A floating type wind turbine variable pitch load shedding control method and system

Abstract

The application provides a floating wind turbine variable pitch load reduction control method and system, and belongs to the technical field of offshore wind power control, and comprises the following steps: updating a priori Gaussian model based on the real-time operation state of a floating wind turbine to obtain a posteriori Gaussian model, the posteriori Gaussian model outputs a system future state prediction trajectory; constructing a cost function based on the system future state prediction trajectory, and screening a plurality of groups of variable pitch control trajectories with the best performance in the current iteration period under the condition of meeting physical and operation constraints; obtaining a global optimal control sequence by iteratively optimizing the plurality of groups of variable pitch control trajectories by using a cross-entropy method; and extracting the first component of the global optimal control sequence as a current variable pitch instruction based on the rolling horizon control principle, and entering the next control cycle after the variable pitch instruction is executed. The application realizes efficient closed-loop control of the complex dynamics of the floating wind turbine.

Description

Technical Field

[0001] This invention belongs to the field of offshore wind power generation control technology, and particularly relates to a method and system for variable pitch load reduction control of floating wind turbine generator sets. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] As wind power generation expands into deeper and more distant waters, floating wind turbines (FOWTs) are becoming a development trend due to their ability to overcome the limitations of near-shore seabed geological conditions and utilize superior wind energy resources. A typical floating wind turbine system consists of a top turbine, a floating foundation, and a mooring system. At wind speeds above rated speed, the turbine primarily maintains constant rotor speed and output power through pitch control. However, unlike stationary turbines, floating turbines possess additional platform degrees of freedom, particularly pitch freedom. When the pitch mechanism adjusts the pitch angle to change the aerodynamic torque, it simultaneously alters the aerodynamic thrust on the rotor, leading to severe platform oscillations under the combined effects of waves and wind loads. This complex aerodynamic-structural-hydraulic coupling characteristic necessitates that the control objectives of floating wind turbines not only include power stability but also the suppression of platform motion and the reduction of structural loads (such as tower base bending moment and mooring tension).

[0004] Currently, existing technologies for pitch control and load reduction of floating wind turbines mainly include: compensation control schemes based on gain-scheduled PID, linear model predictive control based on mechanistic modeling, and nonlinear model predictive control based on gradient optimization kernels. However, existing technologies still have shortcomings: First, mechanistic modeling is inaccurate and struggles to capture unmodeled dynamics: Floating wind turbines operate under highly complex stochastic sea conditions, and traditional physical modeling (such as linearized models or simplified nonlinear models) cannot accurately describe the complex interactions between aerodynamics, structure, and fluid, especially wave excitation forces and the nonlinear stiffness of the mooring system. This model uncertainty can cause the predicted trajectory of the MPC controller to deviate from the actual operating state, leading to control failure or closed-loop system oscillation. Second, traditional optimization algorithms are prone to getting trapped in local optima: There is a significant nonlinear game between the control objectives of floating wind turbines (speed control, power stability, platform load reduction, and pitch mechanism wear), and their cost functions are usually non-convex and non-smooth. Existing gradient-based optimization algorithms such as SQP and interior-point methods are prone to getting trapped in local optima and cannot find the optimal control strategy that minimizes structural loads globally. Third, the ability to resist random environmental disturbances is weak: most existing solutions are based on deterministic models and lack utilization of the probability distribution characteristics of wind and wave disturbances. When faced with sudden gusts or giant waves, traditional controllers often overreact or lag, making it difficult to effectively suppress the negative damping effect of platform motion while ensuring power generation efficiency. Summary of the Invention

[0005] To overcome the shortcomings of the existing technologies, this invention proposes a variable pitch load reduction control method and system for floating wind turbine generators. Based on a model predictive control algorithm using Gaussian processes and cross-entropy methods, the Gaussian process directly processes the nonlinear predictive model, solving the problems of decreased control accuracy and system instability caused by model linearization distortion. By constructing a cost function containing multiple penalty terms and utilizing the cross-entropy algorithm for global optimization in a complex constraint space, the fatigue load on key wind turbine structural components is minimized while ensuring power generation efficiency. Furthermore, by introducing a cross-entropy-based sampling evolution strategy, the global optimization capability problem in the optimization process is solved, improving the robustness and real-time response capability of the controller under extreme random sea conditions.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: In a first aspect, the present invention discloses a method for variable pitch load reduction control of a floating wind turbine generator set, comprising: The prior Gaussian model is updated based on the real-time operating status of the floating wind turbine to obtain the posterior Gaussian model, and the posterior Gaussian model outputs the predicted trajectory of the future state of the system. Based on the predicted trajectory of the future state of the system, a cost function is constructed, and several sets of pitch control trajectories that best match the control objective in the current iteration cycle are selected under the constraints of physical and operational conditions. The global optimal control sequence is obtained by iteratively optimizing the several sets of pitch control trajectories using the cross-entropy method. Based on the rolling time-domain control principle, the first component of the globally optimal control sequence is extracted as the current pitch command. After the pitch command is executed, the next control cycle begins.

