Method and system for attitude monitoring and self-adaptive load adjustment in launching process of floating type wind power foundation

Abstract

The invention discloses an attitude monitoring and self-adaptive load adjusting method and system in the launching process of a floating type wind power foundation. The method comprises the following steps: acquiring height change and pressure distribution data of liquid in a ballast water tank, and combining basic six-degree-of-freedom pose data to form a multi-source sensing data set; extracting frequency, amplitude and energy distribution characteristics of free liquid level shaking by adopting time-frequency domain conjoint analysis based on the data set, and constructing a characteristic parameter set representing a shaking state; establishing a nonlinear coupling dynamic model of liquid level shaking in the ballast tank and basic six-degree-of-freedom motion; on the basis of the prediction model, a model prediction control algorithm is adopted to generate a suppression strategy containing a ballast water cross-cabin dispatching flow distribution scheme, and the suppression strategy is converted into a load regulation control instruction set; instructions are executed through a variable-frequency water pump set and a proportional control valve network, active suppression and accurate control over attitude disturbance in the foundation launching process are achieved, and the stability and safety of the launching process are improved.

Description

Technical Field

[0001] This invention belongs to the field of offshore wind power foundation technology, and in particular relates to a method and system for attitude monitoring and adaptive load adjustment during the launching process of a floating wind power foundation. Background Technology

[0002] As offshore wind power expands into deeper waters, floating wind turbine foundations have become an important technological direction. Floating wind turbine foundations face complex marine environmental challenges during launch, particularly the free surface sloshing within the ballast tanks. This sloshing generates additional torque, interfering with the foundation's attitude stability and potentially triggering resonance at certain frequencies, leading to attitude instability and seriously threatening the safety of launching operations.

[0003] Currently, attitude control of floating wind turbine foundations mainly relies on ballast water regulation systems. By controlling ballast pumps and valves in the pipelines, ballast water is transferred between different adjustable ballast tanks to maintain a balance of ballast water volume at the bottom of the column, thereby adjusting the foundation's attitude. However, this method is still based on a simple liquid level control strategy and cannot effectively suppress free surface sloshing, especially under harsh sea conditions.

[0004] Therefore, there is an urgent need for a method that can actively sense and suppress the free surface sloshing of ballast water and achieve intelligent load adjustment based on the dynamic needs during the launching process, so as to enhance the accuracy and robustness of attitude control. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a method and system for attitude monitoring and adaptive load adjustment during the launching process of a floating wind turbine foundation.

[0006] Firstly, this application provides a method for attitude monitoring and adaptive load adjustment during the launching process of a floating wind turbine foundation, including: S1. Collect dynamic change data of liquid height and pressure distribution data in the tanks by a distributed sensor array deployed on the inner wall of each ballast water tank of the floating wind power foundation, and collect six-degree-of-freedom pose data of the foundation by an inertial measurement unit installed on the foundation structure of the floating wind power foundation, so as to obtain the original multi-source sensor dataset containing the dynamic characteristics of the liquid surface and the attitude of the foundation. S2. Based on the original multi-source sensor dataset, the time-frequency domain joint analysis method is used to extract the sloshing frequency characteristics, fluctuation amplitude characteristics and energy distribution characteristics of the free liquid surface in each ballast water tank, and obtain the characteristic parameter set characterizing the sloshing state of the liquid surface. S3. Based on the feature parameter set and basic attitude history data, a nonlinear dynamic model describing the coupling relationship between free liquid surface sloshing and basic six-degree-of-freedom motion in the ballast water tank is established to obtain a real-time coupled prediction model for predicting the influence of specific sloshing modes on attitude. S4. Based on the real-time coupled prediction model, a model predictive control algorithm is used to generate a suppression strategy for the current swaying mode. Based on the suppression strategy, an optimal load adjustment control command set is generated to offset the disturbance torque. Among them, the suppression strategy includes the target flow allocation scheme for ballast water cross-tank transportation. S5. Execute the optimal load adjustment control command set through the variable frequency water pump group and proportional control valve network to establish a controlled water flow channel between the target ballast water tanks in order to adjust the liquid level distribution of each tank.

[0007] Secondly, this application also provides a floating wind turbine foundation attitude monitoring and adaptive load adjustment system for launching process, used to implement the method described in the first aspect, the system comprising: The dynamic multi-source data sensing module is used to collect dynamic changes in liquid height and pressure distribution data in the ballast water tanks of the floating wind power foundation through a distributed sensor array deployed on the inner wall of each ballast water tank. It also collects six-degree-of-freedom pose data of the foundation through an inertial measurement unit installed on the foundation structure of the floating wind power foundation, thus obtaining a raw multi-source sensing dataset containing the dynamic characteristics of the liquid surface and the attitude of the foundation. The liquid surface dynamic feature extraction module is used to extract the sloshing frequency features, fluctuation amplitude features, and energy distribution features of the free liquid surface in each ballast water tank based on the original multi-source sensor dataset and using the time-frequency domain joint analysis method, so as to obtain a set of feature parameters characterizing the liquid surface sloshing state. The coupled dynamics prediction module is used to obtain a real-time coupled prediction model for predicting the impact of a specific swaying mode on attitude by establishing a nonlinear dynamic model that describes the coupling relationship between the free liquid surface swaying and the basic six-degree-of-freedom motion in the ballast water tank, based on the feature parameter set and the basic attitude history data. The adaptive load adjustment decision module is used to generate a suppression strategy for the current swaying mode based on a real-time coupled prediction model and a model predictive control algorithm. Based on the suppression strategy, it generates an optimal load adjustment control instruction set to counteract the disturbance torque. The suppression strategy includes a target flow allocation scheme for ballast water cross-tank transport. The liquid level dynamic adjustment execution module is used to execute the optimal load adjustment control command set through the variable frequency water pump group and proportional control valve network to establish a controlled water flow channel between the target ballast water tanks in order to adjust the liquid level height distribution of each tank.

[0008] The beneficial effects of this invention are as follows: As can be seen from the above scheme, the floating wind turbine foundation attitude monitoring and adaptive load adjustment method and system provided by the present invention collects real-time data on liquid height changes and pressure distribution in the ballast water tank through a distributed sensor array, and obtains six-degree-of-freedom posture data of the foundation by combining it with an inertial measurement unit to form a multi-source sensor dataset. Based on this dataset, the frequency, amplitude and energy distribution characteristics of free liquid surface sloshing are extracted by time-frequency domain joint analysis to construct a set of characteristic parameters characterizing the sloshing state. Then, a nonlinear coupled dynamic model of liquid surface sloshing in the ballast tank and the six-degree-of-freedom motion of the foundation is established to realize real-time prediction of the attitude influence of specific sloshing modes. Based on this prediction model, a model predictive control algorithm is used to generate a suppression strategy including a ballast water cross-tank flow distribution scheme, and converts it into an optimal load adjustment control command set. Finally, the commands are executed by a variable frequency pump group and a proportional control valve network to establish a controlled water flow channel between the target ballast tanks to dynamically adjust the liquid level distribution, thereby realizing active suppression and precise control of attitude disturbances during the floating wind turbine foundation launching process, significantly improving the stability and safety of the launching process. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating a method for attitude monitoring and adaptive load adjustment during the launching process of a floating wind power foundation, provided by the present invention. Figure 2 This is a schematic diagram of the process of constructing a real-time coupled prediction model in one optional embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a floating wind power foundation attitude monitoring and adaptive load adjustment system provided by the present invention. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in this embodiment will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0011] refer to Figure 1 The document presents a flowchart illustrating a method for attitude monitoring and adaptive load adjustment during the launching process of a floating wind turbine foundation, as provided in this application. The method includes the following steps: S1. Data on the dynamic changes in liquid height and pressure distribution inside the ballast tanks of the floating wind turbine foundation are collected by a distributed sensor array deployed on the inner wall of each ballast tank. The six-degree-of-freedom pose data of the foundation is collected by an inertial measurement unit installed on the foundation structure of the floating wind turbine foundation. The resulting raw multi-source sensor dataset contains the dynamic characteristics of the liquid surface and the attitude of the foundation.

[0012] Specifically, the core of this step lies in constructing a multi-source data acquisition system to achieve synchronous and accurate perception of the liquid state and basic attitude within the ballast water tank. Its technical principle is based on the fusion of distributed sensing and inertial measurement, ensuring data validity according to sensor selection principles, installation structure logic, and data acquisition processes. Specifically, the selection of the distributed sensor array is adapted to the marine environment and measurement accuracy requirements. Dynamic liquid height change data acquisition can utilize submersible hydrostatic level sensors, while pressure distribution data acquisition can utilize thin-film pressure sensors. The installation of the sensor array should be combined with the structural characteristics of the ballast water tank. Level sensors can be arranged in layers along the height of the tank wall and fixed to the inner wall using corrosion-resistant brackets. Shock-damping components are added between the brackets and the tank wall to reduce tank vibration interference. Pressure sensors can be evenly pasted onto the central area of ​​the tank bottom and the surrounding side walls. Before pasting, the tank wall surface is derusted and adhesive is applied to ensure a good fit. The inertial measurement unit (IMU) has the capability to measure triaxial acceleration and triaxial angular velocity. The installation location is selected at the steel structure of the naval base directly above the center of gravity of the floating foundation. It is fixed with high-strength bolts, and heat insulation components are installed between the IMU shell and the steel structure to avoid the impact of temperature changes on measurement accuracy.

[0013] During data acquisition, a unified sampling sequence is set. All sensors are connected to the data acquisition card via a bus, and the acquisition card communicates with the central controller via Ethernet. To eliminate noise interference in the raw data, the acquired data is preprocessed. For level and pressure data, a low-pass filter is used to remove high-frequency vibration noise, and for IMU data, a Kalman filter algorithm is used for data fusion. During the Kalman filtering process, the state vectors are set as roll angle, pitch angle, yaw angle, roll angular velocity, pitch angular velocity, and yaw angular velocity. The filter gain matrix is ​​calculated using the following formula:

[0014] In the above formula, K(k) represents the filter gain matrix at time k, which is used to adjust the state estimate to reduce the error; P(k|k-1) represents the state covariance matrix at time k based on the observation results at time k-1, reflecting the uncertainty of the current state estimate; H represents the observation matrix, which is used to establish the mapping relationship between the state vector and the observation value; R represents the observation noise covariance matrix, which reflects the statistical characteristics of noise during the observation process.

[0015] The resulting raw multi-source sensor dataset can contain fields such as sensor number, acquisition timestamp, liquid level value, pressure value, and six-degree-of-freedom pose value, and is stored in industrial-grade storage media for subsequent processing.

[0016] S2. Based on the original multi-source sensor dataset, the time-frequency domain joint analysis method is used to extract the sloshing frequency characteristics, fluctuation amplitude characteristics and energy distribution characteristics of the free liquid surface in each ballast water tank, so as to obtain the characteristic parameter set characterizing the sloshing state of the liquid surface.

[0017] Specifically, this step aims to extract liquid surface sloshing features from the raw data through joint time-frequency domain analysis. Its technical principle is based on the time-frequency analysis theory in signal processing, and the core is to achieve accurate characterization of liquid surface dynamics through multi-domain feature fusion.

[0018] The time-frequency domain joint analysis method uses a combination of wavelet transform (WT) and short-time Fourier transform (STFT). STFT is used to initially obtain the time evolution law of frequency distribution, while wavelet transform is used to improve the frequency resolution of the low-frequency band to capture small-amplitude swaying characteristics. During the implementation of STFT, an appropriate window function is selected, and a reasonable window length and overlap rate are set to avoid frequency leakage. Through STFT, the spectrum of the liquid level height signal at different times can be obtained, thereby determining the approximate range of swaying frequency.

