Wind turbine model-based control and estimation using accurate online models

Abstract

The present invention relates to wind turbine model-based control and estimation using accurate online models. A system for calculating estimated operating parameters and / or control commands for a wind turbine includes: a sensor that monitors a wind turbine; a control processor that implements a linearized evaluation to obtain a model of structural component dynamic behavior, fluid component dynamic behavior, and / or combined structural and fluid component dynamic behavior of wind turbine operation; and a module that performs operations using the linearized evaluation of the structural component dynamic behavior, fluid component dynamic behavior, and / or combined structural and fluid component dynamic behavior. The module is at least one of an estimation module and a multivariable control module. The estimation module generates a signal estimate of a turbine or fluid state. The multivariable control module determines an actuator command, the actuator command including a wind turbine command to maintain the operation of the wind turbine to a predetermined setting in real time. A method and a non-transitory medium are also disclosed.

Description

Technical Field

[0001] The present invention relates to wind turbine model based control and estimation using accurate online models. Background Art

[0002] Models of wind turbines and the surrounding aerodynamics are used in estimation and control applications to capture the physical behavior of the turbine for online computation. In some conventional estimation applications, models can be used to provide approximations of parameter values that are not directly measured. For example, an estimator based on a wind turbine model can provide blade tip deflection even without a tip deflection sensor. In some conventional control applications, the wind turbine model can provide computed estimates of tower base moments, blade root moments, and other factors as inputs to the control system, which can then use these inputs to determine the appropriate command set to adjust the blade pitch angle and generator torque to ensure that the blade root moments and tower base moments do not exceed their permitted limits.

[0003] Wind turbine models can also be used in offline high-fidelity simulations, for example, as required for detailed structural load analysis. These models typically include many degrees of freedom to improve the accuracy of transient behavior predictions. Typical turbine degrees of freedom modeled in a wind turbine simulation environment include drive train rotation, drive train torsion, tower bending modes in the fore-aft and lateral directions, tower torsion modes, blade bending modes in the flap and edge directions, blade torsion modes, tower foundation translation and rotation modes, and nacelle pitch modes.

[0004] A key difference between models used for simulation and those used for estimation or control is that the latter require linear approximations or the operation of derivatives, which are computationally expensive, whereas simulation models only require evaluating equations. Models used for control and estimation applications will be referred to as "control models."

[0005] Due to the limited computing power available in conventional industrial control platforms, the fidelity of characterizing simulation models is difficult to achieve in control models that operate in real time. To capture the most important wind turbine operating characteristics within a limited computing environment, conventional first-generation control models use simplified models of turbine structural dynamics and aerodynamics. Typical elements of these conventional online models feature a simplified quasi-steady-state aerodynamic model in the form of a lookup table derived from blade element momentum ("BEM") theory that employs a number of assumptions (e.g., modeling only the first tower bending mode in the fore-aft and lateral directions, and assuming that aerodynamic forces are generated instantaneously in response to any changes in the wind field or any changes in structural displacement).

[0006] These typical online models of wind turbines may include elements such as, for example, drive train rotation [Equations (1)-(2)]; the first bending mode of the tower in the fore-aft direction [Equations (3)-(4)]; the first bending mode of the tower in the lateral direction [Equations (5)-(6)]; and static aerodynamic tables for power, torque, and thrust [Equations (7)-(8)]. These conventional models may have the following form:

[0007]

[0008]

[0009]

[0010]

[0011]

[0012]

[0013]

[0014]

[0015] Where θ is the blade pitch angle, V eff =V w -v fa is the effective wind speed at the top of the tower, λ = ωR / V eff is the speed ratio of the blade tip to the wind; T gen is the generator torque control input; and, N g is the gearbox ratio. The advantage of model (1-8) is that its derivatives are simple to compute. However, controllers using this type of control model may have limited performance due to its inaccuracies.

[0016] To achieve higher energy production and superior load shedding capabilities, advances in design and operating standards for wind turbine power plants are needed. Operational advances can be achieved by improving control system design and turbine structural and aerodynamic models.

[0017] What is missing in the art is a control model that has higher fidelity than conventional models to provide improved control system design and turbine structural and aerodynamic modeling. These higher fidelity models will still need to operate on the limited computing power available at the wind power platform, but provide a rigorous physical representation of the turbine and aerodynamics. Summary of the Invention

[0018] Technical Solution 1. A wind turbine system comprising:

[0019] a sensor suite comprising a plurality of sensors positioned to monitor wind turbine operating parameters and environmental conditions, the sensor suite in communication with a data store and configured to provide sensor dynamics data to the data store;

[0020] a control processor in communication with the data storage section, the control processor comprising a processor unit;

[0021] a control model configured to perform a linearization evaluation at a turbine operating point, the linearization evaluation applying an analytical differentiation technique to obtain at least one of a structural component dynamic behavior, a fluid component dynamic behavior, and a combined structural and fluid component dynamic behavior of wind turbine operation;

[0022] the control model including a coupling between the structural component dynamic behavior and the fluid component dynamic behavior, the coupling being implemented by introducing generalized fluid forces into a structural model of the wind turbine and including in the fluid component dynamic behavior a dependency of the generalized fluid forces on a structural state of the wind turbine, the structural model including structural features that are accurate with respect to at least one of a first bending natural mode of a tower, a first bending natural mode of one or more blades, and a rotation of a drive train of the wind turbine; and

[0023] A module is configured to perform an operation using at least one of the linearized estimate of the structural component dynamic behavior, the fluid component dynamic behavior, and the combined structural and fluid component dynamic behavior.

[0024] Technical Solution 2. The system according to Technical Solution 1, characterized in that it includes:

[0025] The module is an estimation module configured to generate one or more estimates selected from the group consisting of a turbine state signal estimate, a turbine performance signal estimate, a fluid state signal estimate, and a fluid performance signal estimate; and,

[0026] Each of the one or more estimates is based on one or more elements of sensor dynamic data, the structural member dynamic behavior, and the fluid member dynamic behavior.

