Yaw optimization method suitable for densely-arranged wind power plant

Abstract

The invention relates to the technical field of intelligent control of wind power plants, in particular to a yaw optimization method suitable for densely-arranged wind power plants. The method comprises the following steps of collecting operation information, layout parameters, environment data and unit three-dimensional time sequence operation data of a wind power plant, constructing a digital flow field description system of the wind power plant, and forming a data feature space for yaw decision; introducing a virtual laser radar measurement model based on the data feature space to carry out observability calculation, and screening and judging the optimal installation position of the laser radar; and carrying out fusion processing on the optimal installation position and the three-dimensional time sequence operation data of the unit. According to the method, a high-dimensional digital flow field feature space fusing space topology, time dynamics and environmental sensitivity is constructed, virtual laser radar observability analysis and wake flow influence rule modeling are combined, and the coupling influence of upstream unit yaw on downstream unit power can be accurately recognized.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for wind farms, and more specifically, to a yaw optimization method applicable to densely distributed wind farms. Background Technology

[0002] With the current trend of wind power development towards large-scale and intensive development, densely arranged wind farms are increasingly becoming an important way to improve the efficiency of land and wind energy resource utilization. However, the significantly reduced spacing between turbines leads to a sharp increase in wake interference effects. The wake generated by upstream wind turbines can significantly reduce the inflow wind speed of downstream turbines and increase turbulence intensity, resulting in a decrease in their power generation and severely restricting the overall performance of the wind farm. Traditional yaw control strategies are mostly based on the independent operation logic of a single turbine, aiming only at aligning with the incoming wind direction, ignoring the aerodynamic coupling relationship between turbines, and failing to actively use yaw to induce wake deflection to alleviate downstream wake obstruction. In addition, existing wake models have limited accuracy and lack the ability to perceive and model real-time complex wind conditions (such as wind direction changes, terrain disturbances, and turbine dynamic response delays), and the deployment of feedforward sensing devices such as lidar lacks systematic optimization, making it difficult to support the high-quality flow field observation required for full-field collaborative control. Therefore, this paper proposes a yaw optimization method suitable for densely arranged wind farms. Summary of the Invention

[0003] The purpose of this invention is to provide a yaw optimization method suitable for densely arranged wind farms, so as to solve the problem of low overall power generation efficiency in densely arranged wind farms caused by strong wake interference and lack of coordination in traditional single-unit yaw control, as mentioned in the background art.

[0004] To achieve the above objectives, the present invention aims to provide a yaw optimization method suitable for densely distributed wind farms, comprising the following steps:

[0005] S1. Collect wind farm operation information, layout parameters, environmental data and three-dimensional time-series operation data of the units, and construct a digital flow field description system for the wind farm to form a data feature space for yaw decision-making;

[0006] S2. Based on the data feature space, a virtual lidar measurement model is introduced to perform observability calculations, and the optimal installation location of the lidar is selected and determined.

[0007] S3. The optimal installation location is fused with the three-dimensional time-series operation data of the unit, and a mapping relationship is established between the incoming flow characteristics, wake characteristics and the changes in unit power to identify the yaw effect on the power generation of downstream units.

[0008] S4. Based on the yaw effect law, a yaw optimization strategy including real-time wake reconstruction, upstream yaw compensation and downstream power recovery is constructed to set the unit yaw angle under different wind direction and wind speed conditions.

[0009] S5. Based on the yaw optimization strategy, yaw coordination control is carried out through multi-unit collaborative optimization and closed-loop correction.

[0010] As a further improvement to this technical solution, in step S1, the construction of a digital flow field description system for wind farms includes the following steps:

[0011] S1.1 Collect wind farm operation information, unit layout parameters and environmental data, and establish a unified data input interface;

[0012] The wind farm's operational information must include at least wind speed, wind direction, power, and yaw angle; the turbine layout parameters must include at least relative position, spacing, and azimuth angle; and the environmental data must include at least meteorological tower data and altitude.

