Wind power plant robust yaw control method considering wind direction uncertainty

By constructing a probability characterization model for wind direction uncertainty and robust wake prediction, the yaw control of the wind farm is optimized, which solves the problem of wake interference misjudgment caused by wind direction uncertainty and improves the power generation efficiency and structural safety of the wind farm.

CN121806457APending Publication Date: 2026-04-07DATANG TONGXIN NEW ENERGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511934833.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-12-21
Publication Date
2026-04-07

AI Technical Summary

Technical Problem

Existing yaw control methods for wind farms fail to effectively handle wind direction uncertainty, leading to inaccurate wake prediction, yaw optimization failure, and increased risk to turbine load.

Method used

A probability characterization model for wind direction uncertainty based on circular statistical theory is constructed, and an online update is performed using a sliding time window mechanism. A robust wake prediction model is established, and the yaw angle is optimized using a sequential quadratic programming method, with the goal of maximizing the total power of the wind farm and meeting load constraints.

Benefits of technology

It significantly improves the power generation performance of wind farms under complex wind conditions, enhances the ability to adapt to sudden changes in wind direction, avoids the risk of mechanical overload caused by excessive yaw, ensures the structural safety of wind turbines, and reduces operation and maintenance costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121806457A_ABST
    Figure CN121806457A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of yaw control, in particular to a wind power plant robust yaw control method considering wind direction uncertainty. The method comprises the following steps: acquiring historical and real-time wind direction data of each fan in a wind power plant, and preprocessing the wind direction data; extracting wind direction statistical characteristics based on the preprocessed wind direction data, constructing a wind direction uncertainty representation model based on the wind direction statistical characteristics, and representing wind direction uncertainty in a probability form; and establishing a robust wake flow prediction model based on a wind direction uncertainty representation result in combination with terrain and fan layout information so as to predict a change trend of a fan wake flow superimposed effect under a wind direction uncertainty condition. According to the method, the wind direction uncertainty probability representation model based on the circle statistics theory is constructed, the wind direction distribution parameters are updated online in combination with a sliding time window mechanism, and random fluctuation, rapid change and space inconsistency of the wind direction can be accurately described.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of yaw control technology, and more specifically, to a robust yaw control method for wind farms that takes into account wind direction uncertainty. Background Technology

[0002] In wind farm operation, the wake effect is one of the key factors affecting overall power generation efficiency. Current mainstream wind farm yaw control strategies are mostly based on deterministic wind direction assumptions, assuming the incoming wind direction is known and constant. They aim to reduce wake obstruction by actively yawing to improve overall power output. However, actual wind conditions exhibit significant randomness and dynamic uncertainty, especially in complex terrain or atmospheric instability, where wind direction often shows rapid fluctuations, periodic jumps, and inconsistent spatial distribution. Traditional methods do not fully consider the uncertainty of wind direction, easily leading to wake prediction errors, inaccurate yaw commands, and even excessive yaw causing turbine structural loads to exceed limits. This not only weakens the wake control effect but may also threaten the safe operation of wind turbines. Furthermore, the wind direction angle itself has circular periodicity (e.g., 0° is equivalent to 360°). Using conventional linear statistical methods will introduce serious biases, further reducing the accuracy of wind direction estimation and wake modeling. Therefore, this paper proposes a robust yaw control method for wind farms that considers wind direction uncertainty. Summary of the Invention

[0003] The purpose of this invention is to provide a robust yaw control method for wind farms that takes into account wind direction uncertainty, so as to solve the problems of inaccurate wake prediction, yaw optimization failure and increased unit load risk caused by the existing wind farm yaw control methods mentioned in the background art due to ignoring wind direction uncertainty.

[0004] To achieve the above objectives, the present invention aims to provide a robust yaw control method for wind farms that considers wind direction uncertainty, comprising the following steps:

[0005] S1. Obtain historical and real-time wind direction data for each wind turbine in the wind farm, and preprocess the wind direction data;

[0006] S2. Extract wind direction statistical features based on the preprocessed wind direction data, and construct a wind direction uncertainty characterization model based on the wind direction statistical features to characterize wind direction uncertainty in a probabilistic form.

[0007] S3. Combining topographic and wind turbine layout information, a robust wake prediction model is established based on the wind direction uncertainty characterization results to predict the changing trend of wind turbine wake superposition effect under wind direction uncertainty.

[0008] S4. Based on the prediction results, a robust yaw optimization problem is constructed with the goal of maximizing the total power of the wind farm and the load constraint. The robust yaw optimization problem is solved by the sequential quadratic programming method to obtain the optimal yaw angle reference value of each wind turbine under the condition of wind direction uncertainty.

[0009] S5. Generate yaw control commands based on the optimal yaw angle reference value.

[0010] As a further improvement to this technical solution, step S2, which involves extracting wind direction statistical features based on the preprocessed wind direction data, includes the following steps:

[0011] S2.1 Based on the wind direction data preprocessed in step S1, construct wind direction time series according to wind turbine number;

[0012] S2.2. Within a preset time sliding window, the wind direction time series of each wind turbine is segmented and the continuous wind direction observation values ​​are mapped to a unified angle range.

[0013] S2.3 Within each time window, the wind direction data is processed based on the circular statistical method to calculate the average wind direction.

[0014] S2.4 Calculate the wind direction dispersion index based on the time window, including at least the circular square deviation and the circular standard deviation;

[0015] S2.5. Based on the wind direction difference between adjacent time points, calculate the wind direction change rate characteristics, and extract the mean, maximum value and quantile characteristics of the wind direction change rate within the statistical time window;

[0016] S2.6. Based on the wind direction statistics of multiple wind turbines in the wind farm, calculate the wind direction consistency index between adjacent wind turbines.

[0017] S2.7. Normalize the extracted wind direction mean, wind direction dispersion index, wind direction change rate characteristic, and spatial consistency index, and combine them to form wind direction statistical characteristics.

[0018] As a further improvement to this technical solution, in step S2, a wind direction uncertainty characterization model is constructed based on wind direction statistical characteristics to characterize wind direction uncertainty in probabilistic form, including the following steps:

[0019] S2.8. Based on the statistical characteristics of wind direction, the historical and real-time wind direction data are reconstructed in an equivalent manner, and the original wind direction time series is mapped into an equivalent wind direction sample set that satisfies the circular continuity constraint.

[0020] S2.9 Construct a wind direction probability distribution function based on an equivalent wind direction sample set using a circular statistical distribution model;

[0021] S2.10. Combining real-time wind direction data, a sliding time window mechanism is introduced to update the parameters in the wind direction probability distribution function online;

[0022] S2.11. Based on the updated wind direction probability distribution function, output the probability information representing the uncertainty of wind direction, i.e., the wind direction uncertainty representation result.

[0023] As a further improvement to this technical solution, in S2.10, a sliding time window mechanism is introduced to update the parameters in the wind direction probability distribution function online, including the following steps:

[0024] Set the length on the timeline as A sliding time window with a fixed step size. The wind direction observation samples in the window are updated in chronological order.

[0025] At each update time, the equivalent wind direction sample set within the window is uniformly mapped to the circular statistical space. By calculating the first circular moment of the sample, a composite vector is obtained, and the phase of the composite vector is used as the estimated value of the mean wind direction corresponding to the current window.

[0026] Based on the magnitude of the synthetic vector, the mean wind direction angle in the wind direction probability distribution function Perform recursive correction;

[0027] Based on the magnitude of the synthesized vector within the window, the maximum likelihood estimation method is used to estimate the lumped parameters in the wind direction probability distribution function. Update;

[0028] Updated mean wind direction angle With lumped parameters The parameters are exponentially weighted with those from the previous moment to complete the online update of the wind direction probability distribution function parameters.

