Rapid online identification method for hydrodynamic parameters of dynamic positioning ship

Abstract

The invention discloses a dynamic positioning ship hydrodynamic parameter rapid online identification method, which is characterized in that a dynamic positioning ship hydrodynamic parameter online identification model VMC-PINN suitable for hydrodynamic coefficient identification is designed based on Monte Carlo and a physical information neural network, and in the network, through probability representation of variational distribution learning network weight, the dynamic positioning ship hydrodynamic parameter identification model VMC-PINN is obtained. And Dropout is adopted to carry out Monte Carlo approximate inference, so that the dual-order robustness of PINNs to noise data and model uncertainty is enhanced. Meanwhile, an uncertainty quantification algorithm and a dynamic adjustment mechanism for online parameter updating are further designed, the problem of multi-source uncertainty in ship motion coupling dynamics is effectively solved, and rapid online hydrodynamic parameter identification capable of quantizing uncertainty in a strong interference environment is achieved. The on-line identification framework developed by the invention provides a reference for the self-adaptive parameter updating of the dynamic positioning control system, and the control precision and the safety margin can be remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of ship control, and in particular to a method for quickly identifying hydrodynamic parameters of a dynamically positioned ship on-line. Background Art

[0002] In recent years, with the continuous development and utilization of marine resources, dynamically positioned vessels have played an irreplaceable role in deep-sea operations, marine engineering, and other fields. The core of a dynamic positioning system lies in the real-time offsetting of environmental disturbances through thruster thrust. Accurate identification of hydrodynamic parameters is crucial for ensuring the robustness and accuracy of the control system. Traditional hydrodynamic parameter identification methods are mostly based on offline experimental data, such as planar motion mechanism tests (PMM) and swing arm tests. However, such methods are difficult to reflect the dynamic characteristics of ships in complex sea conditions (Fossen, 2011). Therefore, online identification technology has gradually become a research hotspot. It can update hydrodynamic model parameters in real time during ship operation, providing data support for adaptive control.

[0003] In the early days, classical methods based on system identification theory dominated. For example, the least squares method achieves parameter estimation by minimizing the sum of squared prediction errors (Fossen), and the Kalman filter utilizes state-space models and recursive algorithms to handle noise interference in dynamic systems (Balakrishnan). While these methods are computationally efficient, their core assumption is that the system dynamics equations must be linear or weakly nonlinear, making it difficult to accurately characterize the strongly nonlinear hydrodynamic response of a ship under the coupled effects of waves and currents.

[0004] As the complexity of ship operating environments increases, researchers have begun to explore ways to combine numerical simulation with experimental analysis. Computational fluid dynamics (CFD) constructs high-precision hydrodynamic models by solving the Navier-Stokes equations. However, single simulation calculations that often take several hours seriously restrict real-time requirements, and the contradiction between grid division accuracy and computing resources has never been completely resolved.

[0005] The rise of data-driven approaches has provided new insights for overcoming these bottlenecks. Machine learning algorithms, such as support vector machines, transform nonlinear problems into linear regression in high-dimensional space through kernel function mapping (Cherkassky et al.), while deep learning leverages deep neural networks to automatically extract high-order features from ship motion data (Abadi et al.). These approaches have demonstrated superior adaptability to traditional methods in areas such as wave load prediction and maneuverability modeling. However, the "black box" nature of purely data-driven approaches poses the risk of a lack of physical consistency: neural networks may fit spurious correlations that violate the laws of conservation of mass or momentum. In particular, model predictions under operating conditions outside the scope of the training data may produce results that violate the principles of fluid dynamics. This contradiction is even more acute in online applications of dynamically positioned vessels, where the time-varying nature of the ocean environment makes it difficult for training data to cover all possible operating conditions. Therefore, how to embed physical prior knowledge while maintaining the flexibility of data-driven approaches has become a core issue that needs to be addressed in this field. Summary of the Invention

[0006] The present invention provides a method for rapid online identification of hydrodynamic parameters of a dynamically positioned ship, aiming to effectively solve the above technical problems.

