Intelligent adjusting method for dynamic ballast water of floating wind power semi-submersible foundation

Abstract

The present application relates to the technical field of floating wind power semi-submersible foundation motion control, and discloses a kind of floating wind power semi-submersible foundation dynamic ballast water intelligent adjustment method, for solving the problems of existing ballast water adjustment method adjustment instruction lag, platform motion response violent under complex sea state, stability control deficiency. The present application acquires platform motion and marine environment data in real time, and constructs comprehensive input vector by combining historical ballast water distribution state, inputs trained motion response prediction model to output future motion state prediction. According to this, the total amount of ballast water adjustment and distribution ratio are calculated, and the ballast water distribution is adjusted through the water pump system, and then real-time monitoring and comparing with the predicted value, when the deviation exceeds the threshold, re-predict and adjust, form a closed loop control. The present application is mainly used to improve the stability and safety of semi-submersible floating wind power foundation under complex sea state.

Description

Technical Field

[0001] This invention relates to the field of motion control technology for floating semi-submersible wind turbine foundations. More specifically, this invention relates to a method for intelligent dynamic ballast water regulation of floating semi-submersible wind turbine foundations. Background Technology

[0002] Floating wind power, as an important direction for the development of offshore wind power, has attracted widespread attention due to its semi-submersible foundation platform, which boasts excellent deep-water adaptability and economic efficiency. The semi-submersible foundation platform achieves attitude balance and motion suppression through ballast water regulation, which is crucial for ensuring the normal operation of the wind turbine. However, in actual marine environments, especially under complex and variable sea conditions, existing ballast water regulation methods still have several limitations.

[0003] Ballast water regulation methods often rely on feedback control based on real-time collected marine environmental parameters and platform motion data. However, they fail to adequately consider the temporal correlation between historical motion responses and the current state, resulting in ballast regulation commands lagging behind actual sea state changes. Due to the significant dynamic characteristics of the marine environment, factors such as waves and currents exhibit strong temporal correlation and nonlinearity, making decisions based solely on instantaneous data struggles to accurately predict future platform motion trends. Consequently, regulation behavior often lags behind rapid changes in the external environment, leading to problems such as timing deviations and insufficient stability control.

[0004] Ballast water regulation commands lagging behind actual sea state changes can trigger a chain of hazards, such as motion instability and resonance risks. Delayed regulation commands cause the platform's restoring torque to fail to match the phase of environmental loads such as waves and currents. This not only fails to effectively suppress motion but may even produce a "negative damping" effect, exacerbating the platform's roll, pitch, and heave. At certain frequencies, this can lead to dynamic amplification or resonance, causing the motion amplitude to far exceed design limits and generating huge instantaneous impact loads on the platform structure, tower, and wind turbine.

[0005] For example, it can accelerate structural fatigue. When the platform is continuously subjected to unexpected and violent motion, critical structural components (such as welds, struts, and mooring points) will be subjected to alternating stresses far exceeding the design expectations. Such cyclic loading will significantly accelerate material fatigue damage, shorten the design life of the entire floating wind turbine, increase the risk of structural failure, and may even lead to catastrophic accidents.

[0006] In addition to power generation losses, wind turbines need to maintain a near-vertical orientation to ensure maximum wind energy capture efficiency. Severe platform movement can cause the turbine nacelle to tilt, preventing the rotor plane from aligning directly with the incoming wind direction, significantly reducing the wind energy utilization coefficient and resulting in direct power generation losses. Power quality deterioration is also a concern; platform swaying causes drastic fluctuations in generator output power, introducing low-frequency harmonics, causing voltage fluctuations and flicker on the grid side, severely impacting grid-connected power quality. This can not only violate grid connection regulations, leading to fines or even forced disconnection, but also damage the wind turbine's own inverters and electrical systems.

[0007] Lagging control forces the pump system to start more frequently and make greater adjustments in a shorter time to "catch up" with the platform's movement. This "catch-up" drastic operating mode will significantly increase the energy consumption and mechanical wear of the pump and its power system, reduce equipment life, and increase failure rate and maintenance costs.

[0008] When the platform's movement exceeds a safety threshold, the wind turbine control system will trigger a protective shutdown. Frequent unplanned shutdowns will directly lead to lost power generation revenue. In extreme cases, if the movement becomes uncontrollable, it may even be necessary to dispatch a maintenance fleet to conduct high-risk operations in adverse sea conditions, further increasing operation and maintenance costs and personnel risks.

[0009] The marine environment, especially extreme sea states, is characterized by its immense energy and rapid changes. Lagging control systems are simply unable to respond promptly and effectively to sudden storms or abnormal waves. This lack of responsiveness makes the platform most vulnerable when stability is most needed, making it highly susceptible to loss of control and potentially leading to serious consequences such as mooring system failure, water ingress into equipment, and other serious consequences, threatening the survivability of the entire facility.

[0010] When multiple ballast tanks are coordinated, delayed commands may become coupled with the actual motion state of the platform. For example, a water injection command intended to suppress starboard roll may be delayed and only take effect when the platform has already begun to roll back to port. As a result, it will not only fail to stabilize the platform, but will also exacerbate the next roll, leading to instability of the entire control system and further deterioration of its performance. Summary of the Invention

[0011] One object of the present invention is to solve at least the above-mentioned problems and to provide at least the advantages that will be described later.

[0012] To achieve these objectives and other advantages according to the present invention, a method for intelligent dynamic ballast water regulation of a floating semi-submersible wind turbine foundation is provided, comprising: S1. The semi-submersible base platform is equipped with a sensor system to collect the roll angle, pitch angle, heave displacement, wave height, wave period and ocean current speed in real time. The semi-submersible base platform includes at least three columns, a ballast tank installed inside the columns, a water pump system connected to the ballast tank, a heave plate arranged at the bottom of the columns, and a sensor system for monitoring the motion status of the semi-submersible base platform and marine environmental parameters. The sensor system includes an attitude sensor, an accelerometer, a water level sensor, a wave height sensor and a current meter. S2. Combine the real-time data collected in S1 with the ballast water distribution data extracted from the historical database for the same time period to form a comprehensive input vector. The integrated input vector is fed into the motion response prediction model trained on the historical dataset. The temporal characteristics of the integrated input vector are analyzed, and the predicted motion state of the semi-submersible base platform in the future is output. The historical dataset includes historical sea state data and its corresponding historical motion response data. The motion response prediction model adopts an architecture that combines a long short-term memory network with an attention mechanism. The predicted motion state includes the predicted roll angle, the predicted pitch angle, and the predicted heave displacement. S3. Calculate the total amount of ballast water to be adjusted and its distribution ratio in each ballast tank based on the predicted motion state. S4. Control the water pump system to inject or drain water into the designated ballast tank according to the allocation ratio, adjust the relative position of the center of gravity and the center of buoyancy of the semi-submersible foundation platform, continuously monitor the actual motion state of the semi-submersible foundation platform during the injection or drainage process, and compare it with the predicted value. When the deviation between the actual motion state and the predicted value exceeds the set threshold, the data acquisition step of S1 and the prediction step of S2 are re-executed to update the ballast water adjustment command.

[0013] Preferably, step S3 specifically includes: Based on the predicted roll angle, predicted pitch angle and predicted heave displacement output in step S2, the ballast water adjustment amount required to suppress roll motion, pitch motion and heave motion respectively is calculated. Based on the spatial distribution of each ballast tank in the semi-submersible base platform, the inhibition ratio of the unit water transfer volume of each ballast tank on the platform's roll, pitch and heave motions was calculated. The optimization objective is to minimize the total water transfer volume of all ballast tanks and achieve a balanced state in the suppression of roll, pitch and heave movements in all directions. A multi-objective optimization function with the water transfer volume of each ballast tank as the variable is established. Solve the multi-objective optimization function to obtain the total amount of ballast water to be adjusted and the optimal distribution ratio in each ballast tank.

[0014] Preferably, step S4 specifically includes: Based on the optimal allocation ratio obtained in step S3, specific execution instructions are generated. The execution instructions include the target ballast tank identifier, the direction of water injection or drainage operation, and the water adjustment amount. The execution instructions are grouped and sorted according to the time window, and simultaneously partitioned according to the spatial location of the ballast tanks to form a time-division and partitioned execution sequence. Within the same time window, synchronous water transfer operations are only allowed to be performed on ballast tank groups whose spatial distance exceeds the minimum interference distance. According to the time-sharing and zone-based execution sequence, the water pump system is controlled to perform water injection or drainage operations to the designated target ballast tank; during the water diversion operation, the actual motion state of the semi-submersible foundation platform, including roll angle, pitch angle, and heave displacement, is continuously monitored at a preset frequency. The actual motion state monitored is compared in real time with the predicted motion state output at the same time in step S2. When the deviation between the actual roll angle, actual pitch angle or actual heave displacement and the corresponding predicted value exceeds the set threshold, the currently executing water adjustment operation sequence is paused, and then steps S1 to S3 are re-executed to generate an updated ballast water adjustment command.

