Method and system for operating a wind turbine considering life consumption

Abstract

The present disclosure relates to a method (100) of operating a wind turbine (10) comprising a plurality of actuators (164). The method (100) comprises receiving operational data (166) and estimating an initial state based on the received operation data (166). The method (100) also comprises predicting, using a model (171 ), potential operational states of the wind turbine (10). The method (100) also comprises deriving fatigue indicators indicative of an anticipated life consumption of a wind turbine component. Furthermore, the method (100) comprises optimizing a cost function to determine an optimum trajectory for commands of the actuators (164) and using the first commands of the trajectory to control the actuators (164). The optimization is subject to constraints. Furthermore, the cost function and / or the constraints are adjusted based on the derived fatigue indicators. The disclosure also relates to a control unit (163) for a wind turbine (10) configured to implement such a method (100).

Description

GENERAL ELECTRIC RE OVABLES ESPANA S.L. MARCH 25, 2025 701040-WO-1 P5593PC00METHOD AND SYSTEM FOR OPERATING A WIND TURBINE CONSIDERING LIFE CONSUMPTION

[0001] The present invention relates to wind turbines, and more particularly, to methods and systems for controlling a wind turbine with predictive control methods accounting for life consumption of wind turbine components.BACKGROUND

[0002] Modern wind turbines are commonly used to supply electricity into the electrical grid. Wind turbines of this kind generally comprise a tower and a rotor arranged on the tower. The rotor, which typically comprises a hub and a plurality of blades, is set into rotation under the influence of the wind on the blades. This rotation generates a moment that is normally transmitted through a rotor shaft to a generator, either directly (“directly driven” or “gearless”) or through the use of a gearbox. This way, the generator produces electricity which can be supplied to the electrical grid.

[0003] Variable speed wind turbines are commonly controlled by means of wind turbine actuators which allow, e.g. varying the generator torque or the pitch angle of the blades. In general, one or more control systems are arranged to determine actuator signals. In general, wind turbines are controlled so as to capture maximum power from the wind so that maximum output power, for the prevailing wind conditions, is generated.

[0004] The structure and components of a wind turbine undergo multiple loads of different magnitudes and frequencies during operation. In order to ensure structural integrity and proper operation of the wind turbine, such loads need to be maintained within certain limits which define the operational limits of the wind turbine, e.g. maximum rotational speed or maximum electromagnetic torque. Consequently, apart from optimizing power production, wind turbine control systems are also designed to control wind turbine actuators such that the wind turbine remains within operational limits.

[0005] Furthermore, oscillations and vibrations of different wind turbine components result in fatigue loads, which translate into life consumption of such wind turbine components. Indeed, not only ultimate loads, but also fatigue loads are to be considered while designing the windturbine control systems to prevent premature wear or damage of certain components. Such premature damage or wear can lead to abnormally increased life consumption of the wind turbine components, i.e. life consumption exceeding initially planned lifetime of the wind turbine components, thus requiring early replacement of the same or expensive repair operations.

[0006] Different control approaches have been used for the control of wind turbines. These include classical feedback controllers of the PID type, which have been extensively used. Nevertheless, wind turbines are increasingly complex systems that operate in a varying environment, so advanced control strategies enabling control of systems with multiple control variables are desired. Among advanced modern control strategies, Model Predictive Control (MPC) approaches are particularly well suited for wind turbines because they can handle multivariable control problems while accounting for process constraints. Thus, MPC controllers can manage input, e.g. blade pitch angle or generator torque, constraints and output or operability constraints, e.g., rotational speed, generated electrical power, while taking structural loads, e.g. blade root moments, tower base moments, into account so as to maintain stability of the system while optimizing output power. Such MPC controllers are also particularly suited for the consideration of fatigue loads, as the loads of multiple wind turbine components can be considered in a highly versatile manner by the control system.

[0007] In order to determine the control actions, MPC uses a mathematical model to predict the future behavior, i.e. the output variables, of the wind turbine over a certain time horizon after considering the current or initial state of the wind turbine. The predicted behavior of the wind turbine is not only dependent on the current state of the wind turbine, but also on the evolution of the controllable variables and on the environmental conditions over the prediction period. The MPC controller provides optimal control actions for the actuators of the wind turbine as an output, which ensure that the wind turbine tracks the desired references. To this end, the MPC controller typically comprises an optimizer that solves an online optimization problem at each control step. The optimization problem comprises the minimization of a cost function, / , over an optimization horizon subject to some constraints. The cost function penalizes deviations of the output (also referred to as dependent or controlled) variables from reference values, e.g. output power or rotational speed references.

[0008] The competitiveness of a wind turbine design is dependent on the ability of the controller to handle loads on turbine components and, more specifically, on the ability to handle fatigue loads at specific wind turbine components, whose life consumption may be particularly relevant for the overall performance of the wind turbine.

[0009] MPC controllers for wind turbines with fatigue management have been already proposed, but currently proposed solutions exhibit some drawbacks or limitations. The existingapproaches to manage fatigue with MPC controllers tend to be numerically inefficient. In particular, the optimization problem with fatigue penalties is too expensive from a computational point of view so that online solving becomes challenging. This is especially relevant as the complexity of the system increases, i.e. as more components, subject to fatigue loads, are present in the wind turbine. For this reason, known MPC controllers tend to be oversimplified, i.e. simplified such that computations can be managed, but simplified to such an extent that they do not provide a proper representation of the fatigue behavior of the wind turbine components, or too conservative. Furthermore, existing MPC-based approaches comprising fatigue management are reactive in nature. That is, they are based on the cumulative fatigue damage, i.e. on the lifetime that has been already consumed. Based on the fatigue that has been already consumed by the components, the different parameters of the MPC controller, e.g. the terms of the cost function, may be adjusted. However, such cumulative fatigue may be difficult to maintain and characterize, especially for applications requiring dynamic real time control of complex systems like modern wind turbines.

[0010] The present disclosure seeks to provide improved methods for controlling a wind turbine with a predictive controller while avoiding, or at least reducing, some of the drawbacks of the aforementioned approaches.SUMMARY

[0011] In an aspect of the present disclosure, a method for operating a wind turbine is provided. The wind turbine comprises a plurality of wind turbine actuators. The method comprises receiving operational data and estimating an initial state based on the received operational data. Furthermore, the method comprises using a model to predict, over a prediction period, potential operational states of the wind turbine based on the estimated initial state and depending on operation of the wind turbine actuators. Moreover, one or more fatigue indicators, indicative of an anticipated life consumption of a wind turbine component, are derived based on the predicted potential operational states. The method also comprises optimizing a cost function over an optimization period to determine an optimum trajectory for commands of the wind turbine actuators. The optimization is subjected to one or more constraints. Then, the method comprises using the first wind turbine actuator commands of the determined optimum trajectory to control the wind turbine actuators. The cost function and / or the constraints are adjusted based on the derived fatigue indicators.

[0012] According to this aspect of the disclosure, a method is provided that allows an improved control of a wind turbine while taking fatigue loads into account. The use of fatigue indicators or fatigue surrogates allows important computational savings, thus facilitating realtime operation of the wind turbine controller. Furthermore, a plurality of fatigue indicators can be employed to assess the behavior of different wind turbine components. Consequently, a more accurate representation of the behavior of the wind turbine is achieved.

[0013] Furthermore, unlike in previously known control methods, the employed fatigue indicators are not based on a calculated cumulative fatigue but on an anticipated life consumption. In this manner, a proactive approach is accomplished, which prevents fatigue accumulation when the wind properties promote high oscillatory behaviors as opposed to conventional control methods, which let the fatigue accumulate and then apply corrective actions. Indeed, by taking a predicted life consumption into account, control actions can be implemented in the different wind turbine actuators, which result in a more optimized handling of the life consumption of the wind turbine components. Such improved control results in lower operation and maintained costs and less unplanned outages.

[0014] The model used to predict the wind turbine behavior, i.e. the potential operational states of the wind turbine, may account for the physical behavior of the wind turbine and the wind.