[0027] Technical Solution 3. The system according to Technical Solution 2 is characterized in that the estimation module is configured to generate the one or more estimation values by implementing extended Kalman filtering.

[0028] Technical Solution 4. The system according to Technical Solution 1 is characterized in that at least part of the dynamic behavior of the fluid component included in the lookup table includes a representation of one or more rotor plane fluid properties and their derivatives, and at least one of the one or more blades crosses the rotor plane.

[0029] Technical Solution 5. The system according to Technical Solution 4, wherein the lookup table comprises:

[0030] a respective element comprising, for a respective one of the one or more blades, a data set, each data set comprising a dependency on at least one of one or more of a blade tip to wind speed ratio, a pitch angle, an azimuth angle, and a representative dynamic pressure parameter; and,

[0031] The data set is applicable to determine at least one of blade force and blade twist.

[0032] Technical Solution 6. The system according to Technical Solution 2 is characterized in that the interconnected structural component dynamic behavior model and the fluid component dynamic behavior model are expanded using a wind disturbance dynamic behavior model to estimate wind parameters, wherein the wind parameters include average wind speed, misalignment, shear, steering, harmonic change, and one or more of the effective wind speed of each blade expressed as the harmonic change of wind.

[0033] Technical Solution 7. The system according to Technical Solution 1 is characterized in that the structural model includes blade distortion effects caused by one or more of fluid force, inertial force and mass blade force.

[0034] Technical Solution 8. The system according to Technical Solution 1, characterized in that it includes:

[0035] The module is a multivariable control module configured to determine actuator commands based on one or more control states; and,

[0036] The actuator commands include wind turbine commands to maintain operation of the wind turbine at predetermined settings in approximately real time.

[0037] Technical Solution 9. The system according to Technical Solution 8 is characterized in that the multivariable control module includes a model predictive controller.

[0038] Technical Solution 10. The system according to Technical Solution 1 is characterized in that the control model is configured to perform the linearization evaluation by implementing one of algorithmic differentiation or symbolic differentiation.

[0039] Technical Solution 11. The system according to Technical Solution 1, characterized in that the dynamic behavior of the fluid component includes:

[0040] a contribution rate of local forces applied to the blade segment based on the local aerodynamic properties of the airfoil augmented to apply corrections for three-dimensional effects; and

[0041] Dynamic wake effects from wind turbine operation.

[0042] Technical Solution 12. The system according to Technical Solution 1 is characterized in that the dynamic behavior of the fluid component includes at least one of a wind propagation model upstream of the rotor surface and a wind propagation model downstream of the rotor surface and data from a remote wind measurement system sensor.

[0043] Technical Solution 13. The system according to Technical Solution 1 is characterized in that the control model is configured to obtain an analytical representation of the dynamic behavior of the structural component through multi-agent dynamic modeling technology.

[0044] Technical Solution 14. The system according to Technical Solution 1 is characterized in that the control model is configured to obtain the dynamic behavior of the fluid component by implementing a blade element momentum (BEM) model, wherein the BEM model includes a representation of one or more fluid properties of a rotor plane, and at least one of the one or more blades crosses the rotor plane.

[0045] Technical Solution 15. The system according to Technical Solution 1, wherein the control model is configured as follows:

[0046] approximating a corresponding function for at least one of the structural component dynamic behavior and the fluid component dynamic behavior using a proxy model; and,

[0047] The corresponding analytical derivative is constructed from the corresponding surrogate function.

[0048] Technical Solution 16. A method for calculating at least one of an estimated operating parameter and a control command for a wind turbine, the method comprising:

[0049] The sensor suite monitors wind turbine operating parameters and environmental conditions to acquire sensor dynamic data to a data storage unit;

[0050] performing a linearization evaluation at a turbine operating point, the linearization evaluation applying an analytical differentiation technique to obtain at least one of structural component dynamic behavior, fluid component dynamic behavior, and combined structural and fluid component dynamic behavior of wind turbine operation;

[0051] implementing coupling between the structural component dynamic behavior and the fluid component dynamic behavior by introducing generalized fluid forces into a structural model of the wind turbine and including a dependency of the generalized fluid forces on a structural state of the wind turbine in the fluid component dynamic behavior, the structural model including structural features that are accurate with respect to at least one of a first bending natural mode of a tower, a first bending natural mode of one or more blades, and a rotation of a drive train of the wind turbine; and

[0052] At least one of an estimation module and a multivariable control module performs an operation using at least one of the linearized estimates of the structural component dynamic behavior, the fluid component dynamic behavior, and the combined structural and fluid component dynamic behavior.

[0053] Technical Solution 17. The method according to Technical Solution 16, characterized in that it includes:

[0054] the estimation module generating one or more estimates selected from the group consisting of a turbine state signal estimate, a turbine performance signal estimate, a fluid state signal estimate, and a fluid performance signal estimate; and

[0055] Each of the one or more estimates is based on one or more elements of sensor dynamic data, the structural member dynamic behavior, and the fluid member dynamic behavior.

[0056] Technical Solution 18. The method according to Technical Solution 16, characterized in that it includes:

[0057] The multivariable control module determines actuator commands based on one or more control states; and

[0058] The actuator commands include wind turbine commands to maintain operation of the wind turbine at predetermined settings in approximately real time.

[0059] Technical Solution 19. The method according to Technical Solution 16 is characterized in that it includes using a wind disturbance dynamic behavior model to expand the interconnected structural component dynamic behavior model and the fluid component dynamic behavior model to estimate wind parameters, wherein the wind parameters include average wind speed, misalignment, shear, steering, harmonic change, and one or more of the effective wind speed of each blade expressed as the harmonic change of wind.