[0007] According to a first aspect of the present invention, a method for rapid online identification of hydrodynamic parameters of a dynamically positioned ship is provided, which is characterized in that a hydrodynamic parameter online identification model is constructed for rapid online identification of hydrodynamic parameters of a dynamically positioned ship, and the construction process of the hydrodynamic parameter online identification model includes: step S1, establishing a white box model suitable for online parameter identification; step S2, constructing a variational Monte Carlo physical information neural network VMC-PINN based on the white box model, after probabilistic modeling of the network weights, training is performed based on a dynamic adjustment mechanism for online parameter updates, and confidence assessment is provided for control decisions. After completing the network parameter update and training, the method is used for online identification of hydrodynamic parameters of a dynamically positioned ship; wherein the VMC-PINN network at least includes a fully connected network with a Dropout layer and a loss function.

[0008] Furthermore, the white box model is obtained by removing the common coefficient matrix from the traditional dynamics model.

[0009] Furthermore, the loss function is as follows:

[0010]

[0011] Where, is the neural network loss value term, is the output value of the neural network, where W represents the network weight, θ represents the parameter set, f(θ) is the physical information output value of the hydrodynamic equation, and N d and N pRepresents the amount of data and physical information respectively, is the optimal estimated value of the parameter, and λ is the weight of the physical loss term.

[0012] Furthermore, in step S2, variational inference is used to perform probabilistic modeling on the network weights W:

[0013]

[0014] Where, and is the loss term of the lower bound of evidence and the optimal estimate of the parameters, is the physical residual term, q φ (W),q θ (W) is the Bernoulli mask distribution, p(W) and represents the prior distribution and the posterior distribution.

[0015] Furthermore, during the training phase of the VMC-PINN network, Dropout is randomly enabled in the fully connected layer, which is equivalent to θ (W) Sampling; Keep Dropout activated during the test phase, perform T forward propagation, and generate a parameter set In the inference phase, the network with Dropout is sampled T times by Monte Carlo to generate the posterior distribution of the parameters:

[0016]

[0017] Where, is the posterior distribution of θ forward propagation.

[0018] Furthermore, the reasoning stage can decompose epistemic uncertainty and aleatoric uncertainty to provide confidence assessment for control decisions. The uncertainty of the identification results can be decomposed into:

[0019]

[0020] Where V ar (θ) is the uncertainty regression value, μ θ and σ θ are the expectation and variance.

[0021] Furthermore, the dynamic adjustment mechanism can reduce the strength of physical constraints and avoid false convergence under low signal-to-noise ratio conditions. Specifically:

[0022] λ(t)=λ0·exp(-γ·SNR -1 (t))

[0023] Where λ(t) is the adaptive coefficient and λ0 is the initial coefficient. The adaptive coefficient λ(t) changes with the signal-to-noise ratio (SNR) of the data in the sliding window.

[0024] According to a second aspect of the present invention, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-mentioned methods for rapid online identification of hydrodynamic parameters of a dynamically positioned vessel are implemented.

[0025] According to a third aspect of the present invention, the present invention further provides a storage medium storing a plurality of instructions suitable for being loaded by a processor to execute the steps of any of the above methods for rapid online identification of hydrodynamic parameters of a dynamically positioned vessel.

[0026] By using one or more of the above-mentioned embodiments of the present invention, at least the following technical effects can be achieved: First, while traditional models have a common coefficient matrix that leads to divergent identification results, the present invention reconstructs the traditional MMG dynamics model, eliminating its common coefficient matrix to create a white-box model suitable for online parameter identification, improving the uniqueness and accuracy of parameter identification. Second, by embedding the dynamics equations with a physical information neural network (PINN), the present invention ensures that parameter estimation conforms to the principles of fluid mechanics and exhibits physical consistency. This effectively addresses the technical issues of existing purely data-driven methods, which are prone to fitting spurious correlations, and traditional models that fail to optimize physical constraint integration. Furthermore, the present invention also designs a dynamic adjustment mechanism for online parameter updates, which achieves rapid convergence while effectively preventing false convergence and providing robustness and real-time performance against external forces and data noise. Furthermore, addressing the existing lack of confidence assessment and inability to address sensor noise and model mismatch, the present invention provides a 95% confidence interval to quantify the dual uncertainty of model and data noise. In summary, the fast online identification method of hydrodynamic parameters based on Monte Carlo physical information neural network of the present invention is highly integrated and easy to use. It provides a reference for the adaptive parameter update of the dynamic positioning control system and can significantly improve the control accuracy and safety margin. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The technical solutions and other beneficial effects of the present invention will be made apparent by describing in detail the specific embodiments of the present invention in conjunction with the accompanying drawings.