[0015] Overall, the enhanced control permits a more accurate design of the wind turbine components, which results in a more optimal tradeoff between fatigue loads, i.e. life consumption, and energy production.

[0016] In another aspect of the disclosure, a control unit for a wind turbine is provided. The wind turbine has a plurality of wind turbine actuators. The control unit is configured for receiving operational data and for estimating an initial state based on the received operational data. The control unit is also configured for predicting, using a model, potential operational states of the wind turbine based on the estimated initial state and depending on operation of the wind turbine actuators over a prediction period. Furthermore, the control unit is configured for deriving one or more fatigue indicators indicative of an anticipated life consumption of a wind turbine component based on the predicted potential operational states. Then, the control unit is configured for optimizing a cost function over an optimization period to determine an optimum trajectory for commands of the wind turbine actuators. The optimization is subject to one or more constraints. The first wind turbine actuator commands of the determined optimum trajectory are used to actually control the wind turbine actuators. Moreover, the control unit is further configured for adjusting the cost function and / or the constraints based on the derived fatigue indicators.

[0017] According to this another aspect, a control unit can be provided, which allows implementation of a method for controlling a wind turbine like the one described with reference to the previous aspect of the disclosure. Accordingly, the control unit is such that improved andproactive handling of fatigue loads is obtained. This improved management of fatigue behavior enables increased annual energy production (AEP). The improved handling of fatigue loads facilitates a more consistent maintenance work, especially when wind turbines are installed in wind farms comprising a large number of wind turbines. Hence, the use of a control unit according to this aspect of the disclosure, results in a more even distribution of the life consumption among different wind turbines in a wind farm.BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Non-limiting examples of the present disclosure will be described in the following, with reference to the drawings, in which:Figure 1 illustrates a perspective view of one example of a wind turbine;Figure 2 illustrates a simplified, internal view of one example of the nacelle of the wind turbine of the Figure 1 ;Figure 3 shows a flowchart of an example of a method for controlling a wind turbine according to the present disclosure;Figure 4 shows a high level view of an example of a control system for a wind turbine;Figure 5 schematically illustrates operation of a controller for controlling a wind turbine according to an example of the disclosure;Figure 6 schematically illustrates predicted trajectories over a prediction period of time;Figure 7 schematically illustrates a wind turbine model linearization along a state trajectory according to an example;Figures 8A and 8B show a flowchart of another example of a method for controlling a wind turbine according to the present disclosure;Figure 9 schematically illustrates operation of a model predictive controller for controlling a wind turbine according to another example of the disclosure;Figure 10 schematically illustrates a process to assess approximate trajectory invariance according to an example; andFigure 11 schematically illustrates zone tracking penalty according to an example.DETAILED DESCRIPTION OF EXAMPLES

[0019] Reference now will be made in detail to embodiments, one or more examples ofwhich are illustrated in the drawings. Each example is provided byway of explanation only, not as a limitation. In fact, it will be apparent to those skilled in the art that various modifications and variations can be made in the present disclosure. For instance, features illustrated or described as part of one example can be used with another example to yield a still further example. Thus, it is intended that the present disclosure covers such modifications and variations as come within the scope of the appended claims and their equivalents.

[0020] Figure 1 is a perspective view of an example of a wind turbine 10. In the example, the wind turbine 10 is a horizontal-axis wind turbine. Alternatively, the wind turbine 10 may be a vertical-axis wind turbine. In the example, the wind turbine 10 includes a tower 15 that extends from a support system 14 on a ground 12, a nacelle 16 mounted on tower 15, and a rotor 18 that is coupled to nacelle 16. The rotor 18 includes a hub 20 and at least one rotor blade 22 coupled to and extending outward from the hub 20. In the example, the rotor 18 has three rotor blades 22. In an alternative embodiment, the rotor 18 includes more or less than three rotor blades 22. The tower 15 may be fabricated from tubular steel to define a cavity (not shown in Figure 1) between a support system 14 and the nacelle 16. In an alternative embodiment, the tower 15 is any suitable type of a tower having any suitable height. According to an alternative, the tower can be a hybrid tower comprising a portion made of concrete and a tubular steel portion. Also, the tower can be a partial or full lattice tower.

[0021] The rotor blades 22 are spaced about the hub 20 to facilitate rotating the rotor 18 to enable kinetic energy to be transferred from the wind into usable mechanical energy, and subsequently, electrical energy. The rotor blades 22 are mated to the hub 20 by coupling a blade root region 24 to the hub 20 at a plurality of load transfer regions 26. The load transfer regions 26 may have a hub load transfer region and a blade load transfer region (both not shown in Figure 1). Loads induced to the rotor blades 22 are transferred to the hub 20 via the load transfer regions 26.

[0022] In examples, the rotor blades 22 may have a length ranging from about 15 meters (m) to about 90 m or more. Rotor blades 22 may have any suitable length that enables the wind turbine 10 to function as described herein. As wind strikes the rotor blades 22 from a wind direction 28, the rotor 18 is rotated about a rotor axis 30. As the rotor blades 22 are rotated and subjected to centrifugal forces, the rotor blades 22 are also subjected to various forces and moments. As such, the rotor blades 22 may deflect and / or rotate from a neutral, or nondeflected, position to a deflected position.

[0023] Moreover, a pitch angle of the rotor blades 22, e.g. an angle that determines an orientation of the rotor blades 22 with respect to the local wind direction, may be changed by a pitch system 32 to control the load and power generated by the wind turbine 10 by adjustingan angular position of at least one rotor blade 22 relative to wind vectors. Pitch axes 34 of rotor blades 22 are shown. During operation of the wind turbine 10, the pitch system 32 may particularly change a pitch angle of the rotor blades 22 such that the angle of attack of (portions of) the rotor blades are reduced, which facilitates reducing a rotational speed.

[0024] In the example, a blade pitch of each rotor blade 22 is controlled individually by a wind turbine controller 36 or by a pitch control system 80 (see Figure 2). Alternatively, the blade pitch for all rotor blades 22 may be controlled simultaneously by said control systems.

[0025] Further, in the example, as the wind direction 28 changes, a yaw direction of the nacelle 16 may be rotated about a yaw axis 38 to position the rotor blades 22 with respect to wind direction 28.

[0026] In the example, the wind turbine controller 36 is shown as being centralized within the nacelle 16, however, the wind turbine controller 36 may be a distributed control system throughout the wind turbine 10, on the support system 14, within a wind farm, and / or at a remote-control center. The wind turbine controller 36 may include one or more processors 40 configured to perform one or more of the steps of the methods described herein. Further, many of the other components described herein include one or more processors. The wind turbine controller 36 may also include a memory, e.g. one or more memory devices. As used herein, a memory may comprise memory element(s) including, but not limited to, a computer readable medium (e.g., random access memory (RAM)), a computer readable non-volatile medium (e.g., a flash memory), a floppy disk, a compact disc-read only memory (CD-ROM), a magnetooptical disk (MOD), a digital versatile disc (DVD) and / or other suitable memory elements.

[0027] Figure 2 is an enlarged sectional view of a portion of the wind turbine 10. In the example, the wind turbine 10 includes the nacelle 16 and the rotor 18 that is rotatably coupled at the nacelle 16. More specifically, the hub 20 of the rotor 18 is rotatably coupled to a generator 42 positioned within the nacelle 16 by the main shaft 44, a gearbox 46, a high-speed shaft 48, and a coupling 50. In the example, the main shaft 44 is disposed at least partially coaxial to a longitudinal axis (not shown) of the nacelle 16. A rotation of the main shaft 44 drives the gearbox 46 that subsequently drives the high-speed shaft 48 by translating the relatively slow rotational movement of the rotor 18 and of the main shaft 44 into a relatively fast rotational movement of the high-speed shaft 48. The latter is connected to the generator 42 for generating electrical energy with the help of a coupling 50. Furthermore, a transformer 90 and / or suitable electronics, switches, and / or inverters may be arranged in the nacelle 16 in order to transform electrical energy generated by the generator 42 having a voltage between e.g. 400V to 1000 V into electrical energy having medium voltage (e.g. 10 - 35 kV). Offshore wind turbines may have, for example, generator voltages between 650 V and 3500 V.Transformer voltages may for instance be between 30 kV and 70 kV, or even higher, e.g. 132 kV. In particular, higher voltages result in increased challenges when dealing with faults so faster protection requirements are needed. Said electrical energy is conducted via power cables from the nacelle 16 into the tower 15.