[0019] Wavelet transform is used with a wavelet basis possessing good time-frequency localization characteristics. An appropriate number of decomposition levels is set, and the liquid level height signal is decomposed into approximate and detail components through multi-scale decomposition. After reconstructing each detail component, the power spectral density (PSD) of each reconstructed component is calculated. The swaying frequency characteristics are determined by the PSD peak values, where the dominant frequency is the frequency corresponding to the maximum PSD peak, and the secondary dominant frequency is the frequency corresponding to the second largest peak. The specific relationship between the dominant and secondary dominant frequencies is recorded. The fluctuation amplitude characteristics are extracted based on the preprocessed liquid level height data. The difference between the liquid level height and the steady-state liquid level at each sampling time is calculated. The fluctuation amplitude characteristics include the maximum fluctuation amplitude, the average fluctuation amplitude, and the standard deviation of the fluctuation amplitude. Outliers are removed during the calculation process.

[0020] The energy distribution characteristics are calculated based on the energy proportion of each frequency component, and the total energy is calculated using the following formula:

[0021] In the above formula, This represents the total energy of the liquid surface sloshing. This represents the difference between the liquid level at time t and the steady-state liquid level.

[0022] The energy of each frequency component is calculated using the following formula:

[0023] In the above formula, This represents the energy of the i-th frequency component. This represents the value of the reconstructed signal of the i-th detail component at time t.

[0024] The energy percentage is calculated using the following formula:

[0025] In the above formula, This represents the energy percentage of the i-th frequency component.

[0026] Energy distribution characteristic parameters may include the energy proportion of the dominant frequency component, the total energy proportion of the top three frequency components, and the energy distribution entropy. The energy distribution entropy is calculated using the following formula:

[0027] In the above formula, Entropy represents the energy distribution and is used to characterize the degree of energy dispersion.

[0028] The final set of feature parameters can include swaying frequency features, fluctuation amplitude features, and energy distribution features. The ballast water tank number and extraction time interval corresponding to the feature extraction are also labeled to ensure the correlation between the feature parameters and the original data.

[0029] S3. Based on the feature parameter set and basic attitude history data, a nonlinear dynamic model describing the coupling relationship between the free liquid surface sloshing and the basic six-degree-of-freedom motion in the ballast water tank is established to obtain a real-time coupled prediction model for predicting the influence of specific sloshing modes on attitude.

[0030] Specifically, the key to this step is to construct a coupled prediction model of free surface sloshing and basic motion. Its technical principle is based on nonlinear dynamics theory and system identification method. By setting the model structure, parameter identification process and model verification standards, it is ensured that the model can accurately predict the impact of sloshing on attitude.

[0031] The nonlinear dynamic model is constructed based on potential flow theory and combined with the rigid body motion equation of the floating foundation. It considers the two-way coupling effect between free surface sloshing and foundation motion. The state variables of the model are selected as the free surface sloshing amplitude, sloshing frequency, foundation roll angle, pitch angle, roll angular velocity, and pitch angular velocity in the ballast water tank. The input variables of the model are marine environmental excitations, which are collected by wave sensors deployed near the foundation. The output variables are the predicted roll angle and pitch angle values ​​of the foundation in a specific future time period.

[0032] The mathematical expression of the model is based on Taylor expansion to construct nonlinear terms. The coupled motion equations in the roll direction are as follows:

[0033] In the above formula, It represents the moment of inertia of the foundation about the roll axis, reflecting the magnitude of the inertia of the foundation rotating about that axis; It represents the second derivative of the basic roll angle at time t, i.e., the roll acceleration; It represents the damping coefficient in the roll direction, reflecting the magnitude of the damping effect that opposes the roll motion of the foundation; It represents the first derivative of the basic roll angle at time t, i.e., the roll angular velocity; It represents the roll stiffness coefficient, reflecting the foundation's resistance to deformation in the roll direction; Represents the basic roll angle at time t; It represents the coupling torque of the free liquid surface sloshing on the foundation, reflecting the interaction between the liquid surface sloshing and the foundation motion; This represents the disturbance torque of the wave on the foundation at time t, reflecting the influence of the marine environment on the foundation's attitude.

[0034] The coupling torque of the free liquid surface sloshing on the foundation is calculated using the following formula:

[0035] In the above formula, Indicates the density of seawater; Represents gravitational acceleration; Indicates the volume of ballast water; Indicates the amplitude of free surface sloshing; The frequency of free surface sloshing is represented by t; time is represented by t. This indicates the phase difference, reflecting the time lag between the sway and the torque.

[0036] The wave disturbance moment on the foundation is calculated using the following formula:

[0037] In the above formula, It represents the moment of inertia of the waterline surface, reflecting the influence of the waterline surface on the foundation attitude; Represents wave height; Indicates the wave period; It indicates the wave phase, reflecting the time characteristics of wave motion.

[0038] The coupled motion equations in the pitch direction are similar to those in the roll direction, except that the moment of inertia, damping coefficient, stiffness coefficient, and waterplane moment of inertia of the corresponding axes are replaced.

[0039] Model parameter identification employs a combination of least squares and particle swarm optimization (PSO) algorithms. Historical data is selected from monitoring data collected during multiple underwater operations, covering various sea states. During identification, linear parameters are first estimated using least squares. Then, using these initial estimates as initial values, the PSO algorithm is applied to optimize nonlinear parameters. The objective function is set to minimize the root mean square error (RMSE) between predicted and actual values. Model validation utilizes cross-validation, dividing historical data into multiple groups, with each group alternating between the validation and training sets. Validation metrics include RMSE, mean absolute error (MAE), and coefficient of determination. If the verification metrics do not meet the requirements, the model structure is readjusted and the identification process is repeated until the requirements are met. Finally, the obtained real-time coupled prediction model is stored in a callable file format for subsequent steps.

[0040] S4. Based on the real-time coupled prediction model, a model predictive control algorithm is used to generate a suppression strategy for the current swaying mode. Based on the suppression strategy, an optimal load adjustment control command set is generated to offset the disturbance torque. The suppression strategy includes a target flow allocation scheme for ballast water cross-tank transportation.

[0041] Specifically, this step aims to generate the optimal load adjustment command through model predictive control. Its technical principle is based on the rolling time-domain optimization idea of ​​model predictive control (MPC), and the core is to generate a load adjustment strategy that suppresses swaying under the condition of satisfying constraints.

[0042] The rolling time domain of the model predictive control algorithm is set to include the prediction time domain and the control time domain. The prediction time domain corresponds to the time period that needs to be predicted in the future, and the control time domain corresponds to the time period that needs to output control quantities. The optimization cycle is kept consistent with the prediction step size to ensure real-time performance.

[0043] The objective function is constructed by comprehensively considering attitude stability and load adjustment energy consumption, and adopts multi-objective weighted optimization, as shown in the following formula:

[0044] In the above formula, This represents the overall objective function value, used to measure the comprehensive performance of the load adjustment strategy; , , These represent the weighting coefficients for attitude deviation cost, load adjustment energy consumption cost, and control quantity change rate cost, respectively, to ensure that attitude stability is given priority. This represents the cost of attitude deviation, reflecting the degree of deviation between the actual attitude and the target attitude; This indicates the energy consumption cost of load adjustment, reflecting the energy consumption during the load adjustment process; It represents the cost of the rate of change of the control quantity, reflects the stability of the change of the control quantity, and avoids system shocks caused by sudden changes in the control quantity.

[0045] The attitude deviation cost is calculated using the following formula:

[0046] In the above formula, This represents the number of sampling points in the prediction time domain; This represents the predicted roll angle at time i within the prediction time domain; This represents the target roll angle, which can usually be set to the angle corresponding to the horizontal attitude. This represents the predicted pitch angle at the i-th time point in the prediction time domain; This represents the target value for the pitch angle, which can also usually be set to the angle corresponding to the horizontal attitude.

[0047] Load adjustment energy consumption cost is calculated using the following formula:

[0048] In the above formula, This indicates the number of sampling points in the control time domain; This represents the power of the variable frequency pump at the i-th time point in the control time domain, reflecting the energy consumption of the pump.

[0049] The cost of controlling the rate of change of quantity is calculated using the following formula:

[0050] In the above formula, It represents the difference in flow rate between the i-th time point and the (i-1)-th time point within the control time domain, reflecting the stability of flow rate changes.

[0051] The constraints include flow rate constraints, liquid level constraints, valve opening constraints, and torque balance constraints. Flow rate constraints limit the flow rate range of the variable frequency pump to prevent idling or overload. Liquid level constraints limit the liquid level range of the ballast tank to prevent overflow or pump cavitation; the relationship between liquid level and flow rate can be determined by considering the tank volume and time step. Valve opening constraints limit the opening range of the proportional control valve to ensure normal valve operation. Torque balance constraints require that the restoring torque generated by load adjustment is not less than the disturbance torque, with a safety margin to ensure stability. The restoring torque is calculated using the following formula:

[0052] In the above formula, This indicates the restoring torque generated by load adjustment; Indicates the amount of water to be adjusted; This indicates the distance between the center of gravity of the load-carrying water and the center of gravity of the foundation, reflecting the degree of influence of the load-carrying water on the foundation moment.

[0053] The target flow allocation scheme in the suppression strategy is calculated based on the direction and magnitude of the disturbance torque. The target receiving tank and target discharge tank are determined according to the basic attitude deviation. The required load adjustment water volume is calculated based on the torque balance relationship, and the target flow rate is determined according to the load adjustment time. If multiple ballast tanks are involved, a weighted allocation method is used to determine the target flow rate of each tank based on multi-directional attitude deviations. When generating the optimal load adjustment control instruction set, the target flow allocation scheme is converted into specific equipment instructions, including the target speed of the variable frequency pump, the target opening degree of the proportional control valve, and the instruction execution sequence. The instruction set can include equipment number, target parameters, execution timestamp, and duration, and is stored in a standard file format for subsequent steps by the equipment control system.

[0054] S5. Execute the optimal load adjustment control command set through the variable frequency water pump group and proportional control valve network to establish a controlled water flow channel between the target ballast water tanks in order to adjust the liquid level distribution of each tank.

[0055] Specifically, this step is the physical execution of the load adjustment command. Its technical principle is based on fluid transport and automatic control technology. The core is to achieve precise liquid level regulation through the coordinated work of variable frequency water pump set and proportional control valve network.

[0056] The selection of variable frequency water pump sets is adapted to the requirements of load adjustment flow and head, and is adaptable to marine environments. Mechanical seals are used to adapt to seawater transportation. The arrangement of the water pump sets can be determined in combination with the number of ballast water tanks to ensure that the load adjustment requirements of all tanks are covered. The installation position is located in the equipment compartment at the bottom of the foundation. The water pump inlet and outlet are connected to the pipeline through flexible joints to absorb vibration. The water pump foundation is made of rigid material and fixed by pre-embedded bolts.

[0057] The proportional control valve network features rapid response and corrosion resistance. Each ballast water tank is equipped with one proportional control valve at its inlet and outlet, installed near the bottom of the tank to ensure sufficient fluid flow. Valves are connected to pipelines via flanges, using gaskets to ensure a tight seal, and secured with bolts to guarantee connection strength. The piping system uses seamless steel pipes made of corrosion-resistant materials. The pipe diameter is calculated based on flow requirements, and the wall thickness is calculated based on pressure requirements. Pipe connections are made by welding, and the welds are inspected to ensure connection quality. The piping system is equipped with pressure gauges and flow meters; the pressure gauges monitor pipeline pressure, and the flow meters monitor pipeline flow, ensuring real-time monitoring of the pipeline's operating status.