[0028] Figure 1 1 is a schematic diagram of the architecture of a variational Monte Carlo physical information neural network (VMC-PINN) provided in an embodiment of the present invention;

[0029] Figure 2 is a histogram of parameter identification errors of static parameter experimental condition 3 provided by an embodiment of the present invention;

[0030] Figure 3 is a histogram of parameter identification errors of static parameter experimental condition 4 provided by an embodiment of the present invention;

[0031] Figure 4 is a histogram of parameter identification errors of the dynamic parameter experimental condition 1 provided in an embodiment of the present invention;

[0032] Figure 5 It is a histogram of parameter identification errors of the dynamic parameter experimental condition 2 provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0034] In the description of the present invention, it should be noted that, unless otherwise specified or limited, the term "and / or" herein is merely a description of an association relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, and B exists alone. Furthermore, the character " / " herein, unless otherwise specified, generally indicates that the associated objects are in an "or" relationship.

[0035] In view of the above technical problems, the present invention provides a method for rapid online identification of hydrodynamic parameters of a dynamically positioned vessel.

[0036] Currently, there is no dynamic model developed for online parameter identification of dynamically positioned ships. Due to the existence of a common coefficient matrix, traditional models become ill-conditioned during parameter identification and cannot converge to a unique solution. For this purpose, an embodiment of the present invention develops a dynamic model of dynamically positioned ships suitable for online parameter identification, solving the problem at the level of the online parameter identification model.

[0037] Specifically, this embodiment uses the traditional dynamics model as a white box model for parameter identification after removing the common coefficient matrix.

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044] The above formula is the traditional dynamic model. Where ν=[u,v,r] T is the velocity vector of the ship's three degrees of freedom in the horizontal plane, F is the resultant force vector acting on the ship, m and I z is the ship's mass and moment of inertia, x G is the coordinate of the center of gravity in the direction of the ship's length, M RB and M A is the mass matrix and the additional mass matrix, C RB and C A are the linear matrix and the additional damping matrix, D is the nonlinear damping matrix,

[0045]

[0046] is the hydrodynamic coefficient to be identified.

[0047] Remove the common coefficient matrix from the above formula and transform it into:

[0048]

[0049]

[0050]

[0051]

[0052]

[0053]

[0054] White box model used for parameter identification.

[0055] A variational Monte Carlo physical information neural network (VMC-PINN) is proposed and used for online identification of hydrodynamic parameters of dynamically positioned ships. The network architecture diagram is shown in Figure 1 .

[0056] The loss term of this network is as follows:

[0057]

[0058] Where, is the neural network loss value term, is the output value of the neural network, where W represents the network weight, θ represents the parameter set, f(θ) is the physical information output value of the hydrodynamic equation, and N d and N p Represents the amount of data and physical information respectively, is the optimal estimated (identified) value of the parameter, and λ is the weight of the physical loss term.

[0059] Existing online parameter identification methods use fixed-weight loss function structures that make it difficult to balance the dynamic relationship between data terms and physical constraints. When non-Gaussian noise or outliers exist in sensor measurement data, the model robustness decreases significantly. This paper learns the probabilistic representation of network weights through variational distribution and uses Dropout for Monte Carlo approximate inference, thereby enhancing the dual-order robustness of PINNs to noisy data and model uncertainty.