[0028] The gearbox 46, generator 42 and transformer 90 may be supported by a main support structure frame of the nacelle 16, optionally embodied as a main frame 52. The gearbox 46 may include a gearbox housing that is connected to the main frame 52 by one or more torque arms 103. In the example, the nacelle 16 also includes a forward support bearing 60 and an aft support bearing 62. Furthermore, the generator 42 can be mounted to the main frame 52 by decoupling support means 54, in particular in order to prevent vibrations of the generator 42 to be introduced into the main frame 52 and thereby causing a noise emission source.

[0029] Optionally, the main frame 52 is configured to carry the entire load caused by the weight of the rotor 18 and components of the nacelle 16 and by the wind and rotational loads, and furthermore, to introduce these loads into the tower 15 of the wind turbine 10. The main shaft 44, generator 42, gearbox 46, high-speed shaft 48, coupling 50, and any associated fastening, support, and / or securing device including, but not limited to, main frame 52, and forward support bearing 60 and aft support bearing 62, are sometimes referred to as a drive train 64.

[0030] In some examples, the wind turbine may be a direct drive wind turbine without gearbox 46. Generator 42 operates at the same rotational speed as the rotor 18 in direct drive wind turbines. They therefore generally have a much larger diameter than generators used in wind turbines having a gearbox 46 for providing a similar amount of power than a wind turbine with a gearbox.

[0031] The nacelle 16 may also include a yaw drive mechanism 56 that may be used to rotate the nacelle 16 and thereby also the rotor 18 about the yaw axis 38 to control the perspective of the rotor blades 22 with respect to the wind direction 28.

[0032] For positioning the nacelle 16 appropriately with respect to the wind direction 28, the nacelle 16 may also include at least one meteorological measurement system which may include a wind vane and an anemometer. The meteorological measurement system 58 can provide information to the wind turbine controller 36 that may include wind direction 28 and / or wind speed.

[0033] In the example, the pitch system 32 (see Figure 1) is at least partially arranged as a pitch assembly 66 in the hub 20. The pitch assembly 66 includes one or more pitch drive systems 68 and at least one sensor 70. Each pitch drive system 68 is coupled to a respectiverotor blade 22 (shown in figure 1) for modulating the pitch angle of a rotor blade 22 along the pitch axis 34. Only one of three pitch drive systems 68 is shown in figure 2.

[0034] In the example, the pitch assembly 66 includes at least one pitch bearing 72 coupled to hub 20 and to a respective rotor blade 22 (shown in figure 1 ) for rotating the respective rotor blade 22 about the pitch axis 34. The pitch drive system 68 includes a pitch drive motor 74, a pitch drive gearbox 76, and a pitch drive pinion 78. The pitch drive motor 74 is coupled to the pitch drive gearbox 76 such that the pitch drive motor 74 imparts mechanical force to the pitch drive gearbox 76. The pitch drive gearbox 76 is coupled to the pitch drive pinion 78 such that the pitch drive pinion 78 is rotated by the pitch drive gearbox 76. The pitch bearing 72 is coupled to pitch drive pinion 78 such that the rotation of the pitch drive pinion 78 causes a rotation of the pitch bearing 72.

[0035] Pitch drive system 68 is coupled to the wind turbine controller 36 for adjusting the pitch angle of a rotor blade 22 upon receipt of one or more signals from the wind turbine controller 36. In the example, the pitch drive motor 74 is any suitable motor driven by electrical power and / or a hydraulic system that enables pitch assembly 66 to function as described herein. Alternatively, the pitch assembly 66 may include any suitable structure, configuration, arrangement, and / or components such as, but not limited to, hydraulic cylinders, springs, and / or servomechanisms. In certain embodiments, the pitch drive motor 74 is driven by energy extracted from a rotational inertia of hub 20 and / or a stored energy source (not shown) that supplies energy to components of the wind turbine 10.

[0036] The pitch assembly 66 may also include one or more pitch control systems 80 for controlling the pitch drive system 68 according to control signals from the wind turbine controller 36, in case of specific prioritized situations and / or during rotor 18 overspeed. In the example, the pitch assembly 66 includes at least one pitch control system 80 communicatively coupled to a respective pitch drive system 68 for controlling pitch drive system 68 independently from the wind turbine controller 36. In the example, the pitch control system 80 is coupled to the pitch drive system 68 and to a sensor 70. During normal operation of the wind turbine 10, the wind turbine controller 36 may control the pitch drive system 68 to adjust a pitch angle of rotor blades 22.

[0037] According to an embodiment, a power generator 84, for example comprising a battery and electric capacitors, is arranged at or within the hub 20 and is coupled to the sensor 70, the pitch control system 80, and to the pitch drive system 68 to provide a source of power to these components. In the example, the power generator 84 provides a continuing source of power to the pitch assembly 66 during operation of the wind turbine 10. In an alternative embodiment, power generator 84 provides power to the pitch assembly 66 only during anelectrical power loss event of the wind turbine 10. The electrical power loss event may include power grid loss or dip, malfunctioning of an electrical system of the wind turbine 10, and / or failure of the wind turbine controller 36. During the electrical power loss event, the power generator 84 operates to provide electrical power to the pitch assembly 66 such that pitch assembly 66 can operate during the electrical power loss event.

[0038] In the example, the pitch drive system 68, the sensor 70, the pitch control system 80, cables, and the power generator 84 are each positioned in a cavity 86 defined by an inner surface 88 of hub 20. In an alternative embodiment, said components are positioned with respect to an outer surface of hub 20 and may be coupled, directly or indirectly, to the outer surface.

[0039] As used herein, the term “processor” is not limited to integrated circuits referred to in the art as a computer, but broadly refers to a controller, a microcontroller, a microcomputer, a programmable logic controller (PLC), an application specific integrated circuit, and other programmable circuits, and these terms are used interchangeably herein.

[0040] Figure 3 shows a flowchart of a method 100 for operating a wind turbine 10. The wind turbine 10 comprises a plurality of wind turbine actuators. The method 100 comprises, in block 110, receiving operational data. An initial state is estimated in block 120 based on the received operational data. Furthermore, block 130 comprises using a model to predict, over a prediction period, potential operational states of the wind turbine 10 based on the estimated initial state and depending on operation of the wind turbine actuators. Moreover, one or more fatigue indicators, indicative of an anticipated life consumption of a wind turbine component, are derived in block 140 based on the predicted potential operational states. Block 150 of method 100 comprises optimizing a cost function over an optimization period to determine an optimum trajectory for commands of the wind turbine actuators. The optimization carried out in block 150 is subjected to one or more constraints. Then, block 160 comprises using the first wind turbine actuator commands of the determined optimum trajectory to control the wind turbine actuators. The cost function and / or the constraints used in block 150 of the method 100 are adjusted based on the derived fatigue indicators.

[0041] In an example of the disclosure, the cost function and / or the constraints used in block 150 may also be adjusted based on one or more features of a prevailing wind condition.

[0042] The method 100 is repeated at every control step and it allows an improved control of a wind turbine 10 while taking fatigue loads into account. The use of fatigue indicators or fatigue surrogates allows online implementation of the method 100 in an operating wind turbine 10.

[0043] Furthermore, the fatigue indicators are based on an anticipated life consumption. In this manner, a proactive response is obtained. In other words, control actions can be implemented in the different wind turbine actuators which result in a more optimized handling of the life consumption of the wind turbine components.