[0058] The execution process is divided into three stages: The first stage is instruction reception and parsing. The equipment control system receives the instruction set generated in the previous steps through the communication network, parses out the target equipment, target parameters, and execution sequence, and sends a ready signal back to the central controller after parsing. The second stage is equipment startup and adjustment. According to the preset sequence, the outlet valve of the target drainage tank is first controlled to open to the target opening degree. After a certain interval, the corresponding strain frequency pump is started. After another certain interval, the inlet valve of the target water receiving tank is controlled to open to the target opening degree. At this time, a controlled water flow channel is established between the drainage tank, the pump, and the water receiving tank. During the adjustment process, the flow meter collects the actual flow rate in real time and feeds it back to the control system. If the deviation between the actual flow rate and the target flow rate exceeds the allowable range, the pump speed is adjusted through the PID algorithm to ensure stable flow. The third stage is liquid level monitoring and shutdown. The liquid level sensor collects the liquid level height of each tank in real time. When the liquid level of the water receiving tank and the drainage tank both reach the target height, the control system first controls the pump speed to reduce to the minimum operating speed, maintains it for a certain period of time, and then shuts down the pump. Then, the inlet valve of the water receiving tank and the outlet valve of the drainage tank are closed in sequence. After the execution process is completed, a completion signal is sent back to the central controller.

[0059] The feedback adjustment mechanism is used to deal with abnormal situations during the execution process: if the pressure gauge detects that the pipeline pressure exceeds the safety threshold, the control system immediately shuts down the water pump and opens the pressure relief valve. After the pressure drops to the safe range, it checks whether the valve opening is abnormal. If there is an abnormality, an alarm signal is issued. If the liquid level sensor detects that the liquid level in a certain compartment is outside the normal range, the control system adjusts the opening of the corresponding valve and reduces the water pump speed until the liquid level returns to the normal range, ensuring the safe and stable loading process.

[0060] The aforementioned method for attitude monitoring and adaptive load adjustment during the launching process of a floating wind turbine foundation utilizes a distributed sensor array to collect real-time data on liquid height changes and pressure distribution within the ballast water tank. Combined with inertial measurement unit (IMU) data on the foundation's six degrees of freedom (DOF) pose, a multi-source sensor dataset is generated. Based on this dataset, time-frequency domain joint analysis is employed to extract the frequency, amplitude, and energy distribution characteristics of free surface sloshing, constructing a set of characteristic parameters representing the sloshing state. Furthermore, a nonlinear coupled dynamic model of ballast tank surface sloshing and foundation six-DOF motion is established to achieve real-time prediction of the attitude impact of specific sloshing modes. Based on this prediction model, a model predictive control algorithm is used to generate a suppression strategy including a ballast water cross-tank flow distribution scheme, which is then transformed into an optimal load adjustment control command set. Finally, the commands are executed through a variable frequency pump group and a proportional control valve network to establish a controlled water flow channel between the target ballast tanks, dynamically adjusting the liquid level distribution. This achieves active suppression and precise control of attitude disturbances during the launching process of the floating wind turbine foundation, significantly improving the stability and safety of the launching process.

[0061] refer to Figure 2In one optional embodiment, based on a set of characteristic parameters and historical basic attitude data, a real-time coupled prediction model for predicting the impact of a specific swaying mode on attitude is obtained by establishing a nonlinear dynamic model describing the coupling relationship between free surface swaying and basic six-degree-of-freedom motion within the ballast tank. This includes the following steps: S11. Based on the geometric configuration parameters and material property data of the ballast water tank, the fluid domain is spatially discretized using the finite volume method to generate a three-dimensional numerical calculation grid containing a boundary layer mesh.

[0062] Specifically, the core of this step is to discretize the fluid domain and generate a mesh using the finite volume method, providing an accurate computational platform for subsequent flow field simulation. Its technical principle is based on the concept of control volume integral in computational fluid dynamics (CFD).

[0063] The geometric configuration parameters of ballast water tanks encompass the three-dimensional dimensional features of the tank (such as tank length, width, height, bulkhead thickness, and the location and size of inlet and outlet pipe interfaces) and the distribution of internal structures (such as the presence of baffles, support ribs, and their geometric parameters). These parameters can be digitally modeled using 3D modeling software (such as SolidWorks and AutoCAD) to form a CAD model containing complete geometric information. Material property data includes the density, elastic modulus, and Poisson's ratio of the ballast water tank shell material (used to subsequently consider the impact of structural deformation on the fluid domain) and the density and dynamic viscosity of the ballast water (used to characterize the physical properties of the fluid and affect the flow field motion).

[0064] The spatial discretization of the finite volume method follows the principle of "control volume conservation," dividing the fluid domain (i.e., the space occupied by the liquid within the ballast water tank) into non-overlapping three-dimensional control volume meshes. In practice, the computational domain boundaries (including the inner surface of the bulkhead, the initial position of the free surface of the liquid, and the inlet and outlet boundaries) are first determined based on the CAD model. Then, a hybrid structured and unstructured mesh strategy is adopted: for the near-bulk wall region, due to the large liquid flow gradient (which easily forms a boundary layer effect), a structured mesh is used for refinement. The thickness of the refined region is determined based on the boundary layer thickness (the boundary layer thickness is calculated using the formula...). Estimate, of which (where t is the kinematic viscosity of the ballast water and t is the flow time). The mesh size gradually transitions from near the wall outwards, for example, ensuring that the near-wall mesh meets the following requirements. The value (dimensionless wall distance) is in the range of 30-100 to accurately capture the wall shear stress; for areas without complex structures inside the cabin, an unstructured tetrahedral mesh is used, and the mesh size is set according to the flow field uniformity requirements, for example, to ensure that the volume difference between adjacent control volumes does not exceed 30%, so as to avoid a decrease in computational convergence due to abrupt mesh changes.

[0065] After the mesh is generated, its quality is verified. Verification indicators include mesh orthogonality (e.g., the angle between adjacent mesh faces should be within the range of 45°-135°), twist rate (e.g., the maximum twist of a mesh cell should be less than 0.8), and aspect ratio (e.g., the aspect ratio of a mesh cell should be less than 5). These indicators can be quantitatively evaluated using mesh quality analysis software (such as Fluent Meshing's mesh inspection tool). If there are unqualified meshes, they can be optimized by local mesh re-meshing, node adjustment, etc. Finally, a three-dimensional numerical computation mesh containing a boundary layer refinement mesh is generated. This mesh is stored in a common mesh format (such as CGNS or ANSYS Mesh format) for subsequent flow field simulation software to use.

[0066] S12. Calculate the disturbance torque vector generated by liquid sloshing on the floating wind power foundation based on the characteristic parameter set, and load the disturbance torque vector as a dynamic boundary condition onto the three-dimensional numerical calculation grid; based on the three-dimensional numerical calculation grid, simulate the unsteady flow field of liquid sloshing by solving the transient Reynolds-averaged Navier-Stokes equations, and output the predicted data of the center of mass motion trajectory of the liquid.

[0067] Specifically, this step aims to obtain the trajectory of the liquid centroid by applying disturbance torque and simulating transient flow field. Its technical principle is based on the law of conservation of momentum and the Navier-Stokes equations in fluid dynamics.

[0068] First, the calculation of the disturbance torque vector is based on the characteristic parameter set obtained in step S2. Parameters in the characteristic parameter set, such as sloshing frequency, fluctuation amplitude, and energy distribution, directly determine the magnitude and direction of the force exerted by the liquid sloshing on the foundation. Specifically, the disturbance torque vector... It is a vector quantity containing three components along the basic roll axis, pitch axis, and heave axis. Its calculation is based on the change in angular momentum of the liquid sloshing, with the component along the roll axis being... The following formula is derived:

[0069] In the above formula, This represents the position vector of the fluid element relative to the basic roll axis. For the density of ballast water, Let the velocity vector be the fluid element. For the fluid domain volume, The unit vector along the roll axis, the integral operation covers the entire fluid domain, and the time derivative... It reflects the rate of change of angular momentum over time, i.e., the disturbance torque.

[0070] In actual calculations, the fluctuation amplitude of the characteristic parameter set can be considered. With shaking frequency Simplifying the derivation, by approximating the liquid sloshing as simple harmonic motion, we obtain... ,in Angular frequency, Let be the distance from the center of mass of the liquid to the roll axis. The phase difference; the calculation logic for the disturbance torque components in the pitch and heave axes is similar, ultimately forming a complete disturbance torque vector. .

[0071] When the disturbance torque vector is applied as a dynamic boundary condition, a momentum source term is defined on the bulkhead boundary of the three-dimensional numerical calculation grid. This is implemented using a user-defined function (UDF) in flow field simulation software (such as ANSY SFluent or Open FOAM), which accurately transmits the time-varying behavior of the torque vector (such as harmonic variation or random fluctuation) to the fluid domain.

[0072] The transient Reynolds-averaged Navier-Stokes (RANS) equations are then solved to simulate the unsteady flow field. The transient RANS equations include the continuity equation and the momentum equation.

[0073] The continuity equation is as follows:

[0074] The momentum equation is as follows:

[0075] In the above formula, t represents time. For hydrostatic pressure, For fluid dynamic viscosity, It is the Reynolds stress tensor (closed by a turbulence model, such as the k-ε two-equation model, where k is the turbulent kinetic energy and ε is the turbulent kinetic energy dissipation rate). It is the gravitational acceleration vector. The volumetric force source term corresponding to the application of the disturbance torque.

[0076] In the solution process, the implicit Euler scheme can be used for time discretization, the second-order upwind scheme can be used for spatial discretization, and the SIMPLE algorithm can be used for pressure-velocity coupling. The pressure and velocity can be solved in a coordinated manner through the pressure correction equation.

[0077] Convergence criteria are set during the simulation iteration process. For example, the residuals between the continuity equation and the momentum equation must be less than [a certain value]. The residuals of the turbulence model equations (k-equation, ε-equation) must be less than Simultaneously monitor the position changes of the free liquid surface to ensure that the liquid surface fluctuations reach a stable state, such as when the liquid surface position deviation between adjacent time steps is less than [value missing]. Double cabin height.

[0078] After the simulation is completed, the predicted trajectory data of the liquid center of mass is output, which includes the three-dimensional position coordinates of the center of mass in the basic local coordinate system. Three-dimensional velocity vector and three-dimensional acceleration vector The curve showing the change over time provides a foundation for subsequent data fusion.

[0079] S13. The predicted data of the center of mass trajectory is spatiotemporally aligned with the six-degree-of-freedom pose data collected in real time by the inertial measurement unit to obtain a prediction-measurement dataset with unified spatiotemporal reference. An extended Kalman filter is used with the prediction-measurement dataset as input, and the predicted value of the liquid center of mass trajectory is used as the state prediction quantity and the measured value of the basic pose is used as the observation quantity to perform data fusion and generate the fused optimized pose state estimation result.

[0080] Specifically, this step achieves data fusion through spatiotemporal alignment and extended Kalman filtering (EKF) to generate optimized pose state estimation results. Its technical principle is based on spatiotemporal consistency calibration of multi-source data and state estimation theory of nonlinear systems.

[0081] The core of spatiotemporal alignment is to unify the temporal and spatial references of the predicted center-of-mass trajectory data and the six-DOF pose data. Temporally, the time step of the predicted center-of-mass trajectory (determined by the time interval of the flow field simulation) is synchronized with the sampling step of the inertial measurement unit (IMU). Linear interpolation is used to resample data with lower sampling frequencies to ensure a one-to-one correspondence between the two types of data at the same timestamp. Spatially, a coordinate transformation is performed between the local coordinate system of the predicted center-of-mass trajectory (with the geometric center of the ballast tank as the origin) and the measurement coordinate system of the IMU (with the center of gravity of the foundation as the origin). The transformation matrix is ​​based on the relative position vectors of the two coordinate systems. With attitude angle The construction and conversion formula is as follows ,in Let the position vector of the centroid in the prediction coordinate system be denoted as . Euler angle rotation matrix (rotation about the x-axis) angular and y-axis rotation Angle and z-axis rotation horn), This is the transformed position vector in the IMU coordinate system. This transformation achieves the unification of the spatial reference, ultimately resulting in a prediction-measurement dataset with consistent spatiotemporal reference.