[0060] First, the network weight W is probabilistically modeled based on variational inference (VI):

[0061]

[0062] Where, and The loss term and the most estimated (identified) value of the lower bound of the evidence is the physical residual term, q φ (W),q θ (W) is the Bernoulli mask distribution p(W) and represents the prior distribution and the posterior distribution.

[0063] During the training phase, Dropout is randomly enabled in the fully connected layer, which is equivalent to θ (W) Sampling; Keep Dropout activated during the test phase, perform T forward propagation, and generate a parameter set In the inference phase, the network with Dropout is sampled T times by Monte Carlo to generate the posterior distribution of the parameters:

[0064]

[0065] Where, is the posterior distribution of θ forward propagation.

[0066] Wave impact noise from ship sensors, data loss from motion capture systems (such as DGPS), and model simplification errors constitute the three major sources of uncertainty in parameter identification. Existing methods have not yet addressed this multi-source uncertainty in the coupled dynamics of ship motion. This paper proposes an uncertainty quantification algorithm and an online parameter update mechanism, enabling rapid online hydrodynamic parameter identification with quantifiable uncertainty in high-interference environments.

[0067] Specifically, the above process can decompose epistemic uncertainty and aleatoric uncertainty to provide confidence assessment for control decision-making. In the embodiment of the present invention, the uncertainty of the identification result can be decomposed into:

[0068]

[0069] Where V ar (θ) is the uncertainty regression value, μ θ and σ θ are the expectation and variance.

[0070] Different from the traditional PINN with fixed weight physical residual, the embodiment of the present invention designs a dynamic adjustment mechanism:

[0071] λ(t)=λ0·exp(-γ·SNR -1 (t))

[0072] In the above formula, λ(t) is the adaptive coefficient, and λ0 is the initial coefficient. The adaptive coefficient λ(t) varies with the signal-to-noise ratio (SNR) of the data within the sliding window. This mechanism reduces the strength of the physical constraints in low SNR conditions (such as severe sea conditions), effectively preventing false convergence.

[0073] The embodiment of the present invention combines specific experimental data to verify the above-mentioned method for rapid online identification of hydrodynamic parameters of dynamically positioned ships from static parameter experimental conditions and dynamic parameter experimental conditions.

[0074] Table 1 below shows the static parameter experimental conditions:

[0075] Working conditions Environmental Force Data quality 1 none Filtering 2 none No filtering 3 Wind speed 5±0.5m / s Filtering 4 Wind speed 5±0.5m / s No filtering

[0076] Table 1

[0077] Figure 2 and Figure 3 The parameter identification error histograms for working conditions 3 and 4 are shown, with the blue color representing the present invention and the green, yellow, and red color representing the existing methods. Experiments show that the present invention is highly robust to data and external interference uncertainty. Under conditions of external interference and inaccurate data, existing methods fail, while the present invention still produces accurate results and confidence intervals.

[0078] Table 2 below shows the dynamic parameter experimental conditions:

[0079] Working conditions Environmental Force Data quality Ship quality 1 none Filtering 96.32kg / 112.32kg 2 Wind speed 5±0.5m / s No filtering 96.32kg / 112.32kg

[0080] Table 2

[0081] The following figures are histograms of the parameter identification errors for Conditions 1 and 2, respectively. The blue color represents the proposed method, while the green and yellow colors represent comparisons with existing methods. The goal of dynamic hydrodynamic coefficient identification is to compare and analyze the convergence speed and accuracy of the online parameter identification algorithm proposed in this study with those of currently available mainstream online parameter identification algorithms when parameters are changed. Because mass is included in the identification parameters, the dynamic identification performance of the algorithm can be verified by varying the mass of the ship model. In this experiment, mass variation was achieved by adjusting the number of ballast masses.

[0082] As can be seen from the figure, the embodiment of the present invention shows extremely strong robustness to dynamic changes in parameters, can converge to new parameter values ​​in a few steps of updates, achieve real-time updates, and can accurately identify even under interference, while other existing methods have become ineffective.

[0083] Based on any of the above embodiments, another embodiment of the present invention further provides an electronic device, which may include: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor may call logic instructions in the memory to execute the above method.

