Control method of offshore wind power system

Abstract

The invention discloses a control method of an offshore wind power system, and relates to the technical field of offshore wind power generation control, and the offshore wind power system comprises a wind generating set, a converter, a vibration and stress sensor and a controller. According to the invention, the technical problem of how to effectively suppress the key structure load while ensuring the power output is solved. According to the technical scheme, the method is characterized in that a front wind speed sequence is previewed through a cabin type laser radar, and loads of a tower drum and blades are collected in real time; establishing a parameterized state space model, and performing online identification and state estimation on aerodynamic parameters and unmodeled dynamics by adopting an unscented Kalman filter algorithm; in each control period, taking power tracking and load suppression as multiple targets, and based on the updated model and the preview wind speed, solving a finite time domain optimal control problem; and finally, the optimal pitch angle and generator torque increment instruction is analyzed and executed. The method is mainly used for improving the service life and stability of the offshore wind turbine generator.

Description

Technical Field

[0001] This invention relates to the field of wind power generation control technology. More specifically, this invention relates to a control method for an offshore wind power system. Background Technology

[0002] Offshore wind power, as an important form of renewable energy utilization, operates in a complex and harsh environment, with turbines subjected to the combined effects of irregular aerodynamic and wave loads over extended periods. This makes critical structural components, such as the tower and blades, susceptible to fatigue damage, thus affecting the overall system's lifespan and reliability. Therefore, effectively suppressing structural loads while pursuing high power generation efficiency remains a persistent technical challenge in the field of offshore wind power system control.

[0003] Traditional wind turbine control strategies primarily rely on feedback control, which uses sensors to measure real-time states such as rotor speed and output power, and then adjusts the pitch angle and generator torque to achieve maximum power point tracking or stable power output. However, this control method inherently suffers from lag. When wind speed changes drastically, intense turbulence and wind shear have already acted on the rotor, generating irreversible load impacts before the control system begins to respond, thus limiting its effectiveness. This lag also exposes the turbine structure to severe wind conditions such as gusts, which is detrimental to long-term stable operation.

[0004] To improve control performance, the concept of feedforward control was introduced, the core of which lies in making control decisions in advance using wind speed preview information. Early studies attempted to use anemometers to measure the wind speed at the nacelle as a feedforward signal, but this measurement point is located behind the wind turbine, and the information obtained is the wind speed after the wind turbine disturbance, not the actual incoming wind acting on the entire wind turbine plane. Therefore, the accuracy and globality of its preview information are insufficient, making it difficult to accurately predict the aerodynamic loads that will occur.

[0005] Furthermore, establishing accurate turbine dynamics models is fundamental for designing advanced control algorithms such as high-performance model predictive control. However, offshore wind turbines are complex multi-body coupled systems, with models encompassing multiple subsystems including aeroelasticity, transmission chain, generator, and control system. Aerodynamic parameters, such as aerodynamic torque and thrust coefficient, vary significantly with pitch angle, tip speed ratio, blade surface roughness, and icing conditions, exhibiting strong time-varying characteristics. Using fixed aerodynamic parameters in the control model leads to a mismatch between the model and the actual system, especially under complex turbulent wind conditions. This mismatch severely impacts the performance of model-based control algorithms and may even cause control instability. While offline system identification can obtain model parameters, it cannot adapt to the time-varying characteristics of parameters during operation. Therefore, online real-time identification of key aerodynamic parameters and unmodeled dynamics presents a significant challenge.

[0006] Finally, the controller design itself is a multi-objective optimization problem, requiring a balance between tracking power generation commands and suppressing structural loads. These two objectives are often contradictory; for example, reducing load through rapid pitch control may cause power fluctuations. Simultaneously, control actions must strictly adhere to the physical constraints of the actuators, such as limits on pitch angle change rate and torque change rate. Solving such a complex optimal control problem in real time, under conditions of wind speed feedforward, model uncertainty, and multiple constraints, places extremely high demands on the controller's computational capabilities. In the past, due to limitations in sensing technology, online identification algorithms, and real-time computing hardware, control methods that could simultaneously and effectively utilize preview wind information, adaptively update the model, and efficiently solve multi-objective constrained optimization problems were difficult to implement in practical engineering. Summary of the Invention

[0007] 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.

[0008] Another objective of this invention is to provide a control method for offshore wind power systems, which addresses the technical shortcomings of traditional offshore wind turbine control methods that rely primarily on feedback control, resulting in response lag and an inability to effectively cope with severe wind conditions such as sudden gusts, leading to severe fatigue damage to key structural components.

[0009] To achieve these objectives and other advantages of the present invention, a control method for an offshore wind power system is provided. The offshore wind power system includes a wind turbine generator, a converter, vibration and stress sensors deployed at key structural points of the wind turbine generator, and a controller integrated within the wind turbine generator. The control method for the offshore wind power system includes: Step 1: Use lidar to scan the wind field in front of the wind turbine to obtain a preview wind speed sequence; at the same time, use vibration and stress sensors to collect real-time load measurements of the bending moments at the front and rear of the tower and the bending moments at the blade roots. Step 2: Establish a parameterized state-space model that includes aeroelastic dynamics and generator drivetrain dynamics. Use a joint state and parameter estimation algorithm based on an unscented Kalman filter, taking the preview wind speed sequence and load measurements as observation inputs, to identify and update the time-varying aerodynamic parameters and unmodeled dynamics in the parameterized state-space model in real time, and simultaneously estimate the state variables of the entire system. Step 3: In each control cycle, with the optimization objectives of minimizing power tracking error and minimizing structural load, based on the updated parameterized state-space model, using the preview wind speed sequence as the known feedforward disturbance input, and the total system state variables estimated in Step 2 as the initial state, under the actuator constraints of the pitch angle change rate, the wind turbine speed change rate, and the generator torque change rate, solve the optimal control problem in the finite time domain and output the optimal control sequence in the future control time domain. Step 4: Parse the first element of the optimal control sequence output in Step 3 into the optimal pitch angle increment command and the optimal generator torque increment command for the current control cycle. Superimpose these commands with the corresponding commands from the previous cycle to generate and execute the actual pitch angle control command and generator torque control command for the current cycle.

