Variable pitch load shedding method and system for floating wind turbine nonlinear model predictive control

Abstract

The application provides a floating wind turbine nonlinear model predictive control variable pitch load reduction method and system, and belongs to the technical field of deep-sea wind power generation control. The method comprises the following steps: obtaining wind turbine state quantity measurement data and historical control input data, constructing a nonlinear candidate function library, solving a coefficient matrix, and constructing an SINDy prediction model capable of representing sparse nonlinear dynamics of the system; the system matrix is loaded into the controller as an internal prediction model, and a future wind speed sequence is input into the controller as a known disturbance; in the controller, the SINDy prediction model and a multi-objective optimization function are used to predict and rollingly optimize the system output in a future time domain under the premise of meeting the physical constraints of the system, and the optimal independent variable pitch control quantity at the current time is calculated; the output of a variable gain PI controller of the unified variable pitch part is calculated, and is superimposed with the optimal independent variable pitch control quantity at the current time to form a complete control signal, which is transmitted to the wind turbine as a pitch angle reference.

Description

Technical Field

[0001] This invention belongs to the field of deep-sea wind power generation control technology, and particularly relates to a method and system for nonlinear model predictive control pitch reduction of floating wind turbines. 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] Unlike stationary wind turbines, floating wind turbines (FOWTs) are installed on floating platforms in the deep sea. They are subjected to the strong coupling effects of complex marine environments such as wind, waves, and currents, exhibiting more complex six-degree-of-freedom rigid body motion characteristics, particularly platform pitch motion. This platform motion, induced by both waves and aerodynamic loads, significantly alters the effective inflow velocity of the blades, causing drastic fluctuations in aerodynamic thrust. This, in turn, greatly increases the fatigue load on the tower base and blade roots, seriously threatening the operational safety and fatigue life of the turbine. To address this challenge, effectively suppressing platform motion and structural loads through pitch control while ensuring stable power generation has become a key technology in floating wind turbine control strategies. Nonlinear model predictive control (NMPC), due to its ability to handle the strong coupling of multi-input multi-output (MIMO) systems, its multi-objective optimization capabilities, its consideration of floating platform motion stability and power generation smoothness, and its ability to directly incorporate system constraints into the optimization problem, has become a mainstream research direction in the field of advanced floating wind turbine control. The NMPC-based pitch control method for floating wind turbines is considered a highly promising control scheme.

[0004] The NMPC-based floating wind turbine pitch control technology is a control method specifically designed to address the load reduction requirements of floating wind turbines in complex marine environments. The core idea of ​​NMPC is to directly employ a nonlinear mathematical model as the prediction model, building upon the traditional MPC's "rolling time-domain optimization" mechanism. Within each control cycle, it uses equations describing the system's nonlinear dynamics to deduce the system's state trajectory over a future period and calculates the optimal control input by solving a nonlinear programming problem. Unlike traditional MPC based on linearized models, NMPC does not rely on linear assumptions at specific operating points. Therefore, it can accurately capture the strongly coupled aerodynamic-hydraulic nonlinear characteristics of floating wind turbines under complex conditions such as large-amplitude motion and strong turbulent winds, thus achieving higher-precision pitch control across the entire wind speed range.

[0005] While NMPC-based pitch control for floating wind turbines exhibits superior performance in handling nonlinear constraints and multi-objective optimization, it also faces some significant challenges in practical applications, as follows: Online computational loads are extremely high, making it difficult to meet real-time requirements. The core of NMPC lies in solving complex nonlinear programming (NLP) problems online within each control cycle. If a high-fidelity physical mechanism model, such as an aerodynamic-hydraulic-elastic coupling model, is used as the prediction model, its mathematical expression is extremely complex, containing a large number of differential-algebraic equations, resulting in excessively long controller computation times, often exceeding the system's sampling period. This makes it difficult to achieve real-time operation of high-performance NMPC on existing industrial controller hardware.