[0044] Furthermore, in examples wherein the cost function and / or the constraints are adjusted based on one or more features of a prevailing wind condition, the influence of the wind condition, e.g. turbulence or direction, on the life consumption may also be considered. Such wind conditions can influence the anticipated life consumption of the wind turbines. Consequently, by incorporating the information of the wind condition into the control actions, a more robust approach may be achieved, which may take into account the expected effect of the wind conditions on the life consumption of wind turbine components.

[0045] Figure 4 illustrates an example of a high level overview of a system comprising a wind turbine 10, a control unit 163 according to an example of the present disclosure, and a wind turbine actuator system 164 comprising a plurality of wind turbine actuators. The system depicted in Figure 4 may be employed to implement a method 100 like the one described with reference to Figure 3. The wind turbine actuator system 164 comprises a plurality of wind turbine actuators, such as the above mentioned blade pitch systems 32, yaw drive mechanism 56, and e.g. an electronic converter enabling torque control. Although depicted separately in Figure 4 for illustration and explanation purposes, the wind turbine actuators 164 are physically arranged in the wind turbine 10 as understood by those skilled in the art. The control unit 163 in this example comprises an estimator 161 and an MPC (Model Predictive Control) module 162. The control unit 163 controls the wind turbine 10 by generating commands 167 to control the wind turbine actuators 164.

[0046] As shown in Figure 4, estimating an initial state of the wind turbine 10 based on the received operational data 166 may comprise, at every control step, using an estimator 161. The estimator 161 may comprise filter equations (e.g. extended Kalman filters) that may be used to calculate the most probable operational state of the wind turbine 10 based on the received operational data 166. The operational state may be dynamically determined at each control step. Specifically, the state determined by the estimator 161 may include structural deflections of blades 22 and tower 15, rotor speed, or wind characteristics.

[0047] Receiving operational data 166 in block 110 of the method 100 operational data may comprise receiving data indicative of the operation of the wind turbine 10, i.e. of the different wind turbine actuators 164. In particular, operational data 166 may include the value of the blade pitch angles, output power, electrical torque, rotational speed of the wind turbine rotor, rotational speed of the electrical generator, tower top position, tower deflection, bladedeflection, blade root moment, etc. In some examples, and as also depicted in Figure 4, receiving operational data 166 may comprise receiving data indicative of one or more features of a prevailing wind condition, e.g. wind turbulence or wind direction.

[0048] Operational data 166 may be obtained via direct measurements with sensors. In particular, in some examples, prevailing wind conditions may be at least partially determined with a LIDAR installed on the wind turbine or neighboring wind turbines. LIDAR sensors can anticipate the wind resource reaching the turbine rotor for several seconds in advance. With this anticipation, the MPC controller has more capability to anticipate its pitch and torque commands to prepare for the incoming wind conditions. As a result, the pitch movements can be smoother and can preserve bearing life, the structural load oscillations can be lowered, and the control effort can be utilized for other objectives, such as better power tracking.

[0049] Nevertheless, in examples of the disclosure, operational data 166 may be obtained via virtual sensors. Thus, in examples of the disclosure, no direct measurement may be required. A virtual sensing system comprising a proper model may be used to calculate the value of the operational variable when no direct measurement of the same is available. For example, when rotor speed, pitch angle, generator torque and power output are known, the prevailing wind speed may be estimated even if it is not directly measured. In another example, a blade root bending moment may be estimated, without directly measuring it, when e.g. loads at the main bearing are measured.

[0050] Figure 5 provides a more detailed view of an example of a control unit 163 configured to carry out a method 100 like the one described with reference to Figure 3. The control unit 163 may comprise a model predictive control (MPC) module 162. As previously indicated, an estimator 161 may be used to characterize the operational state of the wind turbine 10. The operational state of the wind turbine 10 may be fed to a model 171 in the model predictive control (MPC) module 162. The model 171 may be used to predict operational states of the wind turbine 10 based on the estimated initial state and depending on operation of the wind turbine actuators 164 over a prediction period, i.e. the model 171 may be used to implement block 140 of the method 100 described with reference to Figure 3. Furthermore, the model 171 may account for the physical behavior of the wind turbine components and the wind. Such prediction is schematically illustrated in Figure 6. In particular, the predicted trajectories of input 267 (u) variables, e.g. pitch angle, and output 268 variables (y), e.g. tower deflection or blade loads, along a prediction horizon are schematically depicted. The prediction horizon is subdivided in a number of N steps as also shown in Figure 6.

[0051] In an example, the model may comprise accurate component models for a plurality of wind turbine components, specifically for wind turbine components subjected to fatigue loadssuch as blades 22, tower 15, hub 20, main frame 52, or, in the case of floating offshore wind turbines, mooring lines.

[0052] The MPC controller 162 may also comprise an optimization builder and solver 172, which may be provided to create and optimize a cost function, / , in accordance with block 150 of the method 100. Furthermore, the control unit 163 is also configured to derive one or more fatigue indicators indicative of an anticipated life consumption of a wind turbine component based on the predicted potential operational states. The optimization builder and solver 172 is understood as an algorithm that builds a cost function, / , and the different constraints, as well as their relative relevance (e.g. the weights of each constraint in the cost function).

[0053] The cost function, / , to optimize may be of the form shown in equation 1 below.where n is the “stage” in the prediction horizon, un,xnare the control inputs and the performance outputs at the n-th stage, and R, N, V, S, T are weight matrices, and c is a constant.

[0054] As also shown in Figure 5, the optimization builder and solver 172 may be used to find an optimum trajectory. An optimization algorithm may be used to find the optimum solution, i.e. the optimum set of independent or controllable variables that minimizes the cost function, / , at the corresponding control step. The optimum trajectory resulting from the optimization may be fed to a command calculation module 173, which may take the first value of the calculated optimum trajectory for each of the wind turbine actuators 164 to generate the commands 167 to control the wind turbine actuators 164.

[0055] Figure 5 is just an example of an architecture to implement a method according to the disclosure. The details of the control unit 163 may differ in other variants. For instance, as a non-limiting example, some of the actions defined for the different modules may be combined in a single module.

[0056] As also shown in Figure 5, in an example of the disclosure, a separate linearization module 174 may be provided in the control unit 163. The linearization module 174 may be used to linearize a wind turbine control model around an initial state obtained from the estimator 161 and around predicted states at subsequent linearization points within the prediction time horizon. The predicted states may be determined on the basis of the initial state and on the basis of a previously determined optimum trajectory for the wind turbine actuators 164. The linearization module 174 may generate a set of linearization matrices to aid in the optimization process. The provision of a linearization around predicted states at subsequent linearization points is advantageous over conventional methos, which typically comprise a singlelinearization for the whole prediction horizon. In the present disclosure, one linearization per step is used because the accuracy of the prediction trajectory is high.

[0057] Once the linearization is complete, the linearization matrices may be fed to the optimization builder and solver 172 for use at a corresponding cycle step during optimization of the cost function, J. The optimization builder and solver 172 may use a quadratic programming approximation to solve the optimization problem. In this manner, a simpler optimization problem (quadratic) than the original optimization problem (nonlinear) may be obtained. Consequently, the computational requirements of the wind turbine control platform may be reduced and the wind turbine 10 may be effectively controlled in real-time.

[0058] Figure 7 illustrates the linearization principle according to an example. In this example, to and tt represent the start and the end time for the prediction time horizon. As shown in the figure, a state trajectory is defined on the basis of the estimated initial state, x0, and the optimized trajectory for the wind turbine actuators which, in Figure 7, is designated as u. As schematically illustrated in Figure 7, the model may be linearized around a plurality of linearization points 510 within the prediction time horizon. Specifically, in an example, the linearization points may correspond to each of the cycle steps within the prediction time horizon, i.e. the separation between linearization points may correspond to the sample time or cycle step of the controller.

[0059] In some alternative examples, the optimization problem may be formulated as a nonlinear programming problem. This may give place to a more accurate representation of the system. However, these examples may exhibit significantly higher computational expense.