[0010] Preferably, in the control method for the offshore wind power system, step one, the process of acquiring a preview wind speed sequence using lidar scanning data, includes: Step 11: Based on the original scanning data of the lidar, reconstruct the four-dimensional spatiotemporal wind field within a predefined range in front of the wind turbine. The four-dimensional spatiotemporal wind field includes three-dimensional spatial coordinates and one-dimensional time coordinates. Step 12: In each control cycle, based on the system state variables estimated in Step 2 of the previous cycle and the optimal control sequence obtained in Step 3 of the previous cycle, predict the state trajectory of the wind turbine generator in the future preview time domain. The state trajectory includes at least the yaw angle trajectory and the rotor rotation angle trajectory. Step 13: Establish a wind field-wind rotor coupling model. The four-dimensional spatiotemporal wind field obtained in Step 11 is transformed according to the state trajectory predicted in Step 12 and dynamically mapped to a rotating coordinate system that moves with the wind rotor in the future. This generates a dynamic preview inflow wind field sequence that acts on each blade spanwise unit, with the wind rotor rotation angle as the index.

[0011] Preferably, in the control method for the offshore wind power system, step one, the process of acquiring a preview wind speed sequence using laser radar scanning data, further includes, based on the error between the real-time measured rotor rotation angle and the rotor rotation angle trajectory predicted in step 12, performing online correction of the coordinate transformation parameters in the wind field-rotor coupling model, specifically: Step a: The actual rotation angle of the wind turbine is measured in real time using an encoder installed on the main shaft of the wind turbine generator. Step b: Compare the actual rotation angle with the wind turbine rotation angle trajectory value predicted in step 12 at the current moment to generate a wind turbine rotation angle error signal; Step c: Input the wind turbine rotation angle error signal into a preset proportional-integral corrector. The proportional-integral corrector outputs an azimuth offset correction amount to compensate for the dynamic mapping process in the wind field-wind turbine coupling model. Step d: Using the azimuth offset correction, adjust the rotation matrix parameters in the coordinate transformation in real time to generate a corrected dynamic preview of the inflow wind field sequence acting on each blade spanwise unit.

[0012] Preferably, in the control method for the offshore wind power system, step three, solving the optimal control problem in the finite time domain, specifically includes: Step 31: Based on the updated parameterized state-space model, the optimization objectives of minimizing the power tracking error and minimizing the structural load are transformed into objective functions, and considering the actuator constraints, a continuous-time optimal control problem is constructed. Step 32: Using the orthogonal collocation method, the state variables and control variables in the future preview time domain are discretized on a series of collocation points, transforming the continuous-time optimal control problem into a nonlinear programming problem; Step 33: Solve the nonlinear programming problem using a real-time iterative algorithm. In the solution process of each control cycle, the optimal control sequence and state trajectory obtained in the previous control cycle are used as the initial guess values ​​for the current solution to perform iterative calculations in order to accelerate convergence. Step 34: After meeting the preset solution accuracy or maximum number of iterations, output the optimal control sequence for the current control cycle.

[0013] Preferably, in the control method for the offshore wind power system, the joint state and parameter estimation algorithm in step two introduces an adaptive mechanism to adjust the process noise covariance matrix and measurement noise covariance matrix of the unscented Kalman filter in real time based on the innovation sequence or residual sequence.

[0014] Preferably, the control method for the offshore wind power system further includes, before step four: filtering the optimal pitch angle increment command and the optimal generator torque increment command output in step three through a first-order inertial element, wherein the time constant of the first-order inertial element is obtained by identifying the dynamic response characteristics of the pitch actuator and the converter.

[0015] Preferably, in the control method of the offshore wind power system, in step two, the joint state and parameter estimation algorithm based on the unscented Kalman filter adjusts the process noise covariance matrix by adaptively scaling the turbulence intensity value of the preview wind speed sequence.

[0016] Preferably, in the control method for the offshore wind power system, the time domain length for solving the finite-time optimal control problem in step three is dynamically adjusted based on the dominant frequencies of wind shear and tower shadow effects in the preview wind speed sequence.

[0017] Preferably, in the control method for the offshore wind power system, the time-varying aerodynamic parameters identified in real time in step two include at least the equivalent time constant of the dynamic stall model and the inducible factor relaxation time constant of the dynamic inflow model.

[0018] Preferably, in the control method of the offshore wind power system, in step four, before generating the actual control command for the current cycle, the optimal pitch angle increment command is independently checked and balanced based on the measured pitch angle of the three blades.

[0019] The present invention has at least the following beneficial effects: 1. This invention uses lidar to preview the wind field, enabling forward perception of disturbances and providing an information basis for feedforward control. Combined with joint state and parameter estimation based on unscented Kalman filtering, it can track changes in the dynamic characteristics of the system online, maintain model accuracy, and finally integrate feedforward information, updated model, and multi-objective optimization through finite-time optimal control. This can significantly reduce the fatigue load of key structural components while maintaining good power point tracking performance, thereby effectively extending the unit's lifespan and improving operational stability and economic benefits, all while satisfying actuator constraints.

[0020] 2. This invention reconstructs a four-dimensional wind field and predicts the turbine's state trajectory, enabling precise mapping of wind field information under fixed coordinates onto each blade unit of the future rotating wind turbine. This makes the previewed wind speed sequence no longer a sequence of a single point in space, but an inflow wind field that precisely corresponds to the turbine's rotation position and can be directly used for aerodynamic load calculation. This greatly improves the accuracy and usability of feedforward information and lays a solid foundation for high-performance independent pitch control and other precise control methods.

[0021] 3. By introducing a closed-loop feedback correction mechanism based on real-time wind turbine rotation angle measurement, this invention can effectively compensate for coordinate transformation errors caused by model prediction deviations and external disturbances, ensuring that the preview wind field sequence and the actual wind turbine orientation are always kept in high-precision synchronization. This significantly improves the reliability and robustness of feedforward control, avoids the negative control effects that may be caused by mapping misalignment, and ensures the stable performance of control.

[0022] 4. This invention uses the orthogonal collocation method to transform the continuous-time optimal control problem into a nonlinear programming problem. Combined with a real-time iterative algorithm and using the solution of the previous cycle as an initial guess, it can significantly accelerate the convergence speed of the solution process. It effectively solves the contradiction between the large amount of computation and the short control cycle of complex optimization problems, ensuring that the optimal control problem can obtain a high-precision solution within the time allowed by engineering applications, and realizing the high-frequency real-time operation of the control algorithm.

[0023] 5. This invention adjusts the noise covariance matrix of the unscented Kalman filter in real time through an adaptive mechanism, enabling the filter to dynamically adjust its confidence in model prediction and measurement information according to the actual operating conditions of the system. When the system undergoes drastic dynamic changes or measurement anomalies occur, it can automatically adjust the filter gain, significantly enhancing the robustness and estimation accuracy of the state and parameter estimation algorithm in the face of uncertainty and interference.

[0024] 6. This invention filters the theoretically optimal control increment command based on the dynamic characteristics of the actuator, making the final command sent to the actuator change smoothly and physically feasible. This avoids placing excessive demands on the actuator, reduces lag and overshoot during execution, and ensures that the control command is executed accurately and stably, thereby improving the performance and stability of the overall control loop.