[0060] Technical Solution 20. The method according to Technical Solution 16 is characterized in that the structural model includes components representing blade distortion effects caused by one or more of fluid forces, inertial forces, and mass blade forces.

[0061] Technical Solution 21. The method according to Technical Solution 16 is characterized in that it includes performing the linearization evaluation by implementing one of algorithmic differentiation or symbolic differentiation.

[0062] Technical Solution 22. The method according to Technical Solution 16, characterized in that it includes modeling the dynamic behavior of the fluid component to take into account:

[0063] a contribution rate of local forces applied to the blade segment based on the local aerodynamic properties of the airfoil augmented to apply corrections for three-dimensional effects; and

[0064] Dynamic wake effects from wind turbine operation.

[0065] Technical Solution 23. The method according to Technical Solution 16 is characterized in that the dynamic behavior of the fluid component includes at least one of a wind propagation model upstream of the rotor surface and a wind propagation model downstream of the rotor surface and data from a remote wind measurement system sensor.

[0066] Technical Solution 24. The method according to Technical Solution 16 is characterized in that it includes obtaining an analytical representation of the dynamic behavior of the structural component by implementing multi-agent dynamic modeling technology.

[0067] Technical Solution 25. The method according to Technical Solution 16 is characterized in that it includes implementing a blade element momentum (BEM) model to obtain at least part of the dynamic behavior of the fluid component, wherein the BEM model includes a representation of one or more fluid properties of a rotor plane, and at least one of the one or more blades crosses the rotor plane.

[0068] Technical Solution 26. The method according to Technical Solution 16, characterized in that it includes:

[0069] approximating a respective proxy function for at least one of the structural member dynamic behavior and the fluid member dynamic behavior using a neural network; and

[0070] The corresponding analytical derivative is constructed from the corresponding surrogate function.

[0071] Technical Solution 27. The method according to Technical Solution 16 is characterized in that at least a portion of the dynamic behavior of the fluid component is included in a lookup table, the lookup table containing representations of one or more rotor plane fluid properties and their derivatives, and at least one of the one or more blades crosses the rotor plane. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 schematically depicts an aeroelastic model structure according to an embodiment;

[0073] Figure 2 Depicted is a process for calculating control signals for dynamic operation of a wind turbine according to an embodiment; and

[0074] Figure 3 A system for calculating control signals for dynamic operation of a wind turbine according to an embodiment is described. DETAILED DESCRIPTION

[0075] The embodying system and method provides a high-fidelity control model of a wind turbine that provides a physically rigorous representation of the turbine structural dynamics and aerodynamics while operating on the limited computing power available at a wind power generation platform.

[0076] Previous attempts to improve conventional modeling have been hampered by modeling errors arising from a set of simplifying assumptions, including but not limited to: using only the first bending mode to represent tower bending; ignoring tower torsion; ignoring blade flexibility; representing deterministic wind as only the mean wind speed; ignoring dynamic wake effects; and assuming a rigid tower foundation.

[0077] Accurate wind turbine aeroelastic model for estimation and control

[0078] The implemented modeling techniques include an approach that differs from conventional models by implementing more rigorous mechanical equations to represent the dynamic interactions between multiple bodies, including flexible representations of the blades, tower, and main shaft. The implemented model also includes accurate aerodynamic equations, including representations of induced velocities and fluid / structure interactions.

[0079] The algorithms that provide these practical solutions to the mechanical and aerodynamic equations produce rigorous and accurate results in real time (ie, during operational control of the wind turbine) when executed on the limited computing power available at the wind turbine platform.

[0080] Figure 1 An aeroelastic model structure 100 according to an embodiment is schematically depicted. To manage the complexity of the overall aeroelastic model 100, it is convenient to divide the overall aeroelastic model 100 into a structural component model 110 and an aerodynamic (fluid) component model 120. The structural component 110 addresses the dynamics of the rigid and flexible bodies; and the fluid component 120 addresses the behavior associated with aerodynamic forces (and fluid dynamics for offshore wind applications). The following discussion only concerns the aerodynamic fluid forces at the interface between the structural component model and the aerodynamic component model. However, one of ordinary skill in the art can readily incorporate fluid dynamics for offshore wind applications.

[0081] Structural Model

[0082] Conventional wind turbine structural models are analytically formulated as a multi-agent system with flexible components. This multi-agent system can be mathematically expressed using the Lagrange method, D'Alembert method, or Newton-Euler method. Applying this conventional method to the Lagrange formula for the wind turbine system, the Lagrangian function L is calculated as the kinetic energy (T 涡轮 ) and potential energy (V 涡轮 ):

[0083] L=T 涡轮 -V 涡轮

[0084] Here, kinetic and potential energy are calculated as the sum of the contributions of each component. This is expressed mathematically as:

[0085] T 涡轮 =T 地基 +T 塔架 +T 机舱 +T 驱动系 +T 叶片

[0086] V 涡轮 =V 地基 +V 塔架 +V 机舱 +V 驱动系 +V 叶片

[0087] Then, the structural equation of motion of the wind turbine can be obtained from the standard Lagrangian formulation:

[0088]

[0089] where q is the generalized coordinate of the system, Q is the generalized force, and the notation denotes the partial derivative with respect to the variable v, and, denotes the derivative with respect to time. Standard mathematical manipulations are applied to (9) to obtain a structural equation of the form:

[0090]

[0091]

[0092] where u is the control input (e.g., blade pitch angle, generator torque); M is the mass matrix; and the vector h represents the generalized forces from different sources (h = h 惯性 +h 重力 +h 空气动力 +h 施加), the generalized forces include inertial forces, gravity, aerodynamic forces, and applied forces such as electrical torque at the main shaft and torque at the blade pitch motor.