[0029] As a further improvement to this technical solution, in step S3, a robust wake prediction model is established based on the wind direction uncertainty characterization results, combining terrain and wind turbine layout information, to predict the changing trend of the wind turbine wake superposition effect under wind direction uncertainty conditions. This includes the following steps:

[0030] S3.1 Obtain the topographic elevation data and surface roughness parameters of the area where the wind farm is located, construct a digital model of the wind farm topography, and establish a wind turbine spatial layout model based on the geographical coordinates, height information and turbine spacing of the wind turbines.

[0031] S3.2. Based on the wind direction probability distribution function, perform uncertainty probability weighting on the wind direction variable to obtain a set of effective incoming wind directions and their corresponding occurrence probabilities;

[0032] S3.3. Combine the digital model of wind farm terrain to perform terrain correction calculations for wind farms under different incoming wind direction conditions;

[0033] S3.4 Establish a wake velocity deficit model for a single wind turbine, and correct the wake spread angle and velocity deficit coefficient based on the yaw angle change under the condition of wind direction uncertainty.

[0034] S3.5 Based on the wind turbine spatial layout model, determine the upstream and downstream relationships between wind turbines under different wind direction conditions, and use energy superposition to form a comprehensive wake effect;

[0035] S3.6. Based on the probability of occurrence, the wake prediction results under different wind directions are weighted and fused to obtain the wake velocity deficit distribution and its statistical characteristics under the condition of wind direction uncertainty.

[0036] S3.7 Based on the probability-weighted wake superposition results, extract the robust wake prediction index as the prediction result.

[0037] As a further improvement to this technical solution, in step S3.1, the topographic elevation data and surface roughness parameters of the area where the wind farm is located are obtained, a digital model of the wind farm topography is constructed, and a wind turbine spatial layout model is established based on the geographical coordinates, height information, and turbine spacing of the wind turbines, including the following steps:

[0038] S3.11. Obtain topographic elevation data covering the wind farm and its upstream influence range, and perform coordinate system-1 and projection transformation on the topographic elevation data so that it adopts the same spatial reference system as the geographical coordinates of the wind turbine.

[0039] S3.12. Based on land use type, assign corresponding surface roughness length parameters to different surface types in the wind farm area, and calibrate them according to spatial location to form a surface roughness parameter field.

[0040] S3.13. Integrate the topographic elevation data with the surface roughness parameters after unification of coordinates to construct a wind farm topographic digital model that includes topographic relief information and roughness distribution information.

[0041] S3.14. Obtain the turbine parameters of each turbine in the wind farm and perform consistency verification on the turbine parameters;

[0042] S3.15. Map the geographic coordinates of each wind turbine to the digital model of the wind farm terrain;

[0043] S3.16. Based on the wind turbine geographic coordinate mapping results, calculate the horizontal distance, vertical height difference and relative azimuth between any two wind turbines to form a wind turbine spacing matrix and a relative position relationship matrix.

[0044] S3.17. Establish a spatial layout model for wind turbines by integrating the geographical coordinates of the wind turbines, the spacing matrix of the wind turbines, and the relative position relationship matrix.

[0045] As a further improvement to this technical solution, in S3.4, a wake velocity deficit model is established for a single wind turbine, and the wake spread angle and velocity deficit coefficient are corrected according to the yaw angle change under the condition of wind direction uncertainty, including the following steps:

[0046] For a single wind turbine, a basic wake velocity deficit model is constructed under the equivalent incoming flow conditions after terrain correction, and the equivalent yaw angle under different incoming flow conditions is determined by combining the probability distribution of wind direction uncertainty. The wake centerline deflection is modeled based on the yaw angle, and the wake extension angle and velocity deficit coefficient are corrected. By probabilistically weighting the correction parameters under different wind direction conditions, a wake velocity deficit model for a single turbine under wind direction uncertainty is formed.

[0047] As a further improvement to this technical solution, in step S4, based on the prediction results, a robust yaw optimization problem is constructed with the objectives of maximizing the total power of the wind farm and load constraints, including the following steps:

[0048] S4.1. Using the yaw angle of each wind turbine in the wind farm as the optimization decision variable, construct the yaw angle vector as the set of optimization variables, and set the allowable range of each yaw angle according to the structural safety and control system capability of the unit, thus forming the decision space of the yaw optimization problem.

[0049] S4.2 Based on the prediction results, the output power of each wind turbine is predicted under different effective incoming wind directions, and the prediction results of each wind turbine power are weighted according to the probability distribution of wind direction uncertainty to obtain the expected value of the total power of the wind farm under the condition of wind direction uncertainty.

[0050] S4.3 Using the probability-weighted result of the total power of the wind farm as the objective function, construct an optimization objective with the goal of maximizing the total power of the wind farm;

[0051] S4.4. Based on the robust wake prediction results and yaw angle changes, set load constraints.

[0052] S4.5. Combining the objective function of maximizing the total power of the integrated wind farm with load constraints, construct a robust yaw optimization problem that considers the uncertainty of wind direction.

[0053] As a further improvement to this technical solution, in step S4, the robust yaw optimization problem is solved using a sequential quadratic programming method to obtain the optimal yaw angle reference value for each wind turbine under uncertain wind direction conditions, including the following steps:

[0054] S4.6. Set an initial yaw angle vector based on the sequential quadratic programming method according to the current operating status of the wind turbine or the historical best yaw angle, and use it as the initial step for the sequential quadratic programming iteration.

[0055] S4.7 At the current iteration point, perform a second-order Taylor expansion on the objective function of step S4.3, linearize the nonlinear constraints into first-order linearization, and construct the corresponding quadratic programming subproblem.

[0056] S4.8 Solve the constructed quadratic programming subproblem to obtain the search direction of the yaw angle variable at the current iteration point;

[0057] S4.9. Based on the search direction, a line search strategy is used to determine the update step size, and the yaw angle vector is updated accordingly.

[0058] S4.10 After the yaw angle is updated, the load index corresponding to each wind turbine is evaluated in real time to determine whether the updated yaw angle meets the preset load safety threshold a.

[0059] S4.11 Determine whether the optimization process meets the convergence condition;

[0060] S4.12 After the convergence condition is met, the yaw angle vector obtained in the current iteration is used as the optimal yaw angle reference value for each wind turbine under the condition of uncertain wind direction.

[0061] As a further improvement to this technical solution, step S5, generating yaw control commands based on the optimal yaw angle reference value, includes the following steps:

[0062] S5.1 Smoothing and limiting of the yaw angle reference value;

[0063] S5.2 Compare the optimal yaw angle reference value after smoothing and limiting with the current actual yaw angle of the wind turbine, and calculate the yaw angle deviation corresponding to each wind turbine;

[0064] S5.3 Based on the yaw angle deviation, generate yaw control quantity according to the preset yaw control strategy, convert the yaw angle deviation into yaw drive control command, and send the yaw drive control command to each wind turbine yaw actuator.

[0065] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0066] 1. This invention relates to a robust yaw control method for wind farms considering wind direction uncertainty. By constructing a probabilistic representation model of wind direction uncertainty based on circular statistical theory and combining it with a sliding time window mechanism to update wind direction distribution parameters online, it can accurately characterize the random fluctuations, rapid changes, and spatial inconsistencies of wind direction. Based on this, a robust wake prediction model integrating terrain influences and wind turbine layout is established, and yaw optimization is performed with the objective of maximizing the expected total power of the wind farm, effectively mitigating the problem of wake interference misjudgment caused by wind direction uncertainty. Compared with traditional deterministic yaw strategies, this method significantly improves the overall power generation performance of wind farms under complex wind conditions, while enhancing the adaptability of the control system to uncertain disturbances such as sudden changes in wind direction.