[0013] S1.2, and preprocess the wind farm's operation information, layout parameters and environmental data;

[0014] S1.3. Generate a spatial connection matrix based on the relative positions and spacing of the units, and extract spatial features based on the spatial connection matrix;

[0015] S1.4 Utilize historical operational data to construct a time series window, extract operational trend features that change over time, and form a dynamic expression in the time dimension;

[0016] S1.5 Integrate environmental parameters to construct environmental sensitivity features that describe the changes in wind field operation characteristics under different climatic and surface conditions;

[0017] S1.6. The spatial characteristics, operational trend characteristics, and environmental sensitivity characteristics are weighted and integrated to form the data feature space of the digital flow field of the wind farm.

[0018] As a further improvement to this technical solution, in step S2, a virtual lidar measurement model is introduced based on the data feature space to perform observability calculations, and the optimal installation location of the lidar is selected and determined, including the following steps:

[0019] S2.1 Introduce a virtual lidar measurement model in the data feature space to digitally simulate the observation capabilities of different installation points in the feature space;

[0020] S2.2. Generate a set of candidate installation locations based on the relative positions and spacing of the units, and map all candidate installation location points into the feature space to form an observation node set;

[0021] S2.3 Construct the observability coverage vector for each candidate installation location and form an observability matrix;

[0022] S2.4 Calculate the comprehensive observability score of candidate points using the observability matrix;

[0023] S2.5. Based on the comprehensive observability score, a greedy algorithm is used to select a subset of 16 points from all candidate installation locations that maximizes overall observability.

[0024] S2.6 Output the final 16 optimal installation positions and generate the corresponding global observability results as input data for yaw decision.

[0025] As a further improvement to this technical solution, in step S2.1, a virtual lidar measurement model is introduced into the data feature space to digitally simulate the observation capability of different installation points on the feature space, including the following steps:

[0026] S2.11. Establish a virtual measurement model based on the measurement method of real lidar;

[0027] S2.12 Unify the coordinates of the scanning geometry and data feature space of the virtual lidar;

[0028] S2.13. Construct a virtual measurement response function in the feature space to describe the sensitivity variation of the lidar to different spatial regions, and introduce an observation sensitivity function during the construction process. , used to describe distance attenuation, view occlusion, and wake effects;

[0029] S2.14. Discretize the target area according to the scanning path of the virtual lidar, and map the observation results of each sampling point into a measurement vector in the feature space to form the digital measurement output corresponding to the installation location point;

[0030] S2.15. Repeatedly generate digital measurement outputs for all candidate installation locations.

[0031] As a further improvement to this technical solution, in S2.13, a virtual measurement response function is constructed in the feature space to describe the sensitivity changes of the lidar to different spatial regions, and an observation sensitivity function is introduced during the construction process. The specific steps involved are as follows:

[0032] The characteristic space of the wind farm is discretized into a series of observation nodes;

[0033] For each candidate installation point, define the scanning parameters of the virtual LiDAR and map the LiDAR scanning point to the feature space coordinate system;

[0034] By combining virtual measurement responses with nodal eigenvalues, a construction is made that includes an observation sensitivity function. The virtual measurement response function;

[0035] Virtual measurement response values ​​are generated based on the virtual measurement response function.

[0036] As a further improvement to this technical solution, in step S2.5, based on the comprehensive observability score, a greedy algorithm is used to select a subset of 16 points from all candidate installation locations that achieve the optimal overall observability. This includes the following steps:

[0037] S2.51. Identify all candidate installation location points and obtain the observability matrix corresponding to each candidate point;

[0038] S2.52. Initialize the set of selected installation locations to be empty;

[0039] S2.53. Filter installation location points through iterative operations;

[0040] S2.54. Repeat the iterative operation until the number of selected installation points reaches the preset 16;

[0041] S2.55 After the iteration is completed, the selected set is the optimal subset of installation points.

