accuracy and stability of the physical information neural network model PINN; the present invention designs an improved loss function based on physical constraints and data-driven. The loss function includes two parts: one is the physical constraint loss based on the manipulation motion model, which is used to ensure that the output of the physical information neural network model PINN conforms to the laws of physics; the other is the prediction error loss based on data, which is used to adjust the parameters of the physical information neural network model PINN to minimize the prediction error. By considering both physical information and data-driven characteristics, the prediction accuracy and stability of the physical information neural network model PINN are optimized.

[0027] Exemplarily, in step S4, the physical information neural network model PINN after the introduction of the improved total loss function is further optimized by using a combined training strategy of optimization algorithms to obtain the optimized physical information neural network model PINN, including: in the network training process of the physical information neural network model PINN after the introduction of the improved total loss function, a combined strategy of Adam and LBFGS optimization algorithms is adopted. The Adam optimizer is used for fast convergence and improved training efficiency; the LBFGS optimizer is used for fine adjustment in the later stage of training to further improve the accuracy of the optimized physical information neural network model PINN. This combined optimization strategy effectively improves the efficiency of network training and the accuracy of the optimized physical information neural network model PINN, making it adaptable to complex manipulation motion prediction requirements.

[0028] Exemplarily, in step S5, the optimized physical information neural network model PINN model is verified and iteratively optimized, and the application of the model to the prediction analysis of the hydrodynamic derivatives of ship maneuvering includes: The optimized physical information neural network model PINN is verified using actual ship maneuvering test data and CFD numerical simulation data to ensure its accuracy and adaptability under different working conditions. According to the verification results, the structure, loss function and training strategy of the physical information neural network model PINN are iteratively optimized to improve the generalization ability of the optimized physical information neural network model PINN. The optimized physical information neural network model PINN can be applied to ship design and maneuvering performance analysis, providing support for the development of intelligent ship technology.

[0029] Another example, in step S1, for the problem of intelligent prediction of hydrodynamic derivatives of ship maneuvering, an autoencoder is first constructed to perform dimensionality reduction processing of high-dimensional parameters. The architecture of the autoencoder consists of an encoder and a decoder. The encoder is responsible for compressing high-dimensional input data into a low-dimensional space, and the decoder is used to reconstruct the original data. The initial input high-dimensional data contains 151 parameters related to the three-dimensional geometry of the ship. After training, these high-dimensional data are compressed to a 20-dimensional low-dimensional representation. This dimensionality reduction process not only effectively reduces the computational complexity, but also improves the training efficiency and stability of the physical information neural network model PINN.

[0030] Since the numerical ranges of different input parameters vary greatly, such as the scale parameters of ship geometry, which vary significantly in order of magnitude, all parameters are standardized before being input into the autoencoder. Standardization can scale the values ​​of different parameters to the same scale to avoid model instability caused by amplitude differences. This process ensures that the encoder can fully extract the key information of each parameter and provide reliable low-dimensional representation data input for subsequent physical information neural networks.

[0031] In another exemplary embodiment, in step S3, the improved loss function specifically includes: In view of the linear and nonlinear changes of hydrodynamic derivatives at low and high ship speeds during ship motion, conventional methods set hydrodynamic derivatives to fixed values, which easily leads to prediction errors. To solve this problem, the present invention introduces a physical information neural network model PINN, which dynamically predicts hydrodynamic derivatives by combining physical constraints and data-driven methods. The loss function consists of two parts: the first part is the mean square error (MSE) between the estimated value and the actual value of the hydrodynamic derivative, which ensures the accurate estimation of the hydrodynamic derivative by the physical information neural network model PINN; the second part is the mean square error MSE of the ship maneuvering motion equation based on the partial differential equation of ship maneuvering motion (Modern Management Group, MMG model), which is used to ensure that the model prediction results conform to the physical laws of ship maneuvering motion. The construction of this loss function effectively balances data fitting and compliance with physical laws, and improves the prediction accuracy of the model for hydrodynamic derivatives at different speeds.

