Method for controlling a wind turbine farm using an optimization method

Abstract

The present invention relates to a method for controlling a wind farm that implements a data acquisition (QAC) of a wind speed and direction distribution, as well as a real-time wind speed and direction acquisition (Vac), a wind farm model (MOD F), and a load model (MODC) for each wind turbine. Subsequently, an optimization step (OPT) determines target operating points for each wind turbine. This optimization step optimizes the expected energy generated for the entire wind speed and direction distribution as a function of the expected load for each wind turbine across the same distribution. These target operating points (e.g., target yaw angles) are then applied to the wind turbines in the farm (CON). Figure 2 to be published

Description

Title of the invention: Method for controlling a wind turbine farm using an optimization method technical field

[0001] The present invention relates to the field of wind farm control to maximize power output and to reduce wind turbine fatigue.

[0002] A wind farm, also called a wind park or wind power plant, is a site comprising a number of wind turbines that produce electricity. This site can be on land or at sea. A distinction is thus made between onshore wind farms and offshore wind farms, that is to say, at sea.

[0003] The wind turbines in these farms are generally horizontal axis wind turbines equipped with a system to orient the horizontal axis of rotation in the direction of the wind, in order to maximize the energy recovered by the turbine. A wind turbine transforms the kinetic energy of the wind into electrical or mechanical energy. For the conversion of wind into electrical energy, it consists of the following elements:

[0004] - a mast allowing a rotor to be placed at a sufficient height to allow its movement (necessary for horizontal axis wind turbines) or placing this rotor at a height allowing it to be driven by a stronger and more regular wind than at ground level. The mast may house some of the electrical and electronic components (modulator, control, gearbox, generator, etc.);

[0005] - a nacelle mounted at the top of the mast, housing mechanical components, tire The nacelle can rotate to orient the rotor in the correct direction;

[0006] - a rotor, fixed to the nacelle, comprising several blades (generally three) and the nose of The wind turbine. The rotor is driven by wind energy; it is connected by a mechanical shaft, either directly or indirectly (via a gearbox and mechanical shaft system), to an electric machine (electric generator, etc.) which converts the collected energy into electrical energy. The rotor may be equipped with control systems such as variable-angle blades or aerodynamic brakes;

[0007] - possibly a transmission, composed in particular of two axes (shaft (mechanical components of the rotor and mechanical shaft of the electric machine) connected by a multiplier (gearbox).

[0008] Since the early 1990s, wind energy has experienced a resurgence of interest, particularly in the European Union where the annual growth rate is around 20 This growth is attributed to the inherent potential for carbon-free electricity production. To sustain this growth, the efficiency of wind turbines and wind farms must continue to improve. The prospect of increased wind power production necessitates the development of efficient production tools and advanced control tools to enhance machine performance. Wind turbines are designed to generate electricity at the lowest possible cost.

[0009] For this power regulation, controllers are designed for variable speed wind turbines. The objectives of the controllers are to maximize the electrical power recovered, minimize rotor speed fluctuations, and minimize fatigue and extreme loads on the structure (blades, mast, and platform). Previous technique

[0010] Wind farms are subject to a phenomenon commonly called the "wake effect," where disturbances created by turbines located upstream of the wind farm create suboptimal electricity production conditions for the other turbines. Indeed, a vortex wake forms downstream of the wind turbine, and within this wake, the average wind speed is reduced because the wind turbine has captured some of the wind's kinetic energy, and the intensity of turbulence is increased.

[0011] A common strategy for maximizing the energy production of a wind turbine is to orient its rotor so that it faces the wind. The angle between the rotor and the wind direction, called the yaw angle, is then 0°. Figure 1 schematically illustrates the yaw angle in a non-limiting manner. Figure 1 is a top view of a wind turbine. The turbine comprises blades 1 and a nacelle 2, oriented in direction AA. The wind is represented by arrow U, which has a direction DD. The angle y between direction AA and direction DD is the yaw angle. When the turbine of the wind turbine is aligned with the wind direction, this angle y is zero.

