Intelligent towing tank device based on Gaussian process regression and control method
By using an intelligent towed water tank device based on Gaussian process regression, the device enables adaptive selection of experimental parameters and automatic data analysis, solving the problems of low efficiency and insufficient intelligence in traditional towed water tank experiments. It achieves efficient coverage of high-dimensional parameter space and is suitable for complex fluid mechanics experiments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WESTLAKE UNIV
- Filing Date
- 2026-01-28
- Publication Date
- 2026-05-12
AI Technical Summary
Traditional towed pool experiments are inefficient, limited in parameter exploration, and lack intelligence, failing to meet the needs of complex fluid mechanics experiments for high-dimensional parameter exploration and efficient data utilization.
An intelligent towing pool device based on Gaussian process regression is adopted. Through the collaboration of robots, computers and machine learning algorithms, it realizes adaptive selection of experimental parameters, automatic data acquisition and analysis, and closed-loop control of the experimental process. Combined with multi-axis force sensors and data acquisition cards, it carries an experimental model and executes multi-dimensional motion trajectories to construct a Gaussian process regression model for decision-making.
It significantly improves experimental efficiency, can efficiently cover a wide parameter space, reveal systematic laws that traditional methods cannot reach, and is suitable for complex experiments such as vortex-induced vibration and flapping wing optimization, becoming a general-purpose intelligent experimental platform in the field of fluid-structure interaction.
Smart Images
Figure CN122016232A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of fluid mechanics experiment equipment, in particular to an intelligent towing tank device based on Gaussian process regression and a control method. BACKGROUND
[0002] Fluid-structure interaction is a core research direction in the field of fluid mechanics. The complex nonlinear response characteristics of many typical research objects need to be verified through a large number of experiments, such as vortex-induced vibration and flapping wing trajectory parameter optimization. The traditional method needs to carry out experiments through sequential hypothesis testing and trial-and-error method, which is difficult to cover high-dimensional parameter space. Vortex-induced vibration research involves vibration frequency, amplitude, Reynolds number and other independent parameters. System sampling in high-dimensional parameter space requires a large number of experiments, which is not feasible by traditional methods. The selection of experimental parameters depends on artificial experience, and the experimental scheme cannot be adjusted adaptively according to the existing data. Data collection and analysis are disconnected, and it is difficult to dig out hidden physical laws. Artificial operation is easy to introduce human error, and the experimental gap cannot effectively eliminate fluid disturbance, leading to cross contamination of experimental results.
[0003] Most of the existing laboratory automation equipment can only realize single repeated operation, lacks autonomous decision-making ability based on machine learning, and cannot meet the needs of complex fluid mechanics experiments for high-dimensional parameter exploration and efficient data utilization.
[0004] Therefore, it is urgent to develop an intelligent towing tank device with autonomous learning and decision-making ability to realize full automation and intelligentization of experimental process. SUMMARY
[0005] The present application aims to overcome the defects of low efficiency, limited parameter exploration and insufficient intelligence of traditional towing tank experiments, and provides a full-automatic intelligent towing tank device based on Gaussian process regression. Through the cooperation of robots, computers and machine learning algorithms, the experimental parameters are adaptively selected, the data are automatically collected and analyzed, and the experimental process is closed-loop controlled, which greatly improves the experimental efficiency, expands the parameter exploration range and ensures the accuracy of experimental results. An intelligent towing tank device based on Gaussian process regression comprises: An experimental execution module for carrying an experimental model in a towing tank and executing motion corresponding to experimental parameters; A data collection module for collecting fluid dynamic parameters acting on the experimental model in the motion process in real time; An intelligent decision-making module for constructing and updating a Gaussian process regression model based on the fluid dynamic parameters collected by the data collection module and the corresponding experimental parameters, and deciding the next set of experimental parameters to be executed based on the updated Gaussian process regression model; A control module for controlling the experimental execution module to complete the corresponding experiment according to the next set of experimental parameters output by the experimental parameter optimization unit.