[0060] As also shown in Figure 5, the control unit 163 may comprise a supervisory module 169. The supervisory module 169 may be used to adjust the cost function and / or the constraints used in the optimization builder and solver 172. Such adjustment may have no impact on the cost of solving the optimization problem in the optimization builder and solver 172. The reconfiguration of the cost function and / or the constraints, i.e. the reconfiguration of the MPC optimization problem, may be defined by a plurality of reconfiguration variables 165. Furthermore, the supervisory module 169 may obtain such reconfiguration variables 165 based on the current fatigue indicators, i.e. fatigue indicators derived in the current control step from the received operational data 166. Furthermore, in some examples, features of a prevailing wind condition, e.g. wind speed, wind turbulence, wind direction, wind shear, or wind veer, may also be received by the supervisory module 169 to adjust the cost function and / or constraints. Furthermore, as shown in Figure 5, the supervisory module 169 may also receive inputs representative of a wind turbine state machine 183. Besides, wind farm data 184 may also be received from, e.g. a wind farm controller.

[0061] In examples of the method 100 comprising adjusting the cost function and / or the constraints based on the prevailing wind conditions, such an adjustment may comprise adjusting based on a local wind condition. Furthermore, in variants of this example, the adjustment may be carried out in a wind turbine controller 36, i.e. at local level.

[0062] In these examples, a precise response may be achieved individually at each wind turbine 10. Accordingly, the wind conditions experienced by the respective wind turbine 10 may be accounted for and an optimized response may be achieved.

[0063] Furthermore, in other cases, the wind turbine 10 may be arranged in a wind farm, i.e. in an installation comprising a plurality of wind turbines 10. In these cases, examples of the method 100 comprising adjusting the cost function and / or the constraints based on the prevailing wind condition, may comprise adjusting the cost function and / or the constraints based on the features of the prevailing wind condition may comprise adjusting based on a wind condition at the wind farm level. Furthermore, in these examples, the adjustment may be initiated in a wind farm controller. Indeed, as shown in Figure 5, a supervisory module 169 may be configured to receive wind farm data 184 and these may be used to adjust the reconfiguration variables 165 of the MPC problem.

[0064] In a variant of examples comprising wind farms, at least one of the fatigue indicators may be at least partially based on an internal characteristic of the wind farm. Such internal characteristics may be either stationary, e.g. layout of the wind turbines 10 in the wind farm, or dynamic, e.g. real-time operational states of different wind turbines.

[0065] In these examples, a more comprehensive control of a complete wind farm may be achieved. Specifically, a balancing action may be implemented, such that multiple wind turbines 10 in the wind farm may be fatigued in a substantially uniform manner. Furthermore, by taking wind conditions at wind farm level into account, wake loads and, more particularly, life consumption induced by wake loads may be considered in a more efficient manner. Hence, knowing the wind direction and the turbulence generated by wakes, as well as the location of the wind turbines within the wind farm, appropriate adjustments may be carried out at the wind farm level. Accordingly, the distribution of life consumption in the wind farm may exhibit a lower standard deviation, thus enabling more consistent maintenance operations.

[0066] In still other variants, both approaches may be combined. Hence, both wind conditions at local level and at wind farm level may be considered and an improved response, taking into account both inputs, may be envisaged.

[0067] While implementing a method 100 like the one described with reference to Figure 3, different approaches may be envisaged to adjust the cost function and / or the constraints based on the derived current fatigue indicators.

[0068] In an example, at least one of the derived fatigue indicators may be included as a term in the cost function, J. A weight may be assigned to the corresponding term. Thus, in an example wherein the cost function contains only penalties for the control amplitude and for fatigue life consumption of a certain component, as characterized by the fatigue indicator Fatty , the following cost function may be provided.where 14^ corresponds to the weight associated with the corresponding fatigue indicator. The weight may be a constant in this example. Furthermore, in a variant of this example, the cost function may be adjusted based on one or more features of a prevailing wind condition and, specifically, the weight W±may be adjusted based on the features of the prevailing wind condition. In this manner, a more optimized response may be achieved.

[0069] Specifically, different components and / or fatigue mechanisms may exhibit varying sensitivity to certain wind conditions. Accordingly, a higher weight 14^ may be assigned if the fatigue indicator is associated with a load condition that may be particularly critical at some prevailing wind conditions. On the contrary, a low weight may be given if the correlation between the prevailing wind conditions and the fatigue indicator is minimal. In this manner, overall performance of the wind turbine 10 may not be unnecessarily constrained.

[0070] Different wind conditions may be considered while adjusting the weight in different implementations. Hence, in some examples, the weight may be adjusted based on an axially asymmetric property of the wind condition. In this manner, wind conditions such as vertical wind shear, horizontal wind shear, or wind veer, may be considered. Such wind conditions are known to induce significant fatigue loads in some components. By specifically targeting axially asymmetric properties of the wind condition, a more optimized response may be thus achieved.

[0071] Nevertheless, some other components may be more affected when faced with axially symmetric properties of the wind, i.e. wind speed or longitudinal turbulence intensity. Consequently, in some variants, the weight may be adjusted based on a fluctuation of a symmetric property of the prevailing wind conditions.

[0072] As understood by the skilled person, both approaches are not mutually exclusive and some further variants may be envisaged in which both axially asymmetric and axially symmetric properties may be taking into account in order to assign a value to the weight of acertain fatigue indicator include in the cost function. In particular, assuming a given fatigue indicator for which fatigue consumption depends on multiple wind properties PllP2l... ,Pp, the corresponding term may be of the form shown in expression 3.

[0073] In an example, this may be implemented as a lookup table of p dimensions in the supervisory module 169 of the control unit 163 of the wind turbine 10.

[0074] In still other examples, the weight may be dependent on a plurality of simultaneously occurring properties of both the prevailing wind conditions and the operational state of the wind turbine 10. In these examples, a look-up table of a dimension equal to the number of properties considered may be used to select the most appropriate value for the weight of the corresponding fatigue indicator.

[0075] In some examples, a plurality of fatigue indicators may be considered simultaneously. In such case, a cost function may adopt a form like the one shown in equation 4.where F different fatigue indicators may be considered. Each of the fatigue indicators Fat y^ may be given a respective weight 14 .

[0076] The use of a plurality of fatigue indicators may allow an improved characterization of the predicted life consumption in different components of the wind turbine 10. Furthermore, weights Wtmay be adjusted to account for the relative importance of the different fatigue indicators Fat y . Finally, the weights may be dynamically adjusted based on, e.g. wind conditions. In this manner, different fatigue response of different wind turbine components may be accommodated to the prevailing wind conditions in real time.

[0077] In some examples of the disclosure comprising a plurality of fatigue indicators Fat(yi), these may be representative of fatigue loads at different sections of a same component of the wind turbine 10. In this manner, a more precise control of the life consumption of certain critical components, e.g. blades 22 or tower 15, may be achieved than when considering a singular location of the components. Specifically, different fatigue mechanisms, which may coexist in the same wind turbine component, may be individually accounted for and their contribution may be adapted to prevailing operation state of the wind turbine 10 and to prevailing wind conditions.

[0078] In order to include multiple fatigue indicators for a single wind turbine component, accurate models for subsections of this wind turbine component may be utilized. The accurate models may allow capturing fatigue loads throughout the wind turbine component at multiple locations of the same. This may become particularly relevant for large wind turbine components. As an examples, potential differences may exist in the wind condition across the blade span. Therefore, different parts of the blade 22 may be subjected to different fatigue loads, leading to potential failure at a blade point other than the blade root, which is the point most commonly considered in conventional control methods.

[0079] In some examples of the method 100, at least one of the fatigue indicators Fat y^ may be included as one of the constraints of the optimization problem. The constraint may be assigned a lower and / or an upper bound value. In some examples, the lower and / or upper bound values may be adjusted as function of the features of the prevailing wind conditions.

[0080] Hence, according to these examples, which may be combined with previous examples comprising use of the fatigue indicators Fat(yj) as terms in the cost function, J, a further mechanism may be provided to account for the predicted fatigue of the wind turbine 10. Specifically, model predictive controllers are known to be particularly appropriate to enable control of a wind turbine 10 while taking different operational constraints into account. In these examples, by incorporating fatigue-based constraints, an improved operation may be obtained which may result in an optimized life consumption of the wind turbine 10.