[0067] 2. This invention relates to a robust yaw control method for wind farms that considers wind direction uncertainty. It explicitly introduces load constraints based on wake prediction results (such as tower bending moment, yaw load, and equivalent fatigue indices), and solves for the optimal yaw angle using a sequential quadratic programming algorithm while meeting structural safety thresholds. This mechanism avoids the mechanical overload risk caused by excessive yaw in pursuit of short-term power gain, ensuring that yaw control commands, while improving power generation revenue, always remain within the safe boundary of the wind turbine's structural load-bearing capacity. This extends equipment lifespan, reduces operation and maintenance costs, and achieves coordinated operation of the wind farm with both high efficiency and high reliability. Attached Figure Description

[0068] Figure 1 This is a flowchart of the overall method of the present invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0070] Example: Please refer to Figure 1 As shown, this embodiment provides a robust yaw control method for wind farms that considers wind direction uncertainty, including the following steps:

[0071] S1. Obtain historical and real-time wind direction data for each wind turbine in the wind farm, and preprocess the wind direction data;

[0072] Preprocessing involves: first, synchronizing the original historical and real-time wind direction observation data and handling missing values; then, correcting missing or outlier data using interpolation, smoothing, or removal methods; second, uniformly mapping the wind direction angle to the same circular interval (e.g., ...). – or – To eliminate statistical bias caused by angular periodicity, the data is then denoised using low-pass filtering or moving average methods to suppress short-term random disturbances. Finally, the wind direction data of each wind turbine is standardized into a time series by number or location, forming clean, continuous, and consistent input data for statistical feature extraction and wind direction uncertainty modeling, providing a reliable foundation for subsequent sliding time window processing and probability distribution construction.

[0073] S2. Extract wind direction statistical features based on the preprocessed wind direction data, and construct a wind direction uncertainty characterization model based on the wind direction statistical features to characterize wind direction uncertainty in a probabilistic form.

[0074] In this embodiment, wind direction statistical features are extracted based on the preprocessed wind direction data, including the following steps:

[0075] S2.1 Based on the wind direction data preprocessed in step S1, construct wind direction time series according to wind turbine number;

[0076] S2.2. Within a preset time sliding window, the wind direction time series of each wind turbine is segmented and the continuous wind direction observation values ​​are mapped to a unified angle range to eliminate statistical deviations caused by angle periodicity.

[0077] S2.3 Within each time window, wind direction data is processed using circular statistical methods to calculate the mean wind direction (to avoid statistical bias caused by the periodicity of wind direction angles, used to characterize the mean wind direction within that time window). Specifically, within each time window, the observed wind direction values ​​are first uniformly represented as circular angle variables, and each wind direction angle is mapped to a two-dimensional vector on a unit circle. Then, the two-dimensional vectors are summed within the window, and their composite vector is calculated. By normalizing the composite vector, an average direction vector reflecting the concentration trend of wind direction within the window is obtained. Finally, the phase angle of this average direction vector is used as the mean wind direction for that time window, thereby effectively avoiding the bias caused by the periodicity of wind direction angles (such as 0° being equivalent to 360°) in the calculation process, and achieving a robust estimate of the mean wind direction.

[0078] S2.4 Calculate wind direction dispersion index based on time window, including at least circular variance and circular standard deviation, to quantify the intensity of wind direction fluctuations and provide a basis for modeling wind direction uncertainty; specifically, within each time window, after obtaining the mean wind direction, the dispersion of wind direction samples on a unit circle within the window is quantitatively characterized based on circular statistical theory. Specifically, the magnitude of the composite vector corresponding to each wind direction sample is calculated to characterize the concentration of samples on the circumference; then, based on the relationship between the composite vector magnitude and the number of samples, the circular variance is defined as the normalized complementary quantity of the composite vector magnitude, used to reflect the dispersion level of wind direction samples around the mean direction, and on this basis, the circular standard deviation is obtained by nonlinear transformation of the circular variance, thereby forming a quantitative description of the intensity of wind direction fluctuations, so that the dispersion index can accurately reflect the random fluctuation characteristics of wind direction within the time window;

[0079] S2.5. Based on the wind direction difference between adjacent time points, calculate the wind direction change rate characteristics, and extract the mean, maximum, and quantile characteristics of the wind direction change rate within the statistical time window to characterize rapid or abrupt wind direction changes. Specifically, within each statistical time window, first calculate the difference between the wind direction observations at adjacent time points within the window, and process the wind direction data using the principles of circular angle expansion and minimum angle difference before differencing to avoid wind direction crossing. The system introduces false abrupt changes; then, it divides the wind direction angle difference between adjacent time points by the corresponding sampling time interval to obtain a wind direction change rate sequence that characterizes the magnitude of wind direction change per unit time; based on this, it performs statistical analysis on the wind direction change rate sequence within a time window, calculates its mean to characterize the overall level of wind direction change, calculates its maximum value to identify extreme conditions of rapid wind direction change, and further extracts the change rate feature corresponding to the preset quantile to characterize the distribution characteristics of wind direction change intensity, thereby providing a quantitative basis for identifying conditions of rapid or abrupt wind direction change;

[0080] S2.6. Combining the wind direction statistics of multiple wind turbines within the wind farm, calculate the wind direction consistency index between adjacent wind turbines, such as the wind direction correlation coefficient or average direction deviation, to reflect the synergy and dispersion of wind direction in the spatial dimension. Specifically, within each statistical time window, based on the wind direction mean or equivalent wind direction mean independently calculated by each wind turbine, map the wind direction results of adjacent wind turbine pairs to the same circular statistical space, and calculate the wind direction deviation between adjacent wind turbines using the minimum angle difference principle. On this basis, the direction deviation can be further converted into a consistency metric, for example, by calculating the circular correlation coefficient between the wind direction vectors of adjacent wind turbines or by normalizing the direction deviation to obtain the average direction deviation index, to quantify the synchronicity and consistency of wind direction changes of different wind turbines within the same time window. By summarizing or weighting the consistency indices of all adjacent wind turbine pairs within the wind farm area, a wind direction consistency statistical feature reflecting the synergy and dispersion of wind direction propagation in the spatial dimension is formed.

[0081] S2.7. Normalize the extracted wind direction mean, wind direction dispersion index, wind direction change rate characteristic, and spatial consistency index, and combine them to form wind direction statistical characteristics.

[0082] A wind direction uncertainty characterization model is constructed based on wind direction statistical characteristics, representing wind direction uncertainty in probabilistic form, including the following steps:

[0083] S2.8. Based on the statistical characteristics of wind direction, historical and real-time wind direction data are reconstructed equivalently, mapping the original wind direction time series into an equivalent set of wind direction samples that satisfy the circular continuity constraint. Specifically, based on the extracted statistical characteristics such as wind direction mean, dispersion, and rate of change, consistency processing is performed on the historical and real-time wind direction time series within each statistical time window. First, the wind direction mean direction within the window is used as a reference benchmark, and the relative angle of the original wind direction observations is transformed so that the wind direction samples are expanded within a continuous angle interval around the reference direction. Then, the wind direction samples crossing the circular boundary are equivalently expanded and corrected according to the principle of minimum angle difference, eliminating... The discontinuity caused by periodic jumps; based on this, the corrected wind direction samples are uniformly normalized and resampled to form an equivalent wind direction sample set that is continuous, comparable and meets the requirements of circular statistical modeling in the sense of a circle, thus providing a stable and consistent sample input for the subsequent construction of the wind direction probability distribution function and parameter estimation.

