Method for obtaining incident wind field and modeling it using lidar sensor

Abstract

A subject of the present invention relates to a method for detecting outliers of an incident wind field in a space upstream of a lidar sensor, characterized in that the method comprises: a step of acquiring a measurement value rws(k) of the incident wind field using a lidar sensor and modeling it, a step of estimating in real time a median mr(k) and a mean absolute deviation dr(k) of the measurement value of the incident wind field, and a step of detecting outliers in real time using the estimated median mr(k) and the mean absolute deviation dr(k).

Description

Technical Field

[0001] The present invention relates to the field of Lidar (Light Detection and Ranging) sensors used as remote sensing devices for measuring wind speed. It also relates to the field of wind turbines equipped with Lidar sensors and to the control of such wind turbines. Background Art

[0002] The detection of outliers is ubiquitous in many data processing tasks, covering a wide range of areas such as signal processing, industrial control, etc. A good definition of an outlier is a measurement that differs so much from the other measurements that it is suspected to be caused by a different mechanism.

[0003] In the field of wind energy, the detection and observation of light using lidar has been recognized over the past few years as a reliable and viable remote sensing technology for measuring and predicting wind speed.

[0004] Specifically, recent advances in LiDAR technology have facilitated the deployment of LiDAR in real-time control applications. LiDAR is able to provide measurements characterizing upstream wind volume with high spatiotemporal resolution. Patent application FR 3013777 (US2015145253) describes such an application. Using LiDAR, measurements that differ significantly from the normal range of detected data are considered outliers. These outliers may be caused by sensor errors or data transmission errors.

[0005] Wind forecasts related to measurements are highly dependent on the quality of LiDAR measurements. Since LiDAR measurements inevitably contain abnormal data, including erroneous data generated due to reasons such as device failure or valid data representing special circumstances such as adverse meteorological conditions, the forecast is affected and is usually not optimal.

[0006] Furthermore, wind reconstruction includes estimated wind speeds and confidence intervals, and typically both rely on the quality of the LiDAR measurements.

[0007] Therefore, the detection of outliers provides valuable insights needed to make LiDAR data applications viable.

[0008] Motivated by the need to collect more useful information from LiDAR data, the present invention aims to mitigate the above mentioned drawbacks and provide a method that allows outliers to be removed in real time. The approach has been validated with real measurement data for a LiDAR system mounted on a nacelle. Summary of the invention

[0009] To this end, a first aspect of the invention relates to a method for detecting anomalies of an incident wind field in a space upstream of a lidar sensor, and the method is characterized in that it comprises:

[0010] a) The steps of obtaining the incident wind field measurement rws(k) using a lidar sensor and modeling it;

[0011] b) a step of estimating in real time the median mr(k) and the mean absolute deviation dr(k) of the measured values ​​of the incident wind field; and

[0012] c) a step of detecting outliers in real time using the estimated median mr(k) and mean absolute deviation dr(k), said detection step being performed using a formula of the following type:

[0013] |rws(k)-mr(k)|≤σdr(k), where σ is a positive scalar.

[0014] According to one aspect of the invention, the real-time mean absolute deviation dr(k) of the incident wind field is given by a formula of the following type:

[0015]

[0016] According to one implementation of the invention, the method comprises a step of reconstructing the measurements of the lidar sensor by removing the detected outliers from the modelled measurements rws(k).

[0017] Therefore, a clean measurement signal without outliers is obtained.

[0018] According to one aspect of the present invention, a method for acquiring an incident wind field in a space upstream of the lidar sensor using a lidar sensor and modeling the incident wind field comprises the following steps:

[0019] a) A step of generating a grid of the space located upstream of the lidar sensor, wherein in this step, a grid of the space is generated using a set of discrete points located in a predefined three-dimensional grid, which includes a set of cells consisting of estimated points and measured points.

[0020] The grid generation step allows the space upstream of the LiDAR sensor to be discretized (or sampled) to form a three-dimensional grid composed of discrete points, and it is possible to make these different discrete points coincide with the measured points or estimated points required by the modeling method. In addition, this allows the measured points and the estimated points to be located relative to each other, and the distance separating all these discrete points is known.

[0021] b) a step of measuring the amplitude and direction of the wind along at least three measurement axes at various measurement points located in the upstream space and at at least two different distances from the lidar sensor,

[0022] The measurements performed in this step allow obtaining initial data reliable enough to feed the algorithm aimed at estimating the amplitude and direction of the wind at the estimation point.

[0023] c) a step of estimating the amplitude and direction of the wind at all estimation points at any time, the estimation being carried out by means of optimizing a cost function J using a weighted recursive least squares method, the cost function J using at least the data of the measurement points, the spatial wind speed coherence data, the temporal wind speed coherence data and data identifying the quality of the measurements performed at the measurement points.

[0024] Taking into account these various parameters in the cost function to be optimized will allow achieving an estimation of the amplitude and direction of the wind at each estimation point of the grid.

[0025] d) The step of reconstructing the incident wind field in three dimensions (3D) in real time and in a defined coordinate system from the estimated and measured wind amplitude and direction for each point.

[0026] This step allows the incident wind field to be reconstructed in 3D in a volume sampled by a three-dimensional grid. In this step, a history of LiDAR measurements is generated, which allows the past state of the wind field to be known, and the history is incorporated into a synthesis of current and future estimates of the 3D wind field, which allows reconstruction in real time.

[0027] An advantage of using an optimization method, using a recursive form of a weighted least squares technique, is that this makes it possible to determine a three-dimensional (3D) complete picture of the incident wind propagating through the space upstream of the lidar sensor.