[0082] Extended Kalman filtering is applied to address the nonlinear coupling characteristics between liquid sloshing and fundamental motion, by modifying the system state vector... Defined as the fusion vector of the predicted trajectory of the liquid centroid and the basic pose state, i.e. ,in Based on the roll angle, pitch angle, and yaw angle This corresponds to the angular velocity.

[0083] The EKF algorithm process is divided into a prediction step and an update step: 1) In the prediction step, the state vector and covariance matrix are predicted based on the system dynamics model.

[0084] The state prediction equation is as follows:

[0085] The covariance prediction equation is as follows:

[0086] In the above formula, This is the nonlinear state transition function of the system, used to describe the dynamic relationship between the center of mass motion and the basic pose, and is constructed based on the dynamic relationship of the flow field simulation in S12; This is the control input at time k-1, which is the disturbance torque vector. The noise is the process noise, which follows a mean of 0 and a covariance of . The Gaussian distribution reflects the uncertainty in the flow field simulation; for exist The Jacobian matrix at the given location is used for linearization of nonlinear systems; Let be the state covariance matrix at time k-1. Let be the prediction covariance matrix at time k.

[0087] 2) In the update step, the predicted state is corrected based on the observed values ​​to obtain the optimal estimate.

[0088] The observation equation is as follows:

[0089] The filter gain is calculated as follows:

[0090] The state update equation is as follows:

[0091] The covariance update equation is as follows:

[0092] In the above formula, The observation value at time k is the six-degree-of-freedom pose data acquired by the IMU. This is the observation function, used to establish the mapping relationship between the state vector and the observation values. Here, the pose parameters in the state vector are directly extracted. To observe the noise, it follows a pattern with a mean of 0 and a covariance of . The Gaussian distribution reflects the IMU measurement error; for exist Jacobian matrix at the location; This is the filter gain matrix, used to balance the confidence levels of predicted and observed values; The result is the optimized pose state estimation at time k; It is the identity matrix; Let be the updated covariance matrix at time k.

[0093] The optimized pose state estimation results not only include the corrected basic six-DOF pose parameters, but also the corresponding covariance matrix. The diagonal elements of this matrix reflect the variance (uncertainty) of the estimated values ​​of each state parameter, while the off-diagonal elements reflect the correlation between the parameters, providing a quantitative basis for uncertainty in subsequent model accuracy evaluation.

[0094] S14. Based on the residual vector of the optimized pose state estimation results and the centroid motion trajectory prediction data, calculate the covariance matrix of the output error of the fluid dynamics model, and output a confidence quantification index that characterizes the prediction accuracy of the model.

[0095] Specifically, this step achieves a quantitative assessment of the model's prediction accuracy by calculating the covariance matrix and confidence quantification index. Its technical principle is based on statistical error analysis theory.

[0096] residual vector The definition of is the deviation between the optimized pose state estimation result and the predicted center of mass trajectory data, i.e. ,in Let be the state prediction vector corresponding to the trajectory of the liquid center of mass at time k, containing the center of mass position, velocity, and the derived basic pose parameters. Each component of the residual vector corresponds to the prediction deviation of each parameter in the state vector, such as... The first component is the prediction bias of the centroid position in the x-direction, and the seventh component is the prediction bias of the basic roll angle. The residual vector can intuitively reflect the difference between the model prediction and the actual state after fusion.

[0097] Covariance matrix of output error of fluid dynamics model Calculations based on the statistical properties of residual vectors, assuming the residual sequence... (N is the number of data samples) The covariance matrix of the data satisfies the characteristics of a stationary random process. (The element in the i-th row and j-th column) is calculated using the following formula:

[0098] In the above formula, Let be the value of the i-th component of the residual vector at time k. Let be the mean of the i-th residual component. This formula quantifies the distribution characteristics of the model output error by calculating the covariance between different residual components, i.e., the diagonal elements. The off-diagonal elements represent the variance of the prediction error for the i-th state parameter (the larger the variance, the higher the prediction uncertainty for that parameter). Let N be the covariance of the prediction errors of the i-th and j-th parameters (reflecting the correlation between errors), and N be the total number of time steps.

[0099] The confidence metrics characterizing the model's prediction accuracy are derived from the covariance matrix, with core metrics including root mean square error (RMSE), confidence interval coverage (CICR), and normalized error (NE).

[0100] The root mean square error (RMSE) is used to measure the overall magnitude of the residuals, and its calculation formula is as follows:

[0101] in, Let be the RMSE value of the i-th state parameter. The smaller the value, the higher the prediction accuracy of the parameter. The RMSE threshold of each parameter can be set according to engineering requirements.

[0102] The confidence interval coverage (CICR) is used to assess the degree of calibration of model predictive uncertainty, for example, by calculating a 95% confidence interval based on the covariance matrix. The proportion of samples in the residual sequence that fall within this interval is:

[0103] Ideally The value should be close to 95%. If it is significantly lower than 95%, it indicates that the model underestimates the uncertainty of the prediction; if it is significantly higher than 95%, it indicates that the model overestimates the uncertainty.

[0104] The normalization error (NE) is used to eliminate the influence of dimensions and facilitate cross-parameter comparisons. The calculation formula is as follows:

[0105] in and These are the upper and lower limits of the physical value range for the i-th state parameter, respectively. The smaller the value, the smaller the proportion of the prediction error of the parameter relative to its value range, and the better the prediction accuracy.

[0106] By comprehensively calculating the above indicators, the model prediction accuracy can be quantified from three dimensions: error magnitude, uncertainty calibration, and relative accuracy, providing a basis for decision-making in subsequent model parameter correction.

[0107] S15. When the confidence quantification index is lower than the preset threshold, the equivalent damping coefficient and stiffness coefficient in the transient Reynolds-averaged Navier-Stokes equation are dynamically corrected based on the recursive least squares method to obtain a parameter-adaptive real-time coupled prediction model.

[0108] Specifically, this step uses recursive least squares (RLS) to dynamically correct model parameters and construct a parameter-adaptive real-time coupled prediction model. Its technical principle is based on online parameter estimation methods in adaptive control theory.

[0109] The preset threshold can be set according to the actual prediction accuracy requirements of the project. It can be based on the confidence quantification index in S14, such as setting the maximum allowable value of RMSE (e.g., the basic roll angle RMSE threshold is set to 1 / 3 of the allowable attitude deviation of the project) and the reasonable range of CICR (e.g., 85%-105%). When any confidence index exceeds the set range (i.e., below the accuracy threshold or deviates from the ideal range), it is determined that the model prediction accuracy is insufficient and the parameter correction process is triggered.

[0110] The parameter that needs to be corrected is the equivalent damping coefficient in the transient RANS equations. With stiffness coefficient These two parameters directly affect the energy dissipation and dynamic response characteristics of liquid sloshing in flow field simulation. (Equivalent damping coefficient) The stiffness coefficient reflects the energy loss during liquid sloshing (such as viscous dissipation and wall friction dissipation). A higher value indicates faster energy dissipation and a more significant decrease in sloshing amplitude. This reflects the fluid's elastic resistance to fundamental motion; the higher the value, the more sensitive the fluid is to changes in fundamental attitude. The core of parameter correction is to update the parameters online using the RLS algorithm and real-time acquired residual data. and This minimizes the model's output error.

[0111] When implementing the recursive least squares method, the parameter correction problem is transformed into a linear regression model, and the regression equation is defined as follows:

[0112] in, The regression output at time k is taken here as the residual vector in S14. Because the residuals directly reflect the deviation between the model parameters and the true values; Let be the regression matrix at time k, derived from the transient RANS equation and . , Related components, such as liquid sloshing velocity gradient and displacement gradient, are calculated based on intermediate variables from flow field simulation. Let be the parameter vector to be estimated. The regression error follows a Gaussian distribution with a mean of 0.

[0113] The core of the RLS algorithm is to update the parameter estimates recursively, avoiding the drawback of traditional least squares methods that require storing all historical data. The algorithm flow is as follows: 1) Initialization: Set initial parameter estimates It can be based on empirical values ​​or offline calibration results; initial gain matrix , For larger positive numbers, such as This ensures greater flexibility in initial parameter updates; forgetting factor A value of 0.95-0.99 is typically used to balance the weights of historical and current data. The closer it is to 1, the greater the weight of historical data.

[0114] 2) Recursive update: For each time step k, perform gain matrix update and parameter estimation update.

[0115] The gain matrix is ​​updated as follows:

[0116] The parameter estimates are updated as follows:

[0117] In the above formula, Let be the covariance matrix at time k, reflecting the uncertainty of parameter estimation; The parameter estimate at time k is used, and through this update equation, the parameters can be adjusted based on the current residuals. Adjustments are made to reduce the output error caused by parameter estimation bias.

[0118] After parameter correction, the updated version will be available. and The transient RANS equations are resubmitted to update the governing equations for the flow field simulation, resulting in a revised fluid dynamics model. Simultaneously, the revised model is combined with the EKF data fusion module in S13 to construct a real-time coupled prediction model. This model possesses parameter adaptive capabilities; when sea state changes (such as increased wave amplitude or frequency) lead to variations in liquid sloshing characteristics, the residuals increase accordingly, triggering a re-correction using the RLS algorithm. and This ensures that the model always matches the actual working conditions, maintains high prediction accuracy, and ultimately achieves accurate prediction of the impact of basic attitude on specific swaying modes.

[0119] In one optional embodiment, calculating the disturbance torque vector generated by liquid sloshing on the floating wind turbine foundation based on a set of characteristic parameters includes the following steps: S21. Based on the feature parameter set, the current swaying pattern type is identified by a support vector machine classifier, resulting in pattern recognition results including standing wave, traveling wave, or vortex.

[0120] Specifically, the selection of the feature parameter set focuses on parameters that can characterize the essential differences between different swaying modes. Key input features are selected from the feature parameter set obtained in step S2, including swaying frequency features (dominant frequency, secondary dominant frequency, and the difference between the dominant and secondary dominant frequencies), fluctuation amplitude features (maximum fluctuation amplitude, standard deviation of fluctuation amplitude), energy distribution features (energy distribution entropy, energy proportion of the dominant frequency component), and spatiotemporal characteristic parameters of liquid surface motion (phase difference between monitoring points on the liquid surface, motion synchronization coefficient). These parameters together constitute the input feature vector of the SVM classifier. Each component corresponds to the main frequency, secondary main frequency, main frequency difference, maximum fluctuation amplitude, amplitude standard deviation, energy distribution entropy, main frequency energy ratio, monitoring point phase difference, and synchronization coefficient, respectively. Multi-dimensional feature fusion ensures the significance of mode differentiation.

[0121] To eliminate the impact of different units on classification accuracy through feature preprocessing, standardization can be used to transform each feature component to a uniform numerical range. The standardization formula is as follows: ,in These are the original eigenvalues. This is the mean of the feature in the training set. The standard deviation is used to balance the weights of each feature, thus preventing the classification result from being dominated by a single feature whose value is too large.