[0084] S2.9. Construct a wind direction probability distribution function based on an equivalent wind direction sample set using a circular statistical distribution model (the circular statistical distribution model is a von Mises distribution or its extended form, and uses the wind direction mean angle). and lumped parameters As a key probability distribution parameter describing the concentration and dispersion of wind direction, it enables a quantitative characterization of the randomness of wind direction.

[0085] ;

[0086] In the formula, The mean angle, This is a lumped parameter used to characterize the degree to which wind direction samples are concentrated around the mean angle; a larger value indicates a more concentrated wind direction and less fluctuation. As the distribution approaches zero, it gradually tends towards a uniform distribution. For zero-order modified Bessel functions of the first kind, Represents a random variable of wind direction The probability density function on the unit circle is used to characterize the probability distribution of wind direction in different directions;

[0087] S2.10. Combining real-time wind direction data, a sliding time window mechanism is introduced to update the parameters in the wind direction probability distribution function online;

[0088] Furthermore, a sliding time window mechanism is introduced to update the wind direction probability distribution function parameters online. This primarily addresses the impact of dynamic fluctuations and rapid changes in wind direction on short-term timescales on yaw optimization accuracy. By recursively estimating the latest observation data within the sliding window, the mean wind direction angle and lumped parameters can be adjusted in real time. This allows the wind direction probability distribution function to reflect the uncertainty characteristics of wind direction under current operating conditions, providing more accurate input for wake prediction and yaw optimization, thus improving robustness and response speed. The advantage of this sliding time window mechanism is its ability to dynamically capture the time-varying characteristics of wind direction, rather than relying solely on historical statistics or fixed probability distribution assumptions. This avoids the problem of yaw strategy lag or failure under rapid wind direction changes or meteorological disturbances. Its function is to reflect the uncertainty of wind direction in the wind farm in real time, providing an adaptive probability weighting basis for multi-turbine wake superposition analysis and robust yaw optimization, enabling the wind farm to maximize power generation revenue while ensuring unit load safety under different wind direction disturbances.

[0089] A sliding time window mechanism is introduced to update the parameters in the wind direction probability distribution function online, including the following steps:

[0090] Set the length on the timeline as A sliding time window (unit: seconds) with fixed steps. (Unit: seconds) The wind direction observation samples in the window are updated in chronological order. The timeframe is selected based on the characteristics of wind field fluctuations, ranging from 10 minutes to 1 hour. The control cycle is 1 to 10 seconds;

[0091] At each update time, the equivalent wind direction sample set within the window is uniformly mapped to the circular statistical space. By calculating the first circular moment of the sample, a composite vector is obtained, and the phase of the composite vector is used as the estimated value of the mean wind direction corresponding to the current window.

[0092] Based on the sample concentration information reflected by the magnitude of the synthetic vector, the mean wind direction angle in the wind direction probability distribution function A recursive correction is performed (at each sliding time window update time, firstly, the equivalent wind direction samples within the window are mapped to vector representations on the unit circle, and all sample vectors are summed to obtain a composite vector; then, the magnitude of the composite vector is calculated and used as a metric to characterize the concentration of wind direction samples within the current window; based on this, the phase angle of the composite vector is extracted as the estimated instantaneous mean wind direction for the current window, and the confidence weight of this estimate is determined according to the magnitude of the composite vector; finally, the estimated instantaneous mean wind direction is compared with the mean wind direction angle of the previous time step). Weighted fusion is performed according to the aforementioned reliable weights to achieve recursive correction of the mean wind direction angle, resulting in an updated... It responds quickly to wind direction changes when the sample concentration is high and remains smooth and stable when the sample dispersion is high, thereby improving the robustness of wind direction mean estimation under uncertain conditions.

[0093] Based on the magnitude of the synthesized vector within the window, the maximum likelihood estimation method is used to estimate the lumped parameters in the wind direction probability distribution function. To update (first calculate the magnitude of the composite vector of the unit circle vectors based on the equivalent wind direction samples within the window). The modulus reflects the concentration of the samples on the circumference; subsequently, using the maximum likelihood estimation relationship of the von Mises distribution, the modulus of the sample mean vector is calculated. With lumped parameters Establishing a correspondence, that is, by solving equations ;in, and The instantaneous lumped parameter estimates for the current window are obtained by solving for the zeroth-order and first-order modified Bessel functions, respectively. To suppress the effects of measurement noise and short-term fluctuations, the instantaneous estimate is compared with the lumped parameters from the previous time step. The updated lumped parameters are obtained by fusing the data using either an exponentially weighted or recursive smoothing formula. );

[0094] Updated mean wind direction angle With lumped parameters The parameters of the wind direction probability distribution function are exponentially weighted with those of the previous time step to complete the online update of the wind direction probability distribution function parameters. This step aims to suppress the influence of measurement noise and short-term abrupt changes, thereby enabling the online or adaptive update of the wind direction probability distribution function parameters so that they can continuously reflect the wind direction uncertainty changes under the current operating conditions.

[0095] S2.11. Based on the updated wind direction probability distribution function, output probability information characterizing wind direction uncertainty, i.e., wind direction uncertainty characterization results, including wind direction probability density function, confidence interval, and probability of occurrence within a specific yaw angle range; specifically: after obtaining the updated wind direction probability distribution function, firstly, perform analytical or numerical evaluation on the probability distribution function within the complete wind direction angle interval to obtain a continuous wind direction probability density distribution; then, calculate the wind direction mean, variance, and corresponding confidence interval based on the probability density distribution, and determine the coverage interval of the wind direction angle at a given confidence level by integrating the probability density function, which is used to quantify the range of wind direction uncertainty; further, according to the requirements of yaw control and wake analysis, perform probability integration on the preset wind direction angle interval to obtain the probability of occurrence within a specific wind direction deviation range; finally, output the above probability density function, confidence interval, and interval occurrence probability results uniformly as probability information characterizing wind direction uncertainty under the current operating conditions, which is used for subsequent robust wake prediction and yaw optimization decision-making.

[0096] S3. Combining topographic and wind turbine layout information, a robust wake prediction model is established based on the wind direction uncertainty characterization results to predict the changing trend of wind turbine wake superposition effect under wind direction uncertainty.

[0097] In this embodiment, a robust wake prediction model is established based on the wind direction uncertainty characterization results, combining terrain and wind turbine layout information, to predict the changing trend of the wind turbine wake superposition effect under wind direction uncertainty conditions, including the following steps:

[0098] S3.1 Obtain topographic elevation data and surface roughness parameters of the area where the wind farm is located, construct a digital model of the wind farm topography, and establish a wind turbine spatial layout model based on the geographical coordinates, height information and turbine spacing of the wind turbines to describe the constraint relationship between topographic undulations and the relative position of the wind turbines on the propagation paths of incoming flow and wake.

[0099] The process includes acquiring topographic elevation data and surface roughness parameters of the wind farm area, constructing a digital topographic model of the wind farm, and establishing a spatial layout model of the wind turbines based on their geographical coordinates, height information, and turbine spacing. This includes the following steps:

[0100] S3.11. Obtain topographic elevation data covering the wind farm and its upstream influence range, and perform coordinate system-1 and projection transformation on the topographic elevation data to make it adopt the same spatial reference system as the wind turbine geographic coordinates, so as to ensure the spatial consistency between topographic information and wind turbine location data.

[0101] S3.12. Based on land use type or on-site survey results, assign corresponding surface roughness length parameters to different surface types in the wind farm area, and calibrate them according to spatial location to form a surface roughness parameter field, which is used to reflect the influence of surface conditions on the near-surface wind field.