[0006] The dilemma of balancing model accuracy and model complexity. In order to reduce the amount of computation, existing solutions often have to oversimplify the model, such as linearizing and ignoring some degrees of freedom. However, this will cause the model to fail to accurately reflect the nonlinear dynamic behavior of the floating wind turbine under strong turbulent wind and large-amplitude motion, resulting in model mismatch and seriously affecting the load reduction effect of the controller and the stability of the system.

[0007] Furthermore, traditional black-box data-driven modeling methods, such as deep neural networks, while able to approximate nonlinearity, are complex in structure, have redundant parameters, and lack physical interpretability, further increasing the difficulty of solving the problem. Summary of the Invention

[0008] To overcome the shortcomings of the prior art, this invention provides a method and system for nonlinear model predictive control pitch load reduction of floating wind turbines. This method can significantly reduce the solution dimension and computation time of nonlinear programming problems, thereby greatly improving the real-time performance and engineering applicability of the control strategy while ensuring the accuracy of pitch load reduction control of floating wind turbines, and achieving efficient suppression of platform motion and tower base load.

[0009] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect discloses a nonlinear model predictive control method for pitch control and load reduction of floating wind turbines, including: The acquired wind turbine state variable measurement data and historical control input data are used to construct a nonlinear candidate function library, solve the coefficient matrix, and construct a SINDy prediction model that can characterize the sparse nonlinear dynamics of the system. The system matrix is ​​loaded into the controller as an internal prediction model, and the future wind speed sequence is used as a known disturbance input to the controller. In the controller, based on the SINDy prediction model and multi-objective optimization function, the system output in the future time domain is predicted and rolled optimization is performed under the premise of satisfying the physical constraints of the system, and the optimal independent pitch control quantity at the current moment is calculated. The output of the variable gain PI controller of the unified pitch section is calculated and superimposed with the optimal independent pitch control quantity at the current moment to form a complete control signal, which is then transmitted to the wind turbine as a pitch angle reference.

[0010] As a further technical solution, the multi-objective optimization function is as follows:

[0011] In the formula, N is the prediction time domain length; Q is the state weight matrix, used to penalize speed deviation, blade tip deformation, and platform motion; R is the control weight matrix, used to limit the pitch control amplitude; F(x N () is a terminal penalty item used to ensure system stability.

[0012] As a further technical solution, the variable gain PI controller can be specifically represented as follows:

[0013] The controller parameters are characterized as the output pitch angle. The function of offsetting the effect of small perturbations in the pitch angle at different wind speeds on the power sensitivity is:

[0014] These are the controller parameters when pitch is not being adjusted. For gain:

[0015] in, The coefficients are the experimental fit coefficients.

[0016] As a further technical solution, the system physical constraints include pitch angle amplitude constraints and pitch rate constraints.

[0017] As a further technical solution, a SINDy prediction model capable of characterizing the sparse nonlinear dynamics of the system is constructed, specifically including: Simulations were conducted under turbulent wind and irregular wave conditions to obtain system state data. Based on the nonlinear characteristics of the system, a candidate function library is constructed, which includes the state variable x, the control input u, and their nonlinear interaction terms. ; The dynamic equation of a floating wind turbine can be expressed as follows: The coefficient matrix is ​​solved using the sparse regression algorithm. .

[0018] As a further technical solution, the system state vector x and the control input vector u are:

[0019] In the formula, x1, x2, and x3 represent the deformation of the blade tips in the flapping direction, respectively; x4 represents the generator speed; and x5 to x 10These represent the six degrees of freedom motion of the floating platform, and the control input vector is defined as the pitch angle of the three blades.

[0020] Secondly, a nonlinear model predictive control pitch reduction system for floating wind turbines is disclosed, including: The SINDy prediction model building module is configured to: acquire wind turbine state quantity measurement data and historical control input data, build a nonlinear candidate function library, solve the coefficient matrix, and build a SINDy prediction model that can characterize the sparse nonlinear dynamics of the system. The independent pitch control calculation module is configured to load the system matrix as an internal prediction model into the controller, and simultaneously input the future wind speed sequence as a known disturbance into the controller. In the controller, based on the SINDy prediction model and multi-objective optimization function, the system output in the future time domain is predicted and rolled optimization is performed under the premise of satisfying the physical constraints of the system, and the optimal independent pitch control quantity at the current moment is calculated. The complete control signal generation module is configured to: calculate the output of the variable gain PI controller of the unified pitch section, and superimpose it with the optimal independent pitch control quantity at the current moment to form a complete control signal, which is then transmitted to the wind turbine as a pitch angle reference.