[0093] vector y s Contains output from the structural model that is useful for control applications or estimation applications. For example, for estimation applications, y s Includes simulated sensor measurements; for control applications, y s Typically includes signals that can be used to evaluate feedback actions (e.g. thrust, tower moment and blade moment). The concatenation of the coordinates and their derivatives is generally used as the state vector of the structural system. And adopt.

[0094] These conventional formulas can be used to obtain a simulation environment for system and / or controller design or to obtain load certification. However, these conventional simulation models cannot be effectively implemented to operate in real time.

[0095] The embodied systems and methods utilize the described systematic approach to deriving the equations of motion for a wind turbine system to further advance model definition to include the ability to analytically generate the model Jacobian to implement embodiments in an online, real-time process (i.e., during operational control of the wind turbine).

[0096] Aerodynamic model

[0097] Aerodynamic forces can be derived from a blade element momentum (BEM) model to estimate the induced velocity field around the wind turbine and the local aerodynamic forces at each blade. The blade is divided into a finite number of elements, and then the lift, drag, and aerodynamic moments are calculated at each element based on the local wind speed, the local displacement and velocity of the blade element, and the airfoil properties.

[0098] In order to have a more accurate physical representation, the BEM equations may be modified to take into account the effects of blade tip losses and blade root losses (eg by using Prandtl-Glauert correction factors).

[0099] The BEM model can be combined with a wake model that describes the dynamics of the inflow wind field by means of induced velocity. The dynamic inflow model can be used to acquire the time constant of the induced velocity to adapt to new operating conditions. If the dynamic wake effect is ignored, for example, using a static or steady-state wake model (where the induced velocity can be assumed to achieve instantaneous balance with the aerodynamic force), the controller based on the corresponding model may have an inappropriate representation of the fluid-structure interaction and may lead to instability. The complete aerodynamic model that combines a model such as the BEM with a dynamic wake model can be written as follows:

[0100]

[0101] h 空气动力 =g(x a ,z a ,x s ,w p ) (13)

[0102] 0 = g 代数 (x a ,z a ,x s ,w p ) (14)

[0103] Among them, x a is the aerodynamic state vector that may include the induced axial velocity and the induced tangential velocity; z a is the vector of the aerodynamic algebraic equation; w p is a vector of deterministic wind parameters, which may include mean wind speed, misalignment, shear, steering, and harmonic variation around azimuth; x s is the state of the structural member; h 空气动力 represents the generalized aerodynamic forces and moments on the blades (typically 3 blades); and the function g 代数 Explicit algebraic constraints expressing the aerodynamic force magnitudes. Alternative forms can be expressed in the absence of algebraic constraints that make these forms implicitly defined within the functions f and g. The formulas typically appear as computer programs rather than as a collection of closed-form expressions (12-14).

[0104] The coupling between the aerodynamic model and the structural model is achieved through the structural state x as input to the aerodynamic model. s and the generalized aerodynamic force h as input to the structural model 空气动力 And implement.

[0105] If the wind turbine uses remote wind sensors (e.g., lidar, sodar, radar) to measure the upstream wind speed component or the downstream wind speed component, the aerodynamic model may include an extended representation of the wind to collect inflow and wake velocities in front of the turbine and / or behind the turbine (not just at the rotor face). To obtain a consistent aerodynamic model for estimation applications, some embodiments define a three-dimensional grid that includes the rotor face and the remote wind measurement locations and extend the aerodynamic model in (12-14) to include as the aerodynamic state x a The wind speed at these points is part of and includes the propagation of the wind speed along the direction of the mean wind speed as part of the dynamic equation (12).

[0106] During wind turbine operation, aerodynamic, gravitational, and inertial forces on the blades cause flexible twisting of the blades which tends to reduce aerodynamic performance and structural loads. The effects of blade twist or flexible distortion are typically modeled by including the blade torsional degree of freedom in the dynamic equations, or by employing a quasi-static approach and calculating the steady-state deformation of the blade section that would be in equilibrium with the blade forces at each instant in time. Since the addition of the torsional degree of freedom to the dynamic equations makes the model computationally expensive, and due to the fact that the torsional degree of freedom typically has a natural frequency that falls outside the control bandwidth, it is generally convenient to employ a quasi-static modeling approach. In this case, the calculation of the quasi-static blade distortion based on the blade forces and blade stiffness can be added to equations (12-14) by including z as an algebraic variable. a The blade is twisted and deformed in part, and includes the quasi-static torsional equilibrium equation as part of equation (14).

[0107] The aerodynamic force also depends on the wind parameter w p In general, these parameters are not accurately measured by standard generation sensors used in wind turbine systems. Therefore, for online real-time (i.e., during operational control of the wind turbine) estimation and control applications, these parameters are estimated by augmenting the interconnected wind turbine structural model (9-10) and aerodynamic model (12-14) with a wind disturbance model of the form:

[0108]

[0109] w p =g w (x w ) (16)

[0110] Among them, x w is the wind state vector, w w is a white noise random process, and f w and g w is a nonlinear function. In the practical system, the wind parameter w p The wind parameters w may include mean wind speed, misalignment, shear, turning angle, and harmonic variation around azimuth. p The effective wind speed for each blade may be included, where the wind state x w represents the harmonic variation of wind around the azimuth angle (ie, 1P, 2P, 3P, etc., where XP refers to a multiple of the rotor rotation frequency).