[0028] According to one aspect of the invention, the measured values ​​m of the amplitude and direction of the wind at the measuring point are given by the following relationship:

[0029] m j，x (k) = a j v j，x (k)+b j v j，y (k)+c j v j，z (k)

[0030] Herev j,x (k),v j,y (k),v j,z (k) is the wind speed value projected into the coordinate system x, y, z at the initial time (k), and a j ,b j ,c j (where j = 0, 1, 2, 3, 4) is the measurement coefficient, which is given by

[0031]

[0032] Here θj, are the zenith and azimuth of the measurement axis in the spherical coordinate system.

[0033] In this way, at each sampling time, for all points in space, the wind vector consists of three components that will allow a complete image to be determined in three dimensions. Furthermore, these measurement coefficients depend only on the angle of the beam and not on the measurement distance, which facilitates the computational programming of the cost function J.

[0034] According to one aspect of the invention, the cost function J at any time (t) is written as follows:

[0035]

[0036] Here ω is an ordered vector consisting of all velocity components at the point in space where the wind is estimated, is the estimate of wind speed at time 0, P 0 ,Q,R s and R m is a weight matrix of appropriate size, and C s ,C m is a matrix that takes into account wind speed and measurement noise.

[0037] Using this cost function, it is possible to estimate the wind speed at the estimation point. In addition, this function enables the implementation of the weighting matrix P 0 ,Q,R s and R m A clear explanation is possible.

[0038] According to one aspect of the invention, the measurement of the amplitude and direction of the wind at each measurement point is performed at a sampling rate of at least 0.25 Hz. The use of such a sampling frequency range results in multiple measurements being obtained simultaneously on a given measurement axis, but these measurements are reliable and accurate.

[0039] According to one aspect of the invention, the measurement of the amplitude and direction of the wind at various measurement points is performed at at least two different distances along the measurement axis. Performing the measurements at at least two distances allows a three-dimensional volume to be defined that is sufficient to encompass the blades of a wind turbine, as will be described below.

[0040] According to an aspect of the invention, the amplitude and direction of the wind are measured along at least three measurement axes. The use of at least three measurement axes enables the generation of a fine grid of the upstream space and also allows obtaining a sufficient number of measurements for the step of estimating the wind speed.

[0041] According to one aspect of the invention, the spatial coherence of the wind speed along the x, y and z axes of a Cartesian coordinate system is estimated using a formula of the following type:

[0042] C s ω≈0

[0043] in

[0044] Where:

[0045] οC l represents the wind speed variation for the estimation domain along the longitudinal axis x and

[0046] οC t represents the wind speed variation for the estimation domain along the horizontal axis y and

[0047] οC v Characterizes the wind speed variation for the estimation domain along the vertical axis z.

[0048] The effect of this representation is that it is possible to computationally encode such functions.

[0049] According to one aspect of the invention, the spatial coherence of wind speed along the x, y and z axes of a Cartesian coordinate system is estimated under the following assumptions:

[0050] The wind speed along the longitudinal axis x varies little, and the partial derivative dv along the longitudinal axis x / dx is relatively small,

[0051] The wind varies continuously along the horizontal axis y, and the partial derivative dv along the horizontal axis y x / dy is small,

[0052] ο The wind varies along the vertical axis z according to a power law, which is given by:

[0053]

[0054] Here α is the exponent of the power law, v l is the longitudinal wind at altitude z above the ground, and z r is the reference altitude.

[0055] Such an assumption is realistic and allows obtaining reliable and accurate wind speed estimates.

[0056] According to one aspect of the invention, the quality of the measurements performed by the lidar is represented using a model of the following form:

[0057] C m ω=m m +∈ m

[0058] Here E m Describes the measurement noise.

[0059] This type of model allows taking into account the inaccuracies of LiDAR measurements.

[0060] According to one aspect of the invention, the estimates of the amplitude and direction of the wind field at all estimation points at time (t) are given by:

[0061] ω(t)=ω(t-1)+K(y(t)-Cω(t-1))

[0062] The advantage of the above formula is that it can link wind speed estimates for estimation points over time.

[0063] The present invention also relates to a computer program product, which comprises code instructions arranged to implement the steps of the above method for detecting outliers. The program is executed by a processing unit of a laser radar.

[0064] The invention also relates to a lidar sensor comprising in a memory code instructions of a computer program product as described above and arranged to execute such a computer program product.

[0065] In this way, a lidar sensor executing such a computer program product will return reliable information about the incident wind field in three dimensions and in real time.

[0066] A subject of the invention also relates to a wind turbine comprising a lidar sensor as described above.

[0067] According to an aspect of the invention, a lidar sensor is placed on a nacelle of the wind turbine.

[0068] Finally, the present invention also relates to a method for controlling and / or monitoring a wind turbine equipped with a LiDAR sensor and a programmable logic controller, the method comprising the following steps:

[0069] a) the step of generating an anticipated control strategy for controlling said wind turbine, utilizing the detection of outliers achieved by reconstructing the incident wind field in three dimensions and in real time,

[0070] b) A step of performing control in conjunction with the generated control strategy, which includes controlling the angle of the blades or the orientation of the nacelle.

[0071] Providing sufficiently robust and accurate information about the state of the incident wind adjacent to the rotor allows a new way of control combining dynamic and preventive pre-positioning items. Furthermore, the ability to reconstruct the incident wind field adjacent to the rotor plane online in a real-time manner opens up many potential applications: quantification of wind turbine misalignment, power curves, transfer functions of the nacelle, detection of gusts, monitoring and diagnosis of loads and of fatigue risks, optimization of preventive maintenance, resource analysis, optimization of production. This therefore allows to increase the efficiency of wind turbines through design optimization, reduce maintenance costs, extend the life of components and reduce investment costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Other characteristics and advantages of the method according to the invention will become apparent on reading the following description of a non-limiting example embodiment with reference to the attached drawings.