[0021] The above one or more technical solutions have the following beneficial effects: This invention proposes a nonlinear model predictive control (NMPC) pitch load reduction control strategy for floating wind turbines based on the SINDy model. It utilizes the Sparse Nonlinear Dynamics Identification (SINDy) algorithm to construct the predictive model for NMPC control, solving the technical challenge of balancing high-precision physical modeling and fast real-time solution in traditional NMPC. This invention automatically mines and filters the sparse nonlinear control equations that dominate the system dynamics from wind turbine operating data using the SINDy algorithm. While preserving the key nonlinear characteristics of the system, it greatly simplifies the mathematical model. Using this lightweight and interpretable SINDy model as the internal predictive model for NMPC significantly reduces the solution dimensionality and computation time of the nonlinear programming problem. Therefore, while ensuring the accuracy of the floating wind turbine pitch load reduction control, it greatly improves the real-time performance and engineering applicability of the control strategy, achieving efficient suppression of platform motion and tower load.

[0022] 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

[0023] 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.

[0024] Figure 1 This invention relates to an NMPC control strategy and GP wind speed prediction. Figure 2 A schematic diagram of data-driven modeling based on SINDy; Figure 3 A schematic diagram of the NMPC pitch control strategy for floating wind turbines based on the SINDy model; Figure 4 This is a block diagram of the GSPI controller. Detailed Implementation

[0025] 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.

[0026] 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.

[0027] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0028] Example 1 See appendix Figure 3 As shown, this embodiment discloses a nonlinear model predictive control pitch control method for floating wind turbines based on the SINDy model, including: Step 1: Measure the fan status x obtained from the sensor. i A library of nonlinear candidate functions was constructed using historical control input data. The sparse nonlinear dynamic equations of the floating wind turbine are obtained by solving the coefficient matrix using a sparse regression algorithm. Thus, a SINDy prediction model capable of characterizing the sparse nonlinear dynamics of the system is constructed. .

[0029] The acquired wind turbine condition measurements should include at least blade deformation, generator speed, and the degrees of freedom of motion of the floating platform. Step 2: The system matrix identified in Step 1... The predictive model is loaded into the NMPC controller as an internal predictive model, and the future wind speed sequence provided by the wind speed prediction module is introduced as a known disturbance input to prepare for predictive control.

[0030] Step 3: In the NMPC controller, based on the SINDy prediction model and the multi-objective optimization function (Equation 4), under the premise of satisfying the system physical constraints, the system output in the future time domain is predicted and rolled optimization is performed to calculate the optimal independent pitch control quantity at the current moment. .

[0031] Step 4: Calculate the output of the CPC part of GSPI according to equations (6) to (8). and with The signals are superimposed to form a complete control signal, which is then transmitted to the wind turbine as a reference for the pitch angle, thus forming a control closed loop.

[0032] Within each control cycle, the SINDy prediction model is used to predict the system state trajectory in the future finite time domain. Combined with a preset objective function and constraints, a nonlinear programming problem is solved to obtain the optimal pitch angle control sequence. The optimal independent pitch control quantity at the current moment is the first element of this optimal pitch angle control sequence. The controller calculates the control sequence for the future time domain through rolling optimization, but only extracts the first element of the sequence as the actual control output at the current moment.

[0033] Rolling control: The first element of the optimal pitch angle control sequence is used as the current pitch control command to apply to the floating wind turbine, and the whole process is repeated at the next moment to achieve rolling control.

[0034] Regarding NMPC control strategies and GP wind speed forecasting.