[0111] Model derivatives

[0112] The application of advanced control and estimation algorithms typically requires mathematical models of the controlled system or partial derivatives of the equations with respect to their inputs in terms of Jacobian determinants. The structural equations, aerodynamic equations, and the interconnection between the structural equations and aerodynamic equations of a wind turbine are differential-algebraic equations of the form:

[0113]

[0114] 0=h(x,z,u) (18)

[0115] y=g(x,z,u) (19)

[0116] Where x is the state vector;

[0117] z is a vector of algebraic variables; and

[0118] y is the output vector. * ,z * ,u * ) has the following form:

[0119]

[0120]

[0121]

[0122] where the Jacobian matrix is the gradient of the functions f, g, and h along the directions of the vectors x, z, and u. To simplify the notation, by assuming that there are no algebraic constraints, the following definition can be used:

[0123]

[0124]

[0125] F=f x (x * ,z * ,u * )-Ax * -Bu * (twenty two)

[0126]

[0127]

[0128] G=g x (x * ,z * ,u * )-Cx * -Du* (25)

[0129] In order to achieve the * ,z * ,u * ) is a simplified form of the approximate local model:

[0130]

[0131] y=Cx+Du+G (27)

[0132] The control model data A, B, C, D, F, G vary with the operating point and need to be evaluated regularly during control operation.

[0133] When the system under consideration is highly nonlinear and the operating space spans a wide range in the domain of variables x, z, u, the Jacobian needs to be evaluated at every time step, and the ability to compute fast Jacobians can become critical for implementing advanced control applications. This is the case for wind power systems, where transient operation can be subject to strong wind variations, characterized by flexible structures and nonlinear kinematics, thus requiring regular model data recalculation.

[0134] The online implementation of the estimation and control algorithms is sometimes expressed as a function of the discrete-time representation of the linearized model equations. Therefore, instead of (Equations 26-27), the discrete-time equivalent of the discrete-time version using the Jacobian determinant can be employed. For the purposes of the following discussion, the following formulas do not distinguish between continuous-time linearization or discrete-time linearization.

[0135] In applications based on data from a single time step, the linearization module of the control model 206 produces a set of Jacobian matrices A, B, C, D and associated linearization errors F and G. For applications such as MHE and MPC, the linearization module produces a sequence of Jacobian matrices corresponding to each step within a pre-specified time range.

[0136] The model data represented by A, B, C, D, F, G varies with the operating point. Implementable systems and methods may evaluate these model elements periodically (e.g., at predetermined, periodic, or specified time intervals) during control operation. When the system under consideration is highly nonlinear and the operating space spans a wide range in the domain of variables x, z, u, embodiments may evaluate the Jacobian at approximately every time step, and the ability to compute fast Jacobians may become critical for implementing advanced control applications. This is the case for wind power systems, where transient operation, subject to strong wind variations and characterized by flexible structures and nonlinear kinematics, requires regular model data recalculation.

[0137] Calculating model derivatives

[0138] Derivatives obtained by numerical differentiation

[0139] Conventional methods for obtaining derivatives of functions include numerical differentiation and exact differentiation. Numerical differentiation techniques approximate derivatives by means of numerical finite difference schemes. For example, for the model (Equation 10), numerical derivatives are typically obtained by central difference operations of the form:

[0140]

[0141]

[0142]

[0143] where δ is the perturbation size; and e j is the j-th canonical vector.

[0144] Due to the computational effort and numerical inaccuracies caused by round-off errors, ill-conditioning, and numerical instabilities, this conventional approach is rarely effective online for real-time control / optimization applications. Other approaches are used instead, as described in the following paragraphs.

[0145] Exact derivatives obtained by analytical differentiation

[0146] Alternative conventional methods include exact or analytical differentiation of derivatives using manual differentiation, algorithmic differentiation, or symbolic differentiation. Manually evaluating derivatives is generally tedious and error-prone, and is only suitable when the underlying system model is very simple.

[0147] Algorithmic differentiation is the process by which a mathematical program for evaluating a function is manipulated to create another program for computing the derivatives of the original function. Conventional tools exist that can automatically (without user intervention) perform this evaluation process by converting the original algorithm into a graphical representation, symbolically manipulating the graph of the function to generate graphs for the derivatives by systematically applying the chain rule for differentiation, and ultimately obtaining a new algorithm for evaluating the corresponding derivatives.

[0148] Symbolic differentiation relies on a computer algebra system that operates on an algebraic expression to generate a set of related algebraic expressions to evaluate the derivatives of the original expression. Tools for implementing symbolic differentiation typically include batch techniques for simplifying algebraic expressions and converting the resulting expressions into code. While symbolic differentiation is effective for handling closed-form expressions, it cannot easily handle functions that contain algorithmic control flow features (e.g., "for" or "while" loops and "if-then" branches).

[0149] Both symbolic and algorithmic differentiation produce "exact" derivatives that, unlike numerical differentiation methods, are not subject to roundoff and truncation issues. The choice of method to obtain analytical derivatives depends on the original equation / program being differentiated. Symbolic tools are often useful for simplifying algebraic expressions, while algorithmic tools can generate efficient code and effectively handle logic branches and control loops.

[0150] In some wind power generation platforms, the computational processing power of the processor is insufficient to implement a real-time implementation of the exact derivative. According to an embodiment, on these platforms, the approximate derivative can be implemented as a lookup table. In this method, the derivative is evaluated in a grid in the input space and the derivative is approximated by an interpolation method. The operation of the derivative is performed offline and the online calculation performs the interpolation. According to an embodiment, in the case where the exact derivative method is overly complex, this separation between offline and online calculations allows the use of numerical differentiation in the offline evaluation. Several numerical differentiation schemes including forward differentiation, backward differentiation and central differentiation can be used to approximate the system derivative.

[0151] Approximate derivatives obtained through the proxy model

[0152] In cases where analytical differentiation produces a highly complex function or expression that is unsuitable for implementation in the target control platform, it is convenient to approximate the original function using a proxy function for which analytical differentiation is simpler. This approximation can be performed by using a neural network (NN) to obtain a proxy model of the original function and using this model to easily construct analytical derivatives. In some embodiments, the neural network can use a sigmoid activation function.