[0122] The training dataset for the SVM classifier can cover typical samples of three swaying modes. The sample sources include two parts: first, physical model test data, such as using a 1:50 scale floating ballast tank model to simulate different sea states (standing waves for calm sea states, traveling waves for regular waves, and vortices for irregular waves with strong currents) in a wave tank, collecting feature parameters under each sea state as real samples; second, numerical simulation data, which uses CFD software to simulate the surface motion under different swaying modes, generating a large number of supplementary samples. The two types of samples together constitute the training set, of which 70% can be used for model training and 30% can be used for the validation set to evaluate classification accuracy.

[0123] To adapt the kernel function to the nonlinear characteristics of sway pattern classification, the radial basis function (RBF) can be used as the kernel function of SVM, and its expression is as follows: ,in , Given two input feature vectors, These are kernel function parameters (used to control the radial influence range of the function). Let be the Euclidean norm. Parameter optimization is achieved through five-fold cross-validation, with classification accuracy as the objective function, and the penalty parameter C (which controls the trade-off between classification margin and misclassification rate) is adjusted accordingly. A grid search was performed to determine the parameter combination that resulted in a classification accuracy of over 95% on the validation set, ensuring that the classifier had good generalization ability for unseen shaking samples.

[0124] The classification decision process consists of two steps: First, the preprocessed feature parameters collected in real-time are input into the trained SVM classifier. The classifier calculates the kernel function values ​​of the input vector and each support vector, combined with the optimal classification hyperplane equation. Output the classification results, where For the number of support vectors, Lagrange multipliers (only support vectors corresponding to each other) (non-zero) b represents the class labels for the support vectors (standing wave type is labeled 1, traveling wave type is labeled 2, and vortex type is labeled 3), and b represents the bias term of the classification hyperplane. The first step is to define the sign function; the second step is to verify the reliability of the classification results by calculating the distance from the input vector to the classification hyperplane. ( (The normal vector of the classification hyperplane) When the distance is greater than the preset threshold, the pattern recognition result is directly output; if the distance is less than the threshold, the risk of misclassification is reduced by fusing the classification results of adjacent time moments (the pattern that appears most frequently in 5 consecutive time moments can be selected), and finally a clear shaking pattern type (standing wave, traveling wave or vortex) is output, which provides a basis for the selection of subsequent torque calculation model.

[0125] S22. Based on the pattern recognition results, select the corresponding torque calculation model and combine the angular acceleration data collected by the inertial measurement unit to solve the disturbance torque vector generated by the liquid sloshing.

[0126] Specifically, the model selection follows the "mode-characteristic-model" matching principle: standing wave sloshing liquid motion is characterized by reciprocating vibration along a fixed direction, with no obvious propagation characteristics on the liquid surface, making a simple harmonic dynamic torque model suitable; traveling wave sloshing liquid motion is characterized by undulation propagating along a specific direction, with phase difference changes on the liquid surface, making a wave propagation torque model suitable; vortex sloshing liquid motion has a rotating flow field, accompanied by energy concentration phenomena, allowing the introduction of a torque model with rotating components. The three models correspond to different mathematical expressions to match the liquid dynamic characteristics under each mode.

[0127] 1) Torque calculation model for standing wave type swaying: When the pattern recognition result is a standing wave type, the liquid sloshing can be approximated as a simple harmonic motion along the base's roll or pitch axis. The disturbance torque is mainly generated by the reciprocating motion of the liquid mass. The torque calculation model is derived based on the change in angular momentum of the simple harmonic motion, and the formula is as follows:

[0128] In the above formula, Let be the disturbance torque component generated by the standing wave sway at time t. V is the density of ballast water, V is the volume of ballast water, and A is the amplitude of sloshing fluctuations (taken from the characteristic parameter set). The angular frequency of the oscillation ( , (to shake the main frequency) Let be the distance from the center of mass of the liquid to the axis of rotation of the foundation. The phase difference between the torque and the liquid surface motion (determined by fitting historical data, reflecting the inertial hysteresis characteristics of the liquid). The moment of inertia of the liquid about the rotational axis (calculated from the geometry of the ballast tank and the volume of the liquid). (where r is the distance from the fluid element to the rotation axis). The second term in the formula is for the basic angular acceleration data (roll or pitch acceleration, determined by the direction of the standing wave vibration) acquired by the IMU. The additional torque caused by the base angular acceleration is used to correct the effect of the base's own motion on the relative motion of the liquid. When the base has angular acceleration, the liquid will generate an additional torque in the opposite direction of the base's motion due to inertia. Therefore, it is included in the total disturbance torque calculation to improve accuracy.

[0129] 2) Torque calculation model for traveling wave-shaped swaying: When the pattern recognition result is a traveling wave, the liquid sloshing manifests as a wave propagating along the length or width of the chamber. There is a phase difference between the liquid surfaces of adjacent monitoring points. The disturbance torque needs to consider the momentum transfer characteristics during wave propagation. The torque calculation model is constructed based on the wave equation of the traveling wave and the law of conservation of momentum, as shown in the following formula:

[0130] In the above formula, Let be the disturbance torque component generated by the traveling wave sway at time t. It is the acceleration due to gravity. The traveling wave wavelength (determined by the dominant frequency in the characteristic parameter set) With liquid wave velocity calculate, wave speed Calculations based on ballast water depth h L is the length of the ballast tank along the wave propagation direction, and k is the wave number ( ), The coordinates of the monitoring point along the propagation direction, The basic angular velocity data collected by the IMU The damping coefficient of the traveling wave (which can be fitted using experimental data, reflecting energy dissipation during wave propagation, and is related to the bulkhead friction coefficient and liquid viscosity) is the second term in the formula. The damping torque, related to angular velocity, characterizes the damping effect generated by friction between the liquid and the bulkhead during traveling wave propagation. The meanings of the other parameters are consistent with those of the standing wave model. This model accurately describes the variation of torque with spatial position and time under traveling wave sway by introducing propagation characteristic parameters such as wave number and wavelength, while also correcting for the influence of the underlying motion by incorporating angular velocity and angular acceleration data.

[0131] 3) Torque calculation model for vortex-type swaying: When the pattern recognition result is vortex-like, the liquid sloshing involves a rotating flow field. The disturbance torque needs to include both rotational and reciprocating vibration components. The model construction needs to combine the angular momentum of the rotating flow field and the angular momentum of the simple harmonic motion, as shown in the following formula:

[0132] In the above formula, Let be the disturbance torque component generated by the vortex-like swaying at time t. The circulation of the vortex (calculated from the energy distribution entropy and fluctuation amplitude in the characteristic parameter set; a larger circulation indicates a higher rotation intensity). The volume of the vortex-affected region is determined based on the hull geometry and the vortex radius, which is calculated from the location corresponding to the maximum value of the liquid surface velocity gradient. The vortex rotation angular frequency (obtained from the spectral analysis of angular velocity data collected by the IMU, corresponding to the dominant frequency of the rotating flow field). The damping coefficient for vortex motion (greater than the damping coefficient for traveling wave motion, due to more significant energy dissipation in the rotating flow field) is the second term in the formula. The additional torque generated by the vortex rotation is used to characterize the torque contribution of the rotating flow field to the foundation; the meanings of the remaining parameters are consistent with those of the aforementioned model. This model compensates for the deviation of the traditional harmonic model in calculating the torque of vortex-type swaying by superimposing the rotational component, ensuring the accuracy of torque calculation in complex swaying scenarios containing rotational characteristics.

[0133] When synthesizing the disturbance torque vector, the torque components in the above modes are decomposed to the roll axis, taking into account the basic six-degree-of-freedom motion characteristics. ), pitch axis ( ) and heave axis ( ), forming a three-dimensional torque vector .

[0134] During the decomposition process, the contribution ratio of the torque on each axis is determined based on the pattern recognition results. Standing wave and traveling wave sloshing typically mainly affect the roll and pitch axes, while the heave axis torque component is relatively small (negligible or can be taken as 5%-10% of the roll axis torque). Vortex sloshing may affect all three axes simultaneously due to the rotating flow field; the torque contribution of each axis can be calculated separately using the three-dimensional components of IMU angular acceleration data. Finally, the obtained disturbance torque vector is smoothed in the time domain (e.g., using a 5-point moving average) to eliminate instantaneous noise interference, ensuring the continuity and stability of the torque signal, and providing accurate dynamic torque input for the boundary condition loading in subsequent flow field simulations.

[0135] In one optional embodiment, a model predictive control algorithm is used to generate a suppression strategy for the current swaying mode based on a real-time coupled prediction model, including the following steps: S31. Based on the real-time coupled prediction model, the continuous-time differential equation is converted into a discrete state-space equation using the explicit Euler discretization method, generating a discrete prediction model for rolling optimization calculation; the discrete prediction model is loaded into the model prediction controller kernel, and the rolling optimization framework is initialized based on the preset prediction step size.

[0136] Specifically, this step achieves the transformation from a continuous system to a discrete system through explicit Euler discretization and initializes the rolling optimization framework, providing a basic model and computational carrier for the subsequent rolling optimization of model predictive control (MPC).

[0137] The core of the real-time coupled prediction model is the continuous-time differential equation describing the coupling relationship between liquid sloshing and the basic motion. These equations are usually nonlinear ordinary differential equations. The rolling optimization of MPC requires a fast solution to the optimization problem in each control cycle. Explicit Euler discretization has become the appropriate choice because of its small computational cost, simple implementation and ability to meet the real-time requirements of the system. Its core idea is to use the first-order difference quotient to approximate the derivative and transform the continuous-time state change into a state recursion under discrete time steps.

[0138] The general form of a continuous-time differential equation is: and ,in Let be the time derivative of the state vector at time t. It is a continuous-time state vector (containing core state variables such as liquid centroid position, basic attitude angle, and angular velocity). This is the continuous-time control input vector (here, the ballast water cross-tank transfer flow rate). The state transition function is constructed based on the nonlinear dynamic relationship of the real-time coupled prediction model. It is a continuous-time output vector (including the output quantities that need to be monitored, such as the basic attitude angle and liquid level height). This is the output function.

[0139] When using explicit Euler discretization, let the discrete time step be... (i.e., the sampling period of MPC), then the discrete time k (corresponding to continuous time) ) state vector With control input vector satisfy:

[0140]

[0141] The above formulas constitute the discrete state-space equations, where Let be the discrete state vector at time k+1, and let be the state vector at time k and... The state derivative at time k is approximated by multiples of the state derivative at time k, and the accuracy of this approximation is determined by... Decisions—for example It needs to be less than 1 / 5 of the system's dominant time constant (the system's dominant time can be obtained through step response analysis of the real-time coupled prediction model, reflecting the time it takes for the system to transition from one steady state to another) to ensure that the discretization error meets the prediction accuracy requirements of MPC.

[0142] If the real-time coupled prediction model can be linearized near the operating point (e.g., when the fluctuation amplitude is small, the influence of the nonlinear term can be ignored), the discrete state-space equation can be further simplified to a linear form:

[0143]

[0144] in For discrete state matrix (dimension and Consistent), reflecting the natural evolution of the state without control input. ( It is the identity matrix. for (Jacobi matrix at the working point) This is a discrete input matrix (its dimension is the state vector dimension × the control input vector dimension), reflecting the influence of the control input on the state. ( for right (partial derivative matrix); It is a discrete output matrix (the dimension is the output vector dimension × the state vector dimension). The direct transfer matrix (dimensions of output vector dimension × control input vector dimension) is derived from the output function. The partial derivatives are calculated to obtain, , .