[0073] [ Figure 1 ]

[0074] Figure 1 Illustrated are lidar measurements for one beam over the course of a day.

[0075] [ Figure 2 ]

[0076] Figure 2 Illustration of lidar measurements for one beam over the course of a day after outliers have been removed.

[0077] [ Figure 3 ]

[0078] Figure 3 A wind turbine equipped with a lidar sensor according to the present invention is illustrated.

[0079] [ Figure 4 ]

[0080] Figure 4 The steps of the acquisition and modeling method using a laser radar sensor according to the present invention are illustrated.

[0081] [ Figure 5 ]

[0082] Figure 5 is a front view of a grid of spaces according to the present invention.

[0083] [ Figure 6 ]

[0084] Figure 6 is a perspective view of a grid of spaces according to the present invention.

[0085] [ Figure 7 ]

[0086] Figure 7The wind field in 3D reconstructed from lidar measurements in one specific case is illustrated.

[0087] [ Figure 8 ]

[0088] Figure 8 The steps of a method for controlling a wind turbine according to the invention are illustrated. DETAILED DESCRIPTION

[0089] mark

[0090] In this manual, the following notations are used:

[0091] x,y,z: Direction of a three-dimensional coordinate system, where z is the vertical axis and x is the main direction of the wind.

[0092] θ and The orientation angles of the laser radar sensor are as follows: Figure 1 As shown: Angle θ is the angle formed by the projection of the laser radar's measurement axis in the plane (y, z), and is the angle made by the lidar's measurement axis in the plane formed by the axis x and the projection of the lidar's measurement axis in the plane (y,z).

[0093] m(t): The measurement value of the lidar sensor at the measurement point.

[0094] V j,x (k),v j,y (k),v j,z (k): Projection of wind speed on x, y, and z.

[0095] ω: An ordered vector consisting of all the components of the wind speed at the point in space where the wind is estimated, on the axes x, y and z of the three-dimensional coordinate system.

[0096] An estimate of (t) at time t.

[0097] P(t): The auxiliary matrix array that varies with time and can be obtained at time t.

[0098] P 0 ,Q,R s and R m is a weighting matrix of appropriate size.

[0099] Σ is a positive scalar.

[0100] Rp: Vector

[0101] mr(k): median

[0102] In the rest of the description, the term "lidar" is used to denote a lidar sensor.

[0103] A method for real-time outlier detection based on online estimation of median and mean absolute deviation is presented. The proposed method has been validated using real LiDAR measurement data, showing that outliers can be detected and must be removed from LiDAR measurements.

[0104] 5-beam pulsed cabin lidar (such as Figure 3 The laser radar (shown in FIG. 1 ) measures the wind speed component corresponding to the wind speed projected along the direction of the laser beams, namely beam b0, then beam b1, beam b2, beam b3, and finally beam b4. An advantageous feature of the laser radar is that it can measure the projection of wind speed at multiple distances along each beam (e.g., at distances from 50 to 200 m) at a sampling rate of 4 Hz. At each sampling time, the measurement value of only one beam can be obtained.

[0105] like Figure 1 As shown in the figure, which shows the lidar measurements for one beam during a day without using a detection method, it can be seen that the lidar does not always provide measurements (specifically, this effect can be seen in [6×10 4 ;7×10 4 ]), and, in most cases, the lidar measurements deviate significantly from the normal range. These measurements are considered outliers.

[0106] As stated in the introduction, the purpose of the present invention is to detect and remove outliers from LiDAR measurements.

[0107] To this end, the first step of the method for real-time detection of outliers includes the steps of obtaining the measured value rws(k) of the incident wind field with a lidar sensor and modeling it. This step will be described in more detail in the rest of the specification.

[0108] The second step performed with LiDAR involves estimating in real time the median and mean absolute deviation of the LiDAR measurements, which will then be used to detect outliers.

[0109] In statistics, the median is the value that separates the lower half of a data sample from the upper half. For a data set, it can be thought of as the middle value. For example, in the data set {1,3,3,6,7,8,9}, the median is 6, the fourth highest and fourth lowest number in the sample. The main advantage of the median in describing data relative to the mean is that it is not distorted by extremely high or low values, which allows for a better understanding of typical values. In other words, the median is much more robust to outliers than the mean. In the case of symmetrical data, the mean and median are equal.

[0110] The present invention includes a process for calculating the median of 2nm+1 lidar measurements.

[0111] In the following example (which is in Figure 1 and 3 ), only one beam is considered for measuring a distance. However, the approach can be directly extended to all beams and all distances.

[0112] The detection method includes an algorithm to estimate the median value in real time. For this purpose, the data used are listed below:

[0113] Parameters: Data quantity 2nm+1

[0114] Input: LiDAR measurement rws(k)

[0115] Output: median mr(k)

[0116] Initialization: j = 1

[0117] The algorithm operates at each time k in the following manner:

[0118] 1. If j≤2nm+1

[0119] a) rp(j)=rws(k)

[0120] b) j = j + 1

[0121] c) Go to 1.

[0122] 2. Otherwise:

[0123] a) rp(1:2nm) = rp(2:2nm+1)

[0124] b) rp(2nm+1)=rws(k)

[0125] c) Sort the vector rp in ascending order.

[0126] 3. The median mr(k) is calculated as mr(k)=rp(nm).

[0127] This algorithm requires only very simple storage and sorting operations.

[0128] For the problem of detecting outliers, we will also have to estimate the mean absolute deviation. In mathematics, the mean absolute deviation of a data set is the average of the absolute deviations relative to the center point, that is, for a set X = {x1, x2, ..., xn}, the mean absolute deviation is calculated as:

[0129]

[0130] The center point m(X) can be the mean, median or mode. For robustness against outliers, the median mr(k) is chosen as m(X).