[0006] The experiment execution module comprises a towing tank body, a guide rail slide frame arranged along the length direction of the towing tank, a four-degree-of-freedom motion platform mounted on the guide rail slide frame, and a driving unit connected with the four-degree-of-freedom motion platform, the four-degree-of-freedom motion platform is used for carrying an experiment model and executing a combined motion trajectory in the flow direction, the cross-flow direction and the rotation direction defined by multi-dimensional experiment parameters, and the driving unit is used for driving the guide rail slide frame and the four-degree-of-freedom motion platform to move and output corresponding actual motion trajectory data.
[0007] The data acquisition module comprises a multi-axis force sensor connected with the experiment model and a data acquisition card, the multi-axis force sensor is used for acquiring fluid force data suffered by the experiment model during the motion, and the data acquisition card is used for synchronously acquiring the fluid force data and the actual motion trajectory data.
[0008] The intelligent decision module comprises a data preprocessing unit, a Gaussian process regression modeling unit, an uncertainty evaluation unit and an experiment parameter optimization unit, wherein: The data preprocessing unit is used for filtering and normalizing the acquired fluid force data and actual motion trajectory data, and constructing a multi-dimensional input-output data set for regression modeling; The Gaussian process regression modeling unit is used for constructing a Gaussian process regression model with experiment parameters as input and fluid force as output based on the data set; The uncertainty evaluation unit is used for calculating the predicted mean and predicted uncertainty of the candidate experiment parameters in the parameter space based on the Gaussian process regression model; The experiment parameter optimization unit is used for determining the next group of experiment parameters in the feasible parameter space based on the predicted mean and predicted uncertainty according to a preset active learning sampling criterion or a Bayesian optimization criterion.
[0009] The second aspect of the present application provides a control method of an intelligent towing tank device based on Gaussian process regression, comprising: Carrying an experiment model in the towing tank and executing the motion corresponding to the experiment parameters; Real-time acquisition of fluid force parameters suffered by the experiment model during the motion; Based on the fluid force parameters acquired by the data acquisition module and the corresponding experiment parameters, a Gaussian process regression model is constructed and updated, and the next group of experiment parameters to be executed is decided based on the updated Gaussian process regression model; According to the next group of experiment parameters output by the experiment parameter optimization unit, the experiment execution module is controlled to complete the corresponding experiment.
[0010] The present application has the following beneficial effects: Faced with complex multi-parameter experiments such as vortex-induced vibration and flapping wing optimization, traditional methods, limited by experimental costs, can only perform sparse and local parameter tests. This invention utilizes Gaussian process regression to actively find the experimental points with the "maximum information content," efficiently covering a wide parameter space with the fewest number of experiments, revealing systematic laws that traditional methods cannot reach.
[0011] By setting parametric and non-parametric modes, this invention enables the device to perform precise experiments on preset trajectories and autonomously explore unknown optimal motion forms. This flexibility allows it to be widely applied to various fundamental research and engineering optimization problems in the field of fluid-structure interaction, making it a versatile intelligent experimental research platform. Attached Figure Description
[0012] Figure 1 A flowchart of a control method for an intelligent towed water tank device based on Gaussian process regression provided in an embodiment of the present invention. Detailed Implementation
[0013] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0014] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The first embodiment of the intelligent towing pool device method based on Gaussian process regression in this invention includes: The experiment execution module is used to mount the experimental model in the towed pool and execute the motion corresponding to the experimental parameters; The data acquisition module is used to collect the hydrodynamic parameters of the experimental model during its motion in real time. The intelligent decision-making module is used to construct and update the Gaussian process regression model based on the fluid dynamic parameters and corresponding experimental parameters collected by the data acquisition module, and to determine the next set of experimental parameters to be executed based on the updated Gaussian process regression model. The control module is used to control the experiment execution module to complete the corresponding experiment based on the next set of experimental parameters output by the experimental parameter optimization unit.