[0032] Exemplarily, the total loss function is constructed. The total loss function of the model proposed in the present invention consists of two parts, which are specifically expressed as follows: In the data prediction loss part, the mean square error (MSE) between the hydrodynamic derivatives predicted by the physical information neural network model PINN and the actual measured values ​​is quantified, ensuring the accuracy of the physical information neural network model PINN in estimating the hydrodynamic derivatives.

[0033] ; In the formula, For data prediction loss, is the number of the hydrodynamic derivative of the ship maneuvering, is a predicted maneuvering hydrodynamic derivative value under the current speed and angular velocity, is the true value of a maneuvering hydrodynamic derivative under the current velocity and angular velocity, The ratio of the change in the hydrodynamic force on the hull caused by changing a certain motion parameter to the motion parameter while keeping other parameters unchanged while keeping the other parameters unchanged. The loss of the motion equation based on the partial differential equation of ship maneuvering motion (MMG model) further ensures the physical rationality of the model for ship motion at low and high ship speeds.

[0034] ; In the formula, is the motion equation loss, is the position derivative, is the coupled derivative; The total loss function expression is: ; In the formula, are the weight coefficients of the two parts of loss respectively.

[0035] The total loss function combines the above two parts of loss, and the weight coefficients of the two parts of loss are It is uniformly set to 1 to balance the model's data fitting ability and compliance with physical laws.

[0036] ; In another exemplary embodiment, in step S4, the network framework and experimental settings of the physical information neural network model PINN are optimized.

[0037] Through the description of the above method and its experimental verification, the present invention provides an effective solution for the prediction of hydrodynamic derivatives of ships under different speed conditions, which has the characteristics of high-precision prediction and physical consistency, and provides a new technical path and research basis for ship maneuverability analysis and intelligent navigation technology.

[0038] Example 2, as another implementation of the present invention, the intelligent prediction method of ship maneuvering hydrodynamic derivatives provided in the embodiment of the present invention includes: In the first step, 151 ship parameters related to the ship are calculated based on the SHF (Singapore Heart Foundation) type value point file of the ship. Then, these parameters are standardized and compressed to 20 dimensions using an encoder to reduce the computational complexity. The second step is to relate the expressions of hydrodynamic force and torque to the speed and angular velocity through Taylor expansion of the partial differential equation (MMG model) of ship maneuvering motion, establish the physical information neural network model PINN of ship motion, and use the data collected in step 1 as the input and output data of the physical information neural network model PINN; The partial differential equation is: ; In the formula, For the quality of the ship, is the ship body coordinate system The acceleration of the ship's speed in the axial direction, is the ship body coordinate system Axis ship speed, The ship's body coordinate system is The angular velocity of the ship about the axis, for Inertial fluid hydrodynamic force in the axial direction, for Viscous fluid hydrodynamic force in the axial direction, for The hydrodynamic force of the fluid acting on the rudder in the axial direction is for The hydrodynamic force of the fluid acting on the propeller in the axial direction is is the ship body coordinate system The acceleration of the ship's speed in the axial direction, for Inertial fluid hydrodynamic force in the axial direction, for Viscous fluid hydrodynamic force in the axial direction, for The hydrodynamic force of the fluid acting on the propeller in the axial direction is for The hydrodynamic force of the fluid acting on the rudder in the axial direction is The ship's body coordinate system is The acceleration of the ship's angular velocity about the axis, For around The hydrodynamic moment of inertial fluid in the axial direction, For around The viscous fluid hydrodynamic moment in the axial direction, For around The hydrodynamic torque of the propeller in the axial direction is: For around The hydrodynamic moment of the fluid acting on the rudder in the axial direction is For ships in The force in the axial direction, is the distance from the center of the ship to the center of gravity. The center of the coordinate system is located at the center of gravity of the ship. ; Hull pair The moment of inertia of the shaft and the additional moment of inertia; ; In the formula, is the resistance of the ship when sailing straight, is the position derivative, is the coupled derivative, is the rotation derivative; The three equations represent Translational motion in the direction and rotation Rotation about the axis, subscript The meanings are: inertial fluid hydrodynamics, viscous fluid hydrodynamics, propeller fluid hydrodynamics and rudder fluid hydrodynamics. Respectively indicate the hull Additional mass in direction; All are linear hydrodynamic derivatives; are the higher-order derivatives of the lateral forces and moments; are the higher-order coupled derivatives of the lateral force and moment.