[0145] When loading the discrete prediction model into the MPC controller kernel, data format conversion and compatibility verification must be performed. First, the matrix of the discrete state-space equations ( ) and vector ( The initial state is converted to a numerical format supported by the controller kernel (such as a floating-point matrix), and loaded through the controller's built-in model import interface (such as the mpc function in MATLAB MPC Toolbox or the MPC function block interface in Siemens SIMATIC S7); then compatibility verification is performed, including matrix dimension matching checks (such as...). The number of rows and columns must equal the dimension of the state vector. The number of rows must equal the dimension of the state vector, and the number of columns must equal the dimension of the control input vector. Stability verification (through calculation) The eigenvalues ​​are selected to ensure that the magnitude of all eigenvalues ​​is less than 1, thus guaranteeing the stability of the discrete model. If the verification fails, the discretization step size is readjusted. Alternatively, linearize the operating point until the model meets the loading requirements.

[0146] The initialization of the rolling optimization framework is based on a preset prediction step size. As core parameters, The settings need to balance prediction accuracy and computational complexity. Too small a value will result in insufficient coverage of the prediction time domain, making it impossible to capture the long-term dynamics of the system in the future (such as the periodic changes of liquid sloshing). If the value is too large, it will increase the computational load in each control cycle, affecting real-time performance. Typically... Can be taken to (like hour, Specifically, it can be determined by combining the response characteristics of the real-time coupled prediction model: if the system response speed is fast (such as the turbulent dynamic changes of vortex-type swaying). Take the larger value; if the system response is smooth (such as standing wave type swaying). Take the smaller value. During initialization, the control step size is also set. (generally This indicates that only the first few points in the prediction time domain are considered. Output control quantity at each moment, then (keeping the last control variable unchanged at each time step), and the initial value of the weight matrix of the optimization objective (reserving parameter interfaces for subsequent multi-objective cost function input), ultimately forming a rolling optimization framework that includes a discrete prediction model, prediction / control step size, and parameter interface, awaiting subsequent cost function and constraint input.

[0147] S32. Based on the roll and pitch deviation values ​​of the floating wind turbine foundation collected in real time by the inertial measurement unit, calculate the attitude stability cost term; combine the real-time total power data of the ballast water pump group to calculate the energy consumption change rate cost term; combine the attitude stability cost term and the energy consumption change rate cost term into a multi-objective cost function and input it into the rolling optimization framework.

[0148] Specifically, the calculation of the attitude stability cost term uses the roll and pitch deviation values ​​collected in real time by the inertial measurement unit (IMU) as the core inputs. These two deviation values ​​directly reflect the degree to which the basic attitude deviates from the safety threshold and are key indicators for assessing the safety of underwater operations. Roll deviation value Defined as the actual roll angle at time k Target roll angle The difference, i.e. Pitch angle deviation value Similarly defined as The target attitude angle is usually set to 0° (horizontal attitude). If there is an optimal attitude angle for a specific sea state (such as a small tilt angle set to balance wave forces), then... and Take the corresponding optimal value.

[0149] Attitude stability cost To reflect the principle that "the smaller the deviation, the lower the cost," and considering the cumulative effect of deviations (to avoid long-term small deviations leading to attitude loss of control), the formula is the sum of the squares of the deviations, as follows:

[0150] In the above formula, This represents the roll angle deviation value predicted at time k+i from time k (calculated by the discrete prediction model), where i is the number of steps in the prediction time domain. ), and These are the weighting coefficients for the roll angle deviation and the pitch angle deviation, respectively. The weighting coefficients can be set based on engineering safety priorities: if the foundation's overturning resistance in the roll direction is weak (e.g., a single-column floating foundation), then... If the pitching direction is more significantly affected by waves (such as in a long, semi-submersible foundation), then The specific values ​​can be determined using the Analytic Hierarchy Process (AHP) to ensure that the weight allocation meets the actual security requirements.

[0151] Energy consumption change rate cost item The calculation is based on the real-time total power data of the ballast water pump set. The core purpose is to avoid sudden increases or decreases in pump power caused by control commands, thereby protecting the equipment and reducing energy consumption. (Real-time total power of the ballast water pump set) The sum of the power of all operating ballast water pumps at time k (collected by the current and voltage monitoring modules of the pump controller, and calculated using the power calculation formula). Conversion, where U is voltage, I is current, (Power factor). Energy consumption change rate. Defined as the difference between the total power at time k and the total power at time k-1, i.e. ,like An excessively large absolute value (such as exceeding 20% ​​of the rated power) will cause current surges in the water pump motor, shortening the equipment's lifespan and increasing energy consumption.

[0152] Therefore, the cost of the rate of change in energy consumption Using the sum of the squares of the power change rates, the formula is:

[0153] In the above formula, This represents the predicted rate of change of total power at time k+i (derived from a discrete prediction model combined with the characteristic curves of pump power and flow rate, which are obtained by fitting the pump's factory test data, and are in the form of...). , , , These are the fitting coefficients. (Water pump flow rate) Weighting coefficients for the rate of change in energy consumption. The value needs to balance attitude control and energy consumption: when the attitude deviation is large (such as exceeding 50% of the safety threshold), it needs to be reduced. Prioritize ensuring attitude stability; when the attitude deviation is small (e.g., less than 10% of the safety threshold), the [measurement] needs to be increased. To reduce energy consumption, therefore It can be designed as a cost term that varies with attitude stability. The dynamically adjusted variable is represented by the formula: ,in Based on the weights, To adjust the coefficients and ensure The larger, The smaller the value, the better the dynamic priority balancing.

[0154] Multi-objective cost function for and The weighted sum is calculated using the following formula:

[0155] This combined form retains the physical meaning of each cost term and facilitates multi-objective optimization through weight adjustment. Before being input into the rolling optimization framework, the cost function is normalized (because...). and Different dimensions, such as The dimension is the square of the angle. The dimension is power squared, and the normalization formula is:

[0156] in When the attitude deviation reaches the safety threshold The value of , When the power change rate reaches the equipment's tolerance limit The values ​​of are normalized to ensure that the two cost terms are of the same order of magnitude, preventing one term from dominating the optimization result. Finally, the normalized multi-objective cost function is... The input rolling optimization framework serves as the objective function of the optimization problem, guiding subsequent constraint optimization solutions.

[0157] S33. Extract the maximum flow rate threshold of the water pump and the valve opening limit from the ballast water system control database, and input them as hard constraints into the rolling optimization framework.

[0158] Specifically, this step extracts hard constraints from the ballast water system control database to provide physical and safety boundaries for rolling optimization, ensuring the feasibility of the optimization results and the safety of the equipment.

[0159] The ballast water system control database is a structured database (such as using MySQL or SQL Server databases, communicating with the MPC controller via an ODBC interface) that stores equipment parameters, safety thresholds, and historical operating data of the ballast water system. Its core data modules include an equipment parameter table, a safety constraint table, and a real-time status table: The equipment parameter table stores the rated flow, maximum flow, and rated power of each ballast water pump, and the maximum and minimum opening and response time of each proportional control valve, etc.; the safety constraint table stores safety thresholds set based on engineering experience and standards, such as the maximum and minimum liquid levels of the ballast water tanks and the maximum allowable attitude angle of the foundation; the real-time status table stores the current operating status of the equipment (such as pump start / stop status and valve fault status), providing a real-time adaptation basis for constraint extraction.

[0160] Maximum flow rate threshold of water pump (m is the pump number, (M is the total number of pumps) can be extracted by combining equipment parameters and real-time status: First, read the rated flow rate of the m-th pump from the equipment parameter table. With maximum flow Then, the operating status of the water pump is read from the real-time status table. If the water pump is in normal operating condition, the maximum flow rate threshold of the water pump is set to... If the water pump has a minor malfunction (such as motor temperature approaching the rated temperature or bearing vibration slightly exceeding the threshold), the threshold should be lowered to protect the equipment, for example, by taking... If the water pump is malfunctioning (e.g., motor overload, seal leakage), then This means that the pump in question is prohibited from participating in load adjustment. Simultaneously, the total flow constraint of the pump group—the sum of the flow rates of all operating pumps—must be considered. It must be less than the maximum conveying capacity of the ballast water system pipeline. (Calculate by reading piping design parameters from the equipment parameter table) A is the cross-sectional area of ​​the pipe. (This refers to the maximum allowable flow velocity in the pipeline), therefore the total flow constraint is also considered as one of the hard constraints, and the extraction formula is as follows: .

[0161] Valve opening limits include maximum opening. With minimum opening (n is the valve number, (where N is the total number of valves). The extraction logic is similar to that of the water pump flow threshold: read the maximum design opening of the nth valve from the equipment parameter table. (Typically it can be 100%) and minimum design opening. (Typically, this can be 5%-10% to avoid jamming when reopening the valve after it has been fully closed); read the valve's operating status from the real-time status table—if the valve is in a normal state (no jamming alarm, normal opening feedback signal), then the opening limit is the design value; if the valve has an opening feedback deviation (e.g., the deviation between the commanded opening and the actual opening exceeds 5%), then the opening range needs to be reduced, such as... , If the valve is in a faulty state (e.g., unable to operate, seal failure), then set... This means closing the valve. Furthermore, there is a coupling relationship between the valve opening and the pump flow rate (the flow rate increases with increasing opening, as shown by the characteristic curve). , Percentage of opening, Therefore, the valve opening limit is converted into the corresponding flow constraint to ensure consistency between the opening constraint and the flow constraint. The conversion formula is as follows: , ,in This is the maximum flow rate when the valve is fully open, matched with the maximum flow rate of the connected water pump.

[0162] Finally, the hard constraints are transformed into a mathematical constraint form recognizable by the rolling optimization framework. The pump flow constraint is: (Normal water pump) or (Faulty water pump); Total flow constraint is Valve opening constraint is (Normal valve) or (Faulty valve), and input the constraint matrix into the rolling optimization framework to define the feasible region for subsequent optimization solutions.

[0163] S34. Call the effective set algorithm to solve the constrained optimization problem defined by the rolling optimization framework, and output the baseline flow allocation vector for multiple future sampling periods.

[0164] Specifically, the selection of the effective set algorithm is based on the mathematical properties of the multi-objective cost function, and the cost function constructed through step S32. Since the cost is a quadratic function (each cost term is the sum of the squares of the deviations or rates of change), the constraints of the rolling optimization framework are linear inequalities or equality constraints (such as upper and lower bounds of flow and opening). Therefore, the entire optimization problem belongs to a quadratic programming problem with linear constraints. The effective set algorithm (also known as the active set algorithm) is an efficient method for solving small and medium-sized QP problems. Its core advantage is that by iteratively maintaining the "effective constraint set" (i.e. the constraints that are effective at the current iteration point), the constrained problem is transformed into an unconstrained problem to be solved, which significantly reduces the amount of computation and is suitable for the real-time requirements of MPC.

[0165] The solution process of the effective set algorithm is divided into three stages: initialization, iterative optimization, and convergence determination. 1) Initialization phase: First, determine the initial feasible solution. (That is, the initial control input vector, which here is the initial flow allocation for each water pump). The initial feasible solution must satisfy all hard constraints. The flow allocation corresponding to "zero deviation" can be selected. If the current attitude deviation is 0, then... Let be a vector where the flow rate of each pump is 0; if attitude deviation exists, Based on empirical rules (e.g., when the roll angle is to the right, the flow rate of the left pump is taken as 50% of the maximum flow rate, and the flow rate of the right pump is taken as 0), the initial effective constraint set is then determined. —Calculate the initial feasible solution The distance to each constraint is considered, and constraints with a distance less than a preset threshold (such as 1% of the maximum flow rate) are included. This indicates that these constraints are in effect at the initial point (e.g., if the initial flow rate of a water pump is equal to its maximum flow rate threshold, then the upper limit constraint on the pump's flow rate is a valid constraint). Simultaneously, the Lagrange multiplier vector is initialized (used to determine the reasonableness of valid constraints), and the Lagrange function is defined as... ,in For constraint function vectors (such as...) ), For Lagrange multipliers, initial multipliers By solving unconstrained optimization problems get.