[0007] Secondly, this invention discloses a floating wind turbine generator pitch reduction control system, comprising: The trajectory prediction module is configured to update the prior Gaussian model based on the real-time operating status of the floating wind turbine to obtain a posterior Gaussian model, wherein the posterior Gaussian model outputs the predicted trajectory of the future state of the system. The control prediction module is configured to: construct a cost function based on the predicted trajectory of the future state of the system, and select several sets of pitch control trajectories that best match the control objective in the current iteration cycle under the constraints of physical and operational conditions. The optimization solution module is configured to use the cross-entropy method to iteratively optimize the several sets of pitch control trajectories to obtain the globally optimal control sequence. The control execution module is configured to: extract the first component of the globally optimal control sequence based on the rolling time domain control principle as the current pitch command, and after executing the pitch command, enter the next control loop.

[0008] Thirdly, the present invention discloses an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when run by the processor, complete the steps of the above-mentioned floating wind turbine pitch load reduction control method.

[0009] Fourthly, the present invention discloses a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-described floating wind turbine pitch load reduction control method.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention utilizes Gaussian processes to directly construct a nonlinear prediction model, enabling more accurate capture of the dynamic response of floating wind turbines under different wind speeds and wave heights through a data-driven approach. Furthermore, the GP model possesses online learning capabilities, spontaneously correcting model biases as wind turbine operating data accumulates, effectively solving the prediction inaccuracies caused by parameter drift and model simplification in traditional mechanistic models.

[0011] This invention introduces the cross-entropy method as the optimization kernel. As an evolutionary search algorithm based on probability distribution sampling, it does not rely on the gradient information of the objective function and has extremely strong global optimization capabilities. In multi-objective games involving "power stability, load suppression, and pitch mechanism wear," this invention can find the globally optimal pitch control sequence that minimizes the overall fatigue load of the system within a complex solution space.

[0012] The cross-entropy algorithm used in this invention has simple logic and naturally supports parallel computing, without involving complex matrix inversion or derivative derivation. Even if the cost function has discontinuities or noise, the cross-entropy algorithm can still stably provide a feasible solution, greatly enhancing the robustness of the control system on embedded hardware and its real-time response speed.

[0013] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0014] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0015] Figure 1This is a flowchart of the floating wind turbine generator pitch reduction control method described in Embodiment 1 of the present invention.

[0016] Figure 2 This is a schematic diagram illustrating the principle of model predictive control in the discrete time domain as described in Embodiment 1 of the present invention.

[0017] Figure 3 This is a schematic diagram of the wind and wave environment prediction and system modeling based on GP as described in Embodiment 1 of the present invention. Detailed Implementation

[0018] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0020] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0021] Currently, the most similar implementation schemes among existing technologies for pitch control of floating wind turbines mainly include the following: 1. Gain-Scheduled PID Compensation Control Scheme: This is currently the most mature scheme in the offshore wind power industry. Based on the traditional onshore wind turbine pitch PID controller, this scheme introduces platform pitch rate feedback as an auxiliary control signal. By linearizing the wind turbine model, a gain scheduling table is designed at different wind speed points, utilizing additional pitch compensation to increase the system's aerodynamic damping.

[0022] 2. Mechanism-based linear model predictive control: This scheme utilizes the Lagrange equations or multibody dynamics software to establish a linearized state-space model of the wind turbine near a specific operating point. Within each control cycle, power fluctuations, speed deviations, and platform pose are constructed as quadratic cost functions, and the optimal pitch angle command is calculated using a quadratic programming (QP) solver under linear constraints.