[0015] Preferably, in step S2, the motion response prediction model outputs a confidence score of its prediction results. When the confidence score is lower than a first confidence threshold, the execution of the adjustment instructions based on the motion response prediction model is paused, and the system switches to a real-time feedback stabilization mode. In the real-time feedback stabilization mode, the following steps are performed: With the goal of maintaining the platform's roll and pitch angles at zero degrees, the current roll and pitch angles collected in real time by the sensor system are obtained, and the real-time deviation e(t) between them and the zero-degree set value is calculated. According to the formula Calculate the restoring torque M required for the roll and pitch channels respectively, where K p K is the proportionality coefficient. i K is the integral coefficient. d These are the differential coefficients; Based on the pre-established ballast tank spatial position-moment influence matrix, the required recovery moment M is allocated and calculated as a specific water adjustment command for each ballast tank; The system controls the water pump system to perform water injection or drainage operations according to the water volume adjustment command until the confidence score recovers to above the second confidence threshold. Then, it exits the real-time feedback stabilization mode and re-executes the adjustment command based on the motion response prediction model, where the second confidence threshold is greater than the first confidence threshold.

[0016] Preferably, the weight coefficients of the multi-objective optimization function in step S3 are dynamic adaptive parameters. Based on the real-time effective wave height and ocean current speed collected in step S1, the preset sea state level-weight mapping table is queried to determine the current sea state level. The sea state level is divided into three levels: calm, normal, and severe, according to the range of effective wave height and ocean current speed. Based on the determined sea state level, the weight coefficients between the two optimization objectives of minimizing total water displacement and minimizing residual energy of motion attitude in the optimization function are dynamically adjusted. The higher the sea state level, the larger the weight coefficient corresponding to the objective of minimizing residual energy of motion attitude. The residual energy of motion attitude is calculated by the sum of the square root mean square values ​​of the predicted roll angle, predicted pitch angle and predicted heave displacement.

[0017] Preferably, before generating the time-sharing execution sequence, the execution instructions are prioritized, where the priority is calculated based on an urgency index E using a formula. ,in, This indicates the ratio of the amount of ballast water that needs to be adjusted to the total capacity of the ballast tanks. This represents the absolute value of the deviation between the current motion angle (roll or pitch) and the target state; α and β are preset weighting coefficients, and α+β=1, setting the emergency threshold E. th When the urgency index of the instruction E≥E th When the time is right, the command will be assigned to the first time window to be executed, and the spatial distance between it and other modules will be less than the minimum interference distance. The real-time roll angle, pitch angle or heave displacement of the semi-submersible base platform after adjustment will not exceed the set threshold.

[0018] Preferably, during ballast water regulation, the tension setpoints of each mooring line in the mooring system are adjusted synchronously based on the predicted motion state output in step S2. The adjustment strategy for the mooring system tension setpoints is as follows: With the goal of suppressing the platform's sway, pitch, and yaw motions, the required restoring force and restoring torque are calculated based on the predicted changes in the platform's horizontal displacement and heading angle during the motion state. The required restoring force and restoring torque are distributed to each mooring line according to the predetermined contribution of each mooring line, thus obtaining the target tension value for each mooring line; The tension of each mooring line is adjusted to its target tension value by the tension servo mechanism of the mooring system.

[0019] Preferably, the sensor system also includes redundant water level sensors installed in each ballast tank. If any water level sensor fails, the redundant sensors are activated to continue monitoring the water level in the ballast tank. The pump system includes a main pump and a standby pump. When the main pump fails, it automatically switches to the standby pump to continue the ballast water regulation operation.

[0020] Preferably, the rate of water injection or drainage is controlled at 5%-10% of the total amount of ballast water to be adjusted in a single adjustment operation per minute, and the threshold is set when the roll angle deviation exceeds 15% or the heave displacement deviation exceeds 20%.

[0021] Preferably, when the sensor system detects that the effective wave height exceeds 4m and the wind speed exceeds 20m / s, it is determined to be an extreme sea state. When it is an extreme sea state, the motion response prediction model outputs a predicted motion state that meets the constraints of roll angle not exceeding 10° and pitch angle not exceeding 8° within the next 300s. Step S3 calculates the required total amount of ballast water and its allocation ratio based on the predicted motion state.

[0022] The present invention has at least the following beneficial effects: First, this invention effectively overcomes the control command lag problem caused by relying on instantaneous data by constructing a comprehensive input vector that integrates real-time and historical state data, and by using a motion response prediction model based on long short-term memory networks and attention mechanisms for forward-looking analysis. This enables accurate prediction of the platform's future motion state, and based on this prediction, ballast water pre-adjustment is performed, significantly improving the attitude stability and safety of the semi-submersible platform under complex and variable sea conditions.

[0023] Secondly, this invention achieves a scientific and precise distribution of ballast water in each compartment by establishing a multi-objective optimization function aimed at minimizing the total water volume and balancing the motion suppression effect. Compared to a crude distribution mode that relies on manual experience or fixed proportions, this invention can accurately quantify the suppression effectiveness of each ballast compartment's unit operation on the platform's different degrees of freedom of motion, thereby efficiently coordinating and suppressing complex coupled motions such as roll, pitch, and heave. While significantly reducing the energy consumption of the pump system, it ensures the balance and efficiency of the overall stability control of the platform.

[0024] Third, this invention effectively solves the problems of hydraulic coupling interference and execution efficiency in the coordinated adjustment of multiple ballast tanks by adopting a time-sharing and zone-based collaborative scheduling mechanism, combined with real-time motion status feedback and rapid replanning capabilities. Spatial partitioning avoids mutual interference from simultaneous operations of adjacent tanks, time windowing optimizes the execution sequence, and real-time comparison and interruption replanning mechanisms address sudden changes in sea state, balancing the inherent contradiction between adjustment efficiency and system stability, and achieving a unity of rapid response and precise adjustment.

[0025] Fourth, this invention establishes an online incremental learning and update mechanism for the model, enabling the motion response prediction model to continuously evolve and self-optimize. It can automatically update model parameters using constantly accumulating new data, allowing it to adapt to long-term changes in the marine environment, minor alterations in platform characteristics, and new sea conditions never encountered before. This effectively overcomes the model "drift" problem, maintains high prediction accuracy and strong generalization ability over the long term, and ensures the durability and reliability of the entire intelligent control system's performance.

[0026] Other advantages, objectives and features of the present invention will be apparent in part from the following description, and in part from the understanding of those skilled in the art through study and practice of the invention. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating one of the technical solutions of the present invention. Detailed Implementation

[0028] The present invention will now be described in further detail with reference to examples, so that those skilled in the art can implement it based on the description.

[0029] It should be noted that, unless otherwise specified, the experimental methods described in the following implementation plan are all conventional methods, and the reagents and materials described are all commercially available unless otherwise specified.

[0030] like Figure 1 As shown, this invention provides a dynamic intelligent ballast water regulation method for floating semi-submersible wind turbine foundations. Existing ballast water regulation systems for floating semi-submersible wind turbine foundation platforms typically rely solely on real-time collected marine environmental parameters and platform motion data for feedback control. Because the temporal correlation between historical motion responses and the current state is not fully considered, ballast water regulation commands often lag behind actual sea state changes. Especially in situations with long wave periods or sudden current changes, decisions based solely on instantaneous data cannot accurately predict the platform's motion trend over the next few hundred seconds, leading to problems such as timing deviations in regulation and insufficient stability control. Existing methods also lack the ability to fuse multi-source data and extract dynamic features, further limiting their adaptability and robustness under complex sea conditions.

[0031] To address the aforementioned issues, this invention provides a method for intelligent dynamic ballast water regulation of a floating semi-submersible wind turbine foundation. A sensor system deployed on the semi-submersible foundation platform collects real-time data on various parameters, including roll angle, pitch angle, heave displacement, wave height, wave period, and ocean current velocity. This real-time data is then combined with ballast water distribution data extracted from a historical database for the same time period to form a comprehensive input vector. This vector includes not only current environmental and motion information but also historical operation records, thus providing a more comprehensive reflection of system dynamics.

[0032] The comprehensive input vector is fed into a motion response prediction model trained on a historical dataset. The motion response prediction model employs an architecture combining a long short-term memory network and an attention mechanism, which can effectively capture the temporal features and dependencies of the input data and output the predicted motion state of the semi-submersible base platform within the next 120-300 seconds, including predicted roll angle, predicted pitch angle, and predicted heave displacement.

[0033] After obtaining the predicted motion state, the total amount of ballast water to be adjusted and its distribution ratio in each ballast tank are calculated. Based on this ratio, the water pump system is controlled to perform water injection or drainage operations to adjust the relative position of the platform's center of gravity and center of buoyancy. Throughout the adjustment process, the actual motion state of the platform is continuously monitored and compared with the predicted value in real time. If the deviation between the actual state and the predicted value exceeds a set threshold, the data acquisition and prediction process is immediately restarted, and the adjustment commands are updated, thus forming a closed-loop control to ensure that the platform remains stable under different sea states.