[0042] As a further improvement to this technical solution, in step S3, the optimal installation location is fused with the unit's three-dimensional time-series operating data, and a mapping relationship is established between the incoming flow characteristics, wake characteristics, and unit power changes. This includes the following steps:

[0043] S3.1. Based on the optimal installation location, generate incoming flow characteristics by calling the measured lidar measurement results;

[0044] S3.2. Time synchronization and alignment of lidar measurement data and unit three-dimensional time-series operation data;

[0045] S3.3. By combining the unit's spatial coordinate system and wind direction sequence with the yaw angle, wake deflection angle and wake diffusion model of the unit ahead, the dynamic wake coverage characteristics of each unit at each moment are generated, including wake velocity deficit, turbulence enhancement degree and wake overlap ratio.

[0046] S3.4. The incoming flow characteristics and dynamic wake coverage characteristics are jointly encoded according to the unit location order to generate a joint feature vector of incoming flow and wake.

[0047] S3.5 Decompose the power change sequence of downstream units and extract the characteristics of downstream power change, including power loss, wake disturbance response time and dynamic impact of yaw state switching on power.

[0048] S3.6. The future flow characteristics, dynamic wake coverage characteristics and downstream power changes are input into a multiple linear regression model, that is, a mapping relationship is constructed, and the degree of yaw impact is output.

[0049] As a further improvement to this technical solution, in step S4, based on the yaw effect law, a yaw optimization strategy is constructed, including real-time wake reconstruction, upstream yaw compensation, and downstream power recovery, comprising the following steps:

[0050] S4.1 Based on the mapping relationship in step S3.6, calculate in real time the wake deflection, wake spread range and wake intensity distribution formed by the current upstream unit under the existing yaw angle conditions;

[0051] S4.2 Based on the wake propagation model and the spatial geometric relationship of the units, the wake coverage state of all units in the wind farm is reconstructed in real time to form a high-resolution wake field;

[0052] S4.3 Identify downstream units located in the strong wake coverage area based on the real-time wake field, and identify the main upstream interference source units.

[0053] S4.4. Conduct a cost assessment for each feasible yaw angle, and jointly quantify the upstream wake deflection benefits and the yaw loss of the unit itself to form a comprehensive benefit function for yaw actions.

[0054] S4.5 Establish a downstream power recovery model based on the downstream unit power loss curve, predict the downstream power recovery amount under different upstream yaw angles, and optimize it in conjunction with the upstream revenue function;

[0055] S4.6. A distributed greedy search strategy is adopted to solve the yaw angle combination of the entire wind field, so as to achieve the best overall power recovery and wake improvement effect.

[0056] As a further improvement to this technical solution, in S4.4, a cost assessment is performed on each feasible yaw angle, and the upstream wake deflection benefit and the unit's own yaw loss are jointly quantified to form a comprehensive benefit function for yaw action, including the following steps:

[0057] S4.41. For each upstream unit, define its permissible yaw angle range, and select the distance range based on unit design limitations and safety requirements. Construct a set of feasible yaw angles;

[0058] S4.42, For each candidate yaw angle The wake deflection and wake spread width are calculated based on the wake propagation model.

[0059] S4.43. Based on the real-time wake field, assess the power changes of downstream affected units after yaw. ;

[0060] S4.44, for each candidate yaw angle According to the unit's power curve or calibration data, the power loss caused by the unit's own yaw is calculated. ;

[0061] S4.45. Quantify the upstream wake revenue and its own loss together to form a comprehensive revenue function.

[0062] As a further improvement to this technical solution, in step S5, based on the yaw optimization strategy, electric field-level yaw coordination control is performed through multi-unit collaborative optimization and closed-loop correction, including the following steps:

[0063] S5.1 Input the yaw optimization strategy results generated in step S4 and the lidar measurement data from step S2 into the group control system in real time to form a data input set for multi-unit collaborative control.

[0064] S5.2 Integrate the yaw optimization problem of all units into an overall optimization objective, namely, maximize the total power of the wind farm;

[0065] S5.3. Error calculation is performed using real-time measurement data and feature space prediction output to form a yaw control error vector;

[0066] S5.4. Use the gain adjustment algorithm to perform closed-loop adjustment of the yaw angle for single and multi-unit aircraft.

[0067] S5.5 Outputs the final yaw angle command for each unit, and generates a global yaw coordination effect index.