[0039] The third step is to use CFD data to train and test the physical information neural network model PINN. The physical information neural network model PINN takes the ship's SHF file as input, and inputs the speed and angular velocity to predict the hydrodynamic derivative value of the ship's maneuvering under the current state. The physical information neural network model PINN achieves accurate prediction of the hydrodynamic derivative by minimizing the combination of data prediction loss and physical loss function.

[0040] ; In the formula, is the loss function value of the physical information neural network model PINN, is the loss function value of the three equations; Exemplarily, the third step is specifically as follows: According to the motion response of the ship under different speeds and angular velocities, the total loss function of the physical information neural network is constructed. Combined with the constraint objectives of the ship's motion equation, the error between the predicted results of the physical information neural network model PINN and the true value is quantified to ensure that the predicted results of the physical information neural network model PINN conform to the actual physical laws.

[0041] ; By separating the training set from the test set, the physical information neural network is optimized using the training set. After the training is completed, the trained neural network is tested using the test set to evaluate its accuracy and effectiveness in predicting the hydrodynamic derivatives of ship maneuvering.

[0042] In summary, the present invention provides a novel and efficient method for predicting hydrodynamic derivatives of ship maneuvering by introducing a combination of encoder and physical information neural network, which has broad application prospects and important theoretical significance.

[0043] Example 3. Taking the KCS ship as an example, the intelligent prediction method for ship maneuvering hydrodynamic derivatives provided by the embodiment of the present invention includes: Step 1. Taking the KCS ship as an example, 151 ship parameters are calculated based on its SHF type value point file, including the three-dimensional geometry of the ship, the speed and angular velocity of the maneuvering motion, etc., as shown in Table 1. Through sensitivity analysis, the parameters that have a greater impact on the prediction results are screened out, and an autoencoder is constructed for data preprocessing. In this process, the 151 ship parameters are first standardized to eliminate the numerical differences between different parameters to ensure the consistency of the data and the stability of the model. Next, the architecture of the autoencoder is defined, including the encoder and decoder. The autoencoder structure is a fully connected neural network. The encoding structure is [input_dim, 64, 32, 20], that is, two hidden layers, the number of neurons in the first layer is 64, and the number of neurons in the second layer is 32, and finally compressed to 20 dimensions; the decoder structure is [20, 32, 64, input_dim], that is, two hidden layers, the number of neurons in the first layer is 32, and the number of neurons in the second layer is 64, and finally restored to the dimension of input_dim. These standardized 151 parameters are compressed into a 20-dimensional low-dimensional representation by training the autoencoder. This process aims to reduce computational complexity and improve network stability. Ultimately, the generated low-dimensional representation will serve as the input of the physical information neural network (PINNs) model to enhance the effect of subsequent predictions; Table 1.1 151-dimensional geometric parameters of the ship

[0044] Table 1 (continued)

[0045] Table 1 (continued)

[0046] Step 2: Based on the MMG model of ship maneuvering motion, the Taylor expansion method of hydrodynamic force and torque is used to associate the expressions of hydrodynamic force and torque with the speed and angular velocity, and a physical information neural network model of hydrodynamic derivatives is established; this model comprehensively considers the influence of inertial force, damping force, etc. on ship motion, and integrates it into the training process to improve the physical interpretability of the prediction results.

[0047] Step 3: Train and test the physical information neural network model PINN that incorporates the neural network loss function through CFD data. The model takes the SHF type value point file of the KCS ship as input, and inputs the speed and angular velocity at the same time to predict the maneuvering hydrodynamic derivative value of the ship in the current state. By comparing with the prediction results of the traditional neural network model, it shows obvious advantages in prediction accuracy and stability.