[0034] Compared to control methods that rely solely on real-time data, this invention introduces historical data and motion response prediction models to achieve advanced prediction and adaptive adjustment of the motion state of the semi-submersible foundation platform. This effectively overcomes the problems of insufficient control precision caused by ballast adjustment lag and single data, thereby improving the overall system stability and reliability.

[0035] The motion response prediction model employed in this invention utilizes a deep learning architecture combining a Long Short-Term Memory (LSTM) network and an attention mechanism. Specifically, the network model comprises three LSTM layers, each with 128 neurons, used to deeply mine the temporal dependencies in the input data. Following this, an attention mechanism layer is connected, which assigns different weights to data at different time steps in the historical sequence, thereby focusing on the historical information most relevant to the current prediction. Finally, a fully connected layer outputs the final prediction result. The network model is trained using the Adam optimizer, with the mean squared error (MSE) loss function chosen to accurately fit the changing trends of the motion state. The model training epochs are set to 100, and 80% of the historical dataset is used for training, with 20% used for validation to ensure its generalization ability.

[0036] The historical database forms the data foundation for effective prediction, storing multidimensional time-series data. It primarily comprises two tables: one for sea state and environment data, with fields including timestamp, wave height, wave period, and current velocity; and another for platform status and operational data, with fields including timestamp, roll angle, pitch angle, heave displacement, and real-time water level or ballast water mass in each ballast tank. Each record is associated and indexed using precise timestamps, ensuring the extraction of environmental and platform status data from the same time period, which are then combined to form a comprehensive vector for model input.

[0037] It should be noted that the motion response prediction model of the present invention can be used to accurately predict the motion state of a semi-submersible base platform over a period of time in the future, for the following reasons: A strong nonlinear, time-varying coupling relationship exists between marine environmental loads (waves, currents, wind) and the platform's motion response, and semi-submersible platforms exhibit significant inertia. Long Short-Term Memory (LSTM) networks, due to their unique gating mechanism, can effectively capture and remember complex dependencies in such long-term sequences, overcoming the limitations of traditional methods in handling temporal correlations. The introduced attention mechanism further endows the model with the ability to dynamically focus on the key periods in the historical sequence most relevant to the current prediction, thereby enhancing the model's adaptability and prediction accuracy to non-stationary and abrupt sea states. This deep integration of LSTM and the attention mechanism fundamentally ensures that the model possesses the ability to describe the complex physical process of the platform's hydrodynamic response.

[0038] The model's input is not single-dimensional real-time data, but rather a comprehensive input vector constructed from multi-source heterogeneous data fusion. This vector simultaneously includes: real-time platform motion status (roll angle, pitch angle, heave displacement), real-time ocean environmental parameters (wave height, wave period, ocean current velocity), and historical ballast water distribution status over the same time period. This ensures that the model can simultaneously perceive three key types of information that determine the future dynamics of the semi-submersible platform: external stimuli (environmental loads), system response (platform motion), and historical control inputs (ballast status), providing a comprehensive and reliable information foundation for accurate predictions.

[0039] The model is not based on idealized theoretical assumptions, but rather trained through supervised learning using a large amount of historical datasets. The training data covers historical environmental data under various sea states (from calm to severe) and their corresponding platform motion response data, enabling the model to learn and internalize the platform's real hydrodynamic characteristics and motion patterns. By employing a reserved validation set for cross-validation, using regularization techniques, and optimizing training strategies (such as the Adam optimizer and Early Stopping), overfitting is effectively avoided, ensuring the model possesses good generalization ability and can reliably predict new sea states that are not previously encountered but belong to the same distribution.

[0040] The model outputs quantitative, time-series predictions of the platform's motion state over a future period (e.g., 120-300 seconds), specifically including predicted roll, pitch, and heave displacements. These outputs are core indicators for directly evaluating platform stability. The purpose of these predictions is to provide forward-looking decision-making support for feedforward control, enabling the calculation and execution of optimal ballast water regulation strategies based on predicted future motion trends. This fundamentally addresses the feedback control lag problem caused by the system's large inertia and actuator delays, achieving a transition from passive response to active suppression.

[0041] Comparative experiments were conducted using professional marine engineering simulation software (such as AQWA and OrcaFlex) or established high-fidelity numerical models. Simulation results show that, under typical operating conditions, the feedforward-feedback composite control strategy based on this predictive model significantly reduces the root mean square (RMS) value of the platform's motion response (e.g., a reduction of approximately 35% in roll angle) compared to pure feedback control (such as PID), and substantially shortens the system's response time (by an average of approximately 40%). These data, from an engineering application perspective, strongly demonstrate that the motion response prediction model possesses the required predictive accuracy and reliability, and can effectively improve overall control performance.

[0042] In the ballast water allocation calculation stage, this invention employs a multi-objective optimization algorithm. An optimization function is established, with the water transfer volume of each ballast tank as the decision variable. The optimization objectives are twofold: first, to minimize the total water transfer volume of all ballast tanks to save energy; and second, to minimize the residual energy of the platform's predicted motion attitude, i.e., to suppress roll, pitch, and heave motions as much as possible. This is a typical trade-off problem. For example, a linear weighted sum method can be used to transform the multi-objective problem into a single objective, i.e., constructing a total objective function as a weighted sum of the two sub-objectives, and then using gradient descent or intelligent optimization algorithms to find the optimal ballast water allocation scheme, thereby obtaining the specific water transfer volume of each ballast tank.

[0043] Simulation comparison experiments verified that, under simulated moderate sea state conditions (significant wave height of 2.5 meters and wave period of 8 seconds), the root mean square value of the roll angle of the semi-submersible base platform was reduced by approximately 35% compared to the PID feedback control method after adopting the adjustment method of this invention. Simultaneously, due to the predictive effect of the predictive model, the average response time for adjusting the platform's motion state was shortened by 40%, significantly improving its adaptability to complex sea conditions and the overall stability of the platform.

[0044] In another embodiment of the present invention, in existing ballast water regulation strategies for floating semi-submersible wind power foundation platforms, the allocation of ballast water often relies on the operator's experience or a pre-set, simple, fixed ratio. This method lacks a precise mathematical description of the complex coupling relationships between the platform's multiple degrees of freedom of motion, making it difficult to simultaneously and efficiently suppress movements in multiple directions such as roll, pitch, and heave. As a result, ballast water regulation often results in inconsistencies; while it may suppress movement in one direction, it may exacerbate swaying in other directions, or it may consume excessive ballast water to achieve a certain stabilization effect, leading to low regulation efficiency and uneven overall stability control of the platform.

[0045] In this implementation, after obtaining the predicted roll angle, predicted pitch angle, and predicted heave displacement output by the motion response prediction model, the required ballast water adjustment amount for each directional motion to be suppressed to the target range is calculated. This step decomposes the overall motion control objective of the platform into specific requirements for each degree of freedom.

[0046] Then, based on the specific spatial coordinates of each ballast tank in the semi-submersible base platform, the change in righting torque or buoyancy generated by a unit water adjustment volume of each ballast tank is accurately calculated, thereby quantifying its effectiveness in suppressing the platform's roll, pitch, and heave motions. A quantitative relationship model between ballast tank operation and platform motion response is established.

[0047] A multi-objective optimization function is established, with the water adjustment volume of each ballast tank as the decision variable. This function has two optimization objectives: first, to minimize the total water adjustment volume of all ballast tanks to maximize energy savings in the pump system; and second, to achieve a balanced state in suppressing roll, pitch, and heave motions, avoiding over-suppression of one motion while neglecting another. By solving this multi-objective optimization problem, the total amount of ballast water to be adjusted under the current sea state and the optimal allocation ratio among the various ballast tanks can be obtained. This method represents a leap from experience-based coarse allocation to model- and optimization-based precise and coordinated allocation, significantly improving the efficiency of ballast water adjustment and the overall stability of the platform under multi-degree-of-freedom motion.

[0048] In the intelligent regulation method for dynamic ballast water of floating semi-submersible wind turbine foundations, the mathematical model, formula, and algorithm for step S3 include: 1. Calculation of ballast water adjustment required for movement in all directions: Based on the predicted roll angle output in step S2 Predicting the pitch angle and predicting heave displacement It is necessary to calculate the amount of ballast water adjustment required to suppress the motion of each degree of freedom to the target state.

[0049] Taking the roll motion as an example, the required restoring torque It can be estimated by the following formula: in, The rolling recovery moment coefficient is related to the platform's displacement, initial stability (GM), and other hydrodynamic characteristics. The target value for the roll angle is usually set to 0°. Similarly, the pitch restoring torque can be calculated. and sag recovery .

