Position domain iterative learning independent pitch control method under sparse sampling
Through position domain modeling with sparse sampling and iterative learning, the problems of low efficiency and high cost of periodic load suppression of wind turbines are solved, a balance between load suppression performance and economy is achieved, component life is extended and system cost is reduced.
Patent Information
- Application Number
- CN202510794956.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-14
- Publication Date
- 2025-09-05
AI Technical Summary
During operation, wind turbines face cyclic loads, which shorten the fatigue life of the blade roots. Traditional control methods have limited effect on suppressing cyclic loads and rely on high-density sampling, resulting in high hardware costs. There is a lack of control solutions that balance efficient load suppression and economy.
An independent variable pitch control method with iterative learning in position domain under sparse sampling is adopted. Through position domain modeling and iterative learning mechanism, the variable frequency periodic load is converted into a fixed frequency disturbance. An iterative learning observer and controller are designed. Combined with sparse sampling optimization, the sampling frequency and hardware cost are reduced while maintaining the control performance.
It effectively reduces the blade root tilt and yaw bending moment loads by 10%-15%, extends component life, reduces the sampling frequency to 1/4 of the traditional one, reduces hardware and computing costs, and adapts to different wind speed conditions.
Smart Images
Figure CN120595599A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of wind power generation control, and specifically relates to a position domain iterative learning independent pitch control method and system under sparse sampling, which are used to suppress the periodic load of a wind turbine generator and optimize the control system cost. Background Art
[0002] During operation, wind turbine systems face severe challenges from cyclical loads such as wind shear, tower shadow effects, and gravity. These loads can shorten the fatigue life of key components, such as blade root bending moments, by 30%-50%. Traditional centralized pitch control (CPC) has limited effectiveness in suppressing cyclical loads. While independent pitch control (IPC) can precisely control individual blade loads, traditional time-domain IPC struggles to adapt to variable-frequency cyclical loads caused by speed fluctuations, significantly reducing its control performance during actual variable-speed operation. Furthermore, existing control methods rely on high-density sampling, resulting in high hardware costs and heavy computational burdens. A control solution that balances efficient load suppression with cost-effectiveness is urgently needed.
[0003] While existing position-domain modeling can convert variable-frequency loads into fixed-frequency disturbances, there is still a lack of a systematic solution for combining iterative learning mechanisms to achieve high-precision control while simultaneously reducing system costs through sparse sampling. Therefore, developing an independent variable pitch control method based on iterative learning in the position domain that balances load suppression performance with sampling cost has important engineering application value. Summary of the Invention
[0004] The present invention aims to solve the problems of low efficiency in periodic load suppression and high cost of control systems for wind turbines in the prior art, and provides a method and system for independent pitch control based on iterative learning in the position domain under sparse sampling. Through position domain modeling, iterative learning mechanism and sparse sampling optimization, the load suppression accuracy is improved and the hardware and computing costs are reduced.
[0005] To achieve the above objectives, the technical solutions adopted by the present invention mainly include: establishing a position domain state space model of the wind turbine, and converting the variable frequency periodic load into a fixed spatial frequency interference through multi-blade coordinate transformation (MBC). Designing an iterative learning observer (ILO), and using the output error to iteratively update the interference estimate to achieve high-precision estimation of periodic interference. Designing an iterative learning independent pitch controller (ILC-IPC), and combining the interference estimate to construct a PD-type iterative learning control law. The controller works in conjunction with the traditional centralized pitch control (CPC), CPC is responsible for power limiting, and ILC-IPC focuses on load suppression. Construct and solve the linear matrix inequality (LMI) optimization problem on the sampling step size to determine the maximum allowable sampling step size to ensure system stability. .
[0006] The present invention also discloses a corresponding control system, which includes a modeling unit, an observer unit, a controller unit and a sampling optimization unit, so as to realize the above method.
[0007] The beneficial effects of this invention include: through position-domain modeling and an iterative learning mechanism, blade root pitch and yaw bending moment loads are reduced by over 10%-15% compared to traditional time-domain IPC, effectively extending component life. Sparse sampling reduces the sampling frequency at the maximum allowable sampling step size to one-quarter that of traditional high-precision sampling, reducing hardware (such as encoder) precision requirements and online computational complexity, thereby lowering system costs. The iterative learning mechanism ensures that errors converge gradually with increasing iterations, and even with sparse sampling, maintains stable control performance, adapting to varying wind speed conditions. BRIEF DESCRIPTION OF THE DRAWINGS
[0008] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0009] Figure 1 This is a block diagram of the ILC-IPC control system under sparse sampling provided by an embodiment of the present invention;
[0010] Figure 2 This is a diagram of the sparse sampling ILC-IPC control framework provided by an embodiment of the present invention.