[0012] In wind farms, however, applying this strategy (zero misalignment angle) to all turbines, according to a so-called "greedy" method, makes the farm subject to what is known as the wake effect: when a wind turbine extracts energy from the wind, the downstream wind speed decreases and its turbulence increases. This leads to suboptimal conditions for the power generation of the downstream turbines, with total power generation losses potentially reaching 40% offshore. A number of controllable actuators can be used to reduce this effect: power capture can be influenced by controlling the blade pitch or generator torque, and the wake of a turbine can be deflected under the downstream turbines by tilting the The rotor's angle of attack, or sideways by changing the yaw, is a technique known as wake steering. Wake steering is a wind farm-wide control strategy that typically maximizes total energy production by coordinating interactions between wind turbines. Unlike standard control strategies that aim to maximize the performance of individual turbines, wake steering sacrifices the energy production of some turbines to achieve better net performance for the entire wind farm. In addition to maximizing production, another objective is to limit or reduce structural fatigue in the wind turbines.This represents an additional compromise between the production gain, the negative impact on turbine load due to misalignment, and the positive impact of redirecting wake away from the rotors of turbines located downstream of the misaligned turbines. A positive consequence, when the compromise is favorable, is the increased lifespan of the turbines and the reduction of maintenance costs.

[0013] One strategy involves using yaw actuators to misalign the turbines with respect to the direction of the incoming wind: this allows for wake redirection to limit the impact on downstream turbines. Finding the optimal yaw angles (which maximize the total electrical power of the wind farm while reducing turbine fatigue) is a complex problem.

[0014] For this complex problem, traditional control approaches can be considered: these use a wind propagation model in a wind farm and optimize the yaw angles with respect to this approximation. Various models using analytical approximations or numerical calculations have been proposed, but such models lack accuracy and ignore turbulent wind dynamics and wake propagation, leading to poor estimation of wake effects in wind farms. Higher-fidelity models exist that take into account advection, deflection, meandering, and wake merging, but they are very costly in terms of time and computational resources, which discourages their use for real-time optimization.

[0015] It is possible to overcome these constraints by using model-free methods. Reinforcement learning (RL) is one example: these methods learn by trial and error and deduce optimal actions solely by observing a system's responses to input changes. This online learning approach is particularly interesting because of modeling uncertainties, which necessitate discarding certain suboptimal behaviors learned in the model in the field. However, combining it with a decentralized approach is not straightforward: the algorithms Decentralized algorithms limit the observability of the problem for each turbine, making their environment non-stationary.

[0016] Reinforcement learning methods have also been used for automatic production control via yaw control for a wind farm: patent application number FR 22 / 12772 illustrates such a reinforcement learning method. However, this method does not allow for an optimal compromise between the energy generated by the wind farm and the fatigue of each turbine in all wind conditions (speed and direction).

[0017] Furthermore, the method described in US patent 9201410 relates to a system and a method for optimizing a metric of a wind farm, in particular the energy generated by the wind farm, taking into account the load on the wind farm. However, this method does not allow for an optimal trade-off between the energy generated by the wind farm and the fatigue of each wind turbine under all wind conditions (speed and direction). Summary of the invention

[0018] The present invention aims to control a wind farm in real time, with a compromise between maximizing energy generation and reducing turbine fatigue, under all wind conditions. To this end, the invention relates to a method for controlling a wind farm that implements the acquisition of a wind speed and direction distribution, as well as real-time wind speed and direction data, a wind farm model, and a loading model for each turbine. Subsequently, an optimization step determines target operating points for each turbine. This optimization step optimizes the expected energy generated for the entire wind speed and direction distribution based on the expected load of each turbine for the same distribution.These target operating points (e.g., target yaw angles) are then applied to the wind turbines in the farm. Considering the expected wind speed distribution and the expected wind direction allows for a compromise to be determined between maximizing energy generation and minimizing turbine fatigue in all wind conditions. In particular, the invention makes it possible to find a balance in control between infrequent winds that can cause significant turbine fatigue and frequent winds that cause minimal fatigue.

[0019] In addition, the invention relates to a wind farm capable of implementing the control method according to the invention.

[0020] The invention relates to a method for controlling a wind farm, each wind turbine of said wind farm comprising an actuator for modifying an operating point of said wind turbine, in particular the misalignment angle of said wind turbine, the misalignment angle being the angle formed between a turbine of said wind turbine and a wind direction. For this process, the following steps are implemented: a. We acquire a distribution of wind speed and direction at the site of said wind farm, as well as the real-time wind speed and direction; b. A wind farm model is constructed, said wind farm model relating the wind speed and direction and said operating point of each wind turbine to a power generated by said wind farm, said wind farm model taking into account a wake effect; c. For each wind turbine, a loading model is constructed, said loading model linking the wind speed and direction and said operating point of each wind turbine to the loading of at least one component of said wind turbine; d. For each wind turbine, a target operating point is determined by an optimization method for the expected power generated by said wind farm, as determined by said wind farm model for the acquired wind speed distribution and wind direction. The expected load of each wind turbine, as determined by said model for each wind turbine for the acquired wind speed distribution and wind direction, is a parameter of the cost function of said optimization method to be optimized, or a constraint of said optimization method. This optimization method takes into account said wind speed and direction in real time. e. The operating point of each wind turbine is controlled by applying, by means of said actuator, the determined target operating point.