[0153] For example, if the function f(x) that maps a vector of size n to a set of real numbers is approximated by the following feed-forward neural network with a single hidden layer:

[0154]

[0155] in (where, for all “k” from 1 to m, W [1] is a matrix with n rows and m columns, w [2] and b [1] is a vector of size m, the derivative of “f(x)” can be approximated by the following expression:

[0156]

[0157] where, for all "k" from 1 to m, Moreover, σ′(z k )=σ(z k)(1-σ(z k )). The evaluation of the approximate Jacobian of f involves only three vector-matrix multiplication operations and the evaluation of the sigmoid function at x. If the function f is complex and the number of neurons in the hidden layer is moderate, the computational cost of this type of approximation can be very convenient.

[0158] Platform computing environment for wind turbines

[0159] The implementation of the system and method may vary depending on the computing power available from the platform's control processor environment and the accuracy required for the application. Each embodiment may provide predictive data for use by the control processor in evaluating the commands required for real-time operational control of the wind turbine under current operating parameters and conditions.

[0160] The implemented model decomposes the sensor data from the wind turbine to obtain the aerodynamic state x a , aerodynamic state x a Includes axial induced velocity and tangential induced velocity at each blade section and each blade. Structural state x s Including the blade flap direction displacement and edge direction displacement representation. Output force vector h a Includes the forces and moments applied from each blade to the hub.

[0161] Wind parameter w p The wind parameter w is a representation of the wind field near the wind turbine rotor. p Includes wind components that characterize the temporal and spatial variability of wind, such as mean wind speed, wind shear, wind direction, and vortex structure.

[0162] Derivative calculations of the coupled aeroelastic model can be performed by operating on the coupled forms (17)-(19), or by evaluating the derivatives of the structural model and the aerodynamic model separately and then integrating the coupled derivatives applying standard manipulations of linear systems.

[0163] Online calculation of derivatives of the structural wind turbine model as represented, for example, by equations (9)-(10) is most conveniently performed using exact derivatives (using algorithmic or symbolic differentiation).The following sections focus primarily on the computation of derivatives of the aerodynamic model.

[0164] Low CPU power platforms

[0165] For extremely limited CPU power, the aerodynamic representation of the model can be implemented as a quasi-static implementation via table interpolation, where the aerodynamic representation of the model for a selected reference structural state is According to an embodiment, the formula for this low CPU power is obtained by finding a steady-state solution, that is, for the selected reference structural state Can be targeted at aerodynamic conditions x a and the algebraic variable z a The resulting force on each blade can be tabulated as a function of a minimal set of parameters to obtain a representation of the form:

[0166]

[0167] in, is the speed ratio of the i-th blade tip to the wind; V i is the effective wind speed for the i-th blade; θ i is the pitch angle of the i-th blade; and, is the azimuth angle of the i-th blade.

[0168] According to an embodiment, an offline process may be performed to estimate the aerodynamic force h at each grid point. a and its (numerical) derivatives (which are For example, (represented by i, j, k), the calculation procedures for both are:

[0169]

[0170]

[0171]

[0172]

[0173] The data obtained by (29A-D) can be used to construct a representation for a list of derivatives of the form:

[0174]

[0175]

[0176]

[0177]

[0178] If the blade twist effect is to be considered, the blade tip to wind speed ratio, azimuth angle, and pitch angle are not sufficient to uniquely characterize the blade twist. In this case, the tables (28, 29A-D) need to be expanded to more accurately characterize the aerodynamic operating point. For this reason, the aerodynamic table can be expanded to include the dynamic pressure parameter for blade i: Or any other parameter that can be used to uniquely characterize the aerodynamic operating point, the aerodynamic force and hence the quasi-static blade twist.

[0179] High CPU power

[0180] For platforms with high CPU capabilities that can provide a higher accuracy control model, the formulation of the BEM model for the wake dynamics can be implemented in different ways. According to an embodiment, the aerodynamic state x a This may include the wind axial induced velocity and the wind tangential induced velocity. Algebraic variables may include, for example, the tip loss factor. Wind parameters for the implementation method may include mean wind speed, vertical and horizontal shear coefficients, vertical and horizontal misalignment angles, and harmonic wind variation depending on the rotor azimuth angle. The derivatives of these control models are obtained by applying automatic differentiation or symbolic differentiation to equations (12, 13, 14, 15, 16).

[0181] Medium CPU power

[0182] For moderate CPU power, different strategies can be used to implement different simplifications. For example, according to embodiments, strategies may include using quasi-steady induced speeds, reducing the number of segments used to represent local aerodynamic forces, reducing the number of wind parameters, and combining the calculation routine with a lookup table. The aerodynamic model can be approximated using proxy models for high-fidelity models and / or static approximations. The approximate real-time aerodynamic model can be used for the aerodynamic force vector component h 空气动力 Each input is obtained by using a neural network.

[0183] This process of mapping the neural network to each force vector component can be performed by defining the number of neurons in the hidden layer(s), evaluating the current real-time model to calculate the aerodynamic forces for a given coverage of the input space, and then training the neural network by defining the optimal values of the weights via a back-propagation algorithm.

[0184] Figure 2 A process 200 for estimating control signals for dynamic operation of a wind turbine is depicted, according to an embodiment. An estimation module 202 receives data from an estimation model 204. A multivariable control module 210 receives data from a control model 206, which may be another instance of a first model. In some embodiments, the same instance of a model may be used for both estimation and control modeling.

[0185] The model performs online operations categorized as operating point operations and linearization operations. Estimation model 204 receives monitored sensor data from a sensor suite monitoring the wind turbine. This sensor data may include sensor dynamics data 220, wind turbine dynamics data 222, and actuator dynamics data 224. The sensor data may be updated at predetermined intervals so that the generated control commands are responsive to changes in the monitored sensor data. Previous estimation results stored in the memory block may be used as feedback for entering the operating point operations.