[0081] Furthermore, in an equivalent manner as when dealing with the weights of the terms in the cost function, the upper and / or lower bounds of the constrains may be dynamically adjusted to enhance adaptation to existing conditions. In particular, more restrictive limits may be imposed in those cases wherein prevailing wind conditions are likely to induce significant damage leading to high fatigue loads. On the contrary, less strict bounds may be provided for wind conditions not leading to significant fatigue loads, or for wind conditions expected to last for only a short duration. Hence, in the latter case, one may prefer to keep relatively aggressive bounds, i.e. large or not very limiting bounds, for the constraints as short duration events may not significantly impact the fatigue behavior of the wind turbine 10.

[0082] As shown above, fatigue indicators Fat y^ can be incorporated in different manners in the optimization problem, i.e. as cost terms in the cost function and / or as constraints. Regarding the nature of the fatigue indicators themselves, different options may be envisaged.

[0083] In an example of the method 100, at least one of the derived fatigue indicators may comprise a pseudo-variance, pVar(y~), of a load signal, y. The pseudo-variance is defined as the variance around a previous average value, i.e. the pseudo-variance of a set of values maybe calculated with the well-known variance formula but, as the name indicates, it does not correspond to the actual variance, which would require the average of that same set of values. Instead, the calculated variable is a approximation to the variance as the average value employed corresponds to the set of values from a previous control step. In particular, a previous predicted trajectory of the load signal over a previous prediction period may be available as a byproduct of optimizing the cost function in a previous control step. The calculation of the pseudo-variance may comprise calculating first an average value of the load signal over a previous prediction period for a previous trajectory predicted in a previous control step. Subsequently, a variance of the load signal may be approximated using values of the load signal over the prediction period for the trajectory predicted at a current control step and the calculated average value. The load signal may be associated to a fatigue consumption of a wind turbine component.

[0084] The use of the pseudo-variance as a fatigue surrogate may be particularly advantageous from a computational point of view. Indeed, the use of the pseudo-variance of a signal may provide for an optimization problem, i.e. for a cost function, not containing coupling terms between stages. That is, all the cost terms in the cost function, J, may include a quadratic form of magnitudes belonging only to a single stage of the prediction period. In other words, by using pseudo-variances instead of standard variances, a stage decoupling is provided. Consequently, efficient solvers may take advantage of the stage decoupling, thus reducing computational effort.

[0085] Different load signals may be used in different examples of the disclosure. Hence, the load signals may include at least one of the following signals: flap moment at a root section of a wind turbine blade, a moment in a nodding direction of a yaw bearing of the wind turbine, or a moment in a yaw direction of a main bearing of the wind turbine.

[0086] Figures 8A and 8B show a flowchart of an example of a method 200 for controlling a wind turbine 10 comprising a plurality of wind turbine actuators, and comprising use of a pseudo-variance as a fatigue surrogate. The method 200 comprises receiving operational data in block 210, and estimating an initial state based on the received operational data in block 220. Furthermore, in block 230, the method comprises predicting, using a model, potential operational states of the wind turbine 10 based on the estimated initial stage and depending on operation of the wind turbine actuators over a prediction period. A pseudo-variance of a load signal associated to a life consumption of a wind turbine component is calculated in block 240. A predicted trajectory of the load signal over the prediction period is available as a byproduct of optimizing the cost function in a previous control step.

[0087] In particular, as more clearly depicted in Figure 8B, the calculation of the pseudovariance may comprise calculating an average value of the load signal over a previous prediction period for a previous trajectory predicted in a previous control step in block 241. This calculation may be calculated with a small computational effort. Subsequently, in block 242, the method 200 comprises calculating a variance of the load signal using values of the load signal over the prediction period for the trajectory predicted at a current control step and the calculated average value derived from block 241.

[0088] Moreover, block 250 of the method 200 comprises optimizing a cost function over an optimization period to determine an optimum trajectory for commands of the wind turbine actuators, the optimization being subject to one or more constraints. The cost function and / or the constraints are adjusted based on the calculated pseudo-variance. Then, in block 260, the method 200 comprises using the first wind turbine actuator commands of the determined optimum trajectory to control the wind turbine actuators.

[0089] Figure 9 illustrates an example of a model predictive controller 162 for controlling a wind turbine 10, wherein a pseudo-variance is used as a fatigue indicator. The arrangement disclosed in Figure 9 is applicable to the system described with reference to Figure 5. Consequently, the same numerals will be used to refer to the corresponding systems or modules.

[0090] The model predictive controller 162 schematically depicted in Figure 9 comprises a model 171 and a linearization module 174. Furthermore, the model predictive controller 162 also comprises an optimization builder and solver 172. However, in this example, two additional modules are depicted. In particular, an averaging module 181 can be seen in Figure 9. The averaging module 181 is configured to calculate an average value over a prediction period of a certain load signal. The load signal may be associated with the predicted fatigue of a wind turbine component and it may be obtained as a byproduct of the optimum operating state trajectory obtained by the optimization builder and solver 172. Indeed, while operating the model predictive controller 162, an optimum trajectory for the commands of the wind turbine actuators 164 may be typically obtained. However, trajectories for other variables, including e.g. tower 15 or blade 22 moments, may also be predicted as part of the prediction carried out while executing the algorithm.

[0091] Given the values of a certain load signal evaluated by the optimizer at step k - 1, the averaging module 181 may calculate an average value over the prediction period. Such average value may be stored in a memory module 182. Accordingly, the 1-step old average value may be retrieved from the memory module 182 while executing the controller in thecurrent, i.e. k, step. Specifically, the average value may be used to calculate a pseudo-variance according to equation 5.where pVark(y) is the pseudo-variance of a load signal, y , y^ is the predicted signal n-steps in the future, calculated at the k-th control step, andis the calculated average value obtained for the signal y, evaluated at the k - 1 control step, i.e. in the previous control step. As seen in Equation 5, the pseudo-variance exhibits stage decoupling, allowing the use of efficient optimization solvers. When the optimization carried out in step k is completed, the new optimum trajectory may be used by the averaging module 181 to calculate an updated value of the average value for the load signal. The resulting average value may then be saved in the memory module 182 for use in the next control step, i.e. in k + 1 step. This process may repeat continuously while the controller is in operation.

[0092] In examples involving the use of a pseudo-variance, the pseudo-variance may be included as a term in the cost function. Hence, a cost function J may adopt the following from:< where the term in unpenalizes the motion of the control amplitude u, and the term in ynpenalizes the pseudo-variance of the magnitude y. As described, the pseudo variance pVar is defined as the variance around a previous mean value calculation. Furthermore, as already explained with reference to equation 2 above, this cost function, J, is also provided only for illustrative purposes. Hence, different cost functions, particularly cost functions comprising terms for setpoint tracking, utilization of wind turbine actuators, and / or different additional loads may be provided.

[0093] Furthermore, it is understood that a plurality of pseudo-variances may be used to account for different load signals. Such load signals may be representative of different loads and / or different wind turbine components. In such a case, a cost function, J, may be defined with multiple terms and different weights for the different terms as shown in equation 7.where the magnitudes yi,y2< -TF represent the different load signals for which fatigue is penalized, and Wc, 14^, ... , WFare weights to emphasize certain components in the cost function. Furthermore, the weight may be dynamically adjusted, i.e. in dependence of prevailing wind conditions. Accordingly, whenever a weight value 14 is increased, the cost associated with the corresponding pseudo-variance pVar(yi) may be increased or emphasized and, consequently,the associated fatigue life consumption may tend to decrease while operating the wind turbine 10.