[0011] Figure 3 The following is an abstract. DETAILED DESCRIPTION
[0012] Below in conjunction with specific embodiment, further set forth the present invention.Should be understood that these embodiments are only used to illustrate the present invention and are not used in limiting the scope of the present invention.In addition, should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the application equally.
[0013] 1. Establishment of wind location domain model
[0014] Using an NREL 5MW benchmark wind turbine as the model, a position-domain model was developed to describe the system's dynamic characteristics. The system primarily consists of the rotor, pitch mechanism, transmission system, and controller. The core concept is to transform variable-frequency loads into fixed-frequency disturbances through multi-blade coordinate transformation (MBC).
[0015] The system model is as follows:
[0016] (1.1)
[0017] State vector : Contains the mechanical state (such as blade azimuth, speed) and electrical state (such as generator torque) of the wind turbine, with dimensions of .
[0018] System Matrix : Describes the state transition characteristics of the system when there is no interference, which is dimensional matrix, which is determined by the wind turbine dynamic parameters (such as moment of inertia and damping coefficient).
[0019] Interference Input Matrix : Correlated periodic interference signal and the state vector, dimensional matrix, where q is the dimension of the interference signal (such as the bending moment component at the blade root).
[0020] Control Input Matrix : Map the state vector to a measurable output (such as the blade root bending moment sensor signal), which is dimensional matrix, is the number of output variables.
[0021] Periodic interference signal :Contains variable frequency loads caused by wind shear, tower shadow effect, etc., which are converted into fixed spatial frequency signals after MBC transformation.
[0022] Control Input : Pitch angle adjustment command generated by the iterative learning independent pitch controller (ILC-IPC).
[0023] 2 Iterative learning independent pitch observer design
[0024] 2.1 Design of iterative learning independent pitch observer
[0025] According to formula (1.1), due to the repeatable dynamic characteristics of the system, for each repeated motion cycle, is the number of iterations. Therefore, the system for each iteration cycle can be rewritten as
[0026] (2.1)
[0027] By designing a disturbance estimator based on the state observer, the state vector and interference signals The constant matrix and vector dimension form of the observer are the same as those of system (2.1), and the specific form is as follows:
[0028] (2.2)
[0029] in, is the state vector of the observer, is the estimated output of the observer, is the interference estimation result, represents the observer gain matrix with appropriate number of dimensions.
[0030] Defining state error , state error change rate , interference estimation error and system output error as follows:
[0031] (2.3)
[0032] (2.4)
[0033] (2.5)
[0034] (2.6)
[0035] For the interference estimator, an iterative learning strategy is introduced, using the state error , and the rate of change of state error Constructing PD-type iterative learning law, the result of interference estimation Iterate and finally achieve the interference estimation result Interference with the system The complete tracking of , the specific form is as follows:
[0036] (2.7)
[0037] 2.2 Convergence and stability conditions of iterative learning independent pitch observer
[0038] According to Lyapunov theorem, if the system is to achieve asymptotic stability, it needs to satisfy
[0039] (2.8)
[0040] , : Observer-aided design parameters, which are the gain matrix of the PD-type iterative learning law, are obtained by solving the LMI optimization problem and are used to adjust the iterative convergence speed of the disturbance estimation.
[0041] : Symmetric positive definite matrix, which is the weight matrix constructed by Lyapunov function and determined by LMI optimization, is used to ensure system stability.
[0042] : A positive scalar parameter, which is the slack variable in the stability constraint and is optimized by LMI solution to ensure the convergence of the error dynamic system.
[0043] : Observer gain matrix, which adjusts the estimation accuracy of the observer through state error feedback, and is composed of the system matrix and LMI conditions determined.
[0044] In the above parameter descriptions, symbols with unclear physical meanings are auxiliary parameters solved through LMI optimization. Their specific values are automatically calculated by MATLAB LMI Toolbox based on the system model to simultaneously meet the requirements of stability, convergence, and sparse sampling performance.
[0045] 3 Iterative Learning Independent Pitch Controller Design
[0046] 3.1 Interference Suppression System
[0047] In order to suppress interference, it is necessary to combine the iterative learning interference suppression controller, design the iterative learning control law, and build a new interference suppression system. The system structure is as follows:
[0048] (3.1)
[0049] in is the iterative learning part of interference suppression, is the interference estimate of the j-th iteration ILO, is the interference estimation error, is the output error compensation part of the iterative learning law, represents a gain matrix with the appropriate number of dimensions.
[0050] Also define .
[0051] The relevant variables of the interference suppression system (3.1) are defined as follows: , state error change rate , system output error , control action error , iterative learning disturbance control law error as follows:
[0052] (3.2)
[0053] (3.3)
[0054] (3.4)
[0055] (3.5)
[0056] (3.6)
[0057] In the interference suppression system, the state error , state error change rate and interference estimation results Construct a PD-type iterative learning law to design the iterative learning part of interference suppression as follows:
[0058] (3.7)
[0059] 3.2 Convergence and stability conditions of iterative learning independent pitch controller
[0060] According to Lyapunov's theorem, for the system to achieve asymptotic stability, it needs to satisfy:
[0061]
[0062] (3.8)
[0063] : Observer-aided design parameters, which are the gain matrix of the PD-type iterative learning law, are obtained by solving the LMI optimization problem and are used to adjust the iterative correction amount of the control input.