[0021] According to one embodiment, said operating point is the misalignment angle, and said optimization method is constrained by a range of variation of said misalignment angle of each wind turbine.

[0022] According to one implementation, a wind speed and direction distribution and / or real-time wind speed and direction are acquired by measurement using at least one LiDAR sensor and / or at least one anemometer and / or at least one control and data acquisition system.

[0023] Advantageously, said optimization method implements a weighted sum of the generated power and the load.

[0024] Advantageously, said optimization method implements an optimization of the power generated under a load-related constraint, in particular under the constraint that, for each wind turbine, the expected load is not greater than the overall nominal load or under the constraint that, for each wind turbine, the expected load is not greater than the maximum nominal load of all the wind turbines combined.

[0025] According to one aspect, said optimization method implements a Lagrangian-style solution, with a penalty method, in particular by means of logarithmic barriers, and possibly by means of the Uzawa algorithm.

[0026] According to one embodiment, said loading model determines an equivalent load in damage, in particular for the blades or the mast of the wind turbine.

[0027] According to one embodiment, said wind farm model is constructed by means of an aerodynamic model of said wind farm and a wake model.

[0028] According to one embodiment, said loading model is a map previously obtained by means of servo-aero-hydro-elastic modeling.

[0029] Furthermore, the invention relates to a wind farm, wherein each wind turbine in said wind farm comprises an actuator for modifying an operating point of the wind turbine, in particular the misalignment angle of said wind turbine, the misalignment angle being the angle formed between the turbine of said wind turbine and a wind direction. Said wind farm comprises computer means for implementing the method of controlling a wind farm according to one of the preceding characteristics.

[0030] Other features and advantages of the process according to the invention will become apparent from the following description of non-limiting examples of embodiments, with reference to the figures attached and described below. List of figures

[0031] [Fig.1]

[0032] Fig. 1, already described, illustrates the misalignment angle of a wind turbine.

[0033] [Fig.2]

[0034] Figure 2 illustrates the steps of the control process according to one embodiment of the invention.

[0035] [Fig.3]

[0036] Figure 3 illustrates, by way of example, an implementation of a wind farm.

[0037] [Fig.4]

[0038] Fig. 4 illustrates a wind rose (speed and direction) for the site of the wind farm example in Fig. 3.

[0039] [Fig.5]

[0040] Fig. 5 is a graph of the LED loading for each wind turbine for the example of Figures 3 and 4, the loading being respectively a nominal loading, a loading determined by a prior art method not taking into account the LED loading, and three loadings determined by three embodiments of the method according to the invention. Description of the implementation methods

[0041] The present invention relates to a method for real-time control of a wind farm. A wind farm, also called a wind farm or wind power plant, is a site comprising a plurality of wind turbines that produce electricity. Each wind turbine (also, somewhat inaccurately, called a turbine) in the wind farm includes an actuator for modifying an operating point of the turbine. An example of an operating point could be the yaw angle of the turbine. Other operating points could include, in particular, the turbine's speed setting or the modification of the turbine's power curve. The position of the wind turbines within the wind farm, also called the turbine layout or turbine implementation, is known beforehand.

[0042] In the following description, only the control of the misalignment angle is described; however, other operating points can be controlled by the method according to the invention.

[0043] In the present application, the terms upstream and downstream are defined according to the direction of the wind (an upstream wind turbine is subjected to the wind before a downstream wind turbine).

[0044] The method according to the invention comprises the following steps: 1. Acquisition of a wind speed and direction distribution 2. Construction of a model of the wind turbine farm 3. Construction of a loading model 4. Determining a target operating point 5. Monitoring of each wind turbine

[0045] Steps 2 to 4 can be carried out by computer means, in particular a computer, a processor or a calculator. The steps are detailed in the remainder of the description.

[0046] Figure 2 schematically and non-limitingly illustrates the steps of the wind farm control process according to one embodiment of the invention. A wind speed and direction distribution at the wind farm site is acquired (ACQ), as well as the current wind speed and direction (Vac). A model of the wind farm (MOD F) is constructed that determines the energy generated by the wind farm as a function of the wind speed and direction, and the operating point of each wind turbine. For each wind turbine, a model of Loading (MOD C) determines the load on each wind turbine based on wind speed and direction, and the operating point of each turbine. Then, using an optimization method (OPT), a target operating point is determined based on the current wind speed and direction (Vac). The optimization (OPT) implements both models (MOD F, MOD C) and the acquired wind speed and direction distribution (ACQ). Finally, the wind turbines in the farm are controlled (CON) based on the target operating point and the current wind speed and direction (Vac).