[0050] Based on the relative positions of the ballast tanks and the platform's center of gravity, the aforementioned restoring moments and restoring forces are converted into mass adjustment amounts for each ballast tank. Let the mass change of the i-th ballast tank be... Its contribution to the roll moment is Where g is the acceleration due to gravity, d x,i This is the lateral distance from the center of gravity of the compartment to the longitudinal axis of the platform. By solving the system of linear equations constrained by the total restoring moment matching, a preliminary ballast water adjustment vector can be obtained. .

[0051] 2. Quantification of the inhibition effect ratio per unit water adjustment volume: To evaluate the effectiveness of each ballast tank's unit water transfer in suppressing platform motion, an effect ratio vector needs to be defined. For the i-th ballast tank: its suppression effect ratio r on roll motion is... roll,i Defined as: , where I roll Let be the moment of inertia of the platform's lateral rotation about its longitudinal axis.

[0052] The ratio of inhibition effect on pitch motion r pitch,i for: , where d y,i I is the distance to the horizontal axis. pitch This is the moment of inertia of pitching rotation.

[0053] The ratio of inhibition effect on heave motion r heave,i It can be modeled as: Where ρ is the density of seawater, and A wp The waterline area of ​​the platform.

[0054] The effect ratio vector of the i-th ballast tank is .

[0055] 3. Establishment and solution of multi-objective optimization functions: Establish a system based on the water volume adjustment of each ballast tank. This is a multi-objective optimization problem with decision variables, and the objective function is as follows: Objective 1: Minimize the total water diversion volume. Objective two: Minimize residual motion inhibition. .

[0056] The multi-objective problem is transformed into a single-objective optimization problem using the linear weighted sum method: Where w1 and w2 are weighting coefficients, satisfying w1+w2=1, and can be dynamically adjusted according to sea conditions.

[0057] The optimization problem can be solved using numerical optimization methods such as gradient descent, quadratic programming, or genetic algorithms. The solution obtained is the optimal ballast water allocation scheme, including the total water transfer volume. and the allocation ratio of each cabin .

[0058] In another embodiment of the present invention, the ballast water regulation system of existing floating semi-submersible wind power foundation platforms typically employs a simple sequential control strategy, i.e., sequentially injecting or dewatering each ballast tank. This sequential execution method is not only inefficient overall, but also prone to additional disturbances in platform movement due to the hydrodynamic coupling effect of ballast water flow between tanks during multi-tank coordinated regulation, causing mutual interference between the regulation effects of each tank. Especially when encountering sudden changes in sea state, this lagging sequential execution mode is difficult to respond quickly, easily leading to control instability and failing to achieve a good balance between regulation efficiency and system stability.

[0059] This implementation method, after generating specific execution instructions based on the optimal allocation ratio, introduces a time-sharing and zone-based collaborative scheduling mechanism. The execution instructions explicitly include the target ballast tank identifier, the direction of water injection or drainage operation, and the water transfer volume. These execution instructions are grouped according to certain time windows and simultaneously divided into regions based on the actual spatial location of the ballast tanks on the platform, forming a time-sharing and zone-based execution sequence. The core principle of the sequence is that, within the same time window, only ballast tank groups whose spatial distance exceeds a preset minimum interference distance are allowed to perform synchronous water transfer operations, thereby effectively avoiding hydraulic coupling interference that may be caused by simultaneous operations of adjacent tank groups.

[0060] During the water diversion operation, the actual motion status of the semi-submersible base platform is continuously monitored once per second, including roll angle, pitch angle, and heave displacement. This real-time monitoring data is rapidly compared with the predicted motion status output by the motion response prediction model at the same moment. If any deviation between the actual roll angle, actual pitch angle, or actual heave displacement and the corresponding predicted value exceeds a set safety threshold, the entire water diversion operation sequence currently in progress will be immediately paused. The entire process of calculation, from data acquisition to motion prediction to optimized allocation, will be automatically restarted to generate and execute updated ballast water regulation commands, forming a highly adaptive closed-loop control loop.

[0061] By combining the above-mentioned time-sharing and zone-based execution with real-time feedback correction, this invention effectively overcomes the inherent defects of existing sequential execution methods, significantly improves the efficiency and accuracy of multi-ballast tank coordinated adjustment, and ensures that the semi-submersible base platform can maintain a high degree of motion stability even in complex and changeable sea conditions.

[0062] In the intelligent regulation method for dynamic ballast water of floating semi-submersible wind turbine foundations, the key algorithm, execution logic, and judgment conditions of step S4 include: 1. Structured definition and generation of execution instructions: Based on the optimal allocation ratio obtained in step S3, specific ballast water adjustment commands are generated. Each command can be formally defined as a structure: Among them, CID j OP is the unique identifier for the target ballast tank. j This indicates the operation type, and its value can be an element from the set {water injection, drainage}. This indicates the volume of ballast water that needs to be adjusted, in cubic meters (m³). 3 ).

[0063] 2. Algorithm for constructing time-sharing and partitioned execution sequences: The execution instruction set must be scheduled according to time windows and spatial partitions, and its generation process follows the following algorithm steps: Time window division: Divide the total operation time into K consecutive time windows. Duration of each window For example, a fixed value. =10s.

[0064] Spatial distance constraints, calculate CID for any two ballast tanks m With CID n The spatial Euclidean distance d(m,n) between them. A minimum interference distance threshold d is set. min (e.g. d) min=15.0m). For a group of instructions to be executed within the same time window, any two modules within the group must satisfy d(m,n)≥d. min .

[0065] The scheduling sequence is generated using a constraint-based scheduling algorithm (such as a greedy algorithm or a constraint programming algorithm) to allocate the instruction set to each time window, ultimately outputting an ordered sequence. G k This is the subset of instructions that can be executed in parallel within the k-th time window.

[0066] 3. Real-time motion status monitoring and feedback correction mechanism: During execution, the actual motion state vector of the platform is acquired at a fixed sampling frequency fs=1Hz: ,in , and These represent the actual roll angle, pitch angle, and heave displacement, respectively.

[0067] Simultaneously, obtain the predicted motion state vector at the same moment: , Calculate the deviation vector: , Set the deviation threshold for each degree of freedom as ϕ th θ th and z th (e.g., ϕ) th =3.0°). The interruption and replanning conditions of feedback control are defined by the following logical predicates: .

[0068] Pause the current execution sequence Seq, re-execute steps S1 to S3, generate new adjustment instructions based on the updated output, and continue execution.

[0069] In another embodiment of the present invention, the motion control of floating semi-submersible wind power foundation platforms in the prior art largely relies on feedback based on real-time data or a single feedforward control strategy. These methods typically lack an evaluation mechanism for the reliability of the control commands themselves. When the motion response prediction model encounters unreliable predictions due to unknown sea conditions or extreme conditions outside the training data, blindly executing adjustment commands based on these predictions will not only fail to stabilize the platform but may also exacerbate its motion due to erroneous operations, potentially leading to control instability, equipment damage, or power generation interruption. There is a lack of an effective backup mechanism to ensure the basic stability of the platform when the model prediction confidence is low.

[0070] This implementation, while outputting the prediction result from the motion response prediction model, also outputs a confidence score characterizing the reliability of that prediction result. This score comprehensively reflects the model's grasp of the current input features and the degree of agreement between the predicted output and historical patterns. A first confidence threshold is preset. When the confidence score is detected to be below this threshold, it is immediately determined that the current environment has exceeded the reliable operating range of the prediction model. All subsequent ballast water regulation commands based on this unreliable prediction are suspended, thereby avoiding secondary risks that may arise from misoperation.

[0071] Then, it automatically switches to a PID control mode that does not rely on long-term predictions but only on the current real-time motion state feedback. In this mode, the control system's direct objective is to keep the platform's roll and pitch angles as close to zero degrees as possible. It acquires the platform's current motion attitude in real time through a sensor system and calculates the real-time deviation between it and the zero-degree setpoint. Based on this deviation, it quickly calculates the instantaneous restoring torque required to offset this deviation through proportional, integral, and derivative calculations. Then, according to the pre-calibrated relationship between the spatial position of the ballast tanks and the torque effect, it distributes this restoring torque as a water adjustment command for specific ballast tanks and controls the water pump system to perform corresponding water injection or drainage operations.

[0072] This PID control mode will run continuously, constantly monitoring the confidence score output by the motion response prediction model. Only when the confidence score recovers and stabilizes above the second confidence threshold is it determined that the external environment has returned to the range that the model can accurately predict. At this point, the system will automatically exit the real-time feedback stabilization mode and return control to the adjustment mode based on the motion response prediction model, thereby restoring the system's optimal performance while ensuring safety.