[0166] 2) Iterative optimization phase: The core of the iterative optimization phase is to gradually approach the optimal solution through a loop of "determining the search direction - calculating the search step size - updating the effective constraint set". The specific steps are as follows: Determine the search direction At that time, in the k-th iteration, based on the current effective constraint set Construct a feasible direction subspace (i.e., directions that satisfy effective constraints), and solve for the Newtonian directions (search directions) with effective constraints. If the current iteration point... If all Lagrange multipliers at a given location are non-negative (indicating valid constraints and no constraint conflicts), then the search direction is... To determine the direction of the cost function's descent, we solve... Received, among which Cost function The Hessian matrix (the second derivative matrix of the quadratic function, here a constant matrix, calculated from the weight coefficients of the cost term, such as...) , (where is the weight matrix); if negative Lagrange multipliers exist, remove the corresponding constraint (indicating that the constraint is unreasonable and needs to be relaxed), and update the effective constraint set to . (c represents an unreasonable constraint), recalculate the search direction.

[0167] Calculate the search step size At that time, the search step size needs to ensure the new iteration point It remains within the feasible region while significantly reducing the cost function. The step size calculation adopts the "maximum feasible step size" principle:

[0168] in, For ineffective constraint i in The function value at that point, Its gradient, To ensure that the new iteration point does not violate the maximum step size of the ineffective constraints; to reduce the cost function, the actual step size is taken as... ,in This is the descent step size determined by the Armijo criterion. For example... ,like ,but Halve the amount until the descent condition is met. This is the Armijo coefficient, which is usually set to 0.1.

[0169] Update the effective constraint set At that time, calculate the new iteration point. The distance to each ineffective constraint, if some ineffective constraint i satisfies ( If a constraint satisfies a threshold, it is included in the effective constraint set. Simultaneously, the Lagrange multipliers are recalculated. In preparation for the next iteration.

[0170] 3) Convergence determination phase: The convergence criterion is based on the "gradient condition" and the "constraint satisfaction condition." The gradient condition states that the gradient projection of the cost function into the feasible region is zero. For all feasible directions The condition is met if and only if there is no feasible direction at the current iteration point that would decrease the cost function; the constraint condition is satisfied, meaning all constraints are satisfied. And all Lagrange multipliers are non-negative If both conditions are met simultaneously, then the current iteration point... If the result is the optimal solution, then continue iterating until the maximum number of iterations is reached.

[0171] After the optimization problem is solved, output the future... Reference flow distribution vector for each sampling period ,in Assign a subvector to predict the flow rate of each pump at time k+i, with dimension M (total number of pumps). Each element of the subvector... This represents the target flow rate of the m-th pump at time k+i. The vector contains a timestamp, pump number, target flow rate, and corresponding valve opening command (derived from the flow-opening characteristic curve). It is output in a structured data format (such as JSON or XML) to provide a precise flow control benchmark for the execution of the variable frequency pump group and proportional control valve network in subsequent steps S5. This ensures that the suppression strategy is implemented according to the predicted timing, effectively offsetting the disturbance torque generated by the current swaying mode.

[0172] In one optional embodiment, based on the roll and pitch deviation values ​​of the floating wind turbine foundation collected in real time by the inertial measurement unit, an attitude stability cost term is calculated; combined with the real-time total power data of the ballast water pump unit, an energy consumption change rate cost term is calculated; the attitude stability cost term and the energy consumption change rate cost term are combined into a multi-objective cost function, including the following steps: S41, Based on roll angle deviation value Deviation value from pitch angle The attitude stability cost term is calculated using a quadratic form function; the expression for the attitude stability cost term is: ,in The preset roll weighting coefficient, t represents the preset pitch weighting coefficient, and t represents time.

[0173] Specifically, the core of this step is to calculate the attitude stability cost term, which aims to quantify the degree to which the floating wind turbine foundation deviates from the target state during launching, providing a safety priority basis for subsequent optimization. In practice, the deviations of the roll and pitch angles relative to the target attitude (usually horizontal) are first determined based on real-time data collected by the inertial measurement unit. These two deviations directly reflect the risk of the foundation overturning; the greater the deviation, the higher the safety hazard.

[0174] To emphasize the "safety first" principle, this step uses a quadratic function to process the deviation data. The quadratic nature of the function causes the cost of larger deviations to increase quadratically, effectively amplifying the impact of severe attitude deviations and preventing loss of control due to the accumulation of small short-term deviations. Simultaneously, roll and pitch weighting coefficients are set based on the characteristics of the foundation structure and the influence of sea conditions: for example, single-column foundations have weaker roll resistance to overturning, so the roll weighting coefficient needs to be greater than the pitch weighting coefficient; if waves mainly propagate along the pitch axis, the pitch is more significantly affected, and the pitch weighting coefficient needs to be increased accordingly.

[0175] Furthermore, by integrating the square of the deviation, the attitude deviation over a period of time can be accumulated, avoiding the risk of ignoring the risk of continuous deviation while only focusing on the instantaneous deviation. This ensures that the cost term can fully reflect the attitude stability level, laying a safety-oriented foundation for subsequent multi-objective optimization.

[0176] S42, Based on the real-time total power of the ballast water pump set The energy consumption change rate cost term is calculated using first-order difference; the expression for the energy consumption change rate cost term is:

[0177] in The preset energy consumption penalty coefficient, t represents the sampling period, and t represents time.

[0178] Specifically, this step aims to calculate the energy consumption change rate cost term. The core objective is to protect the ballast water pump equipment and control energy consumption costs, avoiding the impact of sudden power changes on the system during load adjustment. During implementation, real-time total power data of the ballast water pump group is first collected. Power values ​​of faulty pumps (such as those with motor overload or seal failure) must be excluded to ensure that the total power data accurately reflects the energy consumption status of the operating equipment.

[0179] This step uses first-order difference calculation to determine the power change rate. By combining the power difference between two adjacent sampling periods with the sampling period, the power change amplitude per unit time is obtained. This indicator is directly related to the safe operation of the water pump motor—when the power change rate is too large, the stator winding current of the motor will fluctuate drastically, accelerating the aging of the insulation layer and shortening the equipment life. To suppress drastic changes, an energy consumption penalty coefficient is introduced in this step. This coefficient needs to balance the equipment's tolerance and energy consumption costs: when the risk of sudden power changes is high, the coefficient is increased to strengthen the constraint on fluctuations; when the attitude deviation is small, the coefficient is adjusted appropriately to balance the energy consumption control requirements.

[0180] Meanwhile, the sampling period is kept consistent with the step size of the discrete prediction model mentioned above, ensuring that the timing of the power change rate calculation is coordinated with the overall optimization framework, and avoiding the loss of reference significance of the cost term due to timing misalignment.

[0181] S43, Incorporate attitude stability cost term Cost of energy consumption change rate A multi-objective cost function is generated through linear weighted combination. .

[0182] Specifically, this step constructs a multi-objective cost function by linearly weighting the attitude stability cost term and the energy consumption change rate cost term. The core objective is to achieve a dynamic balance between "safety assurance" and "energy consumption optimization," adapting to the load adjustment requirements under different sea states. During implementation, the key is to determine the weight coefficients of the two cost terms, and the weights must meet the normalization requirement (the sum of the weights must be 1) to prevent any one term from dominating the optimization result due to differences in numerical magnitude.

[0183] The adjustment of the weighting coefficients needs to be based on the real-time attitude deviation level. For example, under high deviation conditions (attitude close to the safety threshold), the attitude stability weighting coefficient is 0.8-0.9 to prioritize basic safety, while energy consumption optimization is secondary. Under medium deviation conditions (attitude within the safety range but still requiring attention), the weighting coefficients are adjusted to 0.6-0.7 for attitude and 0.3-0.4 for energy consumption to achieve a balance between the two. Under low deviation conditions (attitude close to the ideal state), the energy consumption weighting coefficient is increased to 0.5-0.7 to prioritize reducing load adjustment energy consumption and reduce fuel oil power generation costs.

[0184] Through dynamic weighting, the multi-objective cost function can be flexibly adapted to different operating scenarios, ensuring that the optimization results meet both safety requirements and economic efficiency, and providing a reasonable objective guide for subsequent rolling optimization.

[0185] S44. Based on the prediction time domain length defined in the rolling optimization framework. For multi-objective cost functions Integrating in the time domain yields the total cost function; the total cost function is: .

[0186] Specifically, the steps involve integrating the multi-objective cost function over the prediction time domain to obtain the total cost function. The core objective is to ensure that the optimization process takes into account the system dynamics over a future period, preventing short-term optimization from leading to subsequent attitude instability or energy consumption spikes. In implementation, setting the prediction time domain length is crucial; it must cover a complete cycle of liquid sloshing. Because liquid sloshing is periodic, if the prediction time domain is too short, it cannot capture the complete sloshing dynamics, and the optimization may only adjust for the deviation at the current moment, leading to a larger deviation in the next cycle.

[0187] The essence of integral operation is to accumulate the comprehensive cost in the prediction time domain, transforming instantaneous cost into long-term cost, ensuring that the optimization result not only improves the current attitude and energy consumption, but also prevents potential future risks. For example, if the current attitude deviation is small but the prediction is that the deviation will be aggravated by swaying in the future, the total cost after integration will reflect this trend, prompting the optimization process to adjust the load adjustment strategy in advance. Meanwhile, the total cost function is strictly time-series aligned with the rolling optimization framework. The prediction time domain matches the prediction steps and distance step length mentioned earlier (prediction time domain = prediction steps × distance step length), ensuring that the integral result can be directly used as the objective function for rolling optimization, guiding the subsequent effective set algorithm to solve for the optimal traffic allocation, and providing a comprehensive cost reference for load adjustment command generation.

[0188] The aforementioned method for attitude monitoring and adaptive load adjustment during the launching process of a floating wind turbine foundation utilizes a distributed sensor array to collect real-time data on liquid height changes and pressure distribution within the ballast water tank. Combined with inertial measurement unit (IMU) data on the foundation's six degrees of freedom (DOF) pose, a multi-source sensor dataset is generated. Based on this dataset, time-frequency domain joint analysis is employed to extract the frequency, amplitude, and energy distribution characteristics of free surface sloshing, constructing a set of characteristic parameters representing the sloshing state. Furthermore, a nonlinear coupled dynamic model of ballast tank surface sloshing and foundation six-DOF motion is established to achieve real-time prediction of the attitude impact of specific sloshing modes. Based on this prediction model, a model predictive control algorithm is used to generate a suppression strategy including a ballast water cross-tank flow distribution scheme, which is then transformed into an optimal load adjustment control command set. Finally, the commands are executed through a variable frequency pump group and a proportional control valve network to establish a controlled water flow channel between the target ballast tanks, dynamically adjusting the liquid level distribution. This achieves active suppression and precise control of attitude disturbances during the launching process of the floating wind turbine foundation, significantly improving the stability and safety of the launching process.

[0189] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0190] Based on the same inventive concept, this application also provides a system for implementing the above-mentioned method for attitude monitoring and adaptive load adjustment during the launching process of a floating wind turbine foundation. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the floating wind turbine foundation attitude monitoring and adaptive load adjustment system provided below can be found in the limitations of the floating wind turbine foundation attitude monitoring and adaptive load adjustment method described above, and will not be repeated here.