[0131] The mean absolute deviation is a measure used to quantify the amount of variation or dispersion in a set of data values. A low absolute deviation indicates that the data points tend to be closer to the center of the set, while a high absolute deviation indicates that the data points are spread over a wider range of values. The main advantage of the mean absolute deviation over the standard deviation is that it is much more robust to outliers than the standard deviation.

[0132] In the following, an algorithm for calculating the mean absolute deviation of previously measured lidar data np is provided. For greater simplicity, only one distance measurement with one beam is considered.

[0133] To this end, the data used by the algorithm is listed below:

[0134] Parameter: Number of data np

[0135] Input: LiDAR measurement rws(k)

[0136] Output: mean absolute deviation dr(k)

[0137] Initialization: j = 1

[0138] At each time k:

[0139] 1. If j ≤ np

[0140] a) dp(j) = rws(k)

[0141] b) j = j + 1

[0142] c) Go to 1.

[0143] 2. Otherwise

[0144] a) dp(1:np–1)=dp(1:np)

[0145] b) dp(np) = rws(k)

[0146] 3. An algorithm for estimating the median in real time is used to estimate the median mr(k).

[0147] 4. The mean absolute deviation dr(k) is calculated as

[0148]

[0149] In this section, a process is proposed for detecting outliers in a real-time manner using an estimated median mr(k) and an estimated mean absolute deviation dr(k). It will be recalled that the median can be considered as the middle value, whereby the mean absolute deviation dr(k) quantifies the amount of variation or dispersion of a set of data values.

[0150] To detect outliers, it is assumed that the lidar measurements rws(k) cannot change rapidly, i.e. the difference |rws(k)-mr(k) is small. More precisely, it is assumed that:

[0151] |rws(k)–mr(k)|≤σdr(k)

[0152] Here σ is a positive scalar.

[0153] In this way, all lidar measurements that do not satisfy the aforementioned equation are considered outliers.

[0154] Therefore, the method for detecting outliers is included in the algorithm for detecting outliers, which can be summarized as follows:

[0155] The parameters for this algorithm are:

[0156] Input: LiDAR measurement rws(k)

[0157] Output: Detect whether rws(k) is an outlier.

[0158] Therefore, at each time k, the algorithm does the following:

[0159] 1. Use the algorithm to estimate the median mr(k).

[0160] 2. Use the algorithm to estimate the mean absolute deviation dr(k).

[0161] 3. If |rws(k)–mr(k)|≤σdr(k), then rws(k) is not an outlier.

[0162] It can be used in wind field reconstruction algorithms.

[0163] 4. Otherwise: rws(k) is an outlier.

[0164] According to one implementation of the invention, the method comprises a step of reconstructing the measurements of the lidar sensor by removing the detected outliers from the modelled measurements rws(k).

[0165] Therefore, a clean measurement signal without outliers is obtained.

[0166] Figure 4 The various steps of the acquisition and modeling method according to one embodiment of the present invention are shown: other variant embodiments of the acquisition and modeling process known to those skilled in the art are applicable to the method according to the present invention.

[0167] 1. Generate a grid (MA) of the space upstream of the lidar sensor, which includes estimation points (PE) and measurement points (PM).

[0168] 2. Measure (MES) the amplitude and direction of the wind at each measurement point (PM).

[0169] 3. Estimate (EST) the amplitude and direction of the wind for all estimation points (PE) at any time (t).

[0170] 4. Reconstruct the incident wind field in three dimensions (3D) and in real time (MOD 3D) at all discrete points.

[0171] Figure 3 A wind turbine 1 is shown which is equipped with a Lidar sensor 2. The Lidar sensor 2 is used to measure the wind speed at a given distance and at a measurement point PM. In principle, knowing the wind measurements in advance makes it possible to generate a large amount of information.

[0172] There are many types of lidar sensors, such as scanning lidar sensors, continuous lidar sensors or pulsed lidar sensors. In the context of the present invention, pulsed lidar is preferably used. However, other lidar technologies may be used while remaining within the scope of the present invention. Figure 1 As shown in (an example embodiment), the used LiDAR includes 5 beams or measurement axes (b0, b1, b2, b3, b4). Without limitation, the acquisition and modeling methods also operate with LiDARs including three or more beams. The pulsed 5-beam LiDAR sensor is mounted on the nacelle 3 of the wind turbine 1.

[0173] Conventionally, a wind turbine 1 allows the kinetic energy of the wind to be converted into electrical or mechanical energy. In order to convert the wind into electrical energy, a wind turbine consists of the following elements:

[0174] a mast 4 which allows the rotor (not shown) to be placed at a sufficient height to enable it to move (required for a horizontal axis wind turbine) or at a height that allows it to be driven by stronger and more regular winds than at ground level 6. The mast 4 usually houses some electrical and electronic components (modulators, command units, multipliers, generators, etc.);

[0175] - A nacelle 3 mounted at the top of the mast 4, housing the mechanical, pneumatic components and certain electrical and electronic components (not shown) required for the operation of the machine. The nacelle 3 can be rotated to orient the machine in the correct direction.

[0176] - A rotor fixed to the nacelle, comprising a plurality of blades 7 (usually three) and the nose cone of the wind turbine. The rotor is driven by wind energy and is connected through a mechanical shaft directly or indirectly (via a gearbox and a mechanical shaft system) to a motor (generator, etc.) (not shown) that converts the harvested energy into electrical energy. The rotor may be equipped with control systems, such as variable angle blades or pneumatic brakes;

[0177] - A transmission consisting of two shafts (the mechanical shaft of the rotor and the mechanical shaft of the electric machine) connected via a transmission (gearbox) (not shown).