[0073] Specifically, including: 1. Confidence scoring mechanism for motion response prediction models: Motion response prediction models output a confidence score (Cscore) characterizing the reliability of their predictions, along with the predicted motion state. This score is a scalar value between 0 and 1, calculated based on the quantification of the uncertainty of the model's predictions or its consistency with historical patterns. One feasible approach is to calculate the variance of the probability distribution of the predicted sequence, or to utilize the dispersion of the predictions from multiple sub-models in an ensemble learning model. For example, if the model output follows a Gaussian distribution, the confidence score can be inversely proportional to the prediction variance. , where σ 2 This is a variance estimate of the model's predicted output based on the current input. A preset first confidence threshold C is used. th1 (e.g., 0.6) and the second confidence threshold C th2 (For example, 0.75, and C) th2 >C th1 ).

[0074] 2. Logic for mode switching: Real-time comparison confidence score C score The logic for mode switching, relative to a preset threshold, is defined by the following predicate: C score <C th1 Pause the execution of adjustment commands based on the motion response prediction model (i.e., pause steps S3 and S4). Switch to real-time feedback stabilization mode (i.e., PID control mode).

[0075] In real-time feedback stable mode, C is continuously monitored. score .

[0076] C score≤ C th2 It remains in a stable mode with real-time feedback.

[0077] C score >C th2 Exit the real-time feedback stabilization mode and switch back to predictive control mode. Re-execute the adjustment instructions based on the motion response prediction model (i.e., resume execution steps S2 to S4).

[0078] 3. Specific algorithm for real-time feedback stable mode (PID control): In real-time feedback stable mode, the control system aims to maintain the platform's roll angle ϕ and pitch angle θ at zero degrees.

[0079] Real-time deviation calculation obtains the current roll angle ϕ collected in real time by the sensor system. actual (t) and pitch angle θ actual (t), calculate its real-time deviation from the zero-degree setpoint: , .

[0080] The PID controller calculates the required restoring torque M for the roll and pitch channels respectively, based on the standard PID control algorithm. ϕ and M θ : , , Among them, K p,ϕ K i,ϕ K d,ϕ For the proportional, integral, and differential coefficients of the roll channel; K p,θ K i,θ K d,θ These are the proportional, integral, and differential coefficients for the pitch channel. These coefficients need to be pre-tuned through simulation or experimentation based on the specific dynamic characteristics of the semi-submersible foundation platform.

[0081] The torque distribution is calculated based on a pre-established ballast tank spatial location-torque influence matrix T (an n×2 matrix, where n is the number of ballast tanks, and the columns represent the contribution of each ballast tank unit water transfer volume to the roll and pitch moments). The required restoring torque distribution is then converted into specific water transfer volume commands for each ballast tank. (An n×1 vector). This is typically achieved by solving an optimization problem or a system of linear equations, for example: Solve under constraints .

[0082] The command is executed based on the calculated water diversion volume command. Control the water pump system to perform water injection or drainage operations.

[0083] In another embodiment of the present invention, when allocating ballast water using a multi-objective optimization function with fixed weights, the same set of preset weight coefficients is used under different sea states to coordinate the two optimization objectives of minimizing total water transfer and minimizing residual energy of motion attitude. However, the marine environment is complex and variable, and the fixed-weight strategy is difficult to adapt to different sea state conditions, from calm to severe. Under normal or severe sea states, the platform's motion response intensifies. If energy saving is still the primary objective, it may lead to insufficient motion suppression and decreased platform stability. Conversely, if motion suppression is excessively pursued, it may result in excessive energy consumption and overly frequent adjustment actions. Therefore, the fixed-weight strategy lacks responsiveness to actual sea conditions and cannot achieve dynamic optimal adjustment.

[0084] This implementation employs a sea state adaptive dynamic weight adjustment mechanism. Based on real-time collected effective wave height and current velocity, it categorizes the current sea state into three levels: calm, normal, and severe by querying a preset sea state level mapping table. Each sea state level corresponds to a different range of wave height and current velocity, thus objectively reflecting the severity of the external environment.

[0085] After determining the current sea state level, the weight coefficients of the two objectives—minimizing total water transfer and minimizing residual energy in motion attitude—are dynamically adjusted in the multi-objective optimization function. Specifically, at lower sea state levels, energy conservation is prioritized, and the objective of minimizing total water transfer is given a higher weight. As the sea state level increases, the weight of the objective of minimizing residual energy in motion attitude is gradually increased to prioritize platform stability. Residual energy in motion attitude is characterized by the sum of the squared root mean square values ​​of the predicted roll angle, predicted pitch angle, and predicted heave displacement; a smaller value indicates more stable platform motion.

[0086] Through the aforementioned sea state adaptive weight adjustment mechanism, this implementation enables the ballast water regulation strategy to perceive and respond to environmental changes, and can automatically achieve a balance of regulation targets under different sea states. This not only improves the stability of the platform, but also avoids unnecessary energy consumption, and significantly enhances the adaptability and robustness of the overall system.

[0087] Specifically, including: Sea state classification, real-time acquisition of significant wave height (H) measured by sensors. s (unit: m) and ocean current velocity (V) c (Unit: m / s). Sea state levels are determined based on a pre-defined numerical range using conditional logic. Specific classification criteria are as follows: If H s ≤H1 and V c If the sea state is ≤V1, then the sea state level is determined to be "calm"; If H1 <H s ≤H2 or V1 <V c If the sea state is ≤V2, then the sea state level is determined to be "normal". If H s >H2 or V c If the sea state is >V2, then the sea state level is determined to be "severe".

[0088] Wherein, H1, H2, V1, and V2 are pre-set threshold parameters based on marine environmental statistics of the target sea area (e.g., H1=1.0m, H2=2.5m, V1=0.5m / s, V2=1.2m / s). This judgment logic is implemented through simple comparison operations, and finally outputs a discrete sea state level indicator.

[0089] The residual energy of motion posture is calculated as follows: The predicted motion state sequence output by the motion response prediction model for a future period contains data from N time points. For the three degrees of freedom—roll angle, pitch angle, and heave displacement—the root mean square value of each is calculated within this sequence.

[0090] Root mean square value of roll angle (RMS) （roll） The calculation formula is: , Similarly, calculate the root mean square value (RMS) of the pitch angle. （pitch） and the root mean square value of heave displacement (RMS) （heave） .

[0091] Then the residual energy E of the motion posture （residual） The calculation formula is: The calculation result E （residual） It is a scalar; the larger its value, the more violent the predicted platform movement and the worse the stability.

[0092] A multi-objective optimization function and a dynamic adjustment algorithm for weight coefficients are proposed. The general form of the multi-objective optimization function is: minF = w1 × f1 + w2 × f2, where f1 represents the objective of "minimizing the total water transfer volume," and its value is the sum of the absolute values ​​of the water transfer volumes of all ballast tanks, i.e., | f2 represents the objective of "minimizing motion suppression residuals," and its value is the residual energy E of the motion posture calculated according to the above algorithm. (residual) Or it could be some functional form of that energy (such as directly using E). (residual) (Or the value after standardization). w1 and w2 are weighting coefficients, and w1 + w2 = 1.

[0093] The dynamic adjustment algorithm for the weighting coefficients is based on the aforementioned sea state levels: When the sea state level is "calm," energy conservation is prioritized, with w1 set relatively large and w2 relatively small (e.g., w1=0.7, w2=0.3). When the sea state level is "normal," energy conservation and stability are balanced, with w1 and w2 set to similar values ​​(e.g., w1=0.5, w2=0.5). When the sea state level is "severe," stability is prioritized, with w1 set relatively small and w2 relatively large (e.g., w1=0.3, w2=0.7). The specific values ​​of these weighting coefficients and their correspondence with sea state levels are pre-stored in a "sea state level-weight mapping table" for querying. This adjustment process is a lookup-based assignment operation.

[0094] In another embodiment of the present invention, the ballast water regulation process for floating semi-submersible wind power foundation platforms typically employs a scheduling strategy that strictly adheres to spatial zoning and time windowing. To ensure stability, this method mandates that only ballast tank groups spatially separated from the minimum disturbance distance can be operated within the same time window to avoid hydraulic coupling interference. However, this mechanical scheduling rule reveals significant shortcomings in emergency situations such as sudden changes in sea state. When the platform's motion deteriorates rapidly and large-scale, rapid ballast water regulation is urgently needed, critical commands may be forcibly delayed because the target ballast tank is too close to the currently operated tank group, forcing the system to wait for the next available time window. This inherent delay prevents the system from responding promptly to emergencies, creating a high-risk window for platform stability control.

[0095] This implementation uses a time-sharing and zone-based collaborative scheduling mechanism with dynamic priority evaluation to calculate an urgency index for each ballast water adjustment command before generating the execution sequence. This index comprehensively considers two key factors: first, the proportion of ballast water to be adjusted relative to the total capacity of the target ballast tanks, reflecting the scale of the adjustment task; and second, the absolute value of the deviation between the platform's current actual motion angle and the target motion angle, directly reflecting the severity of platform instability. These two dimensionless parameters are weighted and combined to obtain a quantitative urgency evaluation value.