[0102] S3.13. The topographic elevation data with unified coordinates is fused with the surface roughness parameters to construct a wind farm topographic digital model containing topographic relief information and roughness distribution information. The wind farm topographic digital model is represented in the form of a raster or grid and is used to describe the constraint relationship between the topography and wind flow in the wind farm area.

[0103] The wind farm topographic digital model is constructed using a hierarchical spatial data architecture, comprising two basic sub-modules: a topographic elevation layer and a surface roughness layer. The topographic elevation layer, based on rasterized elevation data under a unified spatial reference system, describes the topographic undulations and slope variations within the wind farm area. The surface roughness layer, within the same spatial grid framework, calibrates and maps roughness parameters corresponding to different surface types, characterizing the impact of surface conditions on the near-surface wind field structure. By fusing elevation information and roughness parameters, a topographic digital model is formed that can be used for wind field correction and wake propagation analysis. This model provides consistent topographic constraints and boundary inputs for different incoming wind directions and speeds.

[0104] S3.14. Obtain the wind turbine parameters of each wind turbine in the wind farm, including at least the geographic coordinate information, hub height and turbine model parameters, and perform consistency verification on the wind turbine parameters to ensure that the wind turbine location information and the terrain digital model are expressed in the same spatial reference system.

[0105] S3.15. Map the geographic coordinates of each wind turbine to the digital model of the wind farm terrain to determine the spatial position and relative elevation of each wind turbine in the terrain model, so as to provide a spatial basis for analyzing the impact of terrain undulation on the inflow and wake propagation path of a single turbine.

[0106] S3.16. Based on the wind turbine geographic coordinate mapping results, calculate the horizontal distance, vertical height difference and relative azimuth between any two wind turbines to form a wind turbine spacing matrix and a relative position relationship matrix, which are used to describe the possible upstream and downstream relationships between wind turbines under different wind direction conditions.

[0107] S3.17. Based on the geographical coordinates of the wind turbines, the spacing matrix of the wind turbines, and the relative position relationship matrix, establish a spatial layout model of the wind turbines. This model is stored in the form of a matrix or graph structure and is used to quickly identify the wake influence path and superposition relationship of the wind turbines under different incoming wind direction conditions.

[0108] The wind turbine spatial layout model is constructed based on the geographical coordinates, hub height, and unit spacing information of the wind turbines, adopting a spatial relationship modeling architecture with wind turbines as nodes. Each wind turbine node includes its spatial location and structural parameter attributes. Nodes are connected through relationship parameters such as distance, relative azimuth, and elevation difference, thus forming a spatial layout network describing the relative positional relationships of the wind turbines. This model supports the rapid identification of upstream and downstream relationships and potential wake influence paths between wind turbines under given incoming wind direction conditions, providing structured spatial relationship support for multi-turbine wake superposition analysis and robust yaw control.

[0109] S3.2. Based on the wind direction probability distribution function, perform uncertainty probability weighting on the wind direction variable to obtain a set of effective incoming wind directions and their corresponding occurrence probabilities (for the updated wind direction probability distribution function in the wind direction angle interval). The continuous wind direction angle is discretized into several representative wind direction angle units. Then, the integral value or probability mass of each wind direction unit under the probability density function is calculated to obtain the corresponding occurrence probability, which is used to characterize the incoming flow direction space of the wind turbine under the condition of wind direction uncertainty.

[0110] S3.3. Combining the wind farm topographic digital model, topographic correction calculations are performed on the wind field under different incoming wind direction conditions. The corrections include wind direction deflection, wind speed distribution changes, and turbulence intensity adjustments, thereby obtaining an equivalent incoming flow field that considers the influence of topography, providing boundary conditions for wake modeling. Specifically: First, based on the wind farm topographic digital model, the incoming path and inlet angle are determined for each wind direction condition, considering the influence of topographic undulations, slope changes, and surface roughness on near-surface wind speed and turbulence characteristics; then, based on the interaction between the incoming wind direction and topography... The system calculates the wind direction deflection angle, local wind speed correction coefficient, and turbulence intensity adjustment parameters to spatially correct the original equivalent incoming wind field. Specifically, for each grid cell along the wind path, the local wind speed and turbulence field structure are adjusted by superimposing the terrain drag effect and roughness influence, thereby obtaining the equivalent incoming wind field after terrain correction under the current wind direction conditions. This equivalent wind field maintains the probability weighted characteristics of wind direction and reflects the constraints of terrain on wind speed distribution and wake propagation path, providing accurate boundary conditions for single-aircraft and multi-aircraft wake modeling.

[0111] S3.4 Establish a wake velocity deficit model for a single wind turbine, and modify the wake spread angle and velocity deficit coefficient according to the yaw angle change under the condition of wind direction uncertainty, so that the single-unit wake model can reflect the influence of wind direction disturbance on the wake shape.

[0112] Furthermore, a wake velocity deficit model is established for a single wind turbine, and the wake spread angle and velocity deficit coefficient are corrected based on the yaw angle variation under wind direction uncertainty, including the following steps:

[0113] For a single wind turbine, a basic wake velocity deficit model is constructed under the equivalent incoming flow conditions after terrain correction. This model takes the equivalent incoming flow velocity and impeller characteristics of a single wind turbine under no-yaw conditions as input. If the wake extends axially symmetrically along the incoming flow direction, the energy loss is characterized by describing the attenuation law of the wake centerline velocity gradually recovering with downstream distance. The model uses the incoming flow velocity at the turbine hub height, impeller diameter, and wake diffusion coefficient as key parameters to establish a functional relationship between the wake velocity deficit and downstream distance (a monotonically decaying function of downstream distance, whose attenuation characteristics are jointly determined by impeller size parameters and wake diffusion parameters). This is used to predict the basic impact of a single turbine's wake on the downstream wind field without considering wind direction uncertainty and yaw effects, providing a baseline wake description for subsequent yaw correction and uncertainty modeling. This model describes the velocity distribution of the wake decaying with downstream distance under no-yaw conditions and determines the equivalent yaw angle under different incoming flow scenarios by combining the wind direction uncertainty probability distribution. Based on the yaw angle, the model describes the velocity distribution of the wake centerline velocity under different incoming flow conditions. The deflection is modeled, and the wake propagation angle and velocity deficit coefficient are corrected. (First, the incoming wind direction is discretized based on the probability distribution of wind direction uncertainty to obtain several representative incoming wind directions and their probabilities of occurrence. Each representative incoming wind direction is compared with the current nacelle orientation of the wind turbine to calculate the corresponding equivalent yaw angle. Then, the deflection direction and deflection magnitude of the wake centerline are determined based on the equivalent yaw angle, and the deflection result is used as the correction amount for the wake propagation path. On this basis, the wake propagation angle is incrementally corrected using the equivalent yaw angle as input to reflect the change in the lateral diffusion characteristics of the wake under yaw conditions. At the same time, the wake velocity deficit coefficient is attenuated to characterize the reduction in effective energy extraction caused by yaw operation.) This reflects the impact of yaw operation on wake diffusion characteristics and energy extraction efficiency. By probabilistically weighting the correction parameters under different wind direction conditions, a wake velocity deficit model for a single turbine under wind direction uncertainty is formed, providing a reliable single-turbine wake input for subsequent multi-turbine wake superposition prediction and robust yaw optimization.

[0114] The Jensen model is used for the basic wake velocity deficit model.