[0191] In one exemplary embodiment, such as Figure 3 As shown, a floating wind turbine foundation attitude monitoring and adaptive load adjustment system 30 is provided to implement the methods in the above embodiments. The system includes: The dynamic multi-source data sensing module 31 is used to collect dynamic change data of liquid height and pressure distribution data in the ballast water tanks of the floating wind power foundation through a distributed sensor array deployed on the inner wall of each ballast water tank, and to collect six-degree-of-freedom pose data of the foundation through an inertial measurement unit installed on the foundation structure of the floating wind power foundation, so as to obtain the original multi-source sensing dataset containing the dynamic characteristics of the liquid surface and the attitude of the foundation.

[0192] The liquid surface dynamic feature extraction module 32 is used to extract the sloshing frequency features, fluctuation amplitude features and energy distribution features of the free liquid surface in each ballast water tank based on the original multi-source sensor dataset and using the time-frequency domain joint analysis method, so as to obtain a set of feature parameters characterizing the sloshing state of the liquid surface.

[0193] The coupled dynamics prediction module 33 is used to obtain a real-time coupled prediction model for predicting the influence of a specific swaying mode on attitude by establishing a nonlinear dynamic model that describes the coupling relationship between the free liquid surface swaying and the basic six-degree-of-freedom motion in the ballast water tank based on the feature parameter set and the basic attitude history data.

[0194] The adaptive load adjustment decision module 34 is used to generate a suppression strategy for the current swaying mode based on a real-time coupled prediction model and a model predictive control algorithm, and to generate an optimal load adjustment control instruction set to offset the disturbance torque according to the suppression strategy; wherein, the suppression strategy includes a target flow allocation scheme for ballast water cross-tank transport.

[0195] The liquid level dynamic adjustment execution module 35 is used to execute the optimal load adjustment control command set through the variable frequency water pump group and proportional control valve network to establish a controlled water flow channel between the target ballast water tanks in order to adjust the liquid level height distribution of each tank.

[0196] Embodiments of this application also provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the aforementioned method embodiments.

[0197] Embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0198] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0199] The above are preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for attitude monitoring and adaptive load adjustment during the launching process of a floating wind turbine foundation, characterized in that, The method includes: S1. Collect dynamic change data of liquid height and pressure distribution data in the tanks by a distributed sensor array deployed on the inner wall of each ballast water tank of the floating wind power foundation, and collect six-degree-of-freedom pose data of the foundation by an inertial measurement unit installed on the foundation structure of the floating wind power foundation, to obtain the original multi-source sensor dataset containing the dynamic characteristics of the liquid surface and the attitude of the foundation. S2. Based on the original multi-source sensor dataset, the time-frequency domain joint analysis method is used to extract the sloshing frequency characteristics, fluctuation amplitude characteristics and energy distribution characteristics of the free liquid surface in each ballast water tank, so as to obtain a set of characteristic parameters characterizing the sloshing state of the liquid surface. S3. Based on the feature parameter set and basic attitude history data, a nonlinear dynamic model describing the coupling relationship between the free liquid surface sloshing and the basic six-degree-of-freedom motion in the ballast water tank is established to obtain a real-time coupled prediction model for predicting the influence of a specific sloshing mode on attitude. S4. Based on the real-time coupled prediction model, a model predictive control algorithm is used to generate a suppression strategy for the current swaying mode, and an optimal load adjustment control command set is generated according to the suppression strategy to offset the disturbance torque; wherein, the suppression strategy includes a target flow allocation scheme for ballast water cross-tank transportation. S5. The optimal load control command set is executed through the variable frequency water pump group and proportional control valve network to establish a controlled water flow channel between the target ballast water tanks in order to adjust the liquid level distribution of each tank.

2. The method according to claim 1, characterized in that, Based on the feature parameter set and historical basic attitude data, a nonlinear dynamic model describing the coupling relationship between free surface sloshing and basic six-degree-of-freedom motion within the ballast tank is established. This model yields a real-time coupled prediction model for predicting the impact of specific sloshing modes on attitude, including: S11. Based on the geometric configuration parameters and material property data of the ballast water tank, the fluid domain is spatially discretized using the finite volume method to generate a three-dimensional numerical calculation grid containing a boundary layer mesh. S12. Calculate the disturbance torque vector generated by the liquid sloshing on the floating wind turbine foundation based on the characteristic parameter set, and load the disturbance torque vector as a dynamic boundary condition onto the three-dimensional numerical calculation grid; based on the three-dimensional numerical calculation grid, simulate the liquid sloshing by solving the transient Reynolds-averaged Navier-Stokes equations. The dynamic unsteady flow field outputs predicted data on the trajectory of the liquid's center of mass. S13. The predicted data of the center of mass motion trajectory is spatiotemporally aligned with the six-degree-of-freedom pose data collected in real time by the inertial measurement unit to obtain a prediction-measurement dataset with unified spatiotemporal reference; an extended Kalman filter is used with the prediction-measurement dataset as input, and the predicted value of the liquid center of mass trajectory is used as the state prediction quantity and the measured value of the basic pose is used as the observation quantity to perform data fusion and generate the fused optimized pose state estimation result. S14. Based on the residual vector between the optimized pose state estimation result and the centroid motion trajectory prediction data, calculate the covariance matrix of the fluid dynamics model output error, and output a confidence quantification index characterizing the model prediction accuracy. S15. When the confidence quantification index is lower than a preset threshold, the equivalent damping coefficient and stiffness coefficient in the transient Reynolds-averaged Navier-Stokes equation are dynamically corrected based on the recursive least squares method to obtain the parameter-adaptive real-time coupled prediction model.

3. The method according to claim 2, characterized in that, In step S11: The fluid domain is discretized and meshed using the finite volume method. The geometric parameters of the ballast tank cover the three-dimensional dimensional features and internal structural distribution of the tank. The corresponding parameters are digitally modeled using 3D modeling software to form a CAD model containing complete geometric information. The material property data include the density, elastic modulus, Poisson's ratio of the ballast tank shell material, and the density and dynamic viscosity of the ballast water.

4. The method according to claim 3, characterized in that, The step of calculating the disturbance torque vector generated by the liquid sloshing on the floating wind turbine foundation based on the feature parameter set includes: S21. Based on the feature parameter set, the current swaying mode type is identified by a support vector machine classifier to obtain pattern recognition results including standing wave, traveling wave, or vortex. S22. Based on the pattern recognition result, select the corresponding torque calculation model, and combine the angular acceleration data collected by the inertial measurement unit to solve the disturbance torque vector generated by the liquid sloshing.

5. The method according to claim 4, characterized in that, In step S21: The selection of the feature parameter set focuses on parameters that can characterize the essential differences between different swaying modes. Key input features are selected from the obtained feature parameter set, including swaying frequency features, fluctuation amplitude features, energy distribution features, and spatiotemporal characteristics of liquid surface motion. These parameters together constitute the input feature vector of the SVM classifier. Each component corresponds to the main frequency, secondary main frequency, main frequency difference, maximum fluctuation amplitude, amplitude standard deviation, energy distribution entropy, main frequency energy proportion, monitoring point phase difference, and synchronization coefficient, respectively.

6. The method according to claim 5, characterized in that, The method for generating a suppression strategy for the current swaying mode based on the real-time coupled prediction model and using a model predictive control algorithm includes: S31. Based on the real-time coupled prediction model, the continuous-time differential equation is converted into a discrete state-space equation using the explicit Euler discretization method, generating a discrete prediction model for rolling optimization calculation; the discrete prediction model is loaded into the model prediction controller kernel, and the rolling optimization framework is initialized based on the preset prediction step size. S32. Based on the roll angle deviation and pitch angle deviation values ​​of the floating wind power foundation collected in real time by the inertial measurement unit, calculate the attitude stability cost term; combine the real-time total power data of the ballast water pump group to calculate the energy consumption change rate cost term; combine the attitude stability cost term and the energy consumption change rate cost term into a multi-objective cost function and input it into the rolling optimization framework. S33. Extract the maximum flow rate threshold of the water pump and the valve opening limit from the ballast water system control database, and input them as hard constraints into the rolling optimization framework; S34. Call the effective set algorithm to solve the constrained optimization problem defined by the rolling optimization framework, and output the baseline flow allocation vector for multiple future sampling periods.

7. The method according to claim 6, characterized in that, In S34: The solution process of the effective set algorithm is divided into three stages: initialization, iterative optimization, and convergence determination.

8. The method according to claim 7, characterized in that, The attitude stability cost term is calculated based on the roll angle deviation and pitch angle deviation values ​​of the floating wind power foundation collected in real time by the inertial measurement unit; and the energy consumption change rate cost term is calculated by combining the real-time total power data of the ballast water pump group. The attitude stability cost term and the energy consumption change rate cost term are combined into a multi-objective cost function, including: S41, Based on the aforementioned roll angle deviation value The deviation value of the pitch angle The attitude stability cost term is calculated using a quadratic form function; the expression for the attitude stability cost term is as follows: ,in The preset roll weighting coefficient, t represents the preset pitch weighting coefficient, and t represents time. S42, Based on the real-time total power of the ballast water pump set The energy consumption change rate cost term is calculated using first-order difference; the expression for the energy consumption change rate cost term is: in The preset energy consumption penalty coefficient, The sampling period is t, and time is t. S43, the attitude stability cost term The energy consumption change rate cost term A multi-objective cost function is generated through linear weighted combination. ; S44. Based on the prediction time domain length defined in the rolling optimization framework. For the multi-objective cost function The total cost function is obtained by integration in the time domain; the total cost function is: .

9. The method according to claim 8, characterized in that, In step S42: The power change rate is calculated using the first-order difference method. The power change amplitude per unit time is obtained by combining the power difference between two adjacent sampling periods with the sampling period. An energy consumption penalty coefficient is introduced, which takes into account both equipment tolerance and energy consumption cost: the coefficient is increased when the risk of sudden power change is high to strengthen the constraint on fluctuations; the coefficient is moderately adjusted when the attitude deviation is small to balance the energy consumption control requirements. The sampling period is kept consistent with the step size of the discrete prediction model to ensure that the timing of the power change rate calculation is coordinated with the overall optimization framework.

10. A floating wind turbine foundation attitude monitoring and adaptive load adjustment system for launching process, used to implement the method according to any one of claims 1 to 9, characterized in that, The system includes: The dynamic multi-source data sensing module is used to collect dynamic changes in liquid height and pressure distribution data in the ballast water tanks of the floating wind power foundation through a distributed sensor array deployed on the inner wall of each ballast water tank. It also collects six-degree-of-freedom pose data of the foundation through an inertial measurement unit installed on the foundation structure of the floating wind power foundation, thereby obtaining a raw multi-source sensing dataset containing the dynamic characteristics of the liquid surface and the attitude of the foundation. The liquid surface dynamic feature extraction module is used to extract the sloshing frequency features, fluctuation amplitude features and energy distribution features of the free liquid surface in each ballast water tank based on the original multi-source sensor dataset and using the time-frequency domain joint analysis method, so as to obtain a set of feature parameters characterizing the liquid surface sloshing state. The coupled dynamics prediction module is used to obtain a real-time coupled prediction model for predicting the impact of a specific swaying mode on attitude by establishing a nonlinear dynamic model that describes the coupling relationship between the free liquid surface swaying and the basic six-degree-of-freedom motion in the ballast water tank, based on the feature parameter set and the basic attitude history data. An adaptive load adjustment decision module is used to generate a suppression strategy for the current swaying mode based on the real-time coupled prediction model and a model predictive control algorithm, and to generate an optimal load adjustment control instruction set to offset the disturbance torque according to the suppression strategy; wherein, the suppression strategy includes a target flow allocation scheme for ballast water cross-tank transport. The liquid level dynamic adjustment execution module is used to execute the optimal load adjustment control command set through the variable frequency water pump group and proportional control valve network to establish a controlled water flow channel between the target ballast water tanks in order to adjust the liquid level height distribution of each tank.

Images