[0048] In step 2, the specific steps of using the MMG physical information neural network model and integrating the neural network loss function module are as follows: The MMG model is used to establish the ship maneuvering motion equation. Considering that the heave, trim and roll of the ship are very small when maneuvering on the still water surface, the influence of heave, trim and roll on the motion in the horizontal plane is ignored. The three-degree-of-freedom motion equation is: ; Decomposing the forces and moments on the right side of the above equation, the external forces and moments acting on the ship are decomposed into the fluid dynamics acting on the bare hull, propeller and rudder , and and their interactions, such as Represents the fluid dynamics of the propeller's effect on the hull, etc.

[0049] ; Ignoring the interaction between propeller, hull and rudder in the above formula, the following simplification is obtained: ; Separating the viscous force and the inertial force and ignoring their mutual influence, we can obtain the following formula: ; The subscript The meanings are: inertial fluid hydrodynamics, viscous fluid hydrodynamics, propeller fluid hydrodynamics and rudder fluid hydrodynamics. By substituting them into the expression, the control motion equation of the three-degree-of-freedom MMG model can be obtained as follows: ; in, The mass of the ship (ship model); Respectively indicate the hull Additional mass in direction; and Hull pair The moment of inertia of the shaft and the additional moment of inertia. Then the viscous hydrodynamic force in the conventional ship is expanded using the Taylor expansion, and the Taylor expansion is retained to the third-order term. The nonlinearity of the hydrodynamic force on the hull is expressed as: ; in, The resistance of the ship when sailing straight; is the linear hydrodynamic derivative; are the higher-order derivatives of the lateral forces and moments; are the higher-order coupled derivatives of the lateral force and moment.

[0050] The data prediction loss part quantifies the error between the longitudinal velocity predicted by the model and the experimental data: ; The kinematic equation loss based on manipulating the equation of motion further ensures the physical plausibility of the model predictions.

[0051] ; The total loss function combines the above two parts of loss, and the weight coefficients of the two parts of loss are uniformly set to 1 to balance the model's data fitting ability and compliance with physical laws.

[0052] ; In step 3, the model specifically includes the following steps: The encoder and physical information neural network based model is trained and tested using CFD data. The model takes the SHF type value point file of the KCS ship as input, and also inputs the speed and angular velocity to predict the maneuvering hydrodynamic derivative value of the ship in the current state. During the training process, the model shows good loss convergence and test performance. Specifically, according to Figure 2 From the network model test loss curve, it can be seen that the model based on the encoder and physical information neural network has a test error of less than 0.005 on the KCS standard model, indicating its effective learning ability for input data. The test results show that the model can achieve stable prediction performance under various working conditions, showing good generalization ability and accuracy. These characteristics ensure the reliability of the model in practical applications and provide strong support for the intelligent prediction of hydrodynamic derivatives of ship maneuvering.

[0053] Example 4. Figure 3 As shown, the ship maneuvering hydrodynamic derivative intelligent prediction system provided by the embodiment of the present invention includes: Constructing an autoencoder for data preprocessing module 1, using the defined autoencoder to compress the extracted three-dimensional geometric shape of the ship, the ship maneuvering motion speed and angular velocity parameters into low-dimensional representation data, and perform preprocessing; A physical information neural network model building module 2 is used to build a physical information neural network model based on the preprocessed low-dimensional representation data; The total loss function introduction module 3 is used to introduce the improved total loss function and optimize the physical information neural network model; The combined training strategy module 4 of the optimization algorithm is used to optimize the physical information neural network model after the improved total loss function is introduced, and the combined training strategy of the optimization algorithm is used to obtain the optimized physical information neural network model to adapt to the complex control motion prediction; The model verification and optimization module 5 is used to verify and iteratively optimize the optimized physical information neural network model, and is applied to the prediction and analysis of hydrodynamic derivatives of ship maneuvering.

[0054] It can be seen from the above embodiments that the present invention improves the prediction accuracy of ship maneuverability in complex navigation environments. Traditional methods often rely on model tests or CFD numerical simulations in the identification of hydrodynamic derivatives, which have problems such as high computational cost, large data requirements, and poor adaptability. The present invention breaks through the limitations of traditional methods by combining physical information neural networks with encoder technology.