[0115] ;

[0116] in, For the incoming wind speed, This is the thrust coefficient (dimensionless), with a value ranging from approximately 0.7 to 0.9. This is the wake attenuation coefficient, with a value ranging from 0.04 to 0.08. downstream distance, The impeller diameter; at the yaw angle Below, the deflection angle of the wake centerline ( (Yaw effect coefficient), wake spread angle correction is... The speed loss coefficient is corrected to , The equivalent thrust coefficient (dimensionless) after taking into account the effect of yaw angle is used to correct the thrust reduction caused by yaw. The yaw effect coefficient (dimensionless) represents the degree to which the yaw angle enhances wake diffusion, and its value ranges from 0.01 to 0.05 rad⁻¹. The corrected wake attenuation coefficient (dimensionless) is used to describe the characteristic of faster wake diffusion under yaw conditions;

[0117] S3.5. Based on the wind turbine spatial layout model, determine the upstream and downstream relationships between wind turbines under different wind direction conditions, and use energy superposition to form a comprehensive wake effect to describe the comprehensive wake effect formed by the interaction of the wakes of multiple wind turbines under different incoming wind directions. Specifically, under each effective incoming wind direction condition, firstly, using the wind turbine spatial layout model, calculate the horizontal distance and vertical height difference between wind turbines according to the geographical coordinates, hub height and relative azimuth angle of each wind turbine, identify the upstream and downstream relationships along the wind direction line, that is, determine which wind turbines are within the wake influence range of other wind turbines; then, for each downstream wind turbine, calculate its local wind speed field affected by the wake of the upstream wind turbine according to the single-unit wake velocity loss model, and superimpose the wake influence from multiple upstream wind turbines according to the principle of energy conservation, that is, combine the velocity loss or power loss caused by the wake of each upstream wind turbine in a weighted or cumulative manner to form the comprehensive wake effect of the downstream wind turbine.

[0118] S3.6. Based on the probability of occurrence, the wake prediction results under different wind directions are weighted and fused to obtain the wake velocity deficit distribution and its statistical characteristics under the condition of wind direction uncertainty.

[0119] S3.7. Based on the probability-weighted wake superposition results, robust wake prediction indicators that reflect the most unfavorable or desired operating conditions are extracted as prediction results. These include the expected value of the wake velocity loss, confidence interval, or risk-weighted indicators, used to characterize the changing trend of the wake superposition effect under wind direction uncertainty. Specifically, firstly, the wake velocity loss results under each wind direction condition are weighted and fused according to the probability of occurrence of the corresponding wind direction to obtain the global probability-weighted wake distribution of the wind farm. Then, robust indicators are selected according to the optimization objective. For example, the expected value of the wake velocity loss can be taken as the prediction indicator for the average operating condition, or the high quantile (such as the 95th quantile) of the weighted wake velocity loss can be selected as the prediction indicator for the most unfavorable operating condition. Finally, the expected value, confidence interval, and high quantile are output in a unified manner to form robust wake prediction results under wind direction uncertainty, which are used to guide yaw optimization decisions and ensure that the power generation efficiency and unit load safety of the wind farm are both taken into account under different wind direction disturbances.

[0120] S4. Based on the prediction results, a robust yaw optimization problem is constructed with the goal of maximizing the total power of the wind farm and the load constraint. The robust yaw optimization problem is solved by the sequential quadratic programming method to obtain the optimal yaw angle reference value of each wind turbine under the condition of wind direction uncertainty.

[0121] In this embodiment, based on the prediction results, a robust yaw optimization problem is constructed with the objectives of maximizing the total power of the wind farm and load constraints, including the following steps:

[0122] S4.1. Using the yaw angle of each wind turbine in the wind farm as the optimization decision variable, construct the yaw angle vector as the set of optimization variables, and set the allowable range of each yaw angle according to the structural safety and control system capability of the unit, thus forming the decision space of the yaw optimization problem.

[0123] S4.2 Based on the prediction results, the output power of each wind turbine is predicted under different effective incoming wind directions, and the prediction results of each wind turbine power are weighted according to the probability distribution of wind direction uncertainty to obtain the expected value of the total power of the wind farm under the condition of wind direction uncertainty.

[0124] S4.3 Using the probability-weighted result of the total power of the wind farm as the objective function, construct an optimization objective with the goal of maximizing the total power of the wind farm, which is used to guide the coordinated adjustment of the yaw angle across the entire field under different wind direction uncertainties.

[0125] The objective function is:

[0126] ;

[0127] In the formula, Within the wind farm The yaw angle vector of the typhoon generator is an optimization decision variable. Wind direction angle (continuous variable, unit: radians). For the first A discrete wind direction angle, Wind direction The probability of occurrence, For the direction of the incoming wind and yaw angle Next The power output of the typhoon turbines is calculated by modifying the incoming wind speed using a digital model of the wind farm terrain to obtain a free-flow wind speed that takes into account surface roughness and terrain undulation. Next, the wake effects of upstream turbines are identified based on the turbine spatial layout, and a wake velocity loss model is used to calculate the velocity loss caused by each upstream turbine. The thrust coefficient and wake attenuation coefficient are modified according to the yaw angle to reflect the thrust reduction and enhanced wake diffusion caused by yaw. Then, based on the principle of energy conservation, the wake effects of multiple upstream turbines are superimposed using the sum of squares and the square root to determine the equivalent wind speed of the downstream turbines. Finally, this equivalent wind speed is input into the piecewise power curve function of the turbine and mapped to the actual output power according to the cut-in, rated, and cut-out wind speed conditions, completing the overall modeling from wind conditions to power generation. For discrete wind directions This represents the expected total power of the wind farm under conditions of uncertain wind direction.

[0128] S4.4. Based on robust wake prediction results and yaw angle changes, load constraints are set, including nacelle yaw load, tower bending moment, or equivalent fatigue load indices. Safety thresholds or constraint boundaries are set for these load indices to ensure that the yaw optimization results meet the unit's structural safety and lifespan requirements. Specifically: based on robust wake prediction results, the wind load and local wind speed changes caused by wake interference on each wind turbine under the current yaw angle configuration are calculated. Then, based on the wind turbine structural model and control system parameters, the wind load is converted into corresponding nacelle yaw load, tower bending moment, or equivalent fatigue load indices. Next, these load indices are compared with preset safety thresholds or constraint boundaries to identify over-limit or near-critical load conditions. Finally, the load constraints are added to the yaw optimization problem in the form of inequalities to tighten the yaw angle adjustment range for over-limit wind turbines or correct the objective function search direction through penalty terms, thereby ensuring that the optimized yaw angle scheme maximizes power generation revenue under different wind directions and wake interference conditions.

[0129] S4.5. Combining the objective function of maximizing the total power of the integrated wind farm with load constraints, a robust yaw optimization problem considering wind direction uncertainty is constructed, ensuring that the optimization process can balance power generation revenue and operational safety under different wind direction conditions. Specifically, the expected total power of the wind farm is used as the objective function. The power output of each wind turbine under different wind direction conditions is calculated using the probability-weighted multi-turbine wake prediction results, and these powers are weighted and summed to obtain the expected total power. Subsequently, combined with load constraints, the nacelle yaw load, tower bending moment or equivalent fatigue load index of each wind turbine under the current yaw angle configuration is constrained by inequalities with the safety threshold. On this basis, the wind turbine yaw angle vector is used as the optimization decision variable, and the allowable range of yaw angle values ​​is set to form a complete decision space. Finally, the objective function, load constraints, and yaw angle physical constraints are uniformly expressed as a constrained optimization problem considering wind direction uncertainty, so that the optimization process can pursue the maximization of the total power of the wind farm while ensuring the structural safety and operational reliability of the wind turbines under different wind direction disturbance conditions, providing a standardized mathematical framework for solving the problem using the sequential quadratic programming method.