[0035] NMPC is an advanced control strategy specifically designed for strongly nonlinear systems. It dynamically adjusts the control input by predicting future behavior based on a nonlinear model, so as to achieve the optimal control objective while satisfying system constraints. Figure 1(a) illustrates the NMPC rolling time-domain control principle in the discrete-time domain. The red curve in the figure represents the reference trajectory, which is the target signal that the NMPC expects the floating wind turbine system output to track in real time. In the floating wind turbine scenario, this reference signal is typically determined by multiple control objectives, such as maintaining the generator's rated speed, smoothing output power, and suppressing the movement of the floating platform. NMPC utilizes a high-precision nonlinear mathematical model (in this invention, a sparse nonlinear model based on SINDy identification) to predict the dynamic behavior of the system at future moments. The purple stepped signal in the figure represents the control input sequence (i.e., the pitch angle adjustment command) obtained at the current time t by solving a nonlinear programming (NLP) problem. These control inputs are optimized to make the system's predicted output as close as possible to the reference trajectory in the future time domain. The core mechanism of NMPC lies in rolling optimization. Within each control cycle (from t to t+N in the figure), the NMPC controller uses a nonlinear prediction model to extrapolate the system state for the next N steps and calculates an optimal set of control input sequences based on the set objective function (including penalties for tracking errors and control variables). Although a series of future control inputs are calculated, the system only executes the control action at the current time t. When the next sampling time arrives, the controller rolls the prediction time domain backward one step based on the new system state measurements, and re-predicts and optimizes. This mechanism ensures that the NMPC can respond in real time to random changes in the wind and wave environment and model prediction errors, exhibiting extremely strong robustness. Finally, the blue curve in the figure represents the actual output (or predicted output) of the system. The NMPC's internal nonlinear model accurately captures the aerodynamic-hydraulic-servo elastic coupling characteristics, and the output signal can closely fit the reference trajectory across the entire wind speed range, especially under turbulent winds and large-amplitude motion conditions. At the same time, the NMPC can explicitly handle the system's nonlinear constraints, such as pitch rate limits and physical amplitude limits, ensuring that the wind turbine operates within safe limits.

[0036] Since random fluctuations in wind speed directly determine the aerodynamic load and output power of floating wind turbines, accurate short-term wind speed prediction is a prerequisite for achieving high-performance model predictive control (NMPC). Figure 1 As shown in (b), this invention employs a non-parametric Gaussian process (GP) regression model to handle wind speed uncertainty. The output includes not only the predicted mean of future wind speed but also the predicted variance. This variance is crucial for subsequently evaluating the robustness of the NMPC control strategy. The specific operation steps are as follows: S1: Model Training and Hyperparameter Optimization. The left side of Figure 1(b) uses the collected historical wind speed observation sequence (blue sample points in the figure) as the training set. Maximizing the log-likelihood function is equivalent to maximizing the marginal log-likelihood function of the Gaussian process. Given the historical wind speed sequence, the optimization algorithm is used to find the maximum and minimum values ​​of this function, thereby determining the optimal hyperparameters of the kernel function. This ensures that the established mathematical model best matches the temporal correlation and statistical distribution characteristics of historical wind speeds.

[0037] The kernel hyperparameters of the Gaussian process are optimized. This process aims to capture the temporal correlation and statistical distribution characteristics of wind speed data, such as smoothness and periodicity, thereby constructing a prior Gaussian process model that can describe the prior distribution of wind speed. The initial Gaussian process mathematical model is trained solely on offline collected historical wind speed data. It is fully determined by the kernel hyperparameters optimized in the previous step and includes the initial covariance matrix and mean function of wind speed. It represents a mathematical description of the overall wind speed pattern before current real-time wind speed measurements are obtained.

[0038] S2: Online Update and Posterior Inference. As the system runs (up to the current time t), the controller acquires the latest wind speed measurements in real time. Using Bayesian inference principles, the newly observed data is integrated into the prior model. The prior model, which is the initial model trained and fixed with hyperparameters in step S1, serves as the foundational input for real-time Bayesian inference in this step. The model is then updated in real time. This process transforms the prior distribution into a conditional probability distribution, generating a posterior Gaussian process model. Compared to the prior model, the posterior model incorporates real-time information, significantly reducing prediction errors and correcting the mean and variance of the predictions.