[0096] An urgency threshold is preset. When the urgency index of a command exceeds this threshold, the scheduling system automatically marks it as a critical command. Such critical commands will be allocated to the highest priority execution window, typically for immediate execution. More importantly, to ensure that critical commands can be responded to immediately, spatial distance constraints are temporarily relaxed, allowing them to be within a preset minimum interference distance from other operating modules. Priority scheduling is conducted under the premise of ensuring the overall safety of the platform. The platform's motion status after command execution is monitored in real time to ensure that its roll, pitch, or heave displacement does not exceed the safety threshold, thereby improving response speed while firmly maintaining the bottom line of stability.

[0097] By employing a priority scheduling mechanism based on urgency calculation, this implementation adds a priority dimension to the existing spatial and temporal scheduling dimensions, forming a hybrid scheduling strategy. It can handle routine adjustment tasks methodically and efficiently, maintaining stability, just as it would in normal times. Simultaneously, in emergency situations, it can act like a special forces unit, quickly identifying and prioritizing the most critical tasks. This achieves a higher level of balance between the inherent contradiction of efficiency and stability, significantly enhancing the ability to respond to sudden sea conditions.

[0098] Specifically, this includes: the formula for calculating the urgency index E. , (The ratio of the ballast water volume to be adjusted to the total capacity of the ballast tanks) is a dimensionless scalar quantity. ΔV j This indicates the volume of ballast water required to be adjusted in the target ballast tank as specified in the instruction (unit: m³). 3 A positive value indicates water injection, and a negative value indicates drainage. The absolute value |ΔV| is used in the calculation. j |。V total,j This indicates that the target ballast tank (by CID) j Total design capacity of the sign (unit: m) 3 The ratio ΔV reflects the relative scale of this water diversion operation; a larger value indicates a larger operation scale and potentially higher urgency. |Δθ| (the absolute value of the deviation between the current motion angle and the target angle) is a dimensionless scalar (unit: °), calculated as follows: θactual This represents the platform's current roll or pitch angle (unit: °) as collected in real time by the sensor system. θ target The target angle for this degree of freedom motion is typically set to 0°. The absolute value |Δθ| directly reflects the severity of the platform's current motion instability; a larger value indicates greater instability and a more urgent need for adjustment. α and β (preset weighting coefficients) are pre-set constants greater than 0, used to adjust the relative importance of ΔV and |Δθ| in the urgency assessment. They satisfy the constraint α + β = 1. The specific values ​​of these coefficients (e.g., α = 0.4, β = 0.6) can be pre-tuned through simulation or experimentation based on the platform's specific dynamic characteristics and control strategy.

[0099] The urgency index E is calculated by substituting the calculated ΔV and |Δθ| into the formula. The weighted summation yields a scalar value E between 0 and 1. The larger the value of E, the higher the urgency of the water diversion order.

[0100] Priority scheduling algorithm, urgent threshold E th It is a preset constant threshold (e.g., E). th =0.7), used to determine whether an instruction belongs to the "critical instructions" that require immediate processing. This threshold can be dynamically adjusted according to sea conditions. For example, in severe sea conditions, the threshold can be lowered so that more instructions are classified as critical instructions, thereby improving responsiveness.

[0101] Command traversal and calculation: For all ballast water regulation command sets generated in step S3, traverse each command and calculate its urgency index E according to the above formula.

[0102] Critical instruction determination involves comparing the E value of each instruction with the emergency threshold E. th Compare them. If E ≥ E th If so, then mark the instruction as a "critical instruction".

[0103] During scheduling execution, all operations marked as "critical instructions" will be immediately assigned to the most readily available execution window (usually immediate execution), without waiting for the regular scheduling sequence. For these critical instructions, a temporary exemption is granted for the requirement that "the spatial distance between operational modules within the same time window must be greater than the minimum interference distance d". min The constraint is that the ballast tank space distance between a critical instruction and other operations (whether critical or routine) being performed within the same time window is less than d. min .

[0104] Safety constraints necessitate continuous monitoring of the platform's response, even when executing critical commands. Command execution must adhere to a strict constraint: the platform's real-time roll, pitch, or heave displacement after adjustment must not exceed their respective safety thresholds. This is the safety baseline to prevent instability caused by priority scheduling.

[0105] In another embodiment of the present invention, the motion control of a floating semi-submersible wind power foundation platform typically treats ballast water regulation and mooring system control as two independent subsystems. The ballast water regulation system, by adjusting the distribution of ballast water within the platform, primarily controls the platform's roll, pitch, and heave motions. However, for low-frequency horizontal plane motions caused by wind, ocean currents, and second-order wave forces—namely, sway, pitch, and bow—a single ballast water regulation method is ineffective or even negligible. While the mooring system provides overall positioning and restoring force for the platform, its traditional control strategies are mostly static or slow-responding, unable to effectively coordinate with the rapidly changing ballast water regulation actions. This control method results in a weakness in the platform's overall stability, making it difficult to achieve comprehensive and precise suppression of six degrees of freedom motion.

[0106] This embodiment proposes a coordinated control mechanism for ballast water regulation and mooring system tension setting. While regulating ballast water based on the output of a motion response prediction model, the tension setpoints of each mooring line in the mooring system are simultaneously and dynamically adjusted. Specifically, the coordinated control objective is to suppress the platform's sway, pitch, and bow movements. It rapidly calculates the overall restoring force and restoring torque required to counteract these unwanted low-frequency movements based on the platform's horizontal displacement and heading angle changes included in the predicted motion state.

[0107] After calculating the required total restoring force and restoring moment, these forces and moments are rationally allocated to each mooring line according to a pre-calibrated mooring line contribution allocation matrix. This allocation matrix defines the effectiveness weight of each mooring line in providing restoring force or restoring moment in a specific direction. Through this allocation, a precise target tension value is calculated for each mooring line, representing the tension level that the mooring line should achieve to achieve optimal stability under the current sea state and conditions.

[0108] The mooring system utilizes tension servo mechanisms, such as hydraulic winches or electric actuators, to rapidly and precisely adjust the actual tension of each mooring line to its target tension value. This process is synchronized with the ballast water pumping operations, ensuring that ballast water adjustment and mooring system adjustment are no longer two isolated actions, but rather a highly coordinated whole. The ballast water system primarily suppresses rotational and vertical motion, while the mooring system focuses on suppressing translational and rotational motion in the horizontal plane. Together, they form a complete six-degree-of-freedom motion control system, significantly improving the overall stability and safety of the semi-submersible platform in complex marine environments.

[0109] Specifically, including: Calculation of restoring force and restoring torque: Based on the platform's horizontal displacement (Δx, Δy) and bow angle change (Δψ) in the predicted motion state, calculate the required restoring force (F). x ,F y ) and restoring torque (M ψ ): , , , where k x k y k ψ The restoring force / moment coefficient is pre-calibrated based on the platform's hydrodynamic characteristics; Δx, Δy, and Δψ represent the changes in the platform's horizontal displacement and heading angle during the predicted motion state.

[0110] Mooring line contribution distribution matrix: Suppose there are m mooring lines in the mooring system. The contribution of each line to the platform's restoring force / moment constitutes a contribution matrix. : , Where: c x,i c represents the contribution of the i-th mooring line to the sway restoring force; y,i Indicates the contribution to oscillation recovery; c ψ,i This represents the contribution to the bow roll restoring moment. The contribution matrix C needs to be pre-calibrated using the platform mooring system layout diagram and hydrodynamic simulation.

[0111] Solving for the target tension value: The required restoring force / torque vector is obtained. Distributed to each mooring line according to contribution level, target tension vector Solve the following least squares optimization problem: And suffer Constraints, among which, , Let be the upper and lower limits of the tension of the i-th mooring line; the optimization problem can be solved using quadratic programming (QP) or least squares method.

[0112] Tension servo control: The target tension value T obtained by the solution is... i The tension servo mechanism (such as a hydraulic winch or electric actuator) sent to the mooring system tracks tension via a PID controller: ,in, For tension error; u i (t) represents the control signal for the servo mechanism.

[0113] In another embodiment of the present invention, the existing ballast water regulation system for floating semi-submersible wind power foundation platforms typically employs a single configuration for both sensors and pumps. A water level sensor is installed inside the ballast tank to monitor the water level, while the pump performs the injection and drainage operations. This configuration presents a significant risk of a single point of failure. If a water level sensor malfunctions, the water level information for that ballast tank cannot be accurately obtained, potentially leading to misadjustments based on incorrect data, or even platform instability. Similarly, if the sole pump fails, the entire ballast water regulation operation will be forced to halt, completely losing its regulation capability until the fault is repaired, severely threatening the platform's stability under changing sea conditions.