[0130] Furthermore, by solving the robust yaw optimization problem based on the sequential quadratic programming method, the optimal yaw angle reference value for each wind turbine under the condition of wind direction uncertainty is obtained, including the following steps:

[0131] S4.6. Set an initial yaw angle vector based on the sequential quadratic programming method according to the current operating status of the wind turbine or the historical best yaw angle, and use it as the initial point for the sequential quadratic programming iteration to improve the convergence speed and numerical stability of the algorithm.

[0132] S4.7 At the current iteration point, perform a second-order Taylor expansion on the objective function of step S4.3, linearize the nonlinear constraints to the first order, and construct the corresponding quadratic programming subproblem so that the subproblem locally approximates the original robust yaw optimization problem.

[0133] S4.8 Solve the constructed quadratic programming subproblem to obtain the search direction of the yaw angle variable at the current iteration point, which indicates the optimal descent or ascent direction of the objective function under the constraints. Specifically, in each sequential quadratic programming iteration, the original robust yaw optimization problem is first locally linearized and approximated at the current iteration point, i.e., the objective function is expanded in the second order by Taylor, and the nonlinear load constraints are linearized in the first order to construct the corresponding quadratic programming subproblem. Then, the subproblem is solved using a quadratic programming solver (such as the interior point method or the active set method) to obtain the optimal solution increment vector of the subproblem. This increment vector indicates the yaw angle adjustment direction for the optimal improvement of the objective function under the linearization constraints. Finally, this increment vector is used as the search direction input line search strategy to determine the actual update step size, thereby guiding the yaw angle vector to adjust along the gradient or quasi-Newton direction during the iteration process, maximizing the total power while taking into account the load constraints.

[0134] S4.9. Based on the search direction, a line search strategy is adopted to determine a suitable update step size, and the yaw angle vector is updated accordingly to ensure that the objective function is monotonically improved and the constraints are always satisfied. Specifically, after obtaining the yaw angle search direction of the current iteration point, the step size is first explored in the search direction according to the preset line search strategy (such as the Armijo criterion or the golden section method), and the objective function value and constraint satisfaction under different step sizes are calculated in turn. Then, the optimal step size is selected that can significantly improve the objective function while ensuring that the load constraints and yaw angle physical constraints are not violated. Finally, the step size is multiplied by the search direction vector to update the current yaw angle vector and obtain a new iterative yaw angle vector. At the same time, the load index of each wind turbine is re-evaluated after the update to ensure that the yaw adjustment promotes the optimization of the total power of the wind farm without exceeding the limit.

[0135] S4.10 After the yaw angle is updated, the load index corresponding to each wind turbine is evaluated in real time to determine whether the updated yaw angle meets the preset load safety threshold a. If it does not meet the threshold, the search direction is corrected by tightening constraints or adjusting penalty terms.

[0136] S4.11 Determine whether the optimization process meets the convergence conditions, including the change in the objective function being lower than the preset threshold, the yaw angle update being lower than the set tolerance, or the maximum number of iterations being reached; if not, return to step S4.7 to continue iterating;

[0137] S4.12 After the convergence condition is met, the yaw angle vector obtained in the current iteration is used as the optimal yaw angle reference value for each wind turbine under the condition of uncertain wind direction.

[0138] S5. Generate yaw control commands based on the optimal yaw angle reference value;

[0139] In this embodiment, the yaw control command is generated based on the optimal yaw angle reference value, including the following steps:

[0140] S5.1 Smoothing and limiting the yaw angle reference value (smoothing and filtering the optimal yaw angle reference value obtained by robust yaw optimization to eliminate possible abrupt changes in the optimization result within adjacent control cycles; at the same time, according to the mechanical constraints and safe operation requirements of the wind turbine yaw system, limiting the yaw angle change rate and maximum adjustment range to generate an executable yaw angle adjustment target).

[0141] S5.2 Compare the optimal yaw angle reference value after smoothing and limiting with the current actual yaw angle of the wind turbine, and calculate the yaw angle deviation corresponding to each wind turbine, which is used to characterize the adjustment amount that the yaw system needs to perform;

[0142] S5.3 Based on the yaw angle deviation, generate a yaw control quantity according to the preset yaw control strategy, convert the yaw angle deviation into a control command that the yaw drive system can recognize, drive the yaw motor to rotate, and send the control command that the yaw drive system can recognize to each wind turbine yaw actuator to control the yaw system to complete the nacelle turning according to the command, and monitor the yaw execution status in real time to ensure that the yaw action is consistent with the optimal yaw angle reference value.

[0143] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A robust yaw control method for wind farms considering wind direction uncertainty, characterized in that, Includes the following steps: S1. Obtain historical and real-time wind direction data for each wind turbine in the wind farm, and preprocess the wind direction data; S2. Extract wind direction statistical features based on the preprocessed wind direction data, and construct a wind direction uncertainty characterization model based on the wind direction statistical features to characterize wind direction uncertainty in a probabilistic form. S3. Combining topographic and wind turbine layout information, a robust wake prediction model is established based on the wind direction uncertainty characterization results to predict the changing trend of wind turbine wake superposition effect under wind direction uncertainty. S4. Based on the prediction results, a robust yaw optimization problem is constructed with the goal of maximizing the total power of the wind farm and the load constraint. The robust yaw optimization problem is solved by the sequential quadratic programming method to obtain the optimal yaw angle reference value of each wind turbine under the condition of wind direction uncertainty. S5. Generate yaw control commands based on the optimal yaw angle reference value.

2. The robust yaw control method for wind farms considering wind direction uncertainty according to claim 1, characterized in that: In step S2, wind direction statistical features are extracted based on the preprocessed wind direction data, including the following steps: S2.1 Based on the wind direction data preprocessed in step S1, construct wind direction time series according to wind turbine number; S2.

2. Within a preset time sliding window, the wind direction time series of each wind turbine is segmented and the continuous wind direction observation values ​​are mapped to a unified angle range. S2.3 Within each time window, the wind direction data is processed based on the circular statistical method to calculate the average wind direction. S2.4 Calculate the wind direction dispersion index based on the time window, including at least the circular square deviation and the circular standard deviation; S2.

5. Based on the wind direction difference between adjacent time points, calculate the wind direction change rate characteristics, and extract the mean, maximum value and quantile characteristics of the wind direction change rate within the statistical time window; S2.

6. Based on the wind direction statistics of multiple wind turbines in the wind farm, calculate the wind direction consistency index between adjacent wind turbines. S2.7 Normalize the extracted wind direction mean, wind direction dispersion index, wind direction change rate characteristic, and spatial consistency index, and combine them to form wind direction statistical characteristics.

3. The robust yaw control method for wind farms considering wind direction uncertainty according to claim 1, characterized in that: In step S2, a wind direction uncertainty characterization model is constructed based on wind direction statistical characteristics to characterize wind direction uncertainty in probabilistic form, including the following steps: S2.

8. Based on the statistical characteristics of wind direction, the historical and real-time wind direction data are reconstructed into equivalent wind direction samples that satisfy the circular continuity constraint. S2.9 Construct a wind direction probability distribution function based on an equivalent wind direction sample set using a circular statistical distribution model; S2.

10. Combining real-time wind direction data, a sliding time window mechanism is introduced to update the parameters in the wind direction probability distribution function online; S2.

11. Based on the updated wind direction probability distribution function, output the probability information representing the uncertainty of wind direction, i.e., the wind direction uncertainty representation result.