[0186] By applying monitored data to the model, an operating point can be defined by the prevailing wind speed and other wind and environmental parameters, which may include misalignment and shear. The operating point can also be defined by the structural conditions of the wind turbine, including a reference blade pitch for each blade, reference structural deformations in the flexible components, and azimuthal rotation angles and rates. This operating point serves as a starting point for linearization. The linearization module evaluates Jacobians for both structural and aerodynamic components and then combines these Jacobians into a unified set of Jacobians for the entire model.

[0187] The estimation module receives the linearized model Jacobian from the estimation model and the monitored sensor data provided to the estimation model. The estimation module generates turbine and aerodynamic estimated state and performance signals. These signals are provided to the memory block for access by the estimation model 204. The turbine and aerodynamic estimated state and performance signals are also provided by the estimation module to the control model 206.

[0188] The control model 206 may obtain estimates of the turbine operating point and the aerodynamic operating point by applying estimation applications and / or control applications to the estimated state signals and performance signals generated by the estimation module in real time (i.e., approximately when these operating points affect the operation of the wind turbine).

[0189] Estimation applications and control applications can be categorized as using a single time step or multiple time steps. The former group may include estimation applications such as Kalman filtering and extended Kalman filtering (EKF); and control applications such as linear quadratic regulators (LQR), linear quadratic Gaussian (LQG), state-dependent Riccati equations and controllers obtained as solutions to linear matrix inequalities (LMI) problems. The latter group may include receding horizon estimation (MHE) applications and model predictive control (MPC) applications.

[0190] Real-time estimates of the turbine operating point and aerodynamic operating point from control module 206 are provided to multivariable control module 210. The multivariable control module generates control conditions based on the turbine operating point and aerodynamic operating point, as well as real-time operating parameters as monitored by sensors. The control conditions are provided to command generation module 212, which generates actuator commands that direct the control processor to control wind turbine operation. Wind turbine operating parameters and conditions are monitored and provided back to estimation module 202 and estimation model 204 in a loop. Monitoring can be performed at predetermined intervals or approximately continuously (within the response rate and sampling rate of the sensor suite).

[0191] Figure 3 A system 300 for estimating control signals for dynamic operation of a wind turbine according to an embodiment is depicted. System 300 may include a control processor 312 in communication with a data store 320. The control processor may communicate with the data store directly or indirectly via an electronic communication network (not shown). A processor unit 314 may execute executable instructions 322 that cause the processor to perform a process for estimating control signals. A memory unit 316 may provide local cache memory for the control processor. Data store 320 may include estimation model 204, control model 206, sensor dynamic data 220, wind turbine dynamic data 222, and actuator dynamic data 224. A sensor suite 330 may include sensors positioned relative to wind turbine 340 to acquire corresponding dynamic data.

[0192] According to some embodiments, a computer program application stored in a non-volatile memory or computer-readable medium (e.g., register memory, processor cache, RAM, ROM, hard drive, flash memory, CD ROM, magnetic media, etc.) may include code or executable program instructions that, when executed, may direct and / or cause a controller or processor to perform the methods discussed herein, such as the method of estimating control signals for dynamic operation of a wind turbine, as disclosed above.

[0193] The computer readable medium may be a non-transitory computer readable medium, which includes all forms and types of memory and all computer readable media (except transitory propagating signals). In one embodiment, the non-volatile memory or computer readable medium may be an external memory.

[0194] Although specific hardware and methods have been described herein, it is noted that any number of other configurations may be provided in accordance with embodiments of the present invention. Thus, while the essential novel features of the present invention have been shown, described, and indicated, it will be understood that various omissions, substitutions, and changes in the form and details of the illustrated embodiments and their operation may be made by those skilled in the art without departing from the spirit and scope of the present invention. The substitution of elements from one embodiment to another is also fully anticipated and contemplated. The present invention is limited only with respect to the appended claims and equivalents thereof.

Claims

1. A wind turbine system comprising: a sensor suite comprising a plurality of sensors positioned to monitor wind turbine operating parameters and environmental conditions, the sensor suite in communication with a data store and configured to provide sensor dynamics data to the data store; a control processor in communication with the data storage section, the control processor comprising a processor unit; a control model configured to perform a linearization evaluation at a turbine operating point, the linearization evaluation applying an analytical differentiation technique to obtain at least one of a structural component dynamic behavior, a fluid component dynamic behavior, and a combined structural and fluid component dynamic behavior of wind turbine operation; the control model including a coupling between the structural component dynamic behavior and the fluid component dynamic behavior, the coupling being implemented by introducing generalized fluid forces into a structural model of the wind turbine and including in the fluid component dynamic behavior a dependency of the generalized fluid forces on a structural state of the wind turbine, the structural model including structural features that are accurate with respect to at least one of a first bending natural mode of a tower, a first bending natural mode of one or more blades, and a rotation of a drive train of the wind turbine; as well as A module is configured to perform an operation using at least one of the linearized estimate of the structural component dynamic behavior, the fluid component dynamic behavior, and the combined structural and fluid component dynamic behavior.

2. The system according to claim 1, wherein: include: The module is an estimation module configured to generate one or more estimates selected from the group consisting of a turbine state signal estimate, a turbine performance signal estimate, a fluid state signal estimate, and a fluid performance signal estimate; and, Each of the one or more estimates is based on one or more elements of sensor dynamic data, the structural member dynamic behavior, and the fluid member dynamic behavior.

3. The system according to claim 2, characterized in that The estimation module is configured to generate the one or more estimation values by implementing an extended Kalman filter.

4. The system according to claim 1, wherein: At least a portion of the fluid component dynamic behavior included in the lookup table comprises a representation of one or more rotor plane fluid properties and derivatives thereof, across which at least one of the one or more blades traverses.