[0025] 7. This invention correlates the process noise covariance with the turbulence intensity of the preview wind field, achieving adaptive matching between filter parameters and the intensity of external wind disturbances. Under high turbulence conditions, the process noise covariance is automatically increased, indicating that the model uncertainty increases and the filter trusts the measured values ​​more, thus tracking system changes faster. Under stable wind conditions, the opposite is true, further optimizing the filter's performance under different wind conditions and improving the accuracy of estimation.

[0026] 8. By dynamically adjusting the prediction time domain length with the dominant frequency of wind shear and tower shadow effects, this invention ensures that the controller's prediction range can cover one or more complete cycles of these periodic disturbances. This allows the optimized controller to "see" and plan ahead to deal with disturbances throughout the entire cycle, thereby more effectively suppressing vibrations and loads caused by these periodic sources and improving the targeting and effectiveness of the control.

[0027] 9. This invention identifies the time constants of key aerodynamic processes such as dynamic stall and dynamic inflow online, instead of using fixed values. This allows the model to more realistically reflect the actual response speed of aerodynamics under different operating conditions, significantly improving the calculation accuracy of aerodynamic loads, thereby enhancing the accuracy of state estimation and the control performance based on model prediction. The effect is particularly obvious when dealing with nonlinear complex phenomena such as dynamic stall.

[0028] 10. This invention compensates for actuator differences and asymmetry caused by wind shear by independently verifying and allocating the uniformly calculated optimal pitch angle increment based on the actual position of each blade. This ensures that the commands ultimately received by the three blades can achieve better load balance, further reducing the asymmetric fatigue load on key components such as the hub, and optimizing the overall performance of independent pitch control.

[0029] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Detailed Implementation

[0030] The present invention will be further described in detail below with reference to embodiments, so that those skilled in the art can implement it based on the description.

[0031] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0032] 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.

[0033] In one embodiment of the present invention, a control method for an offshore wind power system is provided. The offshore wind power system includes a wind turbine generator, a converter, vibration and stress sensors deployed at key structural points of the wind turbine generator, and a controller integrated within the wind turbine generator. The control method for the offshore wind power system includes: Step 1: Use lidar to scan the wind field in front of the wind turbine to obtain a preview wind speed sequence; at the same time, use vibration and stress sensors to collect real-time load measurements of the bending moments at the front and rear of the tower and the bending moments at the blade roots. Step 2: Establish a parameterized state-space model that includes aeroelastic dynamics and generator drivetrain dynamics. Use a joint state and parameter estimation algorithm based on an unscented Kalman filter, taking the preview wind speed sequence and load measurements as observation inputs, to identify and update the time-varying aerodynamic parameters and unmodeled dynamics in the parameterized state-space model in real time, and simultaneously estimate the state variables of the entire system. Step 3: In each control cycle, with the optimization objectives of minimizing power tracking error and minimizing structural load, based on the updated parameterized state-space model, using the preview wind speed sequence as the known feedforward disturbance input, and the system state variables estimated in Step 2 as the initial state, under the actuator constraints of the pitch angle change rate, the wind turbine speed change rate, and the generator torque change rate, solve the optimal control problem in the finite time domain and output the optimal control sequence in the future control time domain. Step 4: Parse the first element of the optimal control sequence output in Step 3 into the optimal pitch angle increment command and the optimal generator torque increment command for the current control cycle. Superimpose these commands with the corresponding commands from the previous cycle to generate and execute the actual pitch angle control command and generator torque control command for the current cycle.

[0034] Traditional offshore wind turbine control methods rely on feedback mechanisms, resulting in delayed responses and difficulty in handling severe wind conditions such as gusts. This leads to irreversible fatigue loads on critical structures such as towers and blades over long periods, while also causing insufficient power output stability, affecting the turbine's operational lifespan and reliability. Furthermore, traditional methods lack accurate forward-looking wind speed perception and real-time system model updates, making it impossible to achieve coordinated optimization control of power and load under variable offshore wind conditions.

[0035] In this embodiment, the specific process is as follows: Step 1: Sensing and Acquiring Feedforward Wind Speed ​​and Load Information A nacelle-mounted lidar system installed at the front of the wind turbine nacelle continuously scans and measures the wind field within a certain range in front of the rotor. Based on the raw lidar scan data, a wind field reconstruction algorithm generates a four-dimensional spatiotemporal preview wind speed sequence for the next tens of seconds. Simultaneously, vibration and stress sensors pre-deployed at key structural points on the tower and blade roots are used to synchronously collect bending moment signals in the forward and backward directions of the tower and the root bending moment signals of each blade in real time, serving as measurements reflecting the structural load.

[0036] Step 2: Online Adaptive Update and Full-State Estimation of the System Model A parameterized state-space model integrating the aeroelastic dynamics of the wind turbine and the dynamics of the generator drivetrain is established as the basic model. A joint state and parameter estimation algorithm based on an unscented Kalman filter is employed, using the preview wind speed sequence and load measurements obtained in step one as observation inputs to continuously identify and update key time-varying aerodynamic parameters and unmodeled dynamics in the model online in real time. Through this algorithm, while updating the model parameters, the entire system state variables, including rotor speed, drivetrain torque, tower bending moments, and blade bending moments, are simultaneously estimated, providing the controller with complete current system state information.

[0037] Step 3: Solving multi-objective constraint optimization based on feedforward and update models Within each control cycle, the dual optimization objectives are minimizing power generation tracking error and minimizing structural load. Based on the latest parameterized state-space model updated in step two, the preview wind speed sequence obtained in step one is used as the known feedforward disturbance input, and the total system state variables estimated in step two are used as the initial state of the optimization problem. Under the strict constraint of actuator physical constraints such as the rate of change of pitch angle, the rate of change of wind turbine speed, and the rate of change of generator torque, a finite-time domain optimal control problem is constructed and solved. Through numerical optimization algorithms, the optimal pitch angle and generator torque control sequence for a future control time period is calculated.

[0038] Step 4: Parsing and Execution of Optimal Control Commands The first element of the optimal control sequence obtained in step three, i.e., the control quantity to be executed immediately, is parsed into the optimal pitch angle increment command and the optimal generator torque increment command for the current control cycle. These two increment commands are then superimposed with the actual pitch angle command and generator torque command of the previous control cycle, respectively, to generate the final actual pitch angle control command and generator torque control command for the current cycle. Subsequently, these two commands are sent to the pitch actuator and the converter for execution, respectively, thereby simultaneously achieving precise control of generator power and effective suppression of critical structural loads.