[0178] In the description given below, the acquisition and modeling methods described are theoretical and work independently of the wind turbine 1. However, various examples and developments are given in the case where the laser radar is installed on the nacelle 3 of the wind turbine 1, so that Figure 4 The various steps of the acquisition and modeling method shown are performed at a certain altitude relative to the ground 6 .

[0179] In this section, the various steps of the acquisition and modeling method according to the present invention are described:

[0180] 1. Generate a mesh (MA) of the space upstream of the LiDAR sensor

[0181] In this first step, the space upstream of the LiDAR sensor is defined as a grid, such as Figure 3 , 5 and 6. In this step, the coordinate system in which the lidar performs measurements is defined. The defined coordinate system is Figure 3 and 5 The xy origin of this system is at the level of the position of the lidar on the nacelle 3 , while the z origin is at the level of the ground 6 .

[0182] The x-axis points horizontally in the direction of the wind, the z-axis points vertically upward, and the y-axis is orthogonal to form a straightforward three-dimensional coordinate system (according to the right-hand rule).

[0183] In this step, the generation of the grid of this space includes generating a set of discrete points that are located upstream and define a three-dimensional grid. For each given distance x, the yz plane is divided into cells with no overlap, such as Figure 5 The grid includes measurement points (PM) and estimation points (PE) where the wind speed is measured and estimated, respectively.

[0184] With reference to this spatial grid, base variables called optimization variables are also defined, which are necessary for the estimation step described below. In order to allow a smart and efficient implementation of the optimization algorithm described below, all optimization variables are assembled into an ordered vector, which is denoted as ω. The determined order of these optimization variables is an engineering element that is critical to the feasibility and performance of the encoding algorithm of this method.

[0185] A vector ω is defined for each point of the discretized space and consists of all components v in space at the point (PE) where the wind is estimated x , followed by the components v y and v z , composed of. Estimation of wind speed at n points involves constructing a vector ω of size 3n, where w 1 tow n Contains all v x ,,w n+1 tow 2n Contains all v y , and w 2n+1 tow 3n Contains all v z .

[0186] For the wind speed component v x The following example is given, but it will be understood that the method is applicable to v y and v z As in the initial step, and as Figure 5 As shown, the space is discretized into x, y and z, where there are n x points, there are n in y y points, and there are n z points.

[0187] In this configuration:

[0188] n=n x n y n z

[0189] The coordinates are (x i ,y j ,z k ) is the wind speed component v x By v i,j,kdefinition.

[0190] The corresponding estimated location w 1 The index l is obtained as follows:

[0191] l=(n x -i)n y n z +(k-1)n y +j

[0192] For example, if i = n x , k = 1 and j = 1, then

[0193] l=(n x -i)n y n z +(k-1)n y +j=1

[0194] This corresponds to the upper left corner of the estimation domain, located at the most upstream distance from the rotor plane, as Figure 6 shown.

[0195] 2. Measure (MES) the amplitude and direction of the wind at each measurement point

[0196] In a second step, the LiDAR sensor performs a measurement m(t) related to the wind speed at a measurement point (PM) located upstream of the wind turbine 1. This measurement m(t) corresponds to the signal from the measurement point (PM) received by the sensor in response to the signal emitted by the LiDAR sensor. In particular, part of the laser signal emitted by the LiDAR sensor is reflected by air molecules at the measurement point and by aerosols (dust and particles in suspension) via interferometry and the Doppler effect. The measurement point is determined by the characteristics of the LiDAR sensor (in particular its focal length) and by its orientation. This measurement, depending on the wind speed, is a time and depends on the orientation of the LiDAR sensor.

[0197] For the pulsed lidar study case, the measurements are acquired sequentially on the grid defined in the previous step, starting with the longitudinal beam b0, then the oblique beam b1, up to beam b4. An advantageous feature of this system is that, for a given beam, it allows the projection of the wind speed to be measured simultaneously at multiple distances. It is thus possible, for example, to acquire 10 consecutive distances between 50m and 400m at a sampling rate of 0.25Hz or 1Hz. Of course, it is possible to limit the measurements to two, which is sufficient to reconstruct the 3D model. At each sampling time, only the measurements of the selected current beam are refreshed.

[0198] In a specific case, according to Figure 6, measurements are made for the five beams at seven distances, and in particular at x = [50 80 120 160 200 240 280] m. Thus, for each given x, yz plane is divided into the following cells:

[0199] • Four first points (PM) corresponding to the yz coordinates of the measurement points of beams 1, 2, 3, 4 at a distance x = 280 m.

[0200] Four second points (PM1) corresponding to the yz coordinates of the measurement points of beams 1, 2, 3, 4 at a distance x = 240 m.

[0201] - Four third points (PM2) corresponding to the yz coordinates of the measurement points of beams 1, 2, 3, 4 at a distance x = 200 m.

[0202] - Four fourth points (PM3) corresponding to the yz coordinates of the measurement points of beams 1, 2, 3, 4 at a distance x = 160m.

[0203] - Four fifth points (PM4) corresponding to the yz coordinates of the measurement points of beams 1, 2, 3, 4 at a distance x = 120 m.

[0204] Four sixth points (PM5) corresponding to the yz coordinates of the measurement points of beams 1, 2, 3, 4 at a distance x = 80 m.

[0205] - Four seventh points (PM6) corresponding to the yz coordinates of the measurement points of beams 1, 2, 3, 4 at a distance x = 50 m.

[0206] • The center point (PM7) corresponding to the yz coordinates of the measurement points of beam 0 at all distances.