[0015] The experimental execution module includes a towed pool body, a guide rail carriage arranged along the length of the towed pool, a four-degree-of-freedom motion platform mounted on the guide rail carriage, and a drive unit connected to the four-degree-of-freedom motion platform. The four-degree-of-freedom motion platform is used to carry the experimental model and execute a combination of downstream, transverse, and rotational motion trajectories defined by multidimensional experimental parameters. The drive unit is used to drive the guide rail carriage and the four-degree-of-freedom motion platform to move and output the corresponding actual motion trajectory data.
[0016] The data acquisition module includes a multi-axis force sensor connected to the experimental model and a data acquisition card. The multi-axis force sensor is used to collect fluid force data of the experimental model during its motion, and the data acquisition card is used to simultaneously collect the fluid force data and the actual motion trajectory data.
[0017] The intelligent decision-making module includes a data preprocessing unit, a Gaussian process regression modeling unit, an uncertainty assessment unit, and an experimental parameter optimization unit, wherein: The data preprocessing unit is used to filter and normalize the collected fluid force data and actual motion trajectory data, and to construct a multidimensional input-output dataset for regression modeling. The Gaussian process regression modeling unit is used to construct a Gaussian process regression model based on the dataset, with experimental parameters as input and fluid dynamics as output. The uncertainty assessment unit is used to calculate the predicted mean and predicted uncertainty of the candidate experimental parameters in the parameter space based on the Gaussian process regression model. The experimental parameter optimization unit is used to determine the next set of experimental parameters within the feasible parameter space based on the predicted mean and prediction uncertainty, according to a preset active learning sampling criterion or Bayesian optimization criterion.
[0018] The experimental parameter optimization unit operates according to the active learning sampling criterion, which is: within the feasible experimental parameter space, select the experimental parameter with the largest prediction uncertainty as the next set of experimental parameters.
[0019] The experimental parameter optimization unit calculates the confidence interval width of the prediction results corresponding to the candidate experimental parameters based on the Gaussian process regression model, and uses the confidence interval width as a sampling index to determine the next set of experimental parameters.
[0020] The experimental parameter optimization unit operates according to the Bayesian optimization criterion, which constructs the expected improvement function of the objective function based on the Gaussian process regression model, and determines the next set of experimental parameters by solving the expected improvement function.
[0021] When determining the next set of experimental parameters, the experimental parameter optimization unit considers both the predicted mean and the predicted uncertainty, and constructs an evaluation function for selecting experimental parameters using a weighted approach.
[0022] The control module is set with a convergence criterion based on prediction uncertainty. When the maximum prediction uncertainty of the Gaussian process regression model in the parameter space is lower than a preset threshold multiple times in a row, the experimental parameter optimization process is terminated.
[0023] This invention also relates to a control method for an intelligent towed water tank device based on Gaussian process regression, comprising: The experimental model was mounted in a towed water tank and the motion corresponding to the experimental parameters was executed. The hydrodynamic parameters of the experimental model during its motion are collected in real time. Based on the fluid dynamic parameters and corresponding experimental parameters collected by the data acquisition module, a Gaussian process regression model is constructed and updated, and the next set of experimental parameters to be executed is determined based on the updated Gaussian process regression model. The experimental execution module controls the corresponding experiment to complete the experiment based on the next set of experimental parameters output by the experimental parameter optimization unit.
[0024] The embodiments of the present invention are as follows: a. Experimental Execution Module: This module includes the towed water tank body, a four-degree-of-freedom motion platform, and a drive unit. The towed water tank body has a tow length of 4876 mm and a test cross-section of 1306 mm × 888 mm, providing a stable fluid environment. The four-degree-of-freedom motion platform is mounted on a double-rail carriage arranged along the length of the water tank. It has the capability to execute combined trajectories of large and small amplitude downstream, transverse, and rotational movements. The carriage's movement speed ranges from 0.01 to 0.5 m / s, and it is used to carry the experimental model and execute preset vibration trajectories. The drive unit uses servo motors and precision transmission mechanisms to ensure high-precision control of the motion trajectory. The drive unit can return the trajectory data of the four-degree-of-freedom motors for monitoring their actual motion trajectory.