[0068] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0069] 1. This invention relates to a yaw optimization method applicable to densely arranged wind farms. By constructing a high-dimensional digital flow field feature space that integrates spatial topology, temporal dynamics, and environmental sensitivity, and combining virtual lidar observability analysis and wake influence law modeling, it can accurately identify the coupling effect of upstream turbine yaw on downstream turbine power. Based on this, the proposed collaborative optimization strategy, which includes real-time wake reconstruction, upstream yaw compensation, and downstream power recovery, can dynamically adjust the yaw angle of each turbine. While minimizing its own power loss, it maximizes the downstream gain brought by wake deflection, thereby effectively reducing wake interference and increasing the total power generation of the entire farm. The effect is particularly prominent in densely arranged scenarios with stable wind direction and small turbine spacing.

[0070] 2. This invention relates to a yaw optimization method suitable for densely distributed wind farms. It deeply integrates lidar measurement data, SCADA time-series operational data, and a physical wake model. Through a multi-unit distributed greedy search and closed-loop error correction mechanism, a field-level yaw coordination control system with real-time feedback and adaptive capabilities is formed. This not only overcomes the limitations of traditional single-unit independent yaw control that ignores the interaction between units, but also dynamically corrects yaw commands based on actual operational deviations, ensuring the robustness and convergence of the control strategy under complex wind conditions and equipment response delays, thereby improving the stability and intelligence level of wind farm operation. Attached Figure Description

[0071] Figure 1 This is a flowchart of the overall method of the present invention. Detailed Implementation

[0072] 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.

[0073] Example: Please refer to Figure 1 As shown in the figure, this embodiment provides a yaw optimization method suitable for densely distributed wind farms, including the following steps:

[0074] S1. Collect wind farm operation information, layout parameters, environmental data and three-dimensional time-series operation data of the units (including core indicators such as unit power, speed, pitch angle, yaw angle and its rate of change recorded by historical SCADA), and construct a digital flow field description system for wind farm to form a data feature space for yaw decision-making.

[0075] In this embodiment, the construction of a digital flow field description system for a wind farm includes the following steps:

[0076] S1.1 Collect wind farm operation information, unit layout parameters and environmental data, and establish a unified data input interface;

[0077] The wind farm's operational information must include at least wind speed, wind direction, power, and yaw angle; the turbine layout parameters must include at least relative position, spacing, and azimuth angle; and the environmental data must include at least meteorological tower data and altitude.

[0078] S1.2, and preprocess the wind farm's operation information, layout parameters and environmental data, including data cleaning and time alignment, feature standardization and coding;

[0079] S1.3. Generate a spatial connectivity matrix based on the relative positions and spacing of the generating units, and extract spatial features based on the spatial connectivity matrix, including at least upstream and downstream relationships, distance weights between adjacent generating units, and influence paths under the prevailing wind direction, to form a data structure for describing the mutual influence between generating units; specifically: based on the three-dimensional coordinate information of the generating units, first calculate the relative positions and spacing between each pair of generating units to generate a spatial connectivity matrix under wind direction reference, where the matrix elements represent the topological relationships and influence intensity between generating units; then, combine wind direction information to determine the upstream and downstream relationships of each generating unit, and assign weights to adjacent generating units based on the distance function, while tracing the wake propagation path along the prevailing wind direction, and encode the above information into a structured data representation, thereby forming a spatial feature tensor that can quantify the interaction between generating units, wake coverage, and potential power interference relationships;