[0023] 3. Nonlinear Model Predictive Control Based on Gradient Optimization Kernel: This is a cutting-edge approach that has been extensively studied in academia. This approach directly uses a nonlinear physical model (such as a simplified multi-degree-of-freedom rigid body model) as the prediction model, and utilizes gradient-based optimization algorithms such as Sequential Quadratic Programming (SQP) or interior point methods to perform nonlinear iterative optimization of the control sequence in the prediction time domain to address the nonlinear characteristics of wind turbines under a wide range of varying operating conditions.

[0024] In summary, existing technologies using traditional linear controllers (such as PID) or MPC based on linearized models struggle to accurately describe the highly nonlinear, time-varying, and aerodynamic-structural-fluid highly coupled characteristics of floating wind turbines under complex sea conditions. Existing control strategies often fail to achieve an optimal balance between the three conflicting objectives of maintaining stable output power, reducing tower / mooring loads, and minimizing pitch mechanism wear. Traditional nonlinear MPCs typically employ gradient-based optimization algorithms (such as SQP), which are prone to getting trapped in local optima when dealing with non-convex, non-smooth cost functions and are sensitive to initial values.

[0025] Example 1 To address the shortcomings of the existing technology, in one or more embodiments, a floating wind turbine pitch load reduction control method based on Gaussian process and cross-entropy model predictive control (GP-CEM-MPC) is disclosed, such as... Figures 1-2 As shown, it includes the following steps: Step S0: Collect real-time operating status data from multiple sources.

[0026] The measuring device captures the operating status of the floating wind turbine in real time, including the system state variables at the current moment. Measured pitch angle and environmental disturbance signals (wind speed) With the height of the waves ).

[0027] In this embodiment, the perception of time boundaries and states, such as Figure 2 As shown in the figure, time t is used as the dividing line between the current time and the past time. The area to the left of time t represents the historical interval, indicating the actual state of the system that has already occurred; the area to the right is the "future" interval, representing the time domain for the algorithm's upcoming prediction and optimization. At each sampling time t, the controller first obtains the current actual output of the wind turbine system (the point at time t indicated by the blue solid line in the figure), including key parameters such as rotor position and wind turbine blade pitch angle.

[0028] Step S1: Based on the real-time operating status of the floating wind turbine, update the prior Gaussian model to obtain the posterior Gaussian model, and the posterior Gaussian model outputs the predicted trajectory of the future state of the system.

[0029] In this embodiment, the aforementioned data, along with historically collected training datasets (covering dimensions such as wind speed, waves, rotor azimuth, blade pitch angle, and system state), are input into the Gaussian process (GP) model. Through online learning and parameter updates, the GP model constructs a high-precision nonlinear mapping model for power generation and blade flapping torque, providing a foundation for subsequent predictions.

[0030] The controller uses a Gaussian process (GP) model to predict the system's behavior over a future period, with the forward prediction step size as the time window. For example... Figure 2 As shown, the red solid line represents the reference trajectory. Based on a preset reference signal (such as rated power), this method generates a "reference trajectory" that smoothly transitions from the current actual output to the reference signal. The goal of this invention is to make the system's future predicted output fit this trajectory as closely as possible.

[0031] The specific logic of wind and wave environment prediction and wind turbine dynamics system modeling based on GP, ​​such as Figure 3 As shown, by combining historical statistical characteristics with real-time observation data, a probabilistic prediction of future states can be achieved. Specifically: Step S1-1: Sample collection and offline training.

[0032] Feature data is extracted from the raw data to be predicted (such as random wind speed or wave signals) and used as offline training samples, such as... Figure 3 The blue dots represent samples that are then input into a prior Gaussian model, where hyperparameters are optimized using algorithms such as maximum likelihood estimation.

[0033] This step aims to determine the optimal parameters of the kernel function so that the model can accurately learn the characteristics of wind and waves or the nonlinear mapping laws of the wind turbine system.

[0034] In this embodiment, a discrete-time nonlinear dynamic model based on GP regression is constructed to learn and simulate the complex nonlinear mapping from the current system state and control input to the system output.

[0035] The system at discrete time arrive The dynamic change process can be expressed as the following regression model: (1) in, Represents the output index; Represents power generation capacity; These correspond to the waving direction bending moments at the roots of the three blades of the wind turbine, respectively. In order to target the A GP regression function trained with each output; Independent Gaussian white noise associated with each output is used to characterize random errors in the modeling process.

[0036] To fully capture the higher-order dynamic dependencies of the floating wind turbine system, the input vector of the GP model consists of the following parts: Wind turbine position angle ( ): Captures the periodic characteristics of aerodynamic loads caused by wind shear and tower shadow effects.