[0039] The posterior Gaussian process model utilizes the conditional probability formula of the multivariate Gaussian distribution. It calculates the conditional probability distribution model by substituting the latest real-time wind speed data as known conditions into the prior model. Mathematically, its predicted mean vector and covariance matrix have been calculated and corrected using the newly observed data. Compared to the prior model, the posterior model, by incorporating the actual wind speed at the latest moment, produces a more accurate predicted mean and a smaller predicted variance.

[0040] S3: Rolling Forecast Output. Based on the updated posterior model, the wind speed distribution in the future forecast time domain is calculated. The GP model not only outputs the predicted mean wind speed at future times (marked by red triangles in the figure, serving as the nominal disturbance input for NMPC), but also simultaneously outputs the forecast variance (confidence interval). This variance quantifies the range of forecast uncertainty and provides crucial information for NMPC to evaluate the robustness of the control strategy.

[0041] Data-driven modeling based on SINDy: Figure 2This invention employs the SINDy algorithm to construct a lightweight nonlinear model, addressing the computational complexity and difficulty in meeting the real-time requirements of NMPC associated with traditional physics models. The specific implementation steps are as follows: Step (1): First, use the high-fidelity aerodynamic-hydraulic-servo elasticity simulation tool (OpenFAST) to perform simulations under turbulent wind and irregular wave conditions to obtain system state data. Define the system state vector x and the control input vector u: (1) In the formula, x1, x2, and x3 represent the deformation of the blade tips in the flapping direction, respectively; x4 represents the generator speed; and x5 to x 10 These represent the six degrees of freedom motion of the floating platform (sway, roll, heave, pitch, pitch, and yaw). The control input vector is defined as the pitch angle of the three blades.

[0042] Step (2): Based on the nonlinear characteristics of the system, construct a candidate function library containing state variable x, control input u, and their nonlinear interaction terms. Considering the physical characteristics of aerodynamics, the candidate function library in this embodiment... The function types include, but are not limited to, polynomial functions, trigonometric functions, exponential functions, and their inter-function terms. The order of the functions can be dynamically adjusted according to computational resources and accuracy requirements. Preferably, considering the physical characteristics of aerodynamic forces, the candidate function library in this embodiment contains at least second-order polynomial terms to achieve a balance between model accuracy and sparsity, and wind speed is included as an external disturbance term in the library. When constructing the candidate function library Θ(X,U), the selected basis functions contain at most the squares of the state variable x and the control input u, along with their pairwise inter-function terms. Constant term: 1; Linear term: x1, x2, ..., u1, u 2； Quadratic term: x1 2 x2 2 ,...,u1 2 u2 2 ,x1x2,x1u 1. Step (3): The dynamic equation of the floating wind turbine can be expressed as follows: The coefficient matrix is ​​solved using the sparse regression algorithm. The SINDy algorithm automatically filters out the dominant non-zero coefficient terms and eliminates redundant terms, thus obtaining an explicit set of differential equations that can accurately describe the nonlinear dynamics of the system, such as aerodynamic-hydraulic coupling, and also has a sparse structure (low computational cost). (2) This model will be directly embedded into the subsequent NMPC controller.

[0043] The coefficient matrix is ​​a sparse weight matrix to be solved. Its row count corresponds to the number of basis functions in the candidate function library Θ, and its column count corresponds to the number of state variables. Each element in the matrix represents the weight of a particular candidate basis function in the derivative equation of a particular state variable. Through the sparse regression algorithm, the coefficients of most redundant terms in this matrix will be 0, leaving only a few non-zero coefficients.

[0044] = (3) After solving for the coefficient matrix Ξ, the non-zero elements are extracted, and they are multiplied by the corresponding candidate functions and then added together to directly form the right-hand side term f(x,u) in equation (2). The coefficient matrix Ξ determines the explicit differential equation system.