[0207] The lidar measurement m(k) for beam j=0,1,2,3,4 at distance x meters and at time k is given by the formula mj,x(k), where j=0,1,2,3,4.

[0208] For example, m 0,50 (1) is the lidar measurement of beam j=0 at distance x=50 meters and time k=1. In the context of the present invention, the lidar measurement is then given by a formula of the following type:

[0209] m j，x (k) = a j v j，x (k)+b j v j，y (k)+c j v j，z (k)

[0210] Herevj,x (k),v j,y (k),v j,z (k) is the wind speed value projected into the given coordinate system at the initial time (k), and a j ,b j ,c j (where j = 0, 1, 2, 3, 4) is the measurement coefficient, which is given by:

[0211]

[0212] Here θj, (where j = 0, 1, 2, 3, 4) are the zenith angle and azimuth angle of the measurement axis in the spherical coordinate system, respectively.

[0213] Under the chosen spatial discretization choice, the advantage of defining the lidar measurement equation in the coordinate system defined above is that the equation can be used directly since the coordinates of the measurement point coincide with a specific point of the discretization space.

[0214] 3. Estimate (EST) the amplitude and direction of the wind at all discrete points at any time (t)

[0215] This step consists in obtaining the value of the wind at the estimation point (PE) of the grid.

[0216] To this end, a weighted recursive least squares method is used to perform an estimation by optimizing a cost function that uses the measured lidar data m(k) and also uses spatial wind speed coherence data, temporal wind speed variation data and data identifying the quality of the lidar measurements m(k). This will be explained below.

[0217] 3.1 Spatial Difference

[0218] These subsections aim to define spatial wind coherence data used in the context of the present invention and more specifically in the context of a lidar mounted on the nacelle 3 of a wind turbine 1 .

[0219] In this step, the components of the wind speed on the axes x, y and z of the coordinate system defined above are considered.

[0220] In this estimation step, it is assumed that the wind speed varies relatively little in space and therefore the wind has high spatial coherence in a small spatial volume. x (i.e., the first n variables of ω) are given as follows, where the estimation domain is Figure 6 As shown (this method is for component v y and v z similar), and set n x =n y =n z =3.

[0221] 3.1.1 Vertical Differences

[0222] Longitudinal difference different corresponds to v along the x-axis x In this case, the partial derivative dv x / dx is relatively small. In other words,

[0223]

[0224] The previous equation can be written in compact vector form:

[0225] C xl ω≈0

[0226] Here

[0227]

[0228] It will be noted that C xl Each row of contains a +1 and a -1.

[0229] Similarly, v can be calculated y and v z The changes along the vertical axis are:

[0230]

[0231] Here C yl ,C zl is a coefficient matrix, each row containing only one +1 and one -1.

[0232] definition:

[0233]

[0234] The following equation is obtained:

[0235] C l ω≈0

[0236] It characterizes the wind speed variation for the estimation domain along the longitudinal axis.

[0237] 3.1.2 Horizontal Differences

[0238] The lateral difference is v x along the y-axis. Similarly, since the wind does not change discontinuously, the partial derivative dv x / dy is relatively small. In other words,

[0239]

[0240] The previous equation can be written in compact vector form

[0241] C xt ω≈0

[0242] Here

[0243]

[0244] C xt Each row of contains a +1 and a -1.

[0245] Similarly, along the horizontal axis v y and v z The change can be calculated as:

[0246]

[0247] Here C yt ,C zt is a coefficient matrix, each row containing only one +1 and one -1.

[0248] definition:

[0249]

[0250] Obviously, the equation:

[0251] C t ω≈0

[0252] Characterizes the wind speed variation for the estimation domain along the horizontal axis.

[0253] 3.1.3 Vertical Difference

[0254] The vertical distribution of wind speed is given by a power law, which enables more accurate determination of the wind speed components v at various heights. x Description.

[0255] The vertical distribution of wind speed describes the variation of longitudinal wind speed with altitude relative to the ground. The power law of wind speed distribution is often used to estimate the longitudinal wind speed v at an altitude z above the ground. L , the reference altitude z r Longitudinal wind speed v lr Taking into account, using the following equation,

[0256]

[0257] Here alpha is the exponent of the power law, which is usually specified in terms of stability.

[0258] A constant value of alpha = 1 / 7 is often used, assuming relatively low wind shear. However, it should be noted that treating alpha as a constant is equivalent to neglecting the roughness of the ground surface, the interaction of wind with potential obstacles, and the stability of the atmosphere.

[0259] Using this power law, the vertical difference in wind is given by:

[0260]

[0261] Here j is the height of ω, and α is the exponent of the power law, which is assumed to be 1 / 7.

[0262] The previous equation can be written in compact vector form:

[0263] C xv ω≈0

[0264] Here

[0265]

[0266] Similarly, we can convert v y and v z The change along the vertical axis is quantified as:

[0267]

[0268] However, since the power law of wind distribution applies only to longitudinal wind speed, C yv and C zv Each row contains only one +1 and one -1.

[0269] definition:

[0270]

[0271] The following equation is obtained:

[0272] C v ω≈0

[0273] It characterizes the wind speed variation for the estimation domain along the vertical axis.

[0274] axis

[0275] Finally, use

[0276] C l ω≈0

[0277] and

[0278] C t ω≈0

[0279] The following can be declared:

[0280]

[0281] or equivalently,

[0282] C s ω≈0

[0283] It is an equation that characterizes the total change in wind speed along the x-, y-, and z-axes.