[0084] In addition, the logical instructions in the above-mentioned memory can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling 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 method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0085] On the other hand, an embodiment of the present invention further provides a storage medium storing a plurality of instructions, which are suitable for being loaded by a processor to execute the method for rapid online identification of hydrodynamic parameters of a dynamically positioned vessel provided in the above embodiments.

[0086] On the other hand, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0087] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0088] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0089] In summary, although the present invention has been disclosed above with reference to preferred embodiments, the above preferred embodiments are not intended to limit the present invention. A person skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be based on the scope defined in the claims.

Claims

1. A method for rapid online identification of hydrodynamic parameters of a dynamically positioned ship, characterized by: A hydrodynamic parameter online identification model is constructed for rapid online identification of hydrodynamic parameters of a dynamically positioned ship. The construction process of the hydrodynamic parameter online identification model includes: Step S1, establishing a white box model suitable for online parameter identification; Step S2: Based on the white box model, a variational Monte Carlo physical information neural network (VMC-PINN) is constructed. After probabilistic modeling of the network weights, training is performed based on a dynamic adjustment mechanism for online parameter updates, and confidence assessment is provided for control decisions. After network parameter updates and training are completed, the network is used for online identification of hydrodynamic parameters of dynamically positioned ships. Among them, the VMC-PINN network at least includes a fully connected network with a Dropout layer and a loss function.

2. The method for rapid online identification of hydrodynamic parameters of a dynamically positioned ship according to claim 1, characterized in that: The white box model is obtained by removing the common coefficient matrix from the traditional dynamics model.

3. The method for rapid online identification of hydrodynamic parameters of a dynamically positioned ship according to claim 1, characterized in that: The loss function is as follows: Where, is the neural network loss value term, is the output value of the neural network, where W represents the network weight, θ represents the parameter set, f(θ) is the physical information output value of the hydrodynamic equation, and N d and N p Represent the amount of data and physical information respectively, is the optimal estimated value of the parameter, and λ is the weight of the physical loss term.

4. The method for rapid online identification of hydrodynamic parameters of a dynamically positioned ship according to claim 1, characterized in that: In step S2, variational inference is used to perform probability modeling on the network weight W: Where, and is the loss term of the lower bound of evidence and the optimal estimate of the parameters, is the physical residual term, q φ (W),q θ (W) is the Bernoulli mask distribution, p(W) and represents the prior distribution and the posterior distribution.

5. The method for rapid online identification of hydrodynamic parameters of a dynamically positioned ship according to claim 4, characterized in that: During the training phase of the VMC-PINN network, Dropout is randomly enabled in the fully connected layer, which is equivalent to θ (W) Sampling; Keep Dropout activated during the test phase, perform T forward propagation, and generate a parameter set In the inference phase, the network with Dropout is sampled T times by Monte Carlo to generate the posterior distribution of the parameters: Where, is the posterior distribution of θ forward propagation.

6. The method for rapid online identification of hydrodynamic parameters of a dynamically positioned ship according to claim 5, characterized in that: The reasoning stage can decompose epistemic uncertainty and aleatoric uncertainty to provide confidence assessment for control decisions. The uncertainty of the identification results can be decomposed into: Where Var(θ) is the uncertainty regression value, μ θ and σ θ are the expectation and variance.

7. The method for rapid online identification of hydrodynamic parameters of a dynamically positioned ship according to claim 1, characterized in that: The dynamic adjustment mechanism can reduce the strength of physical constraints and avoid false convergence under low signal-to-noise ratio conditions. Specifically: λ(t)=λ0·exp(-γ·SNR -1 (t)) Where λ(t) is the adaptive coefficient and λ0 is the initial coefficient. The adaptive coefficient λ(t) changes with the signal-to-noise ratio (SNR) of the data in the sliding window.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

9. A storage medium, characterized in that: The storage medium stores a plurality of instructions, which are suitable for being loaded by a processor to execute the steps of the method according to any one of claims 1 to 7.

Images