[0045] In one implementation example, a floating wind turbine NMPC pitch control strategy based on the SINDy model is disclosed, such as... Figure 3 As shown in the control loop, the NMPC controller solves a constrained nonlinear optimization problem in each control cycle.

[0046] To balance power stability and load suppression, the following multi-objective optimization function J is designed: (4) In the formula, N is the prediction time domain length; Q is the state weight matrix, used to penalize speed deviation, blade tip deformation, and platform motion; R is the control weight matrix, used to limit the pitch control amplitude; F(x N This is a terminal penalty item used to ensure system stability. i u is a condition measurement quantity for the wind turbine. i Let Q and R be a control input vector. Q and R are adjustable hyperparameters set manually in the NMPC controller design. They need to be determined through empirical tuning and repeated trial and error optimization, balancing power stability with load and platform motion suppression. Q=diag([400,400,400,15,50,50,50,50,50]); R=diag([60,60,60]). : The control variable output by the controller in the i-th prediction step, i.e., the specific pitch angle command. : Corresponds to pitch angle amplitude constraints. Represents the physical lower and upper limits of the pitch angle allowed by the actuator; wind turbines are typically limited to... between.

[0047] The optimization problem must satisfy physical constraints: (5) It includes pitch angle magnitude constraints and pitch rate constraints. In addition, state constraints (such as generator overspeed limits) are added as soft constraints to the objective function. : Strain gauge increment This refers to the change in pitch angle between the current control step and the previous control step. : Pitch rate constraint. Combined with the control cycle time, it represents the maximum allowable pitch rate / maximum amplitude of the pitch mechanism. To protect the mechanical structure, the pitch rate is typically rigidly limited to within 8° / s.

[0048] Predict the quadratic form function of the tracking error at the final (Nth) state in the time domain.

[0049] The complete control system block diagram of this method is as follows: Figure 3 As shown, the unified pitch control (CPC) uses a variable gain PI (GSPI) controller to control the wind turbine speed, i.e., power; the independent pitch control (IPC) uses the floating wind turbine NMPC pitch load reduction control strategy based on the SINDy model of this invention.

[0050] The GSPI controller block diagram is as follows: Figure 4 As shown, it can be specifically represented as: (6) System tracking error in the unified pitch circuit. This refers to the deviation between the generator reference speed (rated speed) and the actual measured speed.

[0051] The controller parameters are characterized as the output pitch angle. The function of offsetting the effect of small perturbations in the pitch angle at different wind speeds on the power sensitivity is: (7) These are the controller parameters without pitch control. The two parameters represent the base PI gain when the wind turbine is at its rated operating point, i.e., just reaching the rated wind speed and before the pitch angle begins to change. First, near the rated operating point, linearization is performed using OpenFAST to extract the system's first or second-order linear transfer function at that point. Based on the desired dynamic response index of the wind turbine system, the theoretical proportional gain K is analytically calculated using classical control theory. P and integral gain K I The theoretically calculated gain is then fed into the simulation model, and final fine-tuning is performed based on the overshoot, response time, and fatigue load performance of the wind turbine under step wind and turbulent wind conditions. Subsequently, the parameters at other wind speed points are used for adaptive variable gain calculation via equations (7) and (8) in the text. For gain: (8) in, The experimental fit coefficient is approximately .

[0052] The combination of SINDy and NMPC innovatively applies the Sparse Nonlinear Dynamics Identification (SINDy) algorithm to floating wind turbine control modeling. The sparse nonlinear model constructed by SINDy retains the nonlinear accuracy of the physical model while possessing a minimalist mathematical form (sparseness), perfectly resolving the contradiction between high model fidelity and real-time online solution in NMPC.

[0053] Modeling Method: Data-Driven Sparse Regression. A modeling method based on high-fidelity simulation data (OpenFAST) is proposed. By constructing a candidate function library containing state variables, control inputs, and their nonlinear interaction terms, sparse regression is used to automatically filter out the key terms that dominate the system dynamics. This enables the controller to capture the strongly coupled characteristics of aerodynamics, hydrodynamics, and servoelasticity.