[0080] S1.4. Construct a time series window using historical operational data to extract operational trend features that change over time, such as wind direction change gradient, power trend, and yaw response delay, forming a dynamic expression in the time dimension. Specifically: First, organize and align the historical operational data according to a uniform time step to construct a sliding time series window covering multiple time periods. Each window contains key indicators such as wind speed, wind direction, power, and yaw angle at continuous moments. Then, within each window, calculate the wind direction change gradient (by performing differential or finite difference fitting on the wind direction data to obtain the rate of change of wind direction per unit time as the wind direction gradient) and the power change trend (by applying differential or finite difference fitting to the power series data). Using moving linear regression or moving average methods, we extract the power change trend and slope over time, as well as the yaw angle response delay (for the yaw angle response delay, we calculate the time lag of the yaw angle change relative to the wind direction or power change (e.g., cross-correlation analysis or delayed least squares fitting) to obtain the dynamic delay characteristics of the yaw response of each unit). These dynamic characteristics are captured by differentiating, smoothing filtering, or time weighting methods to capture short-term fluctuations and long-term trends. Finally, we combine the operating trend features extracted from each window in chronological order to form a dynamic feature matrix that reflects the changes of the wind farm over time, which is used to express wind farm behavior and make yaw optimization decisions in the time dimension.

[0081] S1.5. Integrate environmental parameters to construct environmental sensitivity features describing the changes in wind farm operation characteristics under different climatic and surface conditions, including wind direction stability indicators and operating segment division labels. Specifically, the collected environmental parameters (such as meteorological tower data, long-term statistical characteristics of wind speed and direction, altitude, roughness, surface type, etc.) are spatiotemporally aligned with the unit operation data to divide the wind farm operating segments according to different climatic and surface conditions. Subsequently, for each segment, indicators such as wind direction stability, wind speed distribution characteristics, turbulence intensity changes, and power response differences are calculated, and the impact of environmental factors on unit performance is quantified into environmental sensitivity features through normalization, weighting, or classification coding methods. Finally, the environmental sensitivity features corresponding to each segment and unit are fused with spatial and temporal series features to form a high-dimensional feature tensor that can reflect the changes in wind farm operation characteristics under different climatic and surface conditions, providing environmental constraint information for yaw optimization strategies.

[0082] S1.6. The spatial characteristics, operational trend characteristics, and environmental sensitivity characteristics are weighted and fused to form the data feature space (i.e., data expression structure) of the digital flow field of the wind farm, which serves as a high-dimensional feature tensor.

[0083] S2. Based on the data feature space, a virtual lidar measurement model is introduced to perform observability calculations, and the optimal installation location of the lidar is selected and determined.

[0084] In this embodiment, a virtual lidar measurement model is introduced based on the data feature space to perform observability calculations, and the optimal installation location of the lidar is selected and determined, including the following steps:

[0085] S2.1 Introduce a virtual lidar measurement model in the data feature space to digitally simulate the observation capabilities of different installation points in the feature space;

[0086] The process involves introducing a virtual lidar measurement model into the data feature space to digitally simulate the observation capabilities of different installation points within the feature space. This includes the following steps:

[0087] S2.11. Establish a virtual measurement model based on the actual measurement method of lidar (to clarify its measurement characteristics such as scanning angle, detection radius, ranging resolution, and field of view at a given installation point):

[0088] ;

[0089] In the formula, The coordinates of the installation location point, Let be the rotation matrix, describing the LiDAR scanning angle. , The distance is the unit vector of the laser beam direction, and its length is the distance measured. It is a sequence of scan points;

[0090] S2.12 Unify the coordinates of the scanning geometry and data feature space of the virtual lidar to ensure that the measurement model can extract and project the flow field features at any position in the feature space.

[0091] S2.13. Based on the completion of coordinate unification, a virtual measurement response function is constructed in the feature space to describe the sensitivity changes of the lidar to different spatial regions, and an observation sensitivity function is introduced during the construction process. , used to describe distance attenuation, view occlusion, and wake effects;

[0092] Furthermore, in densely packed wind farms, the actual observation capability of lidar is constrained by multiple physical factors, including range attenuation, scanning angle deviation, wind turbine structural obstruction, and signal distortion caused by wake disturbances. If lidar measurement data is directly treated as ideally uniformly covered flow field information, its perception accuracy for key areas (such as the wake interaction zone between turbines) will be severely overestimated, leading to distorted yaw decision-making. Existing technologies typically use lidar data as error-free or uniformly weighted input, ignoring its spatial non-uniformity and environmental sensitivity, making it difficult to support high-precision wake reconstruction and yaw optimization. This invention constructs a virtual measurement response function in the feature space and introduces a comprehensive observation sensitivity function that includes distance attenuation, angle response, obstruction factor, and wake interference terms, achieving refined modeling of lidar measurement quality. This method not only accurately assesses the observability of each candidate installation point for key areas of the wind farm but also provides a physically consistent and weighted high-quality data foundation for subsequent greedy point selection, wake feature extraction, and yaw impact law learning, significantly improving the reliability and optimization effect of the perception and decision-making closed loop.