[0037] Controllable input vector ( ): Independent pitch angle of three blades composition.

[0038] Uncontrollable perturbation vector ( ): Includes real-time wind speed and wave height This provides the model with information about the external environment.

[0039] Steps S1-2: Online observation and posterior inference.

[0040] During real-time control, the controller acquires the current time t and the real-time observation points up to that time, such as... Figure 3 The black dots represent the newly observed real-time data. Based on Bayesian inference principles, this data is injected into the optimized hyperparameter prior model to update the posterior Gaussian model. This posterior Gaussian model, by integrating historical experience with current trends, enables the model to adapt to complex environmental changes.

[0041] Specifically, considering that the output of a Gaussian process is a probability distribution, in order to improve the computational efficiency of control sequence optimization, this invention takes the expected value (mean) of its posterior distribution within each prediction step as the deterministic prediction result at that moment, expressed as: (2) in, Indicates the current Time prediction The first moment One output; It is a mean function.

[0042] Steps S1-3: Predicted point output and characteristics.

[0043] Based on the posterior distribution, the posterior Gaussian model outputs the predicted points within the future step size, such as... Figure 3 The red triangle in the middle indicates that the predicted output includes not only the mean, representing the future evolution trend, but also the variance, representing the uncertainty of the prediction.

[0044] To obtain the entire prediction time domain The state trajectory within is used to construct the state prediction equation shown in equation (3). The core logic of this mechanism lies in using the predicted output of the previous time step as the input state of the next time step: (3) in, The wind turbine position angle at a future moment. These forecasts of wind speed and wave height for future moments are all derived from real-time predictions by GP. The pitch angle control command to be optimized.

[0045] In this step, real-time feedback from the measuring device is received, and combined with the previously known future wind speed and wave prediction sequence, the future wind speed is output using the aforementioned regression iteration mechanism (Equation 3). The system output trajectory sequence within a prediction step. This trajectory sequence dynamically reflects the future power fluctuations and structural load evolution trends of the wind turbine system under different candidate control actions.

[0046] In this invention, the Gaussian process prediction process has a dual function: (1) Environmental prediction; predicting wind speed and wave excitation force in the future time domain, providing a basis for predicting the future state of the system. (2) System modeling; replacing the traditional physical mechanism equations, directly establishing a dynamic model between pitch command and wind turbine load response through data-driven methods, providing a model basis for model predictive control.

[0047] Step S2: Construct a cost function based on the predicted trajectory of the future state of the system, and select several sets of pitch control trajectories that best match the control objective in the current iteration cycle while satisfying physical and operational constraints.

[0048] In this embodiment, a control step size is defined within the prediction time domain. (generally ).like Figure 2 The purple stepped line represents the control signal to be optimized (i.e., the pitch command sequence).

[0049] In this invention, global optimization is employed for the control sequences within the time domain. The cross-entropy method (CEM) is used for random sampling and iterative evolution within the constraint space. CEM calculates the optimal set of control sequences by minimizing the deviation between the predicted output and the reference trajectory, while comprehensively considering platform load reduction constraints. Specifically, this includes: Based on the predicted trajectory of the system's future state generated by equation (3), the cost function shown in equation (4) is constructed: (4) in, This represents the control sequence vector to be optimized; This indicates the result predicted by equation (3). The system output at that moment; The system's output reference value is set to... ; and The sum of squared deviations between them is obtained through the weight coefficient matrix. Adjustments are made to achieve comprehensive optimization of power generation stability and blade root bending moment. The cost function also includes adjustments to the control input. The penalty term is determined by the weight matrix. Adjust the weight of the penalty pitching action to avoid frequent and violent oscillations in the actuator; This is the predicted output at time k+j, given at the current time k.

[0050] In the optimization process, this embodiment not only requires the cost function Minimization must also meet a series of stringent physical and operational constraints to ensure the safety of the wind turbine system. Specifically: (1) The derivation of the predicted trajectory must strictly follow the dynamic constraints determined by the Gaussian process, that is, the state output at each moment must satisfy the nonlinear mapping relationship shown in equation (3) to ensure that the predicted behavior is consistent with the actual mechanism characteristics of the floating wind turbine. For prediction System status at all times Input.

[0051] (2) Due to the mechanical and physical limits of the variable pitch actuator, control quantity boundary constraints are set, requiring the pitch angle command in each prediction step to be... All are at the minimum allowed value With the maximum value This not only protects the physical safety of the pitch bearing, but also prevents turbine stall caused by excessive pitch angle.