[0025] b. Data Acquisition Module: Includes a six-axis force sensor and a data acquisition card. The six-axis force sensor (ATI-Gamma model) is used to measure fluid dynamic parameters such as drag force and lift on the experimental model, and the measurement accuracy meets the requirements of fluid mechanics experiments. The data acquisition card (NI DAQ USB6218) is used to acquire sensor data in real time, and the sampling frequency is adapted to the dynamic response requirements of the experiment.
[0026] c. Intelligent Decision Module: Built-in Gaussian Process Regression (GPR) learning algorithm, including data preprocessing unit, model training unit, uncertainty quantification unit and experimental parameter optimization unit.
[0027] Gaussian process regression (GPR) models are used to establish nonlinear mapping relationships between experimental parameters and fluid dynamic parameters. Taking oscillating flow around a cylinder as the background of a fluid dynamics experiment, its core mathematical definitions and physical meanings are as follows: 1.1 Definition of Input / Output Vectors Input vector : A multidimensional vector composed of experimental parameters, with dimension Based on the dynamic expansion (1~8 dimensions) of the experimental scenario, the expression is: The parameters are key control variables in fluid mechanics experiments, specifically including: Basic parameters: Reduced frequency Dimensionless transverse amplitude Dimensionless flow direction amplitude Reynolds number For the incoming flow velocity, For model feature length, (The fluid's kinematic viscosity). Extended parameters: Phase angle Second frequency reduction speed Second frequency phase angle .
[0028] Output vector : A vector composed of target hydrodynamic parameters, expressed as: The parameters are the core measurement indicators of the experiment, and their calculation methods are as follows: Drag coefficient For average drag force, For fluid density, (for model length) Lift coefficient ( The amplitude of the oscillating lift. (The phase difference between lift and motion). Additional quality coefficient The solution is obtained by fitting the fluid force signal and the model acceleration signal.
[0029] 1.2 Basic Form of the Model The GPR model assumes that the output vector The prior distribution follows a Gaussian process, and its mathematical expression is: The meanings of the parameters in the formula are as follows: : basis function vector ( (where is the number of basis functions), used to characterize the global trend between input and output; The basis function coefficient vector is determined through data fitting. Kernel functions are used to characterize the local dependencies between input parameters. The kernel function hyperparameter vector; : Measure the noise variance, which reflects the inherent measurement error of the experimental equipment; The identity matrix ensures the positive definiteness of the covariance matrix.
[0030] II. Design of Basis Functions and Kernel Functions 2.1 Basis Function Selection Based on the linear or nonlinear trend of the physical characteristics of the hydrodynamic parameters, the appropriate basis function type is selected, and the specific combinations are shown in Table 1: Table 1 in, For input vectors The selection of basis functions for each dimension of the data is verified through experimental data to ensure accurate capture of parameter change trends.
[0031] 2.2 Kernel Function Design The kernel function adopted is the Matern kernel function, which is an improvement of ARD (Automatic Relevance Determination), adapted to the smoothness and nonlinearity characteristics of fluid mechanics experimental data. The specific form is as follows: (1) ARD Matern 3 / 2 kernel function (adapted to drag coefficient) (2) ARD Matern 5 / 2 kernel function (adapted to lift coefficient) Additional quality coefficient The meanings of each hyperparameter in the formula are as follows: : Kernel function hyperparameter vector; The kernel function amplitude hyperparameter controls the overall fluctuation range of the output signal. : Length scale hyperparameter, which controls the decay rate of the correlation between input parameters; Correlation matrix Used to adjust the first The relevance weights of the input dimensions enable adaptive differentiation of the importance of each parameter; Weighted Euclidean distance is used to measure the similarity between two input vectors.