[0093] A virtual measurement response function is constructed in the feature space to describe the sensitivity variation of the lidar to different spatial regions, and an observation sensitivity function is introduced during the construction process. The specific steps involved are as follows:

[0094] The characteristic space of the wind farm is discretized into a series of observation nodes (each node represents a characteristic location of the local flow field, used to quantify the observation capability of the lidar).

[0095] For each candidate installation point, define the scanning parameters of the virtual lidar (such as scanning angle range, detection radius, ranging resolution, and field of view), and map the lidar scanning point to the feature space coordinate system to ensure that the scanning point can observe and project the feature space nodes.

[0096] By combining virtual measurement responses with nodal eigenvalues, a construction is made that includes an observation sensitivity function. The virtual measurement response function;

[0097] The virtual measurement response function is:

[0098] ;

[0099] ;

[0100] In the formula, In the feature space The original feature values ​​of the point For virtual measurement at point The overall sensitivity weight was determined through experimental calibration, and its value range is [0,1]. For distance decay weights, , For point Distance to radar installation point This is the distance attenuation coefficient, which controls the rate of attenuation, and its unit is m⁻¹. For the scanning angle response weight, , For laser beam and point The angle of incidence (or the angle of deviation of the scan) between them. The angular decay exponent controls the degree of decay at non-orthogonal angles. The occlusion factor, determined through experimental calibration, has a value range of [0,1]. If the point... When obstructed by wind turbine blades, towers, or other structures, the sensitivity decreases or even drops to zero; when unobstructed, it is 1. For wake or local disturbance interference factors, The value range was determined through experimental calibration and is [0,1].

[0101] Virtual measurement response values ​​are generated based on the virtual measurement response function;

[0102] S2.14. Discretize the target area according to the scanning path of the virtual lidar, and map the observation results of each sampling point into a measurement vector in the feature space to form the digital measurement output corresponding to the installation location point;

[0103] S2.15. Repeatedly generate digital measurement outputs for all candidate installation locations and record their coverage area, effective observation density, and response capability to local feature changes in the feature space.

[0104] S2.2. Generate a set of candidate installation locations based on the relative positions and spacing of the turbines (Based on the three-dimensional coordinates and spacing of each turbine in the wind farm, generate a preliminary grid set of candidate installation locations along the perimeter of the turbines and key observation areas of the wind farm (such as the upstream flow direction and the wake coverage area between turbines); Subsequently, calculate the relative distance, azimuth angle, and possible observation obstruction of each candidate point with respect to the surrounding turbines, and remove points that overlap with the turbine towers, blades, or inaccessible areas; Finally, map the remaining points that meet the requirements of safe distance and observable coverage to the feature space to form a complete set of candidate installation locations), and map all candidate installation location points to the feature space to form a quantifiable set of observation nodes;

[0105] S2.3. For each candidate installation location, construct its corresponding observability coverage vector (to describe the detection range of the point for local and global flow field features) and form an observability matrix covering the entire set of observation nodes. Specifically, for each candidate installation location, firstly, map its virtual lidar scanning parameters (including scanning angle, detection radius, ranging resolution, and field of view) to observation nodes in the wind farm feature space, calculate the set of nodes that the point can cover and the corresponding sensitivity weights, and encode the coverage into an observable coverage vector, where each element represents the effectiveness and contribution intensity of the node being observed by the installation point; then, arrange the coverage vectors of all candidate installation points by column or row to construct an observability matrix covering the entire set of observation nodes, with each row corresponding to an observation node and each column corresponding to a candidate point, thereby quantifying the detection capability of each installation point for local and global flow field features and providing basic data for comprehensive observability scoring and greedy algorithm point selection;