[0039] This implementation first uses a lidar mounted at the front of the nacelle to scan the wind field in front of the rotor, acquiring a preview wind speed sequence. Simultaneously, vibration and stress sensors are used to collect real-time load data on the tower and blades. Then, a parameterized state-space model incorporating aeroelasticity and transmission chain dynamics is established. Using an unscented Kalman filter algorithm, combined with the preview wind speed and measured loads, the aerodynamic parameters and unmodeled dynamics in the model are identified and their states estimated online. Within each control cycle, with the objectives of minimizing power tracking error and structural load, based on the updated model and preview wind speed, and under the condition of satisfying actuator physical constraints, a finite-time optimal control problem is solved to obtain the optimal pitch angle and generator torque increment commands. Finally, the increment commands are superimposed with the commands from the previous cycle to generate the actual control commands, which are then issued and executed.

[0040] Compared to the closest existing technologies, traditional methods typically rely solely on feedback control or simple feedforward, lacking accurate perception of the wind field ahead and the ability to adaptively update the model. Therefore, they are slow to respond and have limited control effectiveness under gusty or turbulent wind conditions. This implementation uses lidar to preview the wind field, combined with online model identification and multi-objective optimization control, significantly improving the system's foresight and adaptability, and enabling it to simultaneously ensure power output and structural safety under complex wind conditions.

[0041] The technical solution provided in this embodiment improves the control system's ability to predict wind disturbances ahead, enhances the accuracy of the model under different operating conditions, achieves power and load balance optimization, reduces fatigue damage to key structures, extends the service life of the unit, improves operational stability and overall economic benefits, and has strong engineering applicability and robustness, making it suitable for intelligent control of various types of offshore wind turbines.

[0042] In another embodiment of the present invention, step one, the process of obtaining a preview wind speed sequence using laser radar scanning data, includes: Step 11: Based on the original scanning data of the lidar, reconstruct the four-dimensional spatiotemporal wind field within a predefined range in front of the wind turbine. The four-dimensional spatiotemporal wind field includes three-dimensional spatial coordinates and one-dimensional time coordinates. Step 12: In each control cycle, based on the system state variables estimated in Step 2 of the previous cycle and the optimal control sequence obtained in Step 3 of the previous cycle, predict the state trajectory of the wind turbine generator in the future preview time domain. The state trajectory includes at least the yaw angle trajectory and the rotor rotation angle trajectory. Step 13: Establish a wind field-wind rotor coupling model. The four-dimensional spatiotemporal wind field obtained in Step 11 is transformed according to the state trajectory predicted in Step 12 and dynamically mapped to a rotating coordinate system that moves with the wind rotor in the future. This generates a dynamic preview inflow wind field sequence that acts on each blade spanwise unit, with the wind rotor rotation angle as the index.

[0043] In traditional methods, nacelle-mounted lidar measures wind speed sequences fixed at a single point or plane in space. However, wind turbines rotate continuously, and the actual inflow wind speed experienced by each blade varies at different azimuth angles. Directly using fixed-point wind speed sequences for control cannot accurately reflect the specific impact of three-dimensional spatially distributed wind disturbances such as wind shear and tower shadow effects on each blade. This results in a significant spatial mismatch between feedforward information and the actual aerodynamic load on the wind turbine, limiting the performance improvement of precise control strategies such as independent pitch control.

[0044] To address the aforementioned technical challenges, this implementation first processes the raw scanning data from the lidar to reconstruct a four-dimensional spatiotemporal wind field within a predefined three-dimensional spatial range in front of the wind turbine over a future period. In each control cycle, based on the system state estimated in the previous cycle and the optimal control sequence solved in the previous cycle, the wind turbine's trajectory in the future preview time domain is predicted, particularly the yaw angle and the wind turbine rotation angle. Subsequently, a wind field-wind turbine coupling model is established. The reconstructed four-dimensional spatiotemporal wind field in a fixed coordinate system is dynamically mapped to a coordinate system that rotates along with the future wind turbine by performing coordinate transformation based on the predicted future wind turbine rotation state. Through this process, a dynamic preview inflow wind field sequence indexed by the wind turbine rotation angle is generated. This sequence accurately describes the actual wind speed acting on each blade spanwise unit when the wind turbine rotates to a specific angle over a future period.

[0045] Compared to the closest existing technology, traditional methods typically use the wind speed measured by lidar at a point ahead or within a linear range directly as the feedforward signal, ignoring the spatial positional changes caused by the rotation of the wind turbine and the non-uniformity of the three-dimensional spatial distribution of the wind field. This implementation solves the core problem of spatial mapping inaccuracies by reconstructing the four-dimensional wind field and performing dynamic coordinate mapping based on the predicted wind turbine trajectory, transforming fixed wind field information into an inflow wind field sequence that precisely corresponds to the rotating wind turbine blade unit.

[0046] The technical solution provided in this embodiment greatly improves the spatial accuracy and availability of feedforward wind speed information, transforming the preview information from "a single point of wind" to "a single leaf of wind," providing directly available and precise input conditions for high-performance independent pitch control. This lays a solid foundation for more effectively suppressing periodic load fluctuations caused by uneven spatial wind distribution, and improves the accuracy and effectiveness of control.

[0047] In another embodiment of the present invention, step one, the process of obtaining a preview wind speed sequence using laser radar scanning data, further includes, based on the error between the real-time measured wind turbine rotation angle and the wind turbine rotation angle trajectory predicted in step 12, performing online correction of the coordinate transformation parameters in the wind field-wind turbine coupling model, specifically: Step a: The actual rotation angle of the wind turbine is measured in real time using an encoder installed on the main shaft of the wind turbine generator. Step b: Compare the actual rotation angle with the wind turbine rotation angle trajectory value predicted in step 12 at the current moment to generate a wind turbine rotation angle error signal; Step c: Input the wind turbine rotation angle error signal into a preset proportional-integral corrector. The proportional-integral corrector outputs an azimuth offset correction amount to compensate for the dynamic mapping process in the wind field-wind turbine coupling model. Step d: Using the azimuth offset correction, adjust the rotation matrix parameters in the coordinate transformation in real time to generate a corrected dynamic preview of the inflow wind field sequence acting on each blade spanwise unit.

[0048] In the process of mapping the preview wind field in a fixed coordinate system to a rotating coordinate system using the predicted rotor rotation trajectory, errors will occur between the predicted trajectory and the actual rotor rotation state due to factors such as model prediction bias, external disturbances, or changes in system parameters. This error will cause the mapped preview inflow wind field sequence to be out of sync with the actual wind field that the rotor blades will experience in terms of spatial orientation, forming a systematic alignment error. If this misaligned feedforward information is directly used for control, it will not only fail to effectively suppress the load, but may even produce erroneous control actions, exacerbate load fluctuations, and reduce the reliability and practicality of feedforward control.