[0032] III. Model Training and Hyperparameter Optimization 3.1 Dataset Construction Assuming it is already completed This experiment constructs the training dataset. ,in: Input matrix Each row corresponds to a set of experimental parameters; Output matrix Each row corresponds to a set of fluid dynamic parameters; Basis function matrix Each row corresponds to a set of basis function mapping results for the input.
[0033] 3.2 Marginal Likelihood Maximization Model parameters , , The marginal likelihood function is obtained by maximizing it, and its expression is: In the formula: The residual vector reflects the remaining error after the basis functions are fitted. The covariance matrix characterizes the correlation among all experimental data.
[0034] The coefficients of the basis functions are obtained by differentiating the marginal likelihood function and setting the derivative to zero. The optimal estimate: Kernel function hyperparameters and measurement noise variance Solve using numerical optimization algorithms: The optimization objective is to maximize the marginal likelihood function value to ensure that the model can best fit the experimental data and minimize the risk of overfitting.
[0035] IV. Posterior Prediction and Uncertainty Quantification 4.1 Single Sample Prediction For new experimental parameters Its corresponding hydrodynamic parameters The posterior distribution follows a Gaussian distribution. ,in: (1) Predicted mean In the formula , where is the covariance vector between the new input and all training inputs.
[0036] (2) Predicting covariance The covariance reflects the degree of uncertainty in the prediction results. The larger the covariance, the higher the degree of uncertainty in the experimental results under the current parameter combination.
[0037] 4.2 Quantitative Indicators of Uncertainty Using the forecast standard deviation as the core uncertainty indicator, the expression is: For multi-output scenarios, simultaneous prediction , , The overall uncertainty is the weighted sum of the standard deviations of each output: in These are weighting coefficients. They are set according to the experimental objectives, with a default value of 1 / 3, used to balance the uncertainty weights of different output parameters.
[0038] V. Model Update Mechanism Each time new experimental data is added Then, the model is quickly updated using incremental learning to avoid retraining on the entire dataset. The update steps are as follows: Expanding the dataset: , ; Updating the covariance matrix: Efficiently updating using matrix block properties The expression is: in ; 3. Iterative optimization: When the amount of new data reaches the threshold (default 5 sets), hyperparameter optimization is re-executed to ensure the model's adaptability to new data.
[0039] d. Control Module: A PLC motion control system combined with Python is used to achieve coordinated control of all modules. The control module has built-in experimental process control logic, including a closed-loop process of initial experimental parameter random generation, experiment execution, data acquisition, model updating, parameter optimization, and experimental convergence judgment. The control module needs to set the fluid settling time between experiments to avoid cross-contamination caused by fluid disturbances between consecutive experiments.
[0040] Specifically: 1. Initialization: The user defines the input parameter vector for the experiment, and specifies the core control conditions and their feasible range that the experiment needs to focus on. These control conditions can be flexibly adjusted according to the research objectives and cover the key factors that affect the experimental results.
[0041] 2. Initial Experiments: Based on the defined conditions, a small number of initial experimental schemes are randomly generated to ensure that the initial schemes can initially cover the core parameter space, providing basic data support for subsequent modeling. The number of initial experiments must meet the basic requirements for subsequent model construction, driving the experiment execution module to carry out initial sparse experiments.
[0042] 3. Data Input: Import all completed experimental data into the system. After importation, the system will automatically perform preprocessing, mainly including identifying and removing abnormal data caused by equipment interference, environmental fluctuations, etc., and converting various physical measurements into standardized analytical indicators to ensure the accuracy, consistency, and comparability of all data, providing high-quality input for subsequent model building.