[0106] S2.4 Calculate the comprehensive observability score of candidate points using the observability matrix. Specifically: First, perform statistical analysis on the coverage vector corresponding to each candidate installation point, including the number of covered nodes, the total coverage sensitivity, and the redundancy coverage. Then, combine indicators such as information gain, mutual information enhancement, and redundancy suppression to quantify the observation capability of each candidate point. Node coverage quality, observation sparsity, and local feature response capability are weighted together to generate the comprehensive observability score for that point. Finally, normalize the score to a unified dimension so that the observability of different candidate points can be directly compared, providing a basis for the greedy algorithm to select the optimal installation point.

[0107] S2.5. Based on the comprehensive observability score, a greedy algorithm is used to select a subset of 16 points from all candidate installation locations that maximizes overall observability.

[0108] Based on the comprehensive observability score, a greedy algorithm is used to select a subset of 16 points from all candidate installation locations that maximize overall observability. This process includes the following steps:

[0109] S2.51. Identify all candidate installation location point sets and obtain the observability matrix corresponding to each candidate point, representing the observation capability of the point to each observation node in the feature space;

[0110] S2.52. Initialize the set of selected installation locations to be empty;

[0111] S2.53. Filtering installation location points through iterative operations: In each iteration, calculate the overall observability gain of each unselected candidate point, select the candidate point that can maximize the improvement of the current overall observability and add it to the selected set, update the overall coverage status, add the observation capability of the newly added points to the coverage status, and remove the selected points from the candidate point set.

[0112] S2.54. Repeat the iterative operation until the number of selected installation points reaches the preset 16;

[0113] S2.55 After the iteration is completed, the selected set is the optimal subset of installation points;

[0114] S2.6 Output the final 16 optimal installation positions and generate the corresponding global observability results as input data for yaw decision.

[0115] S3. The optimal installation location is fused with the three-dimensional time-series operation data of the unit, and a mapping relationship is established between the incoming flow characteristics, wake characteristics and the changes in unit power to identify the yaw effect on the power generation of downstream units.

[0116] In this embodiment, the optimal installation location is fused with the unit's three-dimensional time-series operating data, and a mapping relationship is established between incoming flow characteristics, wake characteristics, and unit power changes. This includes the following steps:

[0117] S3.1 Based on the optimal installation location, generate incoming flow characteristics by calling virtual or actual measured lidar measurement results, including at least wind speed, wind direction, shear, and turbulence intensity;

[0118] S3.2. Time synchronization and alignment of lidar measurement data and unit three-dimensional time-series operation data are performed, and interpolation, noise reduction and missing data processing methods are used to construct a fused data sequence under a unified time scale;

[0119] S3.3. Using the unit spatial coordinate system and wind direction sequence, combined with the yaw angle, wake deflection angle, and wake diffusion model (Jensen) of the upstream units, the dynamic wake coverage characteristics of each unit at each moment are generated, including wake velocity deficit, turbulence enhancement, and wake overlap ratio. Based on the three-dimensional coordinates of each unit in the wind farm and historical or real-time wind direction sequences, a unit spatial coordinate system is established, and the azimuth and distance of each unit relative to the upstream units are calculated. Subsequently, using the yaw angle of the upstream units and the wake deflection angle and diffusion width calculated by the Jensen wake diffusion model, the coverage range of the wake at the downstream unit location is determined. On this basis, the velocity deficit caused by the wake is calculated for each downstream unit (calculated based on the power output and rotor diameter of the upstream units, to determine the wake velocity attenuation). ,in, For the incoming wind speed, The disturbed wind speed at the downstream unit is used to convert the velocity deficit into a power loss index by combining the power-wind speed relationship. Turbulence intensification is calculated by superimposing the turbulence intensity generated by the upstream unit onto the wake diffusion model to determine the total turbulence intensity at the downstream location. ,in For environmental turbulence, The information is integrated into a dynamic wake coverage feature sequence by time step to provide input data for subsequent yaw optimization and power recovery modeling. This information includes the contribution of the upstream wake and the overlap ratio of the wake with the wake of other units (the area of ​​the downstream unit hub plane covered by one or more upstream wakes is calculated by geometric projection and wake coverage area to obtain the wake overlap ratio).