[0047] 1) Acquisition of a wind speed and direction distribution

[0048] During this step, a distribution of the velocity and direction of the Wind at the wind farm site. The wind speed and direction distribution is a historical record of wind speed and direction, which can be acquired for a predetermined period, for example, at least one year to account for seasonal weather patterns. A distribution is a statistical breakdown of wind speed and direction. This distribution can be graphically represented by a polar histogram, also called a wind rose. Figure 4 is an example of a wind rose applied to the scenario described below. This wind speed and direction distribution allows us to determine the occurrences of wind configurations (speed, direction) at the wind farm site and to deduce the probabilities of these wind configurations. Furthermore, during this step, we acquire the current wind speed and direction.In other words, we acquire the wind configuration (speed, direction) in real time, in order to apply real-time control.

[0049] According to one aspect of the invention, the data acquisition system can provide statistics on quantities of interest, such as the mean or joint standard deviation, obtained periodically and calculated over a predetermined period. By way of example, the predetermined period can be between one minute and one hour, preferably between five and twenty minutes. It is thus possible to acquire time series of average wind speed and direction over a defined period, for example, one or more years. This dataset allows for the deduction of a statistical distribution of wind speeds and directions, which can be visually represented by a wind rose displaying a distribution associated with a classification (binning) of wind conditions (direction, speed, number of occurrences).

[0050] According to one embodiment, wind speed and direction can be measured (whether for historical or real-time purposes), in particular by means of at least one LiDAR sensor (an acronym for "light detection and ranging," which can be translated as laser remote sensing), and / or at least one Anemometer, and / or measurements using at least one SCADA (Supervisory Control and Data Acquisition) real-time control and data acquisition system, or any similar sensor. A SCADA real-time control and data acquisition system is a large-scale remote management system that processes a large number of telemetry measurements in real time and remotely controls technical installations. It is an industrial technology in the field of instrumentation, the implementations of which can be considered as instrumentation structures including a middleware layer. From these measurements, the undisturbed wind speed at the inlet of the wind farm can be deduced, the inlet of the wind farm being defined according to the wind direction.

[0051] 2) Construction of a model of the wind turbine farm

[0052] In this step, a model of the wind farm is constructed. The wind farm model links the upstream wind speed and direction, the operating point of each wind turbine, and its position within the wind farm, to the power generated by the wind farm. In other words, the wind farm model takes as inputs the upstream wind speed and direction, as well as the location, dimensions, and operating point of each wind turbine, and outputs the power generated by the wind farm. Furthermore, the wind farm model is representative of the wake effects induced by the operation of the wind turbines. In other words, the wake farm model allows the modeling of the wake effect. In this way, the model makes it possible to determine the operation of each wind turbine, even when a wind turbine is located in the wake of an upstream wind turbine.Thus, this model is representative of the physical phenomena at play within a wind farm, in particular aerodynamic, and possibly aeroelastic, phenomena.

[0053] According to one embodiment of the invention, the power generated by the wind farm, as determined by the wind farm model, can be the annual energy production (AEP). Thus, the method according to the invention makes it possible to optimize the energy generated by the wind farm over a year, and in this way to properly take into account seasonal climatic variations.

[0054] According to one embodiment of the invention, the wind farm model can be constructed using an aerodynamic model of the wind farm and a wake model. For example, the wind farm model can be constructed using a wind farm simulator to calculate the aerodynamic characteristics of the wind farm. In particular, the modeling of the wind turbines, especially their rotor, can exploit the geometry and aerodynamic profile of the blades, as well as maps of power and thrust coefficients as a function of wind speed. The wake model can be included in the wind farm simulator to model the aerodynamic interactions of the wind turbines. The wake model can, for example, be a super-Gaussian model, or any similar model. Then, a local superposition of the sum of the wakes can be implemented, among other possibilities, to represent the combined wake effects of several successive wind turbines. 3) Construction of a loading model

[0055] During this step, a load model is constructed for each wind turbine. The load model links the wind speed and direction and the operating point of the wind turbine to a load on at least one component (e.g., blades, transmission, actuators or tower, or anchors and moorings for floating wind turbines) of the wind turbine. In other words, the load model takes as inputs the wind direction and speed as well as the operating point of the wind turbine, and has as output the wind turbine load. A load is defined as a load on a wind turbine, this load generating fatigue in the wind turbine component.