[0043] 4. Gaussian Process Regression: Based on preprocessed experimental data, a Gaussian process regression model is constructed. First, the original experimental conditions are transformed into a feature form suitable for modeling through specific mapping rules. Then, a kernel function that adapts to the nonlinear characteristics of the experimental objects is selected to describe the correlation between different experimental conditions. Finally, based on the feature mapping results and the kernel function calculation results, a complete covariance matrix is constructed to comprehensively characterize the complex relationship between experimental conditions and results.
[0044] 5. Hyperparameter Optimization: For the constructed Gaussian process regression model, key parameters in the model (such as feature mapping coefficients, kernel function parameters, and measurement noise-related parameters) are solved by maximizing the marginal likelihood function of the data. This optimization process comprehensively considers the data fitting effect, the impact of noise, and the model stability, ensuring that the optimized parameters enable the model to accurately capture the inherent laws between experimental conditions and results, thereby improving the model's predictive ability.
[0045] 6. Predicting Experimental Results and Assessing Reliability: Based on the uncertainty assessment results provided by the prediction module, the next experimental condition that can supplement experimental information to the greatest extent is selected. The core principle is to prioritize the experimental condition with the lowest prediction reliability and the most missing information. This approach achieves a balance between "exploring unknown areas" and "optimizing the accuracy of known areas," ensuring that each supplementary experiment maximizes the performance of the overall model.
[0046] 7. Automated Experiment and Data Acquisition: After receiving the selected optimal experimental conditions, the intelligent towed pool device automatically initiates the experimental process. The device precisely adjusts various experimental parameters, including key conditions such as motion speed, vibration frequency, amplitude, and phase, through a servo control system. Subsequently, it collects core data in real time during the experiment using high-precision sensors, including fluid forces and flow field distribution information. The collected data undergoes preliminary processing to remove invalid data and obtain valid experimental results.
[0047] 8. Verify if the model has reached a stable state: The system sets a convergence reference standard based on the inherent accuracy of the experimental equipment. This standard is determined through multiple repeated benchmark experiments and can effectively reflect the measurement error range of the equipment. The maximum uncertainty of the current model prediction result is compared with this reference standard. If, after several consecutive iterations, the maximum uncertainty is consistently lower than the reference standard, the model is considered to have reached a stable state, and the experimental process can be terminated. If the standard is not met, newly collected valid experimental data is integrated into the original dataset, the dataset is updated, and the system returns to the data input module to start the next round of model building and prediction.
[0048] 9. Output the final model and experimental results: Once the model reaches the convergence criterion, the system terminates the iterative experiment and outputs the complete results. The results include a predictive model that accurately describes the patterns of the experimental objects, which can be directly used for subsequent analysis and prediction of related problems, as well as detailed data records of the entire experimental process, covering all experimental conditions and corresponding results. A model reliability assessment report is also provided, clearly explaining the confidence level and convergence status of the prediction results, providing a basis for subsequent applications.
[0049] Results output: Automatically generate an experimental report, including the function approximation results of the target quantitative index, the spatial distribution characteristics of the parameters, and the analysis of physical laws.
[0050] This device supports both parametric and non-parametric operation modes, can efficiently cover high-dimensional parameter spaces, and improves experimental efficiency by more than an order of magnitude compared with traditional methods. It solves the problems of low efficiency, limited parameter exploration, and insufficient intelligence in traditional towed pool experiments. It is suitable for complex fluid mechanics experiments such as vortex-induced vibration, flapping wing trajectory parameter optimization, and fluid-structure interaction, providing an efficient and reliable experimental platform for scientific research in related fields.
[0051] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A smart towing pool device based on Gaussian process regression, characterized in that, The device includes: The experiment execution module is used to mount the experimental model in the towed pool and execute the motion corresponding to the experimental parameters; The data acquisition module is used to collect the hydrodynamic parameters of the experimental model during its motion in real time. The intelligent decision-making module is used to construct and update the Gaussian process regression model based on the fluid dynamic parameters and corresponding experimental parameters collected by the data acquisition module, and to determine the next set of experimental parameters to be executed based on the updated Gaussian process regression model. The control module is used to control the experiment execution module to complete the corresponding experiment based on the next set of experimental parameters output by the experimental parameter optimization unit.