[0052] (3) To reduce the mechanical fatigue of the pitch mechanism and suppress the high-frequency oscillation of the system, a pitch rate constraint is further introduced. As shown in equation (4), the control increment between two adjacent sampling times is... It must be limited to the maximum permissible rate of change. Within.

[0053] Finally, by minimizing the cost function within the aforementioned constraint space... This invention transforms the pitch control problem into a nonlinear predictive optimization problem under multiple constraints. Due to the nonlinearity of the Gaussian process model and the complexity of the cost function, this embodiment uses CEM to solve the optimization problem globally. By performing probabilistic sampling and iterative evolution in the solution space, CEM can avoid the defect of traditional gradient optimization algorithms that are prone to getting trapped in local optima, thereby selecting the pitch command sequence that achieves the optimal overall system performance while satisfying physical constraints.

[0054] Step S3: Use the cross-entropy method (CEM) to iteratively optimize the several sets of pitch control trajectories to obtain the globally optimal control sequence.

[0055] In this embodiment, a parallel search based on probability distribution is employed, based on the generated predicted trajectory. This includes: sampling a large-scale solution set within the current control space, evaluating the cost function value using the cost function (Equation 4), and selecting elite individuals according to ranking logic. Through multiple iterations and updates to the sampling distribution, this module ultimately locks down the globally optimal control sequence that optimizes the system performance indicators. Specifically: Each sampling time Internal control instruction sequence to be optimized Consider it as following a multivariate Gaussian distribution: (5) Among them, the mean vector This represents the expected center of the currently searched control sequence; Let be the covariance matrix, representing the distribution range of the search space. Before the optimization iteration begins, the system uses the mean sequence obtained from the previous time step as initial experience, and initializes the mean at the current time step through a time shift operation.

[0056] Based on this, CEM achieves iterative evolution of the control strategy through an internal loop process. In the... In this iteration, the algorithm first draws from the Gaussian distribution based on the current mean and covariance. Group candidate control sequence samples: (6) in, The control instruction sequence of the nth candidate control sequence sample; To control the mean of the sequence in the i-th iteration at the current time k; To control the variance of the sequence.

[0057] For each set of samples, the autoregressive GP state prediction equation shown in equation (3) is used, and the cost function value corresponding to each candidate sequence is calculated in combination with equation (4). The sorted sample sequence is defined as follows: The corresponding cost function values ​​satisfy the following ordering relationship: (7) Here, the subscript (n) indicates the order of the sample in the current population after sorting by cost function value from smallest to largest, rather than the original order of sample generation represented by the superscript (n) in equation (6). The top performers are selected based on their performance. These samples constitute an "elite set". : (8) This elite group This includes the performance that best matches the control objectives (i.e., the most stable power and the best load reduction effect) within the current iteration cycle. Group pitch control trajectory.

[0058] To ensure the search distribution continuously converges to the low-cost region, this invention updates the parameters of the sampling distribution by minimizing the cross-entropy between the parameterized Gaussian distribution and the empirical distribution of the elite set. Specifically, this involves re-estimating the next iteration's parameters using the mean and covariance of the elite set. and : (9) In the formula, This is the mean of the next iteration; This is the covariance for the next iteration.

[0059] Through a predetermined number of iterations, the search window continuously converges towards the global optimum, and the mean vector... It eventually stabilized near the optimal pitch control trajectory.

[0060] This embodiment addresses the aforementioned optimization problem of nonlinear, nonconvex cost functions based on the GP prediction model, employing the cross-entropy method as the core solution kernel. Due to the non-analytical nature of the GP model, traditional gradient-based algorithms struggle to handle its complex search space. In contrast, CEM, a stochastic optimization algorithm based on sampling evolution, can achieve global optimization without derivative information and boasts strong parallelization capabilities and real-time computational performance.

[0061] Step S4: Based on the rolling time domain control principle, extract the first component of the global optimal control sequence as the current pitch command and send it to the OpenFAST wind turbine model (or the actual wind turbine actuator). After the pitch command is executed, the measuring device senses the new system state again and enters the next control cycle.

[0062] Specifically, after completing the iterative optimization at the current moment, this invention follows the rolling time-domain principle of model predictive control and extracts only the first control vector from the optimal mean sequence. ( Figure 2 The portion circled in red (indicated by the dotted red circle) acts on the pitch actuator of the wind turbine. The remaining control components are discarded, and the above probabilistic evolution-based search process is repeated in the next sampling cycle, incorporating the latest sensor feedback.