[0055] The present invention introduces an intelligent prediction method for ship maneuvering hydrodynamic derivatives based on encoders and physical information neural networks, which is specially designed for predicting the changes in hydrodynamic derivatives and maneuvering motion responses of ships at different speeds, and shows significant innovation and practical value. Compared with the prior art, the present invention has the advantages of simple structure and clear algorithm, and provides accurate and reliable decision support for ship maneuvering performance analysis and safe navigation. In addition, the method effectively improves the prediction accuracy of ship motion under complex sea conditions and speed conditions by accurately capturing the nonlinear changes of hydrodynamic derivatives, and solves the limitations of traditional models in dealing with nonlinear motion. Therefore, the present invention not only provides advanced technical means for ship maneuvering performance analysis and intelligent navigation systems, but also opens up new paths for research and application in related fields, and has important theoretical significance and broad application prospects.

[0056] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0057] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for intelligent prediction of ship maneuvering hydrodynamic derivatives, characterized in that: This method uses CFD simulation data and MMG model to optimize network parameters by minimizing the loss function based on physical equations and data prediction loss, thus achieving accurate prediction of the hydrodynamic derivative values ​​of ship maneuvering. The specific steps include: S1. Using the defined autoencoder, the extracted three-dimensional geometric shape of the ship, the ship maneuvering motion speed and angular velocity parameters are compressed into low-dimensional representation data and preprocessed; S2. Construct a physical information neural network model based on the preprocessed low-dimensional representation data; S3, introduce an improved total loss function to optimize the physical information neural network model; S4. Based on the physical information neural network model after the improved total loss function is introduced, the combined training strategy of the optimization algorithm is used for optimization to obtain the optimized physical information neural network model, which is suitable for complex manipulation motion prediction; S5. Verify and iteratively optimize the optimized physical information neural network model and apply it to the prediction analysis of hydrodynamic derivatives of ship maneuvering.

2. The intelligent prediction method for ship maneuvering hydrodynamic derivatives according to claim 1 is characterized in that: In step S1, the autoencoder includes an encoder and a decoder, which extracts 151 ship parameters according to the geometric shape of the ship; by training the autoencoder, the 151 ship parameters are compressed into a 20-dimensional low-dimensional representation; the 20-dimensional representation is combined with the ship speed as the input of the physical information neural network model; the original data sources include ship water tank test data and CFD numerical simulation data.

3. The intelligent prediction method for ship maneuvering hydrodynamic derivatives according to claim 1 is characterized in that: In step S2, a physical information neural network model is constructed, including: establishing a physical information neural network model based on the partial differential equation of the ship maneuvering motion, embedding the physical law into the structure of the network, and capturing the physical law of the ship maneuvering motion; The partial differential equation is: ； The above three equations represent Translational motion in the direction and rotation Rotation about the axis, subscript The meanings are: inertial fluid hydrodynamics, viscous fluid hydrodynamics, propeller fluid hydrodynamics and rudder fluid hydrodynamics. For the quality of the ship, is the ship body coordinate system The acceleration of the ship's speed in the axial direction, is the ship body coordinate system Axis ship speed, The ship's body coordinate system is The angular velocity of the ship about the axis, for Inertial fluid hydrodynamic force in the axial direction, for Viscous fluid hydrodynamic force in the axial direction, for The hydrodynamic force of the fluid acting on the rudder in the axial direction is for The hydrodynamic force of the fluid acting on the propeller in the axial direction is is the ship body coordinate system The acceleration of the ship's speed in the axial direction, for Inertial fluid hydrodynamic force in the axial direction, for Viscous fluid hydrodynamic force in the axial direction, for The hydrodynamic force of the fluid acting on the propeller in the axial direction is for The hydrodynamic force of the fluid acting on the rudder in the axial direction is The ship's body coordinate system is The acceleration of the ship's angular velocity about the axis, For around The hydrodynamic moment of inertial fluid in the axial direction, For around The viscous fluid hydrodynamic moment in the axial direction, For around The hydrodynamic torque of the propeller in the axial direction is: For around The hydrodynamic moment of the fluid acting on the rudder in the axial direction is For ships in The force in the axial direction, is the distance from the center of the ship to the center of gravity. The center of the coordinate system is located at the center of gravity of the ship. ; Hull pair The moment of inertia of the shaft and the additional moment of inertia; ； In the formula, is the resistance of the ship when sailing straight, is the position derivative, is the coupled derivative, is the rotation derivative; Respectively indicate the hull Additional mass in direction; All are linear hydrodynamic derivatives; are the higher-order derivatives of the lateral forces and moments; are the higher-order coupled derivatives of the lateral force and moment.