2. The intelligent towing pool device according to claim 1, characterized in that, The experimental execution module includes a towed pool body, a guide rail carriage arranged along the length of the towed pool, a four-degree-of-freedom motion platform mounted on the guide rail carriage, and a drive unit connected to the four-degree-of-freedom motion platform. The four-degree-of-freedom motion platform is used to carry the experimental model and execute a combination of downstream, transverse, and rotational motion trajectories defined by multidimensional experimental parameters. The drive unit is used to drive the guide rail carriage and the four-degree-of-freedom motion platform to move and output the corresponding actual motion trajectory data.
3. The intelligent towing pool device according to claim 2, characterized in that, The data acquisition module includes a multi-axis force sensor connected to the experimental model and a data acquisition card. The multi-axis force sensor is used to collect fluid force data of the experimental model during its motion, and the data acquisition card is used to simultaneously collect the fluid force data and the actual motion trajectory data.
4. The intelligent towing pool device according to claim 3, characterized in that, The intelligent decision-making module includes a data preprocessing unit, a Gaussian process regression modeling unit, an uncertainty assessment unit, and an experimental parameter optimization unit, wherein: The data preprocessing unit is used to filter and normalize the collected fluid force data and actual motion trajectory data, and to construct a multidimensional input-output dataset for regression modeling. The Gaussian process regression modeling unit is used to construct a Gaussian process regression model based on the dataset, with experimental parameters as input and fluid dynamics as output. The uncertainty assessment unit is used to calculate the predicted mean and predicted uncertainty of the candidate experimental parameters in the parameter space based on the Gaussian process regression model. The experimental parameter optimization unit is used to determine the next set of experimental parameters within the feasible parameter space based on the predicted mean and prediction uncertainty, according to a preset active learning sampling criterion or Bayesian optimization criterion.
5. The intelligent towing pool device according to claim 4, characterized in that, The experimental parameter optimization unit operates according to the active learning sampling criterion, which is: within the feasible experimental parameter space, select the experimental parameter with the largest prediction uncertainty as the next set of experimental parameters.
6. The intelligent towing pool device according to claim 4, characterized in that, The experimental parameter optimization unit calculates the confidence interval width of the prediction results corresponding to the candidate experimental parameters based on the Gaussian process regression model, and uses the confidence interval width as a sampling index to determine the next set of experimental parameters.
7. The intelligent towing pool device according to claim 4, characterized in that, The experimental parameter optimization unit operates according to the Bayesian optimization criterion, which constructs the expected improvement function of the objective function based on the Gaussian process regression model, and determines the next set of experimental parameters by solving the expected improvement function.
8. The intelligent towing pool device according to claim 4, characterized in that, When determining the next set of experimental parameters, the experimental parameter optimization unit considers both the predicted mean and the predicted uncertainty, and constructs an evaluation function for selecting experimental parameters using a weighted approach.
9. The intelligent towing pool device according to claim 4, characterized in that, The control module is set with a convergence criterion based on prediction uncertainty. When the maximum prediction uncertainty of the Gaussian process regression model in the parameter space is lower than a preset threshold multiple times in a row, the experimental parameter optimization process is terminated.
10. A control method for an intelligent towed water tank device based on Gaussian process regression, characterized in that, The method includes: The experimental model was mounted in a towed water tank and the motion corresponding to the experimental parameters was executed. The hydrodynamic parameters of the experimental model during its motion are collected in real time. Based on the fluid dynamic parameters and corresponding experimental parameters collected by the data acquisition module, a Gaussian process regression model is constructed and updated, and the next set of experimental parameters to be executed is determined based on the updated Gaussian process regression model. The experimental execution module controls the corresponding experiment to complete the experiment based on the next set of experimental parameters output by the experimental parameter optimization unit.