[0049] This implementation method uses a high-precision encoder to measure the actual rotation angle of the wind turbine's main shaft in real time, obtaining the true azimuth of the wind turbine. In each control cycle, this measured actual rotation angle is compared in real time with the predicted wind turbine rotation angle trajectory value based on information from the previous cycle, generating a wind turbine rotation angle error signal representing the prediction deviation. This error signal is input to a preset proportional-integral corrector, which outputs a corresponding azimuth offset correction. This correction is used to compensate for and adjust the rotation matrix parameters relied upon for coordinate transformation in the wind field-wind turbine coupling model in real time, thereby correcting the dynamic mapping process online. Finally, using the corrected parameters, a set of dynamic preview inflow wind field sequences, synchronized with the actual rotation azimuth of the wind turbine and acting on each blade spanwise unit, is generated.

[0050] Compared to the closest existing technologies, traditional methods typically perform the mapping transformation from fixed coordinates to rotating coordinates in an open-loop manner, relying entirely on the model's prediction accuracy and lacking the ability to verify and correct the accuracy of the mapping results in real time. Once a prediction deviation occurs, the quality of the entire feedforward information degrades without the user's knowledge. This implementation introduces a closed-loop feedback correction loop based on direct physical measurements, creatively combining the feedforward path with a fast feedback correction loop, which can automatically and continuously eliminate mapping errors caused by imperfections in the prediction model and external interference.

[0051] The technical solution provided in this embodiment significantly improves the synchronization accuracy and long-term reliability between feedforward wind speed information and the actual spatial orientation of the wind turbine. It effectively avoids the risk of decreased feedforward control performance or even negative effects due to model prediction inaccuracies, enhances the robustness of the entire control system, and ensures that the feedforward control strategy based on preview information can stably perform its expected performance in the actual operating environment.

[0052] In another embodiment of the present invention, solving the optimal control problem in the finite-time domain in step three specifically includes: Step 31: Based on the updated parameterized state-space model, the optimization objectives of minimizing the power tracking error and minimizing the structural load are transformed into objective functions, and considering the actuator constraints, a continuous-time optimal control problem is constructed. Step 32: Using the orthogonal collocation method, the state variables and control variables in the future preview time domain are discretized on a series of collocation points, transforming the continuous-time optimal control problem into a nonlinear programming problem; Step 33: Solve the nonlinear programming problem using a real-time iterative algorithm. In the solution process of each control cycle, the optimal control sequence and state trajectory obtained in the previous control cycle are used as the initial guess values ​​for the current solution to perform iterative calculations in order to accelerate convergence. Step 34: After meeting the preset solution accuracy or maximum number of iterations, output the optimal control sequence for the current control cycle.

[0053] The control of offshore wind turbines is a complex, multi-objective, multi-constraint continuous-time dynamic optimization problem. Solving such optimal control problems directly in the continuous-time domain results in excessively high computational complexity, making it difficult to obtain solutions within the finite control cycles required in practical engineering. Traditional discretization methods may suffer from insufficient accuracy or numerical instability, while conventional optimization algorithms are slow and cannot meet the high-frequency refresh requirements of real-time control. This hinders the application of many advanced control theories in practical engineering, limiting further improvements in control performance.

[0054] This implementation method is as follows: First, based on the latest updated parameterized state-space model, the multi-objective optimization problem of minimizing power tracking error and structural load is formally constructed as a continuous-time optimal control problem, taking into account the physical constraints of various actuators. Then, using the orthogonal collocation method, a numerical calculation technique, the continuous state variables and control variables in the future preview time domain are discretized at a series of selected collocation points, thereby transforming the original continuous-time optimal control problem into a structured and more easily handled nonlinear programming problem. To efficiently solve this nonlinear programming problem, a real-time iterative algorithm is employed. During the initialization of each control cycle, the algorithm uses the optimal control sequence and state trajectory obtained in the previous control cycle as the initial guess values ​​for the current iteration. This leverages the continuity of the system state and the smoothness of the control commands to significantly reduce the required number of iterations and accelerate the convergence process. Finally, after meeting the preset solution accuracy requirements or reaching the maximum iteration limit, the optimal control sequence for the current control cycle is output.

[0055] Compared to the closest existing technologies, traditional methods may suffer from accuracy loss due to simple discretization or require solving each cycle from scratch using conventional optimization algorithms, resulting in long computation times and difficulty in ensuring real-time control. This implementation effectively resolves the contradiction between computational accuracy, numerical stability, and solution speed in complex optimization problems by combining the orthogonal collocation method, a high-precision numerical transformation method, with a real-time iterative algorithm utilizing historical solution information.

[0056] This implementation achieves high-precision and high-efficiency numerical solutions to complex continuous-time optimal control problems, ensuring that multi-objective constrained optimization algorithms can obtain reliable solutions within the strictly limited time window of engineering applications. This makes it possible for advanced high-frequency model predictive control to operate in real time on actual wind turbine controllers, laying a core computational foundation for improving control performance.

[0057] In another embodiment of the present invention, in the joint state and parameter estimation algorithm in step two, an adaptive mechanism is introduced to adjust the process noise covariance matrix and measurement noise covariance matrix of the unscented Kalman filter in real time based on the innovation sequence or residual sequence.

[0058] When using unscented Kalman filters for joint state and parameter estimation, their performance is highly dependent on the settings of the process noise covariance matrix and the measurement noise covariance matrix. Traditional filters typically use a fixed noise covariance matrix. However, the operating environment of offshore wind turbines is complex, and their system dynamic characteristics and sensor measurement environment change significantly with factors such as turbulence intensity, weather conditions, and mechanical wear. Fixed noise statistical assumptions cannot adapt to these time-varying characteristics, leading to estimation lag or divergence when operating conditions change drastically, or over-reliance on erroneous data when measurements are disturbed. This severely impairs the accuracy and robustness of state and parameter estimation, and consequently affects the control performance based on this model.

[0059] Specifically, this implementation involves continuously monitoring and calculating the innovation sequence or residual sequence of the filter during the operation of the joint state and parameter estimation algorithm based on an unscented Kalman filter. The innovation sequence reflects the difference between the actual observations and the filter's predictions. By introducing an adaptive mechanism, the statistical characteristics of the innovation sequence or residual sequence can be analyzed online, such as the deviation between its actual covariance and theoretical covariance. Based on this analysis, the mechanism dynamically and in real-time adjusts the values ​​of the preset process noise covariance matrix and measurement noise covariance matrix in the unscented Kalman filter. When the innovation or residual sequence indicates an increased deviation between the model prediction and the actual observation, the adaptive mechanism will correspondingly increase the process noise covariance, indicating greater trust in the measured values; conversely, when abnormal noise appears in the measured values, the measurement noise covariance will be increased, indicating greater trust in the model predictions, thereby achieving online optimization of the filter gain.