4. The intelligent prediction method for ship maneuvering hydrodynamic derivatives according to claim 3 is characterized in that: In step S3, the total loss function consists of two parts: data prediction loss and motion equation loss based on the partial differential equation of the ship maneuvering motion; The data prediction loss expression is: ； In the formula, For data prediction loss, is the number of the hydrodynamic derivative of the ship maneuvering, is a predicted maneuvering hydrodynamic derivative value under the current speed and angular velocity, is the true value of a maneuvering hydrodynamic derivative under the current velocity and angular velocity, The ratio of the change in the hydrodynamic force on the hull caused by changing a certain motion parameter to the motion parameter while keeping other parameters unchanged while keeping the other parameters unchanged. The loss expression of the motion equation based on the partial differential equation of ship maneuvering motion is: ； In the formula, is the motion equation loss, is the position derivative, is the rotation derivative; The total loss function expression is: ； In the formula, are the weight coefficients of the two parts of loss respectively.

5. The intelligent prediction method for ship maneuvering hydrodynamic derivatives according to claim 1 is characterized in that: In step S4, based on the physical information neural network model after the introduction of the improved total loss function, a combination strategy of the Adam and LBFGS optimization algorithms is adopted in the network training process of PINN; the Adam optimizer is used for convergence; the LBFGS optimizer is used for fine-tuning in the later stage of training; in the PINN training process, when the training step reaches 20,000 steps or the loss value decreases by less than 10% compared with the previous step, the Adam method is switched to the LBFGS method for fine-tuning.

6. The intelligent prediction method for ship maneuvering hydrodynamic derivatives according to claim 1 is characterized in that: In step S5, the optimized physical information neural network model is verified and iteratively optimized, including: By separating the training set from the test set, the physical information neural network is optimized using the training set. After the training is completed, the trained neural network is tested using the test set to evaluate its accuracy and effectiveness in predicting the hydrodynamic derivatives of ship maneuvering.

7. An intelligent prediction system for ship maneuvering hydrodynamic derivatives, characterized in that: The system implements the intelligent prediction method for ship maneuvering hydrodynamic derivatives according to any one of claims 1 to 6, and the system comprises: Constructing an autoencoder for data preprocessing module (1), using the defined autoencoder to compress the extracted three-dimensional geometric shape of the ship, the ship maneuvering motion speed and angular velocity parameters into low-dimensional representation data, and perform preprocessing; A physical information neural network model building module (2), used to build a physical information neural network model based on the preprocessed low-dimensional representation data; A total loss function introduction module (3) is used to introduce an improved total loss function and optimize the physical information neural network model; The combined training strategy module (4) of the optimization algorithm is used to optimize the physical information neural network model after the improved total loss function is introduced, and the combined training strategy of the optimization algorithm is used to obtain the optimized physical information neural network model, which is suitable for complex control motion prediction; The model verification and optimization module (5) is used to verify and iteratively optimize the optimized physical information neural network model, and is applied to the prediction and analysis of hydrodynamic derivatives of ship maneuvering.

8. The intelligent prediction system for ship maneuvering hydrodynamic derivatives according to claim 7 is characterized in that: The system is mounted on a computer device, which includes: at least one processor, a memory, and a computer program stored in the memory and running on the at least one processor. When the processor executes the computer program, the functions of the above-mentioned ship maneuvering hydrodynamic derivative intelligent prediction system are realized.

9. The intelligent prediction system for ship maneuvering hydrodynamic derivatives according to claim 7 is characterized in that: The system is mounted on a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the functions of the above-mentioned ship maneuvering hydrodynamic derivative intelligent prediction system can be realized.

Images