[0056] According to one embodiment of the invention, the load can be a Damage Equivalent Load (DEL). The DEL can be defined as the amplitude of a sinusoidal load of a specified frequency around a specified fixed average load that would affect the ability of a structure to withstand the evaluated load.

[0057] Advantageously, the loading model can determine the load density (LD) of the wind turbine blades. This allows for a representation of the fatigue of the structural elements of interest in the wind turbine. Alternatively, the loading model can determine the load density (LD) of the tower, the transmission, or the actuators of the wind turbine. In the case of evaluating floating offshore technologies, the loading model can determine the load density (LD) of the float, moorings, and anchors of the wind turbine.

[0058] According to one aspect of the invention, the loading, in particular the LED loading, can be determined by means of a response surface, or a multidimensional map, preferably synthesized using coupled servo-aero-hydro-elastic simulations.

[0059] 4) Determination of a target operating point

[0060] In this step, a target operating point (for example, the target misalignment angle) is determined for each wind turbine using an optimization method. The target operating point corresponds to an operating point setpoint to be applied to the wind turbine. The optimization method implements the wind turbine farm model built in step 2, the loading model built in Step 3, along with the wind speed and direction distribution acquired in step 1, is used. To do this, the optimization method is applied to the expected power generated, determined by the wind farm model for the entire acquired wind speed and direction distribution, as a function of the expected load of each wind turbine, determined by the loading model for the entire acquired wind speed and direction distribution. The expression "as a function of an expected load" means that the expected load is a parameter of the cost function to be optimized, or a constraint of the optimization method. Recall that the expected value of a random variable corresponds to the average of the possible values ​​weighted by the probabilities associated with those values. The probabilities considered for the expected values ​​correspond to the probabilities derived from the wind speed and direction distribution.In this way, the optimization method depends on the expected distribution of wind speed and direction. In particular, the invention makes it possible to find a balance in control between infrequent winds that can cause significant strain on wind turbines and frequent winds that cause little strain on the turbines. Once the optimization is performed, the target operating point is determined for the current acquired wind conditions (current wind speed and direction).

[0061] For the embodiment in which the operating point is the misalignment angle, the optimization method can be constrained by the range of variation of the misalignment angle of each wind turbine. The range of variation of the misalignment angle is the interval within which the misalignment angle can vary for each wind turbine. This range of variation is limited by a minimum and a maximum bound for the misalignment angle. This constraint prevents the process from determining an unattainable operating point, thus ensuring faster optimization requiring fewer computing resources, thanks to a limitation of the optimization domain.

[0062] Advantageously, the optimization method can be applied to a discrete probability optimization problem. Indeed, the optimization problem allows us to calculate the optimal yaw angles of the wind turbine farm for all wind configurations. If we denote w as the random wind variable defined by the wind speed and wind direction wd, Nw as the number of acquired wind speed and direction data points, and (w,) ,_j y as the sequence of acquired wind speed and direction data points, we can write:

[0063] With P!V the expected production generated by the wind farm, pw the pro probability of occurrence of wind w (from the distribution) and <5Wj the power produced by the wind farm for the wind configuration w;.

[0064] We can define & by: $ ...... . ... '.....

[0065] With S being the vector of misalignment instructions for the wind turbines in the wind farm. The expected value of the function f can then be defined as:

[0066] With E^ 1' expectation of with respect to the random variable w.

[0067] According to one embodiment of the invention, the solution of the optimization method can be in Lagrangian form, with a constraint penalization method, in particular by means of logarithmic barriers, or any similar penalization method. Furthermore, the solution can be obtained using Uzawa's algorithm (which is a fixed-step gradient descent algorithm with projection for iteratively solving a dual problem), or any similar algorithm.

[0068] According to a first embodiment of the invention, the optimization method can be based on a weighted sum of the power generated (determined by the wind farm model) and the loading (determined by the loading model).

[0069] According to an example of this first embodiment, the weighted sum optimization model can be used to solve the following problem: r • ... . ; 1' " / . -------- . i = ' ^4- $ • ' P ' '—' .j&wF WK

[0070] Subject to the following constraint: t> <>» . P , ' v

[0071] With w the random variable of the wind defined by the wind speed and the wind direction w<\ E,^ the expectation of with respect to the random variable w, 6 the vector of misalignment instructions for the wind turbines in the wind farm, d and 0+ the minimum and maximum variation limits of the misalignment, 0 corresponding to no misalignment of the wind turbines in the wind farm, Jws the weighted sum of the two objective functions, P the power produced by the wind farm, t the index of the wind turbine in the farm, NT the number of wind turbines in the wind farm, F the wind turbine load, and a the weighting function between 0 and 1 and set to the maximum value that satisfies: ® Ht■ -fe-ïÿ W j h.