[0284] in:

[0285]

[0286] 3.2 LiDAR measurements

[0287] For ease of calculation, it is important to rewrite the measurement equation in vector form for W. In the previous example of a five-beam lidar, and for seven measurements per beam, j = 0, 1, 2, 3, 4, and x = [50, 80, 120, 160, 200, 240, 280],

[0288]

[0289] Combined with:

[0290] m j，x (k) = a j v j，x (k)+b j v j，y (k)+c j v j，z (k)

[0291] The following is obtained:

[0292] m j，x =C j，x ω

[0293] Here

[0294]

[0295] which can be rewritten in compact vector form:

[0296] C m ω=m m

[0297] Here

[0298]

[0299] To account for measurement noise, a more realistic lidar measurement model can be introduced as follows,

[0300] C mω=m m +∈ m

[0301] Here ε m Describes the measurement noise.

[0302] 3.3 Weighted Recursive Least Squares Method

[0303] It is assumed that wind speed varies little not only in space but also over time. Below, a method is provided to account for this information in an optimization approach. is the estimate of wind speed at time 0. At each time, the optimization problem is as follows:

[0304]

[0305] in:

[0306]

[0307] There are four terms in the previous cost function.

[0308] The first term penalizes the knowledge of the initial wind speed ω(0).

[0309] The second term penalizes the variation of wind speed over time.

[0310] The third term penalizes the variation of wind speed in the space.

[0311] The fourth term penalizes the quality of lidar measurements.

[0312] Using the above formula, we can achieve the weighting matrix P 0 ,Q,R s and R m A clear explanation is possible. Thus:

[0313] If the wind speed ω(t) at time t=0 is well known, then Then P 0 is small. Otherwise, P 0 For greatness.

[0314] If the wind speed varies a lot over time, Q is large.

[0315] Otherwise, Q is small.

[0316] If the wind speed changes rapidly, R s is large. Otherwise, R s For small.

[0317] If there is a lot of noise in the lidar measurements, R m is large. Otherwise, R m For small.

[0318] Let the following three limiting cases be considered:

[0319] No information about the initial wind speed is available. Therefore, P 0 Very large.

[0320]

[0321] can therefore be ignored in the cost function.

[0322] There is no relationship between the wind speed at time t and the wind speed at time t-1. In this case, Q can be chosen to be very large. The following terms can be ignored:

[0323]

[0324] The wind speed in the space varies little. In this case, R s is very small. The following terms are important in the cost function:

[0325]

[0326] The following formula is defined:

[0327]

[0328] The weighted recursive least squares method used to solve the optimization problem is defined as follows:

[0329] Optimization variables are initialized in the following way:

[0330]

[0331] At each time t:

[0332] ■The following formula is defined:

[0333]

[0334] where 0 is a zero vector of appropriate size.

[0335] ■The auxiliary matrix K is calculated so that

[0336] K=(P(t-1)+Q)C(C T (P(t-1)+Q)C+R) -1

[0337] ■The matrix P(t) is calculated such that

[0338] P(t)=(I-KC)P(t-1)

[0339] where I is the identity matrix of appropriate size.

[0340] ■The wind speed at time t is then estimated as:

[0341] ω(t)=ω(t-1)+K(y(t)-Cω(t-1))

[0342] 4. Reconstruct the incident wind field in three dimensions (3D) and in real time

[0343] In this step, the processor integrated in the LiDAR sensor collects all wind amplitude and wind direction data measured and estimated during the previous steps. The collection of these data is performed in real time for each previously defined measurement and estimation point (PM, PE). Thus, the LiDAR sensor is able to reconstruct all wind fields incident on the LiDAR, such as Figure 7 shown.

[0344] In the Figure 7 , the reconstructed wind field is shown for a period of 68 seconds. The y-axis represents the altitude relative to the ground (in meters), and the x-axis represents the distance from the nacelle (in meters) and the lateral position relative to the lidar (in meters).

[0345] Secondly, the present invention relates to a method for controlling and / or monitoring a wind turbine equipped with a lidar sensor as described above and an associated programmable logic controller 10, the method comprising the following steps:

[0346] i) a step of generating a desired control strategy (CON) for controlling said wind turbine 1, using the incident wind field reconstructed in three dimensions and in real time obtained using the method according to the invention,

[0347] ii) a step of performing control (PIL) in conjunction with the generated control strategy, which specifically includes controlling the angle of the blades 7 or the orientation of the nacelle 3 .

[0348] Figure 8 The overall operation of such a wind turbine 1 is shown. To this end, the wind turbine 1 comprises a lidar sensor 2 according to the invention and its processing unit, a computing device including a software solution for reconstructing the wind field in 3D, a programmable logic controller incorporating a control strategy and devices for controlling the wind turbine blades and / or the nacelle. Figure 8 , the present invention applied to wind turbines works in the following manner:

[0349] First, the lidar performs the steps of acquiring and modeling the incident wind field as described above to reconstruct the 3D incident wind field ( Figure 8 Steps ME, MA, EST and MOD 3D),

[0350] Secondly, the programmable logic controller 10 generates a control strategy (CON) and controls (PIL) the units of the wind turbine 1 taking the generated control strategy into account.

[0351] This method according to the invention enables real-time analysis of incident wind or detection of gusts, power curves and turbulence intensity, which may be applied to adjust or monitor wind turbines in order to obtain better alignment of wind turbines, which leads to optimization of production and minimization of loads and fatigue.

Claims

1. A method for detecting outliers of an incident wind field in space upstream of a lidar sensor, It is characterized in that The method comprises: a) obtaining a measurement of an incident wind field using the lidar sensor and modeling the measurement; b) the step of estimating in real time the median and mean absolute deviation of said measurements of said incident wind field; and c) a step of detecting outliers in real time using the estimated median and mean absolute deviation, said detection step being performed using a formula of the following type: | rws ( k ) − mr ( k )| ≤ σ dr ( k ), here rws ( k ) represents the measured value of the incident wind field, mr ( k ) represents the median value of the measured values ​​of the incident wind field, dr ( k ) represents the mean absolute deviation of the measured values ​​of the incident wind field, and σ is a positive scalar.