[0094] The accuracy of the pseudo-variance may depend on the invariance in the behavior of the respective load signal, y, i.e. on the so-called approximate trajectory invariance. As explained above, in order to calculate the pseudo-variance, use is made of an average value which is calculated on the basis of a trajectory determined in the previous control step. This is based on the assumption that such a calculated value is still representative of the average value in the current control step. In other words, an assumption is made that the optimum trajectories computed by the optimizer and solver 172 at one control step are close to the timeshifted version of the optimal trajectories computed by the optimizer and solver 172 in the previous control step. In an example of the disclosure comprising the use of the pseudovariance as a fatigue indicator, the average value of the load signal calculated for the previous trajectory predicted in the previous control step may be compared with an average value calculated for the load signal over the prediction period for the trajectory predicted in the current control step. In other words, the average value employed for the calculation of the pseudo-variance, which is expected to be substantially similar to the average value obtained in the current control step, is compared with the latter.

[0095] In this manner, a verification on the robustness of the implemented method may be carried out in an online manner. In particular, in examples wherein the pseudo-variance of the load signal is included as a term in the cost function with a weight, the respective weight may be reduced if a difference between the average values calculated in the previous control step and in the current control step exceeds a predetermined threshold.

[0096] An increasing difference between the average values may be indicative of a less reliable performance and, more particularly, of a less valid pseudo-variance, i.e. a pseudovariance that is less representative of an actual fatigue measure. In other words, this may be an indication that the trajectory invariance property is not maintained. Accordingly, in order to reduce the reliance on such a less reliable fatigue indicator, the corresponding weight may be reduced. In this manner, the weight may not only be dynamically adjusted on the basis of, e.g. operational and wind conditions, but also on the basis of the reliability of the information obtained while implementing the method.

[0097] An implementation of an example of a method comprising verification of the trajectory invariance property and corresponding corrective action, i.e. adaptation of the weight of the pseudo-variance, is schematically depicted in Figure 10. This function may be implemented in the supervisory module 169 of the controller. As shown in Figure 10, the average value calculated in a previous control step may be kept in a memory module 191 , andthis may be used in a comparator 192 to check if the difference between the average values obtained in the previous and in the current step are below a certain threshold, 3. If that is the case, a standard or default weight value may be assigned in block 194 to the corresponding term in the cost function. Otherwise, the corresponding weight value may be reduced in block 193. The applied weight value may be updated in block 195. Finally, the updated weight value may be included in the reconfiguration variables 165 provided from the supervisory module 169 of the controller to the MPC controller 162.

[0098] Although a pseudo-variance of a load signal may be particularly suitable for analysis of the predicted fatigue, other fatigue indicators may also be utilized. In some examples of the disclosure, different fatigue indicators may be combined in a single implementation of the method. Hence, in an example of the disclosure, the derived fatigue indicator may comprise a damage equivalent load DEL(y, m) of a load signal yof a load signal for a Wohler slope coefficient m. The damage equivalent load may be calculated using the pseudo-variance of the corresponding load signal. The following function may be then used as a surrogate for the fatigue life consumption.where DEL(y,m) represents the damage equivalent load of load y for a Wohler slope coefficient m. Coefficients a and b may be obtained through regression using data from openloop simulations for each load signal.

[0099] Furthermore, in examples comprising the use of a dynamic equivalent load as shown in equation 8, this may be included as a term in the cost function and the term may be assigned a corresponding weight. The weight may be a function of the Wohler slope coefficient:W = W(m) (9)

[0100] In this manner, the different fatigue behavior of different wind turbine components may be considered. In particular, these examples may be particularly useful for applications in which there is a particular set of slope coefficients to be penalized. This may allow, for instance, to adjust the fatigue penalty online for a component that may have different failure modes depending on the regions of operation of the wind turbine 10, and wherein each failure mode may have an associated slope coefficient.

[0101] In still some other examples, the variance of a certain load signal may be used as a fatigue indicated instead of, e.g. the pseudo-variance. In these examples, enhanced accuracy may be obtained at the expense of higher computation cost.

[0102] As already described, different examples of the method 100 may comprise using a fatigue indicator as a constraint in the optimization problem. Hence, in an example comprising the use of a pseudo-variance as fatigue indicator, a constraint may be included in the optimization problem as shown in equation 10.pVar(y) < Varmax(10)

[0103] Hence, in this example, an upper bound Varmaxmay be assigned to the pseudovariance, which may be included as a hard constraint in the MPC optimization problem. Furthermore, in a variant of this example, the value of the upper bound Varmaxmay be adjusted depending on operational conditions of the wind turbine and / or on a feature of the prevailing wind condition.

[0104] The use of a fatigue indicator as a constraint may be particularly interesting in those cases in which a maximum design equivalent load for a certain component, such as a design envelope, needs to be satisfied. Although equation 10 refers to the use of a pseudo-variance as a fatigue indicator, it is understood that, in other variants, a design equivalent load, DEL(y,m), like the one described in reference to previous equation 8 may also be used.

[0105] Furthermore, in some examples, the fatigue indicator may be employed in a soft-constraint of the optimization problem. In particular, in an example, adjusting the cost function and / or the constraints based on the derived fatigue indicator may comprise calculating an average value of a load signal over a previous prediction period for a previous trajectory predicted for the load signal in a previous control step. The method may further comprise calculating a difference between values of the load signal predicted over the prediction period in a current control step and the calculated average value. Then, the calculated difference may be used as a soft-constraint while optimizing the cost function. The load signal may be associated with a life consumption of a wind turbine component.

[0106] In this manner, an approach comprising a so-called zone tracking penalty may be used as a fatigue surrogate. The zone tracking cost may be implemented in the form of two soft constraints for each step of the prediction. In particular, the cost function of the optimization problem may comprise a term like the one shown in equation 11<

[0107] Furthermore, two additional constraints may be included in the optimization problem as shown below.<> >where W£is the constraint slack penalty, and d is the size the zone centered ati.e. at the average value obtained based on the optimum trajectory derived in the previous control step.

[0108] Figure 11 schematically illustrates the concept of the zone tracking penalty. In particular, the zone tracking penalty+ e„2) 301 is illustrated. In particular, in an example of the disclosure, a deadband may be defined for the calculated difference and the soft constraint may be defined such that only differences exceeding the deadband are penalized. In particular, as shown in Figure 11, the deadband may have a width of d around the average value obtained based on the optimum trajectory of the previous control step:

[0109] The use of zone tracking penalty with a deadband may allow saving computational efforts on the MPC controller, especially when predicted fatigue is relatively small.

[0110] Apart from the prevailing wind conditions, other environmental conditions may also have an impact on the life consumption of wind turbine components. In order to take such influence into account, in an example, the method 100 may comprise adjusting the cost function and / or the constraints based on a hygrothermal environment of the wind turbine 10. Specifically, the cost function and / or the constraints may be based on a temperature and / or a humidity determined at a location where the wind turbine 10 is installed.

[0111] In some examples of the disclosure, historical data of the temperature and / or the humidity at the wind turbine 10 location may be recorded to determine a stress caused by environmental factors. In particular, historical data may be used to determine thermal cycles and, from these, the stress suffered by certain wind turbine components may be derived. The stress information may be used to, e.g. adjust the weight of a fatigue indicator in the cost function and / or to adjust the value of a bound of a constraint employed in the optimization problem

[0112] The method 100 for operating a wind turbine 10 may be employed in both onshore and offshore applications. In particular, when considering offshore applications, not only wind conditions, but also sea conditions may impact the life time consumption of wind turbine components. This may be particularly relevant for floating wind turbines. Accordingly, in an example of the disclosure, the wind turbine 10 may be an offshore wind turbine 10, and the cost function and / or the constraints may be based on one or more features of a prevailing sea condition. Specifically, the frequency, amplitude and / or coherence of prevailing sea waves may be taken into account. Hence, as described above with reference to wind conditions, theconditions of the sea may also be used to adjust the weights of fatigue related terms in the cost function and / or the bounds of fatigue related constraints.

[0113] In some variants of this example, a LIDAR sensor may be provided in the wind turbine 10. The use of a LIDAR system may improve the accuracy and characterization of the sea conditions, thus enhancing the overall performance of the control system. In other examples, radar technology may be employed to measure sea waves.