[0064] : Symmetric positive definite matrices, which are weight matrices of state error, control input error, and observer error, respectively, and are automatically calculated by the LMI tool to meet the stability conditions.
[0065] : A positive scalar parameter, which is the slack variable in the stability constraint and is optimized by LMI solution to ensure the convergence of the error dynamic system.
[0066] : Output error compensation gain matrix, used to correct the response speed of the controller to the system output error, which is determined jointly by the system model and the LMI condition.
[0067] In the above parameter descriptions, symbols with unclear physical meanings are auxiliary parameters solved through LMI optimization. Their specific values are automatically calculated by MATLAB LMI Toolbox based on the system model to simultaneously meet the requirements of stability, convergence, and sparse sampling performance.
[0068] 4 Solving Linear Matrix Inequality (LMI) Optimization
[0069] In the position domain sampling system, sparse sampling aims to increase the sampling interval (i.e., the sampling step size) ) to reduce system cost while ensuring the stability of the control system. Therefore, the core problem is to determine the maximum allowable sampling step size while satisfying the stability constraint.
[0070] The maximum sampling step optimization problem can be expressed as:
[0071] (4.1)
[0072] : Sampling step (angle domain interval), the optimization goal is to maximize this value to achieve sparse sampling.
[0073] : The physical upper limit of the sampling step, determined by the encoder accuracy and the mechanical system bandwidth.
[0074] : Interference rejection performance index, which limits the upper bound of interference estimation error and control output error.
[0075] : The safety threshold of the interference signal energy, ensuring that the system remains stable under external interference.
[0076] :The weight parameters of the iterative learning law are optimized through LMI to balance the convergence speed and overshoot.
[0077] In the above parameter descriptions, symbols with unclear physical meanings are auxiliary parameters solved through LMI optimization. Their specific values are automatically calculated by MATLAB LMI Toolbox based on the system model to simultaneously meet the requirements of stability, convergence, and sparse sampling performance.
[0078] By solving equation (4.1), we can determine a solution that can both ensure system stability and maximize sampling sparsity. .
Claims
1. This is a position domain iterative learning independent pitch control method under sparse sampling, characterized by: The following steps are involved: S1: Establishing a position domain control system model of the wind turbine to convert variable frequency periodic loads into fixed spatial frequency interference; S2: Design an iterative learning observer (ILO) to estimate periodic disturbances and optimize control inputs based on a position-domain model. S3: Design an iterative learning independent pitch controller (ILC-IPC) to construct a PD-type iterative learning control law based on disturbance estimates to improve load suppression accuracy. S4: Construct and solve a linear matrix inequality (LMI) optimization problem to determine the maximum allowable sampling step size that ensures system stability and optimize control performance under sparse sampling.
2. The method according to claim 1, wherein: In step S1, establishing the position domain control system model includes: constructing a wind turbine multi-blade coordinate transformation (MBC) model to convert the blade root bending moment into pitch and yaw direction components in a fixed coordinate system; and establishing a discrete time state space equation.
3. The method according to claim 1 or 2, characterized in that: In step S3, the iterative learning independent pitch controller (ILC-IPC) adopts a PD type control law.
4. The method according to claim 1, wherein: In step S4, the linear matrix inequality (LMI) optimization problem is: in, For the stability constraint matrix including the sampling step h, the maximum allowable sampling step size is solved by MATLAB LMI Toolbox .
5. The method according to claim 5, characterized in that: In step S4, sparse sampling reduces hardware accuracy requirements by increasing the sampling interval while satisfying the system stability condition: ( is the safety threshold).
6. A position domain iterative learning independent pitch control system under sparse sampling, characterized in that: include: Modeling unit, used to establish the wind turbine location domain model to achieve load frequency conversion and state space expression; The observer unit is used to run the iterative learning observer (ILO) and output the periodic disturbance estimation value; the controller unit is used to execute the iterative learning independent pitch control law (ILC-IPC) and generate the pitch angle control command; the sampling optimization unit is used to solve the LMI optimization problem, determine the maximum allowable sampling step size and realize sparse sampling.
7. The system according to claim 8, characterized in that: The observer unit works in conjunction with the controller unit to gradually optimize disturbance estimation and control input through an iterative learning mechanism, thereby reducing the bending moment load at the blade root.
8. The system according to claim 8, characterized in that: The sampling optimization unit supports multi-precision sampling mode switching, including high-precision sampling , medium-precision sampling and low-precision sampling , to balance control performance and hardware cost.