[0060] Compared to the closest existing technologies, traditional methods typically rely on offline debugging based on engineering experience, setting a fixed, compromised set of noise covariance parameters for the filter. This approach cannot cope with changes in the dynamic characteristics of the system during operation, and the filter performance degrades significantly under off-design conditions. This implementation introduces an adaptive mechanism based on innovation or residuals, transforming the filter from a static estimator with fixed parameters into a dynamic estimator capable of self-adjusting according to the real-time operating conditions of the system.

[0061] The technical solution provided in this embodiment significantly improves the adaptability and robustness of the joint state and parameter estimation algorithm in the face of system dynamic changes, model uncertainties and measurement disturbances. It ensures that it can provide continuous, stable and accurate state estimation and parameter identification results in the full range of operating conditions, providing a reliable and trustworthy model foundation for subsequent high-performance model predictive control, and fundamentally enhancing the adaptability and stability of the entire control system.

[0062] In another embodiment of the present invention, before step four, the method further includes: filtering the optimal pitch angle increment command and the optimal generator torque increment command output in step three through a first-order inertial element, wherein the time constant of the first-order inertial element is obtained by identifying the dynamic response characteristics of the pitch actuator and the converter.

[0063] The theoretically optimal control increment command obtained by optimization algorithms is often calculated instantaneously based on an ideal system model, and its changes may be drastic or discontinuous. However, actual wind turbine actuators, such as pitch actuators and converters, have inherent mechanical inertia, response delays, and action rate limitations. Directly sending theoretical commands to the actuators would place excessive demands on them, leading to imperfect command tracking and execution. This could cause actuator lag, overshoot, or even mechanical wear, resulting in actual control performance deviating from the optimization algorithm's expected target and affecting the overall control loop's performance and stability.

[0064] In this embodiment, after obtaining the optimal pitch angle increment command and the optimal generator torque increment command in step three, they are not directly sent to the actuator. Instead, these two increment commands are smoothed through an independent first-order inertial filter. The core parameter of this first-order inertial filter, namely its time constant, is not arbitrarily set based on experience, but is obtained through specialized system identification experiments on the dynamic response characteristics of the actual pitch actuator system and converter system. This identification process aims to accurately reflect the dynamic lag characteristics of these actuators from receiving commands to actual actions. After filtering, the theoretically rapidly changing increment commands are converted into smoothly changing command signals that are physically easier for the actuator to track and implement.

[0065] Compared to the closest existing technology, traditional methods may ignore the dynamic characteristics of the actuator, directly issue the controller's output commands, or use fixed, conservative filtering parameters, sacrificing response speed for stability. This implementation, however, precisely sets the filtering parameters by identifying the dynamic characteristics of the specific physical actuator, achieving a customized filtering solution that matches the hardware characteristics.

[0066] This implementation effectively avoids the problem of decreased control performance caused by the mismatch between theoretical instructions and the dynamic capabilities of the actuator. It makes the issued control instructions smoother and more feasible, reduces the action burden of the actuator, reduces mechanical stress, and ensures that the control intention obtained by optimized calculation can be accurately and stably converted into actual actions, thereby improving the performance reliability and system stability of the entire control loop.

[0067] In another embodiment of the present invention, in step two, the joint state and parameter estimation algorithm based on the unscented Kalman filter adjusts the process noise covariance matrix by adaptively scaling the turbulence intensity value of the preview wind speed sequence.

[0068] When using unscented Kalman filters for state and parameter estimation, the value of the process noise covariance matrix is ​​crucial, as it reflects a measure of model uncertainty. Traditional methods typically set this matrix to a fixed value. However, the turbulence intensity in offshore wind fields varies dramatically. High turbulence means drastic and unpredictable wind speed changes, significantly increasing model uncertainty; low turbulence, on the other hand, implies stable wind conditions and a more reliable model. A fixed process noise covariance cannot adapt to the time-varying characteristics of this external wind disturbance. In high turbulence, it can cause the filter to over-rely on inaccurate model predictions, resulting in estimation lag; in low turbulence, it may over-rely on measurement noise, affecting estimation smoothness and accuracy, ultimately limiting the overall performance of the state and parameter estimator under different wind conditions.

[0069] In this implementation, during filter operation, the preview wind speed sequence acquired by lidar is analyzed in real time to calculate a key indicator characterizing wind field fluctuations—turbulence intensity. This turbulence intensity value is used to directly guide the online adjustment of the process noise covariance matrix. Specifically, a scaling mechanism is designed, using turbulence intensity as an adaptive factor. When the calculated turbulence intensity value is large, it indicates that the current operating condition is highly turbulent, the wind field changes drastically, and the uncertainty of model prediction increases. In this case, the element values ​​in the process noise covariance matrix are automatically increased proportionally. This adjustment means that the filter is given greater flexibility at the algorithm level, making it more inclined to trust new measurement data, thereby enabling it to track the dynamic changes of the system more quickly. Conversely, when the turbulence intensity value is small, it indicates stable wind conditions and high model reliability. The process noise covariance is reduced accordingly, making the filter more confident in the model's predictions, thus smoothing measurement noise and improving estimation accuracy.

[0070] Compared to the closest existing technology, the process noise statistical characteristics of filters in traditional methods are static and disconnected from the current external environment. This implementation creatively correlates the core parameter of the filter—process noise covariance—with turbulence intensity, a physical quantity that directly characterizes the characteristics of the main external disturbance sources, enabling the filter's internal adjustment mechanism to actively match the dynamic changes in the external environment.

[0071] The technical solution provided in this embodiment enables the joint state and parameter estimation algorithm to have a self-adjusting capability that adapts to the intensity of wind disturbance, significantly improving the adaptability and accuracy of state and parameter estimation under different wind conditions, providing the entire control system with more reliable and timely system state information, thereby enhancing the robustness and control performance of the controller in the face of complex and changing wind conditions.

[0072] In another embodiment of the present invention, in step three, the time domain length for solving the finite-time optimal control problem is dynamically adjusted based on the dominant frequencies of wind shear and tower shadow effects in the preview wind speed sequence.

[0073] In model predictive control, the length of the prediction time domain is a critical design parameter. Traditional control strategies typically set it to a fixed value. However, wind shear and tower shadow effects, which have the greatest impact on wind turbine structural loads in offshore wind farms, are disturbances with significant periodic characteristics, whose dominant frequencies dynamically change with rotor speed and average wind speed. A fixed prediction time domain cannot match these changing periodic disturbances. If the prediction time domain is too short, it cannot cover a complete disturbance cycle, and the controller cannot "see" the full picture of the disturbance and perform effective forward compensation; if the prediction time domain is too long, it will unnecessarily increase computational complexity and may reduce control performance due to the accumulation of model prediction errors. This contradiction makes it difficult for a fixed-time-domain control strategy to achieve optimal suppression of periodic loads under all operating conditions.