[0114] According to another aspect of the disclosure, and referring again to previously described Figure 4, a control unit 163 for a wind turbine 10 may be provided. The wind turbine 10 comprises a plurality of wind turbine actuators. The control unit 163 is configured for receiving operational data 166. The control unit 163 is further configured for estimating an initial state based on the received operational data. Furthermore, the control unit 163 is configured for predicting, using a model, potential operational states of the wind turbine 10 based on the estimated initial state and depending on operation of the wind turbine actuators over a prediction period. The control unit 163 is also configured for deriving one or more fatigue indicators indicative of an anticipated life consumptions of a wind turbine components based on the predicted potential operational states. The control unit 163 is configured for optimizing a cost function over an optimization period to determine an optimum trajectory for commands of the wind turbine actuators 164, the optimization being subject to one or more constraints. The control unit 163 is configured for using the first wind turbine actuator commands of the determined optimum trajectory to control the wind turbine actuators. Moreover, the control unit 163 is further configured for adjusting the cost function and / or the constraints base on the derived fatigue indicators. In some examples, the control unit 163 may be provided in the wind turbine controller 36. In other examples, the control unit 163 may be a distributed unit.

[0115] Furthermore, in some examples, the control unit 163 may be configured for adjusting the cost function and / or the constraints based on one or more features of a prevailing condition.

[0116] In order to implement a control method, e.g. according to, but not limited to, any of the above-described examples, the control unit 163 may comprise a processor. Examples of the methods disclosed herein may be implemented with hardware, software, firmware and combinations thereof.

[0117] Those of skill in the art would further appreciate that the various illustrative logical blocks, modules, and algorithm steps described in connection with the disclosure herein may be implemented as electronic hardware, computer software, or combinations of both.

[0118] The various illustrative logical blocks and modules described in connection with the disclosure herein may be implemented or performed with one or more general-purpose processors, a digital signal processor (DSP), cloud computing architecture, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general-purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.

[0119] The present disclosure also relates to a computer program or computer program product comprising instructions (code), which when executed, performs any of the methods disclosed herein.

[0120] The computer program may be in the form of source code, object code, a code intermediate source and object code such as in partially compiled form, or in any other form suitable for use in the implementation of the processes. The carrier may be any entity or device capable of carrying the computer program.

[0121] If implemented in software / firmware, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media may be any available media that can be accessed by a general purpose or special purpose computer. By way of example, and not limitation, such computer-readable media can comprise RAM, ROM, EEPROM, CD / DVD or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code means in the form of instructions or data structures and that can be accessed by a general-purpose or special-purpose computer, or a general-purpose or special-purpose processor. Combinations of the above should also be included within the scope of computer-readable media.

[0122] This written description uses examples to disclose the teaching, and also to enable any person skilled in the art to practice the teaching, including making and using any devices or systems and performing any incorporated methods. The patentable scope is defined by the claims, and may include other examples that occur to those skilled in the art. Such other examples are intended to be within the scope of the claims if they have structural elementsthat do not differ from the literal language of the claims, or if they include equivalent structural elements with insubstantial differences from the literal languages of the claims. Aspects from the various examples described, as well as other known equivalents for each such aspects, can be mixed and matched by one of ordinary skill in the art to construct additional examples and techniques in accordance with principles of this application. If reference signs related to drawings are placed in parentheses in a claim, they are solely for attempting to increase the intelligibility of the claim, and shall not be construed as limiting the scope of the claim.

Claims

29CLAIMS1. A method (100) of operating a wind turbine (10), the wind turbine (10) having a plurality of wind turbine actuators (164), the method comprising:receiving operational data (166);estimating an initial state based on the received operational data (166); predicting, using a model (171), potential operational states of the wind turbine (10) based on the estimated initial state and depending on operation of the wind turbine actuators (164) over a prediction period;deriving one or more fatigue indicators indicative of an anticipated life consumption of a wind turbine component based on the predicted potential operational states;optimizing a cost function over an optimization period to determine an optimum trajectory for commands of the wind turbine actuators (164), the optimization being subject to one or more constraints; andusing first wind turbine actuator commands of the determined optimum trajectory to control the wind turbine actuators (164);wherein the cost function and / or the constraints are adjusted based on the derived fatigue indicators.

2. The method of claim 1 , wherein at least one of the derived fatigue indicators is included as a term in the cost function, and wherein a weight is assigned to the corresponding term.

3. The method of any of claims 1 or 2, wherein the derived fatigue indicator comprises a pseudo-variance of a load signal, the calculation of the pseudo-variance comprising:calculating an average value of the load signal over a previous prediction period for a previous trajectory predicted for the load signal in a previous control step; and calculating a variance of the load signal using values of the load signal over the prediction period for the trajectory predicted at a current control step and the calculated average value.

4. The method of claim 3, wherein the average value of the load signal calculated for the previous trajectory predicted in the previous control step is compared with an average value30calculated for the load signal over the prediction period for the trajectory predicted in the current control step,and wherein the pseudo-variance of the load signal is included as a term in the cost function with a weight, and the weight is reduced if a difference between the average values calculated in the previous control step and in the current control step exceeds a predetermined threshold.

5. The method of any of claims 3 or 4, wherein the derived fatigue indicator comprises a damage equivalent load of the corresponding load signal for a Wohler slope coefficient calculated using the pseudo-variance of the load signal.

6. The method of any previous claim, wherein adjusting the cost function and / or the constraints based on the derived fatigue indicators comprises:calculating an average value of a load signal over a prediction period for a previous trajectory predicted for the load signal in a previous control step;calculating a difference between values of the load signal predicted over the prediction period in a current control step and the calculated average value; andusing the calculated difference as a soft constraint while optimizing the cost function.

7. The method of claim 6, wherein a deadband is defined for the calculated difference and the soft constraint is defined such that only differences exceeding the deadband are penalized.

8. The method of any previous claim, wherein the cost function and / or the constraints are adjusted based on one or more features of a prevailing wind condition.

9. The method of claim 8, wherein at least one of the derived fatigue indicators is included as a term in the cost function, and wherein a weight is assigned to the corresponding term, the weight being adjusted based on the features of the prevailing wind condition.

10. The method of claim 9 wherein the weight is adjusted based on an axially asymmetric property of the prevailing wind condition.

11. The method of any of claims 8 to 10, wherein at least one of the derived fatigue indicators is included as one of the constraints, the constraint being assigned a lower and / or an upper bound value, the lower and / or upper bound value being adjusted as a function of the features of the prevailing wind condition.

12. The method of any of claims 8 to 11, wherein adjusting the cost function and / or the constraints based on the features of the prevailing wind condition comprises adjusting based on a local wind condition and adjusting in a wind turbine controller (36); orwherein the wind turbine (10) is arranged in a wind farm and adjusting the cost function and / or the constraints based on the features of the prevailing wind condition comprises adjusting based on a wind condition at the wind farm level, and wherein adjusting is initiated in a wind farm controller.

13. The method of any previous claim, wherein a plurality of fatigue indicators is derived, the plurality of fatigue indicators being representative of fatigue loads at different sections of a same component of the wind turbine (10).

14. The method of any previous claim, wherein the wind turbine (10) is an offshore wind turbine (10), and the cost function and / or the constraints are based on one or more features of a prevailing sea condition.

15. A control unit (163) for a wind turbine (10), the wind turbine (10) having a plurality of wind turbine actuators (164), the control unit (163) being configured for:receiving operational data (166);estimating an initial state based on the received operational data (166); predicting, using a model, potential operational states of the wind turbine (10) based on the estimated initial state and depending on operation of the wind turbine actuators (164) over a prediction period,deriving one or more fatigue indicators indicative of an anticipated life consumption of a wind turbine component based on the predicted potential operational states,optimizing a cost function over an optimization period to determine an optimum trajectory for commands of the wind turbine actuators (164), the optimization being subject to one or more constraints;using the first wind turbine actuator commands of the determined optimum trajectory to control the wind turbine actuators (164);wherein the control unit (163) is further configured for adjusting the cost function and / or the constraints based on the derived fatigue indicators.

Images