[0114] This implementation employs a redundant design in its sensor system. Inside each ballast tank, in addition to a primary water level sensor, a redundant water level sensor is installed. The primary and redundant water level sensors are physically independent, employing the same measurement principle and accuracy, and jointly monitor the water level of the same ballast tank in real time. The readings of the two sensors are continuously compared to ensure consistency. If an anomaly is detected in the primary water level sensor data of a ballast tank—such as a sudden change in reading, exceeding a reasonable range, or logically conflicting with data from other sensors—the primary sensor is automatically identified as faulty. The system immediately and seamlessly switches to the redundant water level sensor for that tank, which then takes over and continues to provide accurate water level monitoring data. This ensures the continuity and reliability of the input information to the ballast water regulation system, effectively preventing regulation interruptions or misadjustments caused by the failure of a single sensor.

[0115] The pump system also employs a primary / standby redundancy architecture. The system comprises one main pump and one standby pump, both with identical performance parameters and operational capabilities, each capable of independently injecting or dewatering water into any ballast tank. Under normal circumstances, the main pump executes all ballast water regulation commands. The main pump's operating status is continuously monitored, including its motor current, outlet pressure, and flow rate. If any abnormal operating parameters are detected, such as current overload, insufficient pressure, or abnormal flow rate, a main pump failure is diagnosed. The control system immediately issues a command to stop the main pump and almost simultaneously activates the standby pump, transferring any unfinished ballast water regulation commands to the standby pump for continuation. The entire switching process is rapid and automatic, ensuring continuous and uninterrupted ballast water regulation operations and significantly enhancing the system's robustness and survivability in the face of critical component failures.

[0116] By implementing a dual redundancy design for both the sensor and pump systems, this invention constructs a highly reliable intelligent ballast water regulation system. Even in extreme cases of partial hardware failure, it can maintain basic regulation functions, continuously ensuring the stability of the semi-submersible foundation platform and significantly improving the operational safety and reliability of floating wind power units in harsh marine environments.

[0117] In another embodiment of the present invention, the rate of water injection or drainage is often set based on experience, lacking clear theoretical basis and systematic constraints. Operators or control systems may adopt a high water adjustment rate in pursuit of rapid attitude adjustment, which can easily lead to drastic changes in the platform's center of gravity in a short period of time. Changes in the platform's buoyancy and stability require a certain response time. Excessively rapid ballast water transfer can trigger significant inertial forces and hydrodynamic shocks, potentially exceeding the platform's own recovery capacity, thereby inducing instantaneous instability of the platform or even causing dangerous situations such as resonance, posing a serious threat to the platform's structural safety and the operation of the wind turbine.

[0118] Furthermore, the standards for monitoring and judging the platform's motion status during the adjustment process are rather vague. Typically, only a broad allowable deviation range is set, or the judgment of whether the adjustment effect deviates from expectations relies on the operator's subjective experience. This management approach, lacking quantitative and precise thresholds, makes it difficult to capture abnormal changes in the platform's motion status in a timely and accurate manner. Once the adjustment effect fails to meet expectations, the platform's attitude deviation may continue to accumulate and amplify due to the failure to trigger a correction mechanism in a timely manner, ultimately leading to adjustment failure. It may even require longer and more drastic interventions to restore stability, severely impacting the platform's power generation efficiency and equipment lifespan.

[0119] This implementation method employs precise quantitative control of the ballast water regulation process. The rate of water injection or drainage is strictly limited to 5%-10% of the total ballast water volume per minute. This rate range was determined through extensive hydrodynamic simulations and pool tests. It ensures that the ballast water regulation has the necessary response speed to cope with changes in sea state, while also ensuring a sufficiently smooth change in the platform's center of gravity, avoiding excessive inertial disturbance torque caused by rapid water adjustment. This effectively prevents the risk of instantaneous platform instability and lays a solid foundation for the stability of the entire regulation process.

[0120] Furthermore, this invention sets clear and quantifiable motion state monitoring thresholds. It stipulates that during ballast water regulation, if the deviation between the platform's actual roll angle and the predicted value exceeds 15%, or the deviation between the actual heave displacement and the predicted value exceeds 20%, the current regulation effect is immediately determined to have deviated from expectations. These thresholds are determined comprehensively based on the platform's structural characteristics, motion response characteristics, and the control system's response capabilities, enabling sensitive detection of harmful deviation trends. Once a threshold is triggered, the current regulation sequence will be interrupted without delay, automatically restarting the entire calculation process from data acquisition and motion prediction to optimized allocation, and generating and executing updated ballast water regulation commands. This forms a fast and accurate feedback correction closed loop, ensuring that the platform's dynamic stability remains under control.

[0121] By implementing water regulation rate limits and clearly defined deviation threshold management, this invention constructs a safe and efficient ballast water regulation process. It fundamentally overcomes the technical defects caused by excessively rapid regulation and inadequate monitoring, ensuring the safe and stable operation of floating wind turbine semi-submersible foundation platforms under various operating conditions, and significantly improving reliability.

[0122] In another embodiment of the present invention, feedback control is implemented based solely on real-time collected marine environmental parameters and platform motion status data. While such methods can maintain basic stability under normal sea conditions, they reveal significant shortcomings under extreme sea conditions. They lack a dedicated mechanism for identifying and responding to extreme conditions, and their control strategies do not differentiate between sea state levels, relying on the same set of adjustment logic. When encountering extreme sea conditions with significant wave heights exceeding 4m and wind speeds exceeding 20m / s, the platform's motion response intensifies dramatically. Real-time feedback-based control methods, due to response lag and limited adjustment capabilities, often fail to effectively suppress platform motion, posing a risk of roll and pitch angles exceeding safe limits. This could lead to platform structural damage or power generation system shutdown, severely impacting operational safety and reliability.

[0123] The sensor system also includes anemometers, typically mounted high on the platform or atop the wind turbine nacelle to avoid structural obstruction, for real-time measurement of wind speed and direction. Wave height data is acquired by wave radar or accelerometer-type wave sensors, and the effective wave height and wave period are derived by analyzing changes in wave height. Ocean current velocity data is measured using acoustic Doppler current profilers or electromagnetic current meters installed at the bottom or center of the platform. All sensor data is synchronized, filtered, and preprocessed by the platform's data acquisition unit before being uploaded to the central controller, providing real-time and accurate marine environmental input for the motion response prediction model.

[0124] This embodiment uses a sensor system deployed on a semi-submersible foundation platform to collect real-time data on significant wave height and wind speed. When the significant wave height exceeds 4m and the wind speed exceeds 20m / s, it automatically determines that the current sea state is extreme and triggers the corresponding control logic. In this mode, the motion response prediction model uses strict motion constraints, such as the platform's roll angle not exceeding 10° and pitch angle not exceeding 8° within the next 300s, to output a predicted motion state that meets these constraints. This prediction result provides a clear control basis for ballast water regulation under extreme sea conditions.

[0125] After obtaining the predicted motion state that meets the constraints, the ballast water adjustment calculation stage begins. This stage calculates the total amount of ballast water to be adjusted and its distribution ratio in each ballast tank based on the predicted motion state, and controls the water pump system to perform water injection or drainage operations according to this ratio. The core objective of the entire adjustment process is to suppress platform motion and ensure that the roll and pitch angles are maintained within a safe range. By introducing an extreme sea state identification mechanism and constrained motion prediction, this invention achieves a shift from passive response to proactive prevention, significantly improving the survivability and operational stability of floating wind turbine semi-submersible foundation platforms in extreme marine environments.

[0126] In another embodiment of the present invention, the model update step is used to address the problem of decreased prediction accuracy of the motion response prediction model due to environmental changes or model drift during long-term use. Ballast water regulation systems employ prediction models with fixed parameters or models trained based on initial historical data, lacking online learning and adaptive update mechanisms. Such static models struggle to adapt to long-term evolution of the marine environment, subtle changes in platform structural characteristics, or unfamiliar sea states, causing the predicted output to gradually deviate from the actual motion response, thereby affecting the accuracy of ballast water regulation and platform stability.

[0127] This implementation effectively overcomes the aforementioned shortcomings through a model update mechanism. During ballast water regulation, various data collected by sensors are recorded in real time, including the roll angle, pitch angle, heave displacement, wave height, wave period, and ocean current velocity of the semi-submersible base platform, as well as the corresponding actual motion state data. This data continuously accumulates and forms a new dataset, serving as the data foundation for model updates. Two trigger conditions are set to initiate the model update process: first, the amount of data in the new dataset reaches a preset threshold, such as 10,000 records; second, the confidence score output by the motion response prediction model is continuously lower than a third confidence threshold, for example, the confidence score of 20 consecutive predictions is lower than 0.6. Once either condition is met, the system automatically starts the incremental training process.

[0128] During incremental training, the parameters of the existing motion response prediction model are updated using the new dataset, rather than retraining the entire model. This process employs the same network architecture and optimizer as the initial training, but only fine-tunes the model weights based on the new data to preserve existing knowledge while incorporating the latest dynamic features. After training, the updated model replaces the original model in actual control applications, thereby achieving continuous evolution and adaptive improvement of the model's capabilities. This mechanism ensures that the motion response prediction model can maintain high accuracy and strong generalization ability over the long term, significantly improving the reliability and stability of the entire ballast water regulation system in complex and variable marine environments.