[0063] Preferably, this embodiment employs a rolling forward approach. At the next sampling time t+1, the entire prediction window moves one step to the right, and the controller repeats the above process based on the new sensor feedback data. This "rolling optimization" mechanism enables the system to correct prediction errors caused by GP model bias or random wind and wave disturbances in real time.

[0064] Through the above technical solutions, the present invention achieves the following effects: 1. Construct a high-precision data-driven wind turbine prediction model.

[0065] The purpose of this invention is to directly construct a wind turbine dynamics prediction model using GP (Geometry, Physics, and Geometry). By learning from historical operating data, the GP model can directly capture the extremely complex nonlinear mapping relationships of floating wind turbines without deriving complex physical mechanism equations. This simplifies the modeling process and improves the ability to capture nonlinear dynamics while ensuring prediction accuracy.

[0066] 2. Solve the global optimal pitch decision problem through gradient-free optimization.

[0067] Since the GP model itself is not conducive to complex gradient differentiation, this invention introduces CEM (Critical Evolutionary Model) as the optimization kernel. CEM, as a stochastic sampling evolutionary algorithm, does not rely on the gradient information of the prediction model and can directly perform global optimization based on the probability output of the GP model. The purpose of this invention is to solve the problem that traditional control strategies are prone to getting trapped in local optima, and to find the pitch control sequence that truly minimizes the overall fatigue load of the system under complex multi-objective constraints of power, speed, and load.

[0068] 3. Improve the dynamic evolution and adaptive capabilities of the control system.

[0069] This invention leverages the ease of online updating of the GP model, enabling the wind turbine model to continuously evolve with operational data and correct model deviations in real time. Combined with the efficient search capabilities of CEM, this invention aims to realize an intelligent controller with self-learning and self-optimization capabilities, achieving high-precision and stable adjustment of power output while effectively suppressing platform pitch motion.

[0070] In summary, the floating wind turbine pitch control method based on GP-CEM-MPC in this invention deeply integrates data-driven modeling, global stochastic optimization, and rolling time-domain control, possessing a complete "perception-prediction-optimization-execution" chain. It solves the problem of difficult floating wind turbine modeling by utilizing the nonlinear fitting capability of the GP model and solves the non-convex optimization problem by utilizing the global search capability of the CEM algorithm. It achieves efficient closed-loop control of the complex dynamics of the floating wind turbine, ensuring that the floating wind turbine can maintain stable power output under complex environmental disturbances and significantly reducing the fatigue load of key structural components.

[0071] Example 2 In one or more embodiments, a floating wind turbine pitch load reduction control system is disclosed, specifically including: The trajectory prediction module is configured to update the prior Gaussian model based on the real-time operating status of the floating wind turbine to obtain a posterior Gaussian model, wherein the posterior Gaussian model outputs the predicted trajectory of the future state of the system. The control prediction module is configured to: construct a cost function based on the predicted trajectory of the future state of the system, and select several sets of pitch control trajectories that best match the control objective in the current iteration cycle under the constraints of physical and operational conditions. The optimization solution module is configured to use the cross-entropy method to iteratively optimize the several sets of pitch control trajectories to obtain the globally optimal control sequence. The control execution module is configured to: extract the first component of the globally optimal control sequence based on the rolling time domain control principle as the current pitch command, and after executing the pitch command, enter the next control loop.

[0072] Example 3 This embodiment provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps of the above-described floating wind turbine pitch load reduction control method.

[0073] Example 4 This embodiment provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-described floating wind turbine pitch load reduction control method.

[0074] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0075] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0076] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0077] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0078] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for controlling the load reduction of a floating wind turbine generator with variable pitch, characterized in that, include: The prior Gaussian model is updated based on the real-time operating status of the floating wind turbine to obtain the posterior Gaussian model, and the posterior Gaussian model outputs the predicted trajectory of the future state of the system. Based on the predicted trajectory of the future state of the system, a cost function is constructed, and several sets of pitch control trajectories that best match the control objective in the current iteration cycle are selected under the constraints of physical and operational conditions. The global optimal control sequence is obtained by iteratively optimizing the several sets of pitch control trajectories using the cross-entropy method. Based on the rolling time-domain control principle, the first component of the globally optimal control sequence is extracted as the current pitch command. After the pitch command is executed, the next control cycle begins.