[0072] With 0* the operating point (for example the misalignment angle) solution of the optimization problem. Such a formulation makes it possible to obtain a bi-objective optimization problem, expressed as a single objective, by a standard weighted sum scalarization approach.

[0073] According to one embodiment, the constraints of this optimization problem can be addressed using logarithmic barriers. We can then write: — < — k'? -h s J • •X' ; b '1 Ht <*

[0074] It appears that this problem is separable in the domain of Bj- The problems obtained are then independent and can be solved in parallel.

[0075] According to a second embodiment, the optimization method can implement an optimization of the expected power generated for the entire wind speed and direction distribution under a constraint related to the expected load for the entire wind speed and direction distribution.

[0076] According to a first variant of this second embodiment, the optimization method can be a maximization of the expected power generated for the entire wind speed and direction distribution, subject to the constraint that, for each turbine, the expected load is not greater than the expected nominal load (which corresponds to the load without misalignment of the wind turbines). We can then write:

[0077] under the following constraints: < KjpFW, wl / .h «A r~ bSF » Æ'H&t V, «. » tf i » *

[0078] with w the random variable of the wind defined by the wind speed M4 and the wind direction wd, Eh.ç? the expectation of P with respect to the random variable w, 0 the vector of misalignment instructions of the wind turbines of the wind farm, 0 and #+ the minimum and maximum variation bounds of the misalignment, 0 corresponding to no misalignment of the wind turbines of the wind farm, P the power produced by the wind farm, NT the number of wind turbines in the wind farm, F the wind turbine load.

[0079] These constraints limit the permissible loading level to the nominal level (which corresponds to the loading without misalignment of the wind turbines) for each turbine of the farm considered individually.

[0080] For this first variant of the second embodiment, a solution using logarithmic barriers can be implemented. The constraints can be treated within a duality framework. Logarithmic barriers may not be used for the loading constraints, as it can be difficult to initialize the yaw angles in such a way that the constraints are strictly satisfied. The penalized Lagrangian j can be defined as follows: 4.. [ V;- 4

[0081] With 2 being the Lagrange multiplier. We can then use Uzawa's classical algorithm, based on duality, to solve this problem.

[0082] An innovative feature of using Uzawa's algorithm in the context of the invention is that the weighting table can be updated based on the expected load score (e.g., DEL) obtained at the last iteration. In particular, the parameter table z7 can be obtained from a set of linear functions representing the differences between the DEL scores obtained with optimization and the nominal DEL scores. These differences are obtained at each iteration of the algorithm for the wind distribution under study, namely: At max {0; 52.( / ¼. [R7¾) — F(ü. 7 / ¾)J}

[0083] According to a second variant of this second embodiment, the optimization method can be a maximization of the expected power generated for the entire wind speed and direction distribution, subject to the constraint that, for each turbine, the expected load is not greater than the expected maximum nominal load, for all wind turbines combined (which (corresponds to the maximum load of one of the wind turbines in the farm without misalignment of the turbines). We can then write: æM «w ), w 1 "I

[0084] under the following constraints: £ v.

[0085] with w the random variable of the wind defined by the wind speed and the wind direction EH,(p expectation of P with respect to the random variable w, 9 vector of the misalignment instructions of the wind turbines of the wind farm, 6 and 0+ the minimum and maximum variation limits of the misalignment, 0 corresponding to no misalignment of the wind turbines of the wind farm, P the power produced by the wind farm, NT the number of wind turbines in the wind farm, F the load of the wind turbine.

[0086] For a second variant of the second embodiment, a solution using logarithmic barriers can be implemented. The constraints can be treated within a duality framework. Logarithmic barriers may not be used for the loading constraints, as it can be difficult to initialize the yaw angles in such a way that the constraints are strictly satisfied. In other words, for a given wind configuration (speed and direction), when wake redirection is applied, the expected load cannot exceed a given level. This given level is that reached by the wind turbine experiencing the most structural fatigue under standard operating conditions, i.e., the one with the maximum expected load for the same wind configuration.