2. The method according to claim 1, It is characterized in that The real-time mean absolute deviation of the incident wind field is given by a formula of the following type: ， Where dr represents the real-time mean absolute deviation of the incident wind field, n represents the number of the measured values, and x i represents each measurement value, and m(X) represents the mean, median or mode of the measurement values.

3. The method according to any one of claims 1 to 2, It is characterized in that The method comprises the step of reconstructing measurements of the lidar sensor by removing the detected outliers from the modelled measurements.

4. The method according to claim 1, It is characterized in that The steps of acquiring an incident wind field in a space upstream of the lidar sensor using the lidar sensor and modeling the incident wind field include the following steps: a) a step of generating a grid of a space located upstream of said lidar sensor, wherein in this step said grid of said space is generated using a set of discrete points located in a predefined three-dimensional grid, said grid of said space comprising a set of cells consisting of estimated points and measured points, b) a step of measuring the amplitude and direction of the wind along at least three measurement axes at various measurement points situated in the upstream space and at at least two different distances from said lidar sensor, c) a step of estimating the amplitude and direction of the wind at all estimation points at all times, said estimation being carried out by means of optimizing a cost function using a weighted recursive least squares method, said cost function using at least the data of the measurement points, the spatial wind speed coherence data, the temporal wind speed coherence data and data identifying the quality of the measurements performed at the measurement points, d) The step of reconstructing said incident wind field in three dimensions in real time and in a defined coordinate system from the estimated and measured wind amplitude and direction for each point of said grid.

5. The method according to claim 4, It is characterized in that The measured values ​​of the amplitude and direction of the wind at the measuring point are given by the following relations: ， Here m j,x (k) is the measured value of the amplitude and direction of the wind for beam j at a distance x meters and at the initial time k, v j,x (k), v j,y (k), v j,z (k) is the wind speed value projected into a given coordinate system at the initial time k, and a j , b j , c j is the measurement coefficient, which is given by: ， Here θ j , φ j are the zenith angle and azimuth angle of the measurement axis in the spherical coordinate system, respectively, where j = 0, 1, 2, 3,4.

6. The method according to any one of claims 4 and 5, It is characterized in that The cost function at any time t is J(t) All are written in the following form: ， Here ω is an ordered vector consisting of all velocity components at the point in space where the wind is estimated, is the estimate of wind speed at time 0, P 0 , Q , R s and R m is a weighting matrix of appropriate size, and C s , C m is the matrix that takes into account wind speed and measurement noise, m m is a vector of measurement values.

7. The method according to claim 6, It is characterized in that The amplitude and direction of the wind at each measuring point are measured at a sampling rate of at least 0.25 Hz.

8. The method according to claim 6, It is characterized in that The amplitude and direction of the wind at each measuring point are measured at at least two different distances along the measuring axis.

9. The method according to claim 4, It is characterized in that The amplitude and direction of the wind are measured along at least three measurement axes.

10. The method according to claim 4, It is characterized in that Use the following type of formula to estimate the x , y and z Spatial coherence of axial wind speed: ， in: ， Where: C l represents the wind speed variation for the estimation domain along the longitudinal axis x and C t Characterize the horizontal axis y The wind speed variation in the estimation domain and C v Characterization for along the vertical axis z The wind speed variation in the estimation domain and The vector ω is an ordered vector consisting of all wind speed components at the points in space where the wind is estimated.

11. The method according to claim 4, It is characterized in that Use the following assumptions to estimate the x , y and z Spatial coherence of axial wind speed: Along the vertical axis x The wind speed changes little, and the partial derivative along the vertical axis dv x / dx Relatively small, Wind along horizontal axis y Changes continuously and along the horizontal axis y The partial derivative of dv x / dy Small, According to the power law, the wind is along the vertical axis z changes, which is given by: ， Here α is the exponent of the power law, v l is the altitude above the ground z The longitudinal wind speed at z r is the reference altitude, v lr is the longitudinal wind speed at the reference altitude.

12. The method according to claim 6, It is characterized in that The quality of the measurements performed by the LiDAR sensor is represented using a model of the following form: ， Here m Describes the measurement noise.

13. The method according to claim 6, It is characterized in that An estimate of the amplitude and direction of the wind field at all estimated points at time t is given by: ， in ， ， , where 0 is a zero vector of appropriate size, , in I is the identity matrix of suitable size.

14. A computer program product, It is characterized in that It comprises code instructions which are arranged to implement the steps of a method for detecting anomalies of an incident wind field in a space upstream of a lidar sensor according to any one of claims 1 to 13 when the program is executed by a processing unit of the lidar sensor.

15. A laser radar sensor, It is characterized in that The lidar sensor comprises code instructions of a computer program product according to claim 14 in a memory, and the lidar sensor is arranged to execute such a computer program product.

16. A wind turbine, It is characterized in that The wind turbine comprises a lidar sensor according to claim 15 .

17. The wind turbine according to claim 16, It is characterized in that The lidar sensor is placed on the nacelle of the wind turbine.

18. A method for controlling and / or monitoring a wind turbine equipped with a lidar sensor and a programmable logic controller, It is characterized in that Perform the following steps: i) a step of detecting anomalies of the incident wind field in the space upstream of the lidar sensor, and a method for detecting anomalies of the incident wind field in the space upstream of the lidar sensor according to any one of claims 1 to 13, ii) controlling the step of controlling in conjunction with the generated control strategy, including controlling the angle of the blades or the orientation of the nacelle without taking into account said detected abnormal value.

Images