[0054] Innovation in Disturbance Mitigation Strategy: SINDy+GP Synergy. In the model predictive control framework, Gaussian process (GP) regression is introduced to perform short-term probabilistic predictions of future wind speeds. The mean wind speed predicted by GP is used as a feedforward disturbance input to the SINDy model, and the variance information of GP is used to assess uncertainty, thereby significantly enhancing the robustness of the control system under turbulent wind and irregular wave conditions.

[0055] A multi-objective optimization load reduction strategy was developed. A multi-objective cost function incorporating platform motion, blade tip deformation, and power fluctuations was designed. Utilizing the predictive capabilities of the SINDy model, the NMPC can anticipate the platform motion trend caused by waves and adjust the blade pitch angle in advance to achieve proactive load reduction.

[0056] The main objective of this invention is to construct a lightweight prediction model using the Sparse Nonlinear Dynamics Identification (SINDy) algorithm, and combine it with Nonlinear Model Predictive Control (NMPC) to achieve the best balance between model accuracy and computational efficiency. Ultimately, this results in significant enhancement of platform stability and reduction of structural load, thereby improving the operational safety and lifespan of floating wind turbines in complex deep-sea environments.

[0057] Because the marine environment where floating wind turbines are located is more severe and complex, with strong coupling effects of wind, waves and currents, existing linear model predictive control suffers from severe model mismatch and reduced control performance when deviating from the operating point. On the other hand, NMPC based on high-fidelity physical mechanisms has an excessive computational load, making it difficult to meet the millisecond-level real-time control requirements and limiting its engineering applications.

[0058] This invention enables lightweight nonlinear models, improved real-time solution performance, and superior multi-objective control in NMPC control of floating wind turbines, balancing high accuracy with low computational cost to meet real-time control requirements. The control equations mined using the SINDy algorithm exhibit high sparsity (i.e., containing only a few dominant nonlinear terms), significantly reducing the dimensionality and complexity of the model compared to complex physical mechanism models. This dramatically improves the computational speed of the NMPC controller during nonlinear rolling optimization, thus overcoming the technical bottleneck of traditional NMPC's inability to operate online in real time.

[0059] This invention overcomes the shortcomings of "black box" models, possessing clear physical interpretability. Unlike conventional data-driven methods such as neural networks, SINDy identifies explicit systems of ordinary differential equations. This means that the model not only accurately approximates the nonlinear dynamics of the system but also has clear physical meaning and interpretability, facilitating the analysis of system stability and effectively avoiding the poor generalization ability caused by high-order polynomials or overfitting.

[0060] The invention offers significant load reduction and adaptability. Thanks to the SINDy model's accurate capture of the nonlinear characteristics of aerodynamic-hydraulic-servo elastic coupling, combined with Gaussian process (GP) for short-term wind speed prediction, the invention can more accurately predict the motion trend of floating platforms under turbulent winds and irregular waves.

[0061] Example 2 The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method.

[0062] Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0063] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the above method.

[0064] Example 4 The purpose of this embodiment is to provide a nonlinear model predictive control pitch reduction system for floating wind turbines, including: The SINDy prediction model building module is configured to: acquire wind turbine state quantity measurement data and historical control input data, build a nonlinear candidate function library, solve the coefficient matrix, and build a SINDy prediction model that can characterize the sparse nonlinear dynamics of the system. The independent pitch control calculation module is configured to load the system matrix as an internal prediction model into the controller, and simultaneously input the future wind speed sequence as a known disturbance into the controller. In the controller, based on the SINDy prediction model and multi-objective optimization function, the system output in the future time domain is predicted and rolled optimization is performed under the premise of satisfying the physical constraints of the system, and the optimal independent pitch control quantity at the current moment is calculated. The complete control signal generation module is configured to: calculate the output of the variable gain PI controller of the unified pitch section, and superimpose it with the optimal independent pitch control quantity at the current moment to form a complete control signal, which is then transmitted to the wind turbine as a pitch angle reference.

[0065] Example 5 The purpose of this embodiment is to provide a computer program product containing instructions that, when run on a computer, cause the computer to perform the methods and functions involved in any of the above embodiments.