2. The method for controlling the load reduction of a floating wind turbine generator set with variable pitch as described in claim 1, characterized in that, The prior Gaussian model is updated based on the real-time operating status of the floating wind turbine to obtain the posterior Gaussian model, and the predicted trajectory of the system's future state is output as follows: The input vector of the Gaussian model includes the rotor position angle, the independent pitch angles of the three blades, and the uncontrollable disturbance vector, which includes the real-time wind speed and the wave height. Within each prediction step, the expected value of its posterior distribution is taken as the deterministic prediction result at that moment, expressed as: in, Indicates the current Time prediction The first moment One output; In order to target the A GP regression function trained with each output; The position angle of the wind turbine; This refers to the pitch angle. This is an uncontrollable perturbation vector.

3. The method for controlling the load reduction of a floating wind turbine generator set with variable pitch as described in claim 1, characterized in that, The cost function The physical and operational constraints are as follows: in, This represents the control sequence vector to be optimized; Indicates the predicted The system output at that moment; This serves as the system's output reference value. The weight coefficient matrix; This is the weight matrix; , These are the control inputs at time k+j-1 and k+j-2, respectively. It is a mean function; This is the i-th output predicted at time k+j from the current time k. For prediction System status at all times Input; , These are the minimum and maximum values ​​allowed by the pitch angle command, respectively. To control the maximum allowable rate of change of the increment.

4. The method for controlling the load reduction of a floating wind turbine generator set with variable pitch as described in claim 1, characterized in that, The method employs cross-entropy to iteratively optimize the several sets of pitch control trajectories to obtain the globally optimal control sequence. Based on the generated predicted trajectory, a parallel search using probability distributions is employed, including... Within the current control space, a large-scale sampling solution set is performed, the cost function value is evaluated using the cost function, and elite individuals are selected according to the sorting logic. By iteratively updating the sampling distribution, the globally optimal control sequence that optimizes the system performance is determined.

5. The method for controlling the load reduction of a floating wind turbine generator set with variable pitch as described in claim 4, characterized in that, The sequence of control commands to be optimized at each sampling time is regarded as a multivariate Gaussian distribution. The cross-entropy method is used to achieve iterative evolution of the control strategy through an internal loop process, at the... In the next iteration, several sets of candidate control sequence samples are first drawn from the Gaussian distribution based on the current mean and covariance; For each set of samples, the GP state prediction equation is used and the cost function value corresponding to each candidate sequence is combined to define the sorted sample sequence; The top few samples are selected according to the order of the arrangement to form an elite set.

6. The method for controlling the load reduction of a floating wind turbine generator set with variable pitch as described in claim 4, characterized in that, The parameters of the sampling distribution are updated by minimizing the cross-entropy between the parameterized Gaussian distribution and the empirical distribution of the elite set. Specifically: Re-estimate the mean and covariance of the next iteration using the mean and covariance of the elite set: in, This is the mean of the next iteration; The covariance for the next iteration; The number of pitch control trajectories; A gathering of elites; Through a preset number of iterations, the search window continuously converges towards the global optimal solution, resulting in the global optimal control sequence.

7. The method for controlling the load reduction of a floating wind turbine generator set with variable pitch as described in claim 1, characterized in that, The system employs a rolling forward approach, where at the next sampling time, the entire prediction window moves one step to the right, and the controller repeats the control process based on the new sensor feedback data.

8. A floating wind turbine generator pitch reduction control system, characterized in that, include: The trajectory prediction module is configured to update the prior Gaussian model based on the real-time operating status of the floating wind turbine to obtain a posterior Gaussian model, wherein the posterior Gaussian model outputs the predicted trajectory of the future state of the system. The control prediction module is configured to: construct a cost function based on the predicted trajectory of the future state of the system, and select several sets of pitch control trajectories that best match the control objective in the current iteration cycle under the constraints of physical and operational conditions. The optimization solution module is configured to use the cross-entropy method to iteratively optimize the several sets of pitch control trajectories to obtain the globally optimal control sequence. The control execution module is configured to: extract the first component of the globally optimal control sequence based on the rolling time domain control principle as the current pitch command, and after executing the pitch command, enter the next control loop.

9. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the variable pitch load reduction control method for a floating wind turbine generator as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, complete the floating wind turbine generator pitch reduction control method according to any one of claims 1-7.

Images