[0129] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. Other modifications can be easily made by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and examples shown and described herein.

Claims

1. A method for intelligent dynamic ballast water regulation of a floating semi-submersible wind turbine foundation, characterized in that, include: S1. The semi-submersible base platform is monitored in real time by a sensor system, which collects the roll angle, pitch angle, heave displacement, wave height, wave period, and ocean current velocity. The semi-submersible base platform includes at least three columns, a ballast tank inside the columns, a water pump system connected to the ballast tank, a heave plate arranged at the bottom of the columns, and a sensor system for monitoring the motion status of the semi-submersible base platform and marine environmental parameters. The sensor system includes an attitude sensor, an accelerometer, a water level sensor, a wave height sensor, and a current meter. S2. Combine the real-time data collected in S1 with the ballast water distribution data extracted from the historical database for the same time period to form a comprehensive input vector. The integrated input vector is fed into the motion response prediction model trained on the historical dataset. The temporal characteristics of the integrated input vector are analyzed, and the predicted motion state of the semi-submersible base platform in the future is output. The historical dataset includes historical sea state data and its corresponding historical motion response data. The motion response prediction model adopts an architecture that combines a long short-term memory network with an attention mechanism. The predicted motion state includes the predicted roll angle, the predicted pitch angle, and the predicted heave displacement. S3. Calculate the total amount of ballast water to be adjusted and its distribution ratio in each ballast tank based on the predicted motion state. S4. Control the water pump system to inject or drain water into the designated ballast tank according to the allocation ratio, adjust the relative position of the center of gravity and the center of buoyancy of the semi-submersible foundation platform, continuously monitor the actual motion state of the semi-submersible foundation platform during the injection or drainage process, and compare it with the predicted value. When the deviation between the actual motion state and the predicted value exceeds the set threshold, the data acquisition step of S1 and the prediction step of S2 are re-executed to update the ballast water adjustment command. Step S4 specifically includes: Based on the optimal allocation ratio obtained in step S3, specific execution instructions are generated. The execution instructions include the target ballast tank identifier, the direction of water injection or drainage operation, and the water adjustment amount. The execution instructions are grouped and sorted according to the time window, and simultaneously partitioned according to the spatial location of the ballast tanks to form a time-division and partitioned execution sequence. Within the same time window, synchronous water transfer operations are only allowed to be performed on ballast tank groups whose spatial distance exceeds the minimum interference distance. According to the time-sharing and zone-based execution sequence, the water pump system is controlled to perform water injection or drainage operations to the designated target ballast tank; during the water diversion operation, the actual motion state of the semi-submersible foundation platform, including roll angle, pitch angle, and heave displacement, is continuously monitored at a preset frequency. The actual motion state monitored is compared in real time with the predicted motion state output at the same time in step S2. When the deviation between the actual roll angle, actual pitch angle or actual heave displacement and the corresponding predicted value exceeds the set threshold, the currently executing water adjustment operation sequence is paused, and then steps S1 to S3 are re-executed to generate an updated ballast water adjustment command.

2. The method for intelligent dynamic ballast water regulation of floating semi-submersible wind turbine foundations as described in claim 1, characterized in that, Step S3 specifically includes: Based on the predicted roll angle, predicted pitch angle and predicted heave displacement output in step S2, the ballast water adjustment amount required to suppress roll motion, pitch motion and heave motion respectively is calculated. Based on the spatial distribution of each ballast tank in the semi-submersible base platform, the inhibition ratio of the unit water transfer volume of each ballast tank on the platform's roll, pitch and heave motions was calculated. The optimization objective is to minimize the total water transfer volume of all ballast tanks and achieve a balanced state in the suppression of roll, pitch and heave movements in all directions. A multi-objective optimization function with the water transfer volume of each ballast tank as the variable is established. Solve the multi-objective optimization function to obtain the total amount of ballast water to be adjusted and the optimal distribution ratio in each ballast tank.

3. The method for intelligent dynamic ballast water regulation of floating semi-submersible wind turbine foundations as described in claim 1, characterized in that, In step S2, the motion response prediction model outputs a confidence score for its prediction results. When the confidence score is lower than a first confidence threshold, the execution of the adjustment instructions based on the motion response prediction model is paused, and the system switches to real-time feedback stabilization mode. In real-time feedback stabilization mode, the following steps are performed: With the goal of maintaining the platform's roll and pitch angles at zero degrees, the current roll and pitch angles collected in real time by the sensor system are obtained, and the real-time deviation e(t) between them and the zero-degree set value is calculated. According to the formula Calculate the restoring torque M required for the roll and pitch channels respectively, where K p K is the proportionality coefficient. i K is the integral coefficient. d These are the differential coefficients; Based on the pre-established ballast tank spatial position-moment influence matrix, the required recovery moment M is allocated and calculated as a specific water adjustment command for each ballast tank; The system controls the water pump system to perform water injection or drainage operations according to the water volume adjustment command until the confidence score recovers to above the second confidence threshold. Then, it exits the real-time feedback stabilization mode and re-executes the adjustment command based on the motion response prediction model, where the second confidence threshold is greater than the first confidence threshold.

4. The method for intelligent dynamic ballast water regulation of floating semi-submersible wind turbine foundations as described in claim 2, characterized in that, In step S3, the weight coefficients of the multi-objective optimization function are dynamically adaptive parameters. Based on the real-time effective wave height and ocean current speed collected in step S1, the preset sea state level-weight mapping table is queried to determine the current sea state level. The sea state level is divided into three levels: calm, normal, and severe, according to the range of effective wave height and ocean current speed. Based on the determined sea state level, the weight coefficients between the two optimization objectives of minimizing total water displacement and minimizing residual energy of motion attitude in the optimization function are dynamically adjusted. The higher the sea state level, the larger the weight coefficient corresponding to the objective of minimizing residual energy of motion attitude. The residual energy of motion attitude is calculated by the sum of the square root mean square values ​​of the predicted roll angle, predicted pitch angle and predicted heave displacement.

5. The method for intelligent dynamic ballast water regulation of a floating semi-submersible wind turbine foundation as described in claim 1, characterized in that, Before generating the time-sharing and partitioned execution sequence, the execution instructions are prioritized, with the priority calculated based on an urgency index E using a formula. ,in, This indicates the ratio of the amount of ballast water that needs to be adjusted to the total capacity of the ballast tanks. This represents the absolute value of the deviation between the current motion angle and the target state; α and β are preset weighting coefficients, and α+β=1, setting the emergency threshold E. th When the urgency index of the instruction E≥E th When the time is right, the command will be assigned to the first time window to be executed, and the spatial distance between it and other modules will be less than the minimum interference distance. The real-time roll angle, pitch angle or heave displacement of the semi-submersible base platform after adjustment will not exceed the set threshold.

6. The method for intelligent dynamic ballast water regulation of a floating semi-submersible wind turbine foundation as described in claim 1, characterized in that, During ballast water regulation, based on the predicted motion state output in step S2, the tension setpoints of each mooring line in the mooring system are adjusted synchronously. The adjustment strategy for the mooring system tension setpoints is as follows: With the goal of suppressing the platform's sway, pitch, and bow motion, the required restoring force and restoring torque are calculated based on the predicted changes in the platform's horizontal displacement and heading angle during the motion state. The required restoring force and restoring torque are distributed to each mooring line according to the predetermined contribution of each mooring line, thus obtaining the target tension value for each mooring line; The tension of each mooring line is adjusted to its target tension value by the tension servo mechanism of the mooring system.

7. The method for intelligent dynamic ballast water regulation of floating semi-submersible wind turbine foundations as described in claim 1, characterized in that, The sensor system also includes redundant water level sensors, which are installed in each ballast tank. If any water level sensor fails, the redundant sensor is activated to continue monitoring the water level in the ballast tank. The pump system includes a main pump and a standby pump. When the main pump fails, it automatically switches to the standby pump to continue the ballast water regulation operation.

8. The method for intelligent dynamic ballast water regulation of a floating semi-submersible wind turbine foundation as described in claim 1, characterized in that, The rate of water injection or drainage is controlled at 5%-10% of the total ballast water volume to be adjusted in a single adjustment operation per minute, with the threshold set at a roll angle deviation exceeding 15% or a heave displacement deviation exceeding 20%.

9. The method for intelligent dynamic ballast water regulation of a floating semi-submersible wind turbine foundation as described in claim 1, characterized in that, When the sensor system detects that the effective wave height exceeds 4m and the wind speed exceeds 20m / s, it is determined to be an extreme sea state. When it is an extreme sea state, the motion response prediction model outputs the predicted motion state that meets the constraints of the roll angle not exceeding 10° and the pitch angle not exceeding 8° within the next 300s. Step S3 calculates the required total amount of ballast water and its allocation ratio based on the predicted motion state.

Images