[0074] In this embodiment, during controller operation, the preview wind speed sequence acquired by the lidar is continuously analyzed to identify the dominant frequencies of periodic wind speed fluctuations caused by wind shearing and tower shadow effects. Based on the real-time identified dominant frequency values, the solution time domain length of the finite-time optimal control problem in step three is dynamically adjusted. Specifically, the adjustment strategy is to adaptively cover one or more complete cycles corresponding to the dominant frequency. For example, when the dominant frequency is high and the corresponding cycle is short, the prediction time domain is automatically shortened to one or two cycle lengths; when the dominant frequency is low and the cycle is long, the prediction time domain is extended accordingly to ensure coverage of critical low-frequency disturbance processes. In this way, the prediction range is dynamically bound to the periodic characteristics of the main disturbance.

[0075] Compared to the closest existing technology, traditional methods employ a fixed prediction time domain. Their design is based on a compromise of typical operating conditions and cannot adapt to the dynamic changes in the periodic characteristics of disturbances during wind turbine operation. It is a passive and compromised solution. This invention, however, transforms the prediction time domain from a fixed control parameter into a dynamic variable closely related to the physical characteristics of the current main disturbance, realizing an active and adaptive prediction strategy.

[0076] This implementation enables the model predictive controller to always match its look-ahead to the most important periodic disturbance characteristics, ensuring that the optimization algorithm can make more reasonable decisions based on disturbance information covering the entire cycle. This significantly improves the suppression effect on periodic load components caused by wind shear and tower shadow effects, and enhances the adaptability of the control strategy and the consistency of control performance under different operating conditions.

[0077] In another embodiment of the present invention, the time-varying aerodynamic parameters identified in real time in step two include at least the equivalent time constant of the dynamic stall model and the induction factor relaxation time constant of the dynamic inflow model.

[0078] When establishing control models for wind turbines, models describing key aerodynamic phenomena such as dynamic stall and dynamic inflow typically include crucial time constant parameters, such as the equivalent time constant of the dynamic stall model and the inducible factor relaxation time constant of the dynamic inflow model. Traditional methods treat these time constants as fixed values. However, the timescale of these aerodynamic processes actually changes significantly with variations in operating conditions such as angle of attack, pitch angle, and tip speed ratio. Using fixed time constants cannot accurately reflect the true response speed of aerodynamics under different operating conditions, leading to dynamic response mismatch in the model when deviating from the design conditions. This severely reduces the accuracy of aerodynamic load calculations, thereby affecting the accuracy of state estimation and control performance based on the model, especially when dealing with highly nonlinear and transient processes such as dynamic stall, where model errors are particularly prominent.

[0079] In this embodiment, the joint state and parameter estimation algorithm executed in step two purposefully expands the set of parameters to be identified, incorporating key aerodynamic time parameters such as the equivalent time constant of the dynamic stall model and the inducible factor relaxation time constant of the dynamic inflow model into the parameter vector to be identified. The algorithm utilizes continuously input preview wind speed sequences and real-time measured structural load data as observation information, not only estimating the system state and common aerodynamic parameters but also simultaneously identifying and updating these two time-varying aerodynamic time constants online in real time. This allows the response characteristics of the model portion describing the aerodynamic dynamic process to automatically adjust according to the current actual operating conditions, thereby enabling the entire parameterized state-space model to more realistically reproduce the dynamic behavior of the wind turbine aerodynamic system.

[0080] Compared to the closest existing technologies, traditional methods rely on offline identification or theoretical calculations to set a constant, compromised time constant for these dynamic processes. These models are static and cannot capture changes in aerodynamic response characteristics during operation. This implementation, however, elevates these two key dynamic parameters from fixed model coefficients to online-identifiable state variables, enabling the aerodynamic model to describe its time-varying dynamic response characteristics and significantly improving the model's dynamic fidelity.

[0081] The technical solution provided by this embodiment significantly improves the calculation accuracy of aerodynamic loads, especially under complex transient conditions such as dynamic stall, thereby providing a more realistic and reliable system model for the state estimator and predictive controller. This effectively improves the prediction accuracy and control performance of the entire control system when facing complex nonlinear aerodynamic phenomena, and enhances the unit's operating performance across the entire operating range.

[0082] In another embodiment of the present invention, in step four, before generating the actual control command for the current cycle, the optimal pitch angle increment command is independently checked and balanced based on the measured pitch angle of the three blades.

[0083] During wind turbine operation, due to wind shearing and tower shadow effects, the actual aerodynamic loads borne by the three blades rotating to different orientations are different, exhibiting inherent asymmetry. Furthermore, there are slight differences in response characteristics among the three independent pitch actuators. Traditional control methods often calculate and output a unified pitch angle command for the three blades, or, while implementing independent pitch control, fail to fully consider the feedback and balance of the actual execution positions of each blade. This results in the unified calculated optimal pitch angle increment command, after distribution and execution, failing to completely eliminate the imbalance of bending moments at the root of the three blades. This residual unbalanced load is transmitted to critical components such as the hub and main shaft, exacerbating fatigue damage and limiting the potential benefits of independent pitch control strategies in load balancing.

[0084] This implementation method specifically involves the following: After obtaining the unified optimal pitch angle increment command in step three, it is not directly and equally allocated to the three blades. Before generating the final actual control command in step four, an independent verification and allocation step is introduced. This step reads the measured pitch angle feedback signals of each of the three blades in real time and compares and analyzes them with the current control target. Its core is to perform consistency verification and redistribution of the uniformly calculated total pitch angle increment command based on the actual position state of the three blades and the specific requirements for load balance. The purpose of verification is to determine whether the command will lead to an increase in imbalance, while allocation dynamically and independently distributes the total command amount to the three blades with different proportions or offsets. The allocation strategy aims to actively compensate for the asymmetric aerodynamic load caused by wind shear and the response deviation caused by actuator differences.

[0085] Compared to the closest existing technology, traditional independent pitch control, while generating different commands for the three blades, may still focus on addressing uneven wind distribution in space, rather than establishing a closed-loop verification and redistribution mechanism based on actual blade position feedback and with load balance as the direct objective. This implementation adds an intelligent decision-making layer before the final command issuance. Based on real-time feedback, it aims to further optimize and ensure load balance at the execution level.