5. The system according to claim 4, characterized in that The lookup table includes: a respective element comprising, for a respective one of the one or more blades, a data set, each data set comprising a dependency on at least one of one or more of a blade tip to wind speed ratio, a pitch angle, an azimuth angle, and a representative dynamic pressure parameter; and, The data set is applicable to determine at least one of blade force and blade twist.

6. The system according to claim 2, wherein: The interconnected structural component dynamic behavior model and the fluid component dynamic behavior model are augmented with a wind disturbance dynamic behavior model to estimate wind parameters, including one or more of average wind speed, misalignment, shear, turning angle, harmonic variation, and effective wind speed of each blade expressed as harmonic variation of wind.

7. The system according to claim 1, wherein: The structural model includes blade twisting effects caused by one or more of fluid forces, inertial forces, and mass blade forces.

8. The system according to claim 1, wherein: include: The module is a multivariable control module configured to determine actuator commands based on one or more control states; and, The actuator commands include wind turbine commands to maintain operation of the wind turbine to predetermined settings in real time.

9. The system according to claim 8, characterized in that The multivariable control module includes a model predictive controller.

10. The system according to claim 1, wherein: The control model is configured to perform the linearization evaluation by implementing one of algorithmic differentiation or symbolic differentiation.

11. The system according to claim 1, wherein: The dynamic behavior of the fluid component includes: a contribution rate of local forces applied to the blade segment based on the local aerodynamic properties of the airfoil augmented to apply corrections for three-dimensional effects; and Dynamic wake effects from wind turbine operation.

12. The system according to claim 1, wherein: The fluid member dynamic behavior includes at least one of a wind propagation model upstream of the rotor face and a wind propagation model downstream of the rotor face and data from a remote wind measurement system sensor.

13. The system according to claim 1, wherein: The control model is configured to obtain an analytical representation of the dynamic behavior of the structural component through multi-agent dynamic modeling technology.

14. The system according to claim 1, wherein: The control model is configured to capture the fluid component dynamic behavior by implementing a blade element momentum model including a representation of one or more fluid properties of a rotor plane across which at least one of the one or more blades traverses.

15. The system according to claim 1, wherein: The control model is configured as follows: approximating a corresponding function for at least one of the structural component dynamic behavior and the fluid component dynamic behavior using a proxy model; and, The corresponding analytical derivative is constructed from the corresponding surrogate function.

16. A method of calculating at least one of an estimated operating parameter and a control command for a wind turbine, the method comprising: The sensor suite monitors wind turbine operating parameters and environmental conditions to acquire sensor dynamic data to a data storage unit; performing a linearization evaluation at a turbine operating point, the linearization evaluation applying an analytical differentiation technique to obtain at least one of structural component dynamic behavior, fluid component dynamic behavior, and combined structural and fluid component dynamic behavior of wind turbine operation; implementing coupling between the structural component dynamic behavior and the fluid component dynamic behavior by introducing generalized fluid forces into a structural model of the wind turbine and including a dependency of the generalized fluid forces on a structural state of the wind turbine in the fluid component dynamic behavior, the structural model including structural features that are accurate with respect to at least one of a first bending natural mode of a tower, a first bending natural mode of one or more blades, and a rotation of a drive train of the wind turbine; as well as At least one of an estimation module and a multivariable control module performs an operation using at least one of the linearized estimates of the structural component dynamic behavior, the fluid component dynamic behavior, and the combined structural and fluid component dynamic behavior.

17. The method according to claim 16, characterized in that include: the estimation module generating one or more estimates selected from the group consisting of a turbine state signal estimate, a turbine performance signal estimate, a fluid state signal estimate, and a fluid performance signal estimate; and Each of the one or more estimates is based on one or more elements of sensor dynamic data, the structural member dynamic behavior, and the fluid member dynamic behavior.

18. The method according to claim 16, characterized in that include: The multivariable control module determines an actuator command based on one or more control states; and The actuator commands include wind turbine commands to maintain operation of the wind turbine to predetermined settings in real time.

19. The method according to claim 16, wherein The method includes augmenting the interconnected structural component dynamic behavior model and the fluid component dynamic behavior model with a wind disturbance dynamic behavior model to estimate wind parameters, wherein the wind parameters include one or more of average wind speed, misalignment, shear, turning angle, harmonic variation, and effective wind speed of each blade expressed as harmonic variation of wind.

20. The method according to claim 16, wherein The structural model includes components representing blade twisting effects caused by one or more of fluid forces, inertial forces, and mass blade forces.

21. The method according to claim 16, wherein The linearization evaluation is performed by performing one of algorithmic differentiation or symbolic differentiation.

22. The method according to claim 16, wherein This includes modeling the dynamic behavior of the fluid components to take into account: a contribution rate of local forces applied to the blade segment based on the local aerodynamic properties of the airfoil augmented to apply corrections for three-dimensional effects; and Dynamic wake effects from wind turbine operation.

23. The method according to claim 16, wherein The fluid member dynamic behavior includes at least one of a wind propagation model upstream of the rotor face and a wind propagation model downstream of the rotor face and data from a remote wind measurement system sensor.

24. The method according to claim 16, wherein This includes obtaining an analytical representation of the dynamic behavior of the structural component by implementing multi-agent dynamic modeling technology.

25. The method according to claim 16, wherein Implementing a blade element momentum model to capture at least a portion of the fluid component dynamic behavior includes implementing a blade element momentum model comprising a representation of one or more fluid properties of a rotor plane across which at least one of the one or more blades traverses.

26. The method according to claim 16, wherein include: approximating a respective proxy function for at least one of the structural component dynamic behavior and the fluid component dynamic behavior using a neural network; and The corresponding analytical derivative is constructed from the corresponding surrogate function.

27. The method according to claim 16, wherein At least a portion of the fluid component dynamic behavior is included in a lookup table containing representations of fluid properties and their derivatives for one or more rotor planes across which at least one of the one or more blades traverses.

Images