[0066] The steps and methods involved in the apparatus of the above embodiments correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0067] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0068] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A nonlinear model predictive control pitch control method for unloading floating wind turbines, characterized by: include: The acquired wind turbine state variable measurement data and historical control input data are used to construct a nonlinear candidate function library, solve the coefficient matrix, and construct a SINDy prediction model that can characterize the sparse nonlinear dynamics of the system. The system matrix is ​​loaded into the controller as an internal prediction model, and the future wind speed sequence is used as a known disturbance input to the controller. In the controller, based on the SINDy prediction model and multi-objective optimization function, the system output in the future time domain is predicted and rolled optimization is performed under the premise of satisfying the physical constraints of the system, and the optimal independent pitch control quantity at the current moment is calculated. The output of the variable gain PI controller of the unified pitch section is calculated and superimposed with the optimal independent pitch control quantity at the current moment to form a complete control signal, which is then transmitted to the wind turbine as a pitch angle reference.

2. The floating wind turbine nonlinear model predictive control pitch reduction method as described in claim 1, characterized in that, The multi-objective optimization function is as follows: In the formula, N is the prediction time domain length; Q is the state weight matrix, used to penalize speed deviation, blade tip deformation, and platform motion; R is the control weight matrix, used to limit the pitch control amplitude; F(x N () is a terminal penalty item used to ensure system stability.

3. The floating wind turbine nonlinear model predictive control pitch reduction method as described in claim 1, characterized in that, The variable gain PI controller can be specifically represented as: The controller parameters are characterized as the output pitch angle. The function of offsetting the effect of small perturbations in the pitch angle at different wind speeds on the power sensitivity is: These are the controller parameters when pitch is not being adjusted. For gain: in, The coefficients are the experimental fit coefficients.

4. The floating wind turbine nonlinear model predictive control pitch reduction method as described in claim 1, characterized in that, The system physical constraints include pitch angle magnitude constraints and pitch rate constraints.

5. The floating wind turbine nonlinear model predictive control pitch reduction method as described in claim 1, characterized in that, A SINDy prediction model capable of characterizing the sparse nonlinear dynamics of the system is constructed, specifically including: Simulations were conducted under turbulent wind and irregular wave conditions to obtain system state data. Based on the nonlinear characteristics of the system, a candidate function library is constructed, which includes the state variable x, the control input u, and their nonlinear interaction terms. ; The dynamic equation of a floating wind turbine can be expressed as follows: The coefficient matrix is ​​solved using the sparse regression algorithm. .

6. The floating wind turbine nonlinear model predictive control pitch reduction method as described in claim 1, characterized in that, System state vector x and control input vector u: In the formula, x1, x2, and x3 represent the deformation of the blade tips in the flapping direction, respectively; x4 represents the generator speed; and x5 to x 10 These represent the six degrees of freedom motion of the floating platform, and the control input vector is defined as the pitch angle of the three blades.

7. A floating wind turbine nonlinear model predictive control pitch control system for load shedding, characterized in that... include: The SINDy prediction model building module is configured to: acquire wind turbine state quantity measurement data and historical control input data, build a nonlinear candidate function library, solve the coefficient matrix, and build a SINDy prediction model that can characterize the sparse nonlinear dynamics of the system. The independent pitch control calculation module is configured to load the system matrix as an internal prediction model into the controller, and simultaneously input the future wind speed sequence as a known disturbance into the controller. In the controller, based on the SINDy prediction model and multi-objective optimization function, the system output in the future time domain is predicted and rolled optimization is performed under the premise of satisfying the physical constraints of the system, and the optimal independent pitch control quantity at the current moment is calculated. The complete control signal generation module is configured to: calculate the output of the variable gain PI controller of the unified pitch section, and superimpose it with the optimal independent pitch control quantity at the current moment to form a complete control signal, which is then transmitted to the wind turbine as a pitch angle reference.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method described in any one of claims 1-6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it performs the steps of the method described in any one of claims 1-6 above.

Images