[0087] We can then introduce the value F+ such that: rrm^E^FîO, w) wh fel

[0088] The Lagrangian can then be written: L t- :

[0089] We can then use Uzawa's classical algorithm, based on duality, to solve this problem. 5) Monitoring of each wind turbine

[0090] During this step, each wind turbine is controlled by applying the desa-

[0091]

[0092]

[0093]

[0094]

[0095]

[0096]

[0097] The target alignment (or target operating point) is determined in step 4. For this step, the turbine's operating point actuator is controlled for each wind turbine. Specifically, the turbine's misalignment angle actuator can be controlled. According to one embodiment, the control of the misalignment angle can correspond to a control at a precise value of the misalignment angle. Furthermore, the invention relates to a wind farm. Each wind turbine in the wind farm includes an actuator for modifying the turbine's misalignment angle or operating point. In addition, the wind farm includes computing resources, specifically a computer, processor, or calculator, for implementing the calculation steps of the control process according to any of the variants or combinations of variants described below. The wind farm is therefore controlled by these computing resources. In particular, the computing resources enable: - Acquire the distribution of wind speed and direction, - To build and implement a model wind turbine farm, - Build and implement a loading model (in particular equivalent in fatigue), - Develop an optimization method to determine a target operating point for each wind turbine, and - Check the misalignment angle of each wind turbine or the operating point. The IT systems can be centralized: the wind farm has a single computer unit that implements the control process steps and communicates with at least all the misalignment angle actuators. Alternatively, each wind turbine has its own computer unit, with all computer units communicating with each other. According to one embodiment, the wind farm may include a wind measurement sensor, in particular a LiDAR sensor or an anemometer. According to one aspect of the invention, the wind farm may include SCADA measurement means. According to one embodiment option, the wind farm may include means of communication, in particular to transmit the data acquired in stage 1 and / or to communicate the target yaw angles to the controllers. Examples The characteristics and advantages of the process according to the invention will become clearer upon reading the application example below.

[0098] For this example, we consider a wind farm comprising seven wind turbines. Figure 3 illustrates the implementation of the wind turbines as points in an (X; Y) coordinate system in km. This near-aligned arrangement of the wind turbines generates a strong wake effect. The wind turbines in this farm have a rotor diameter of 82 m and a rated power of 2 MW. Furthermore, wind speed and direction measurements were acquired over two years (in 2017 and 2018). These measurements were obtained from SCADA data of the wind farm. These wind speed and direction measurements allow us to determine the wind distribution in the form of a wind rose as illustrated in Figure 4.In this figure, for each wind direction (expressed in a cardinal coordinate system), there is a triangle whose height illustrates the occurrence of that wind direction during the measurement period (in other words, the larger the triangle, the more frequently that wind direction was measured during the measurement period). Furthermore, each wind direction is divided into several grayscale segments to illustrate the wind speed distribution. Thus, each segment of the triangle corresponds to a wind speed. The wind speed scale is indicated in the figure in m / s.

[0099] For the simulations, a wind farm simulator is used, which employs a BEM (Blade Element Momentum) code to calculate the aerodynamic characteristics. The modeling of the wind turbines, particularly their rotors, utilizes the geometry and aerodynamic profile of the blades, as well as maps of the power and thrust coefficients as a function of wind speed. The wind farm simulator also models the aerodynamic interactions of the wind turbines: the wake effects. The wake model is a super-Gaussian model: the parameters of this super-Gaussian wake are a function of the ambient turbulence intensity and the operating point of the wind turbine, which defines the thrust coefficient. A local superposition of the linear sum of the wakes is used to implement the superposition of the wake effects of several successive wind turbines.The wake deflection linked to a misalignment of a wind turbine with respect to the average wind direction is represented by the Jimenez model.

[0100] To determine damage equivalent loads (DELs), a model was developed incorporating a multi-dimensional (6D and 5D) mapped surface. The DEL load model is therefore a response surface resulting from a parametric study of blade root fatigue. This analysis depends on wind speed, misalignment angle, turbulence intensity, the overlap ratio of the downstream wind turbine with the footprint of the upstream wind turbine, the distance between the upstream and downstream wind turbines, and the generation of random signals with distinct seeds to perform several independent realizations. The approach put in high-fidelity servo-aero-elastic modeling tools (e.g., DEEPLINES WIND™ or FAST™) are used to inform the multidimensional response surface.