[0086] This implementation improves the precision and intelligence of independent pitch control. By optimizing and allocating unified commands, it can more effectively compensate for asymmetric factors in actual operation, ensuring that the actions of the three blades can ultimately better coordinate to achieve the load balance target. This further reduces the asymmetric fatigue load on key components such as the hub, and optimizes the overall performance and efficiency of independent pitch control.

[0087] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention. Applications, modifications, and variations of the invention will be readily apparent to those skilled in the art.

[0088] Although the 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. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details.

Claims

1. A control method for an offshore wind power system, the offshore wind power system comprising a wind turbine generator, a converter, vibration and stress sensors deployed at key structural points of the wind turbine generator, and a controller integrated within the wind turbine generator, characterized in that, Control methods for offshore wind power systems include: Step 1: Use lidar to scan the wind field in front of the wind turbine to obtain a preview wind speed sequence; at the same time, use vibration and stress sensors to collect real-time load measurements of the bending moments at the front and rear of the tower and the bending moments at the blade roots. Step 2: Establish a parameterized state-space model that includes aeroelastic dynamics and generator drivetrain dynamics. Use a joint state and parameter estimation algorithm based on an unscented Kalman filter, taking the preview wind speed sequence and load measurements as observation inputs, to identify and update the time-varying aerodynamic parameters and unmodeled dynamics in the parameterized state-space model in real time, and simultaneously estimate the state variables of the entire system. Step 3: In each control cycle, with the optimization objectives of minimizing power tracking error and minimizing structural load, based on the updated parameterized state-space model, using the preview wind speed sequence as the known feedforward disturbance input, and the total system state variables estimated in Step 2 as the initial state, under the actuator constraints of the pitch angle change rate, the wind turbine speed change rate, and the generator torque change rate, solve the optimal control problem in the finite time domain and output the optimal control sequence in the future control time domain. Step 4: Parse the first element of the optimal control sequence output in Step 3 into the optimal pitch angle increment command and the optimal generator torque increment command for the current control cycle. Superimpose these commands with the corresponding commands from the previous cycle to generate and execute the actual pitch angle control command and generator torque control command for the current cycle.

2. The control method for an offshore wind power system as described in claim 1, characterized in that, Step one, the process of obtaining a preview wind speed sequence using LiDAR scanning data, includes: Step 11: Based on the original scanning data of the lidar, reconstruct the four-dimensional spatiotemporal wind field within a predefined range in front of the wind turbine. The four-dimensional spatiotemporal wind field includes three-dimensional spatial coordinates and one-dimensional time coordinates. Step 12: In each control cycle, based on the system state variables estimated in Step 2 of the previous cycle and the optimal control sequence obtained in Step 3 of the previous cycle, predict the state trajectory of the wind turbine generator in the future preview time domain. The state trajectory includes at least the yaw angle trajectory and the rotor rotation angle trajectory. Step 13: Establish a wind field-wind rotor coupling model. The four-dimensional spatiotemporal wind field obtained in Step 11 is transformed according to the state trajectory predicted in Step 12 and dynamically mapped to a rotating coordinate system that moves with the wind rotor in the future. This generates a dynamic preview inflow wind field sequence that acts on each blade spanwise unit, with the wind rotor rotation angle as the index.

3. The control method for an offshore wind power system as described in claim 2, characterized in that, In step one, the process of acquiring a preview wind speed sequence using lidar scanning data also includes, based on the error between the real-time measured wind turbine rotation angle and the wind turbine rotation angle trajectory predicted in step 12, performing online correction of the coordinate transformation parameters in the wind field-wind turbine coupling model, specifically: Step a: The actual rotation angle of the wind turbine is measured in real time using an encoder installed on the main shaft of the wind turbine generator. Step b: Compare the actual rotation angle with the wind turbine rotation angle trajectory value predicted in step 12 at the current moment to generate a wind turbine rotation angle error signal; Step c: Input the wind turbine rotation angle error signal into a preset proportional-integral corrector. The proportional-integral corrector outputs an azimuth offset correction amount to compensate for the dynamic mapping process in the wind field-wind turbine coupling model. Step d: Using the azimuth offset correction, adjust the rotation matrix parameters in the coordinate transformation in real time to generate a corrected dynamic preview of the inflow wind field sequence acting on each blade spanwise unit.

4. The control method for an offshore wind power system as described in claim 1, characterized in that, Step three, solving the optimal control problem in the finite-time domain, specifically includes: Step 31: Based on the updated parameterized state-space model, the optimization objectives of minimizing the power tracking error and minimizing the structural load are transformed into objective functions, and considering the actuator constraints, a continuous-time optimal control problem is constructed. Step 32: Using the orthogonal collocation method, the state variables and control variables in the future preview time domain are discretized on a series of collocation points, transforming the continuous-time optimal control problem into a nonlinear programming problem; Step 33: Solve the nonlinear programming problem using a real-time iterative algorithm. In the solution process of each control cycle, the optimal control sequence and state trajectory obtained in the previous control cycle are used as the initial guess values ​​for the current solution to perform iterative calculations in order to accelerate convergence. Step 34: After meeting the preset solution accuracy or maximum number of iterations, output the optimal control sequence for the current control cycle.

5. The control method for an offshore wind power system according to claim 1, characterized in that, In the joint state and parameter estimation algorithm in step two, an adaptive mechanism is introduced to adjust the process noise covariance matrix and measurement noise covariance matrix of the unscented Kalman filter in real time based on the innovation sequence or residual sequence.

6. The control method for an offshore wind power system according to claim 1, characterized in that, Before step four, the process also includes filtering the optimal pitch angle increment command and the optimal generator torque increment command output in step three through a first-order inertial element. The time constant of the first-order inertial element is obtained by identifying the dynamic response characteristics of the pitch actuator and the converter.

7. The control method for an offshore wind power system as described in claim 1, characterized in that, In step two, the joint state and parameter estimation algorithm based on the unscented Kalman filter adjusts the process noise covariance matrix by adaptively scaling the turbulence intensity values ​​of the preview wind speed sequence.

8. The control method for an offshore wind power system as described in claim 1, characterized in that, In step three, the time domain length for solving the finite-time optimal control problem is dynamically adjusted based on the dominant frequencies of wind shear and tower shadow effects in the preview wind speed sequence.

9. The control method for an offshore wind power system as described in claim 1, characterized in that, The time-varying aerodynamic parameters identified in real time in step two include at least the equivalent time constant of the dynamic stall model and the induction factor relaxation time constant of the dynamic inflow model.

10. The control method for an offshore wind power system as described in claim 1, characterized in that, In step four, before generating the actual control command for the current cycle, the optimal pitch angle increment command is independently checked and balanced based on the measured pitch angle of the three blades.