4. The robust yaw control method for wind farms considering wind direction uncertainty according to claim 3, characterized in that: In step S2.10, a sliding time window mechanism is introduced to update the parameters in the wind direction probability distribution function online, including the following steps: Set the length on the timeline as A sliding time window with a fixed step size. The wind direction observation samples in the window are updated in chronological order. At each update time, the equivalent wind direction sample set within the window is uniformly mapped to the circular statistical space. By calculating the first circular moment of the sample, a composite vector is obtained, and the phase of the composite vector is used as the estimated value of the mean wind direction corresponding to the current window. Based on the magnitude of the synthetic vector, the mean wind direction angle in the wind direction probability distribution function Perform recursive correction; Based on the magnitude of the synthesized vector within the window, the maximum likelihood estimation method is used to estimate the lumped parameters in the wind direction probability distribution function. Update; Updated mean wind direction angle With lumped parameters The parameters are exponentially weighted with those from the previous moment to complete the online update of the wind direction probability distribution function parameters.

5. The robust yaw control method for wind farms considering wind direction uncertainty according to claim 1, characterized in that: In step S3, a robust wake prediction model is established based on the wind direction uncertainty characterization results, combining terrain and wind turbine layout information, to predict the changing trend of the wind turbine wake superposition effect under wind direction uncertainty conditions. This includes the following steps: S3.1 Obtain the topographic elevation data and surface roughness parameters of the area where the wind farm is located, construct a digital model of the wind farm topography, and establish a wind turbine spatial layout model based on the geographical coordinates, height information and turbine spacing of the wind turbines. S3.

2. Based on the wind direction probability distribution function, perform uncertainty probability weighting on the wind direction variable to obtain a set of effective incoming wind directions and their corresponding occurrence probabilities; S3.

3. Combine the digital model of wind farm terrain to perform terrain correction calculations for wind farms under different incoming wind direction conditions; S3.4 Establish a wake velocity deficit model for a single wind turbine, and correct the wake spread angle and velocity deficit coefficient based on the yaw angle change under the condition of wind direction uncertainty. S3.5 Based on the wind turbine spatial layout model, determine the upstream and downstream relationships between wind turbines under different wind direction conditions, and use energy superposition to form a comprehensive wake effect; S3.

6. Based on the probability of occurrence, the wake prediction results under different wind directions are weighted and fused to obtain the wake velocity deficit distribution and its statistical characteristics under the condition of wind direction uncertainty. S3.7 Based on the probability-weighted wake superposition results, extract the robust wake prediction index as the prediction result.

6. The robust yaw control method for wind farms considering wind direction uncertainty according to claim 5, characterized in that: In step S3.1, the topographic elevation data and surface roughness parameters of the area where the wind farm is located are obtained, a digital model of the wind farm topography is constructed, and a spatial layout model of the wind turbines is established based on the geographical coordinates, height information, and spacing between turbines, including the following steps: S3.

11. Obtain topographic elevation data covering the wind farm and its upstream influence range, and perform coordinate system-1 and projection transformation on the topographic elevation data so that it adopts the same spatial reference system as the geographical coordinates of the wind turbine. S3.

12. Based on land use type, assign corresponding surface roughness length parameters to different surface types in the wind farm area, and calibrate them according to spatial location to form a surface roughness parameter field. S3.

13. Integrate the topographic elevation data with the surface roughness parameters after unification of coordinates to construct a wind farm topographic digital model that includes topographic relief information and roughness distribution information. S3.

14. Obtain the turbine parameters of each turbine in the wind farm and perform consistency verification on the turbine parameters; S3.

15. Map the geographic coordinates of each wind turbine to the digital model of the wind farm terrain; S3.

16. Based on the wind turbine geographic coordinate mapping results, calculate the horizontal distance, vertical height difference and relative azimuth between any two wind turbines to form a wind turbine spacing matrix and a relative position relationship matrix. S3.

17. Establish a spatial layout model for wind turbines by integrating the geographical coordinates of the wind turbines, the spacing matrix of the wind turbines, and the relative position relationship matrix.

7. The robust yaw control method for wind farms considering wind direction uncertainty according to claim 5, characterized in that: In step S3.4, a wake velocity deficit model is established for a single wind turbine, and the wake spread angle and velocity deficit coefficient are corrected based on the yaw angle variation under wind direction uncertainty conditions. This includes the following steps: For a single wind turbine, a basic wake velocity deficit model is constructed under the equivalent incoming flow conditions after terrain correction, and the equivalent yaw angle under different incoming flow conditions is determined by combining the probability distribution of wind direction uncertainty. The wake centerline deflection is modeled based on the yaw angle, and the wake extension angle and velocity deficit coefficient are corrected. By probabilistically weighting the correction parameters under different wind direction conditions, a wake velocity deficit model for a single turbine under wind direction uncertainty is formed.

8. The robust yaw control method for wind farms considering wind direction uncertainty according to claim 1, characterized in that: In step S4, based on the prediction results, a robust yaw optimization problem is constructed with the objectives of maximizing the total power of the wind farm and balancing load constraints. This includes the following steps: S4.

1. Using the yaw angle of each wind turbine in the wind farm as the optimization decision variable, construct the yaw angle vector as the set of optimization variables, and set the allowable range of each yaw angle according to the structural safety and control system capability of the unit, thus forming the decision space of the yaw optimization problem. S4.2 Based on the prediction results, the output power of each wind turbine is predicted under different effective incoming wind directions, and the prediction results of each wind turbine power are weighted according to the probability distribution of wind direction uncertainty to obtain the expected value of the total power of the wind farm under the condition of wind direction uncertainty. S4.3 Using the probability-weighted result of the total power of the wind farm as the objective function, construct an optimization objective with the goal of maximizing the total power of the wind farm; S4.

4. Based on the robust wake prediction results and yaw angle changes, set load constraints. S4.

5. Combining the objective function of maximizing the total power of the integrated wind farm with load constraints, construct a robust yaw optimization problem that considers the uncertainty of wind direction.

9. The robust yaw control method for wind farms considering wind direction uncertainty according to claim 1, characterized in that: In step S4, the robust yaw optimization problem is solved using a sequential quadratic programming method to obtain the optimal yaw angle reference value for each wind turbine under uncertain wind direction conditions. This includes the following steps: S4.

6. Set an initial yaw angle vector based on the sequential quadratic programming method according to the current operating status of the wind turbine or the historical best yaw angle, and use it as the initial step for the sequential quadratic programming iteration. S4.7 At the current iteration point, perform a second-order Taylor expansion on the objective function of step S4.3, linearize the nonlinear constraints into first-order linearization, and construct the corresponding quadratic programming subproblem. S4.8 Solve the constructed quadratic programming subproblem to obtain the search direction of the yaw angle variable at the current iteration point; S4.

9. Based on the search direction, a line search strategy is used to determine the update step size, and the yaw angle vector is updated accordingly. S4.10 After the yaw angle is updated, the load index corresponding to each wind turbine is evaluated in real time to determine whether the updated yaw angle meets the preset load safety threshold a. S4.11 Determine whether the optimization process meets the convergence condition; S4.12 After the convergence condition is met, the yaw angle vector obtained in the current iteration is used as the optimal yaw angle reference value for each wind turbine under the condition of uncertain wind direction.

10. The robust yaw control method for wind farms considering wind direction uncertainty according to claim 1, characterized in that: In step S5, generating yaw control commands based on the optimal yaw angle reference value includes the following steps: S5.1 Smoothing and limiting of the yaw angle reference value; S5.2 Compare the optimal yaw angle reference value after smoothing and limiting with the current actual yaw angle of the wind turbine, and calculate the yaw angle deviation corresponding to each wind turbine; S5.3 Based on the yaw angle deviation, generate yaw control quantity according to the preset yaw control strategy, convert the yaw angle deviation into yaw drive control command, and send the yaw drive control command to each wind turbine yaw actuator.