[0101] For this example, several criteria are defined: - Annual drinking water supply energy production: The annual energy production gain: s cor- Gays AEP '*, 100 (ÀEP(^) — AEPUL i / AEPUP responding to , with 0 being the angle of descent optimal alignment for wind configuration w;> The gain from optimizing LED charging: Gain DELpO -hlü | < F;#{■«?),— A<- fRL -Gh j ​​ / -A- iFdf 1*7 The damage ratio: with m the coefficient of DJ / DELiY^ ëë VdelJ

[0102]

[0103]

[0104] Wôhler, which is equal to 4 for the blade, or 10 for the mast. For this example, the following optimizations are applied: - An optimization of the prior art without misalignment of the wind turbines, denoted NAME, - An optimization solely of the power generated from the prior art with misalignment of the wind turbines, without considering fatigue, noted AA, - An optimization according to the first embodiment of the invention, denoted INV1, - An optimization according to the first variant of the second embodiment of the invention, denoted INV2, and - An optimization according to the second variant of the second embodiment of the invention, denoted INV3. Figure 5 illustrates the value of the DEL load in Nm for each wind turbine T numbered from 0 to 6. It is noted that, regardless of the embodiment of the invention, the method according to the invention makes it possible to reduce the load, and therefore the fatigue, of six of the seven wind turbines in the wind farm. Table 1 compares the process according to the invention with the nominal case.

[0105] [Tables 1] Gains Optimization INV1 Optimization INV2 Optimization INV3 AEP (%) 0.925 1.105 1.104 DEL (%) -3.330 -1.764 -1.847 Damage Ratio (%) 28 16 17

[0106] Both embodiments of the invention make it possible to increase the power generated by the wind farm, while reducing the load on the wind turbine blades, with an improved damage ratio. Thus, the method according to the invention makes it possible to increase energy production while limiting wind turbine fatigue.

Claims

Demands

1. A method for controlling a wind farm, each wind turbine of said wind farm comprising an actuator for modifying an operating point of said wind turbine, in particular the misalignment angle (V) of said wind turbine, the misalignment angle (T) being the angle formed between a turbine of said wind turbine (2) and a wind direction (U), characterized in that the following steps are implemented: a. We acquire (ACQ) a distribution of wind speed and direction on the site of said wind farm, as well as the real-time wind speed and direction (Vac); b. A wind farm model (MOD F) is constructed, said wind farm model (MOD F) relating the wind speed and direction and said operating point of each wind turbine to a power generated by said wind farm, said wind farm model (MOD F) taking into account a wake effect; c. For each wind turbine, a loading model (MOD C) is constructed, said loading model (MOD C) relating the wind speed and direction and said operating point of each wind turbine to the loading of at least one component of said wind turbine; d. For each wind turbine, a target operating point (OPT) is determined by an optimization method (OPT) of the expected power generated by said wind farm as determined by said wind farm model (MOD F) for the acquired wind speed distribution and direction, the expected load of each wind turbine as determined by said load model (MOD C) for each wind turbine for the acquired wind speed distribution and direction being a parameter of the cost function of said optimization method to be optimized, or a constraint of said optimization method, and said optimization method (OPT) taking into account said wind speed and direction in real time (Vac); and e. The operating point of each wind turbine is controlled (CON) by applying, using said actuator, the determined target operating point.

2. A method according to claim 1, wherein said operating point is the misalignment angle (F), and said optimization method is constrained by a range of variation of said misalignment angle (X) of each wind turbine.

3. A method according to any one of the preceding claims, wherein a wind speed and direction distribution and / or real-time wind speed and direction (Vac) is acquired (ACQ) by measurement using at least one LiDAR sensor and / or at least one anemometer and / or at least one control and data acquisition system.

4. A method according to any one of the preceding claims, wherein said optimization method (OPT) implements a weighted sum of the generated power and the load.

5. A method according to any one of claims 1 to 3, wherein said optimization method (OPT) implements an optimization of the power generated under a load-related constraint, in particular under the constraint that, for each wind turbine, the expected load is not greater than the overall nominal load or under the constraint that, for each wind turbine, the expected load is not greater than the maximum nominal load of all the wind turbines combined.

6. A method according to any one of the preceding claims, wherein said optimization method (OPT) implements a Lagrangian-form solution, with a penalty method, in particular by means of logarithmic barriers, and optionally by means of the Uzawa algorithm.

7. A method according to any one of the preceding claims, wherein said loading model (MOD C) determines an equivalent damage load, in particular for the blades or the mast of the wind turbine.

8. A method according to any one of the preceding claims, wherein said wind farm model is constructed by means of an aerodynamic model of said wind farm and a wake model.

9. A method according to any one of the preceding claims, wherein said loading model (MOD C) is a map previously obtained by means of servo-aero-hydro-elastic modeling.

10. Wind farm, wherein each wind turbine in said wind farm comprises an actuator for modifying an operating point of the wind turbine, in particular the misalignment angle (F) of said wind turbine, the misalignment angle (Y) being the angle formed between the turbine of said wind turbine and a wind direction, characterized in that said wind farm includes computer means for implementing the method of controlling a wind farm according to one of the preceding claims.

Images