[0120] S3.4. The incoming flow characteristics and dynamic wake coverage characteristics are jointly encoded according to the unit location order to generate a joint feature vector of incoming flow and wake, realizing the digital expression of the coupling effect between units.

[0121] S3.5. Decompose the power change sequence of downstream units and extract downstream power change characteristics, including power loss, wake disturbance response time, and dynamic impact segment of yaw state switching on power. Specifically: First, align the power time series of downstream units by time step, and mark the start and end periods of power change according to the yaw angle change of upstream units and wake disturbance events. Then, compare the baseline power curve or environmentally corrected power prediction with the actual power to calculate the power loss caused by each disturbance event, and determine the delay time of wake disturbance on the power response of downstream units through signal analysis methods (such as cross-correlation or delayed regression). At the same time, divide the power sequence into dynamic impact segments before, during, and after yaw state switching, and quantify the immediate and delayed impact of different yaw angle combinations on power change. Finally, integrate the power loss, disturbance response time, and dynamic impact segment characteristics to form a downstream unit power change feature vector.

[0122] S3.6. Input the future flow characteristics, dynamic wake coverage characteristics and downstream power changes into the multiple linear regression model, that is, construct the mapping relationship and output the degree of yaw impact.

[0123] The multiple linear regression model takes downstream power change characteristics, upstream turbine wake coverage characteristics, and joint feature vectors such as incoming wind speed, direction, and turbulence intensity as inputs. It performs weighted summation through a single-layer linear mapping (an intermediate layer with no hidden layers or only necessary feature transformations), and adds a bias term to output a continuous target variable: the downstream power change or yaw impact degree corresponding to the yaw angle. In other words, the model outputs the target variable through the formula... Mapping multidimensional inputs to continuous outputs, where, For the input feature matrix, For the regression coefficient vector, For bias.

[0124] S4. Based on the yaw effect law, a yaw optimization strategy including real-time wake reconstruction, upstream yaw compensation and downstream power recovery is constructed to set the unit yaw angle under different wind direction and wind speed conditions.

[0125] In this embodiment, based on the yaw effect law, a yaw optimization strategy is constructed, including real-time wake reconstruction, upstream yaw compensation, and downstream power recovery, comprising the following steps:

[0126] S4.1 Based on the mapping relationship established in step S3.6, the wake deflection, wake diffusion range, and wake intensity distribution formed by the upstream unit under the current yaw angle are calculated in real time. Specifically, the multivariate linear regression mapping relationship established in step S3.7 is called, and the current yaw angle, wind speed, wind direction, and environmental characteristics of the upstream unit are used as inputs to predict the power change and wake impact parameters at the downstream unit. Subsequently, the wake deflection and diffusion radius are calculated using the upstream yaw angle, combined with the unit spatial coordinate system and the Jensen wake diffusion model. The wake intensity distribution is further calculated based on the relationship between power loss and wind speed. At each time step, the predicted wake deflection, diffusion range, and intensity distribution are projected onto the downstream unit location to form a real-time gridded representation of the wake field. This achieves a high-resolution quantitative representation of the dynamic evolution of the wake within the wind field, providing input data for real-time yaw optimization and power recovery.

[0127] S4.2. Based on the wake propagation model (Jensen model) and the spatial geometry of the turbine units, the wake coverage state of all turbine units in the wind farm is reconstructed in real time to form a high-resolution wake field that reflects the dynamic evolution of the wake. Specifically, based on the three-dimensional coordinates of all turbine units in the wind farm and the real-time wind direction information, the positional relationship of each turbine unit relative to the upstream turbine units is determined. Then, using the Jensen wake model, the wake attenuation, diffusion radius, and centerline deflection are calculated based on the power output, rotational speed, and yaw angle of each upstream turbine unit, and the wake influence is mapped to the downstream turbine units and the wind farm coverage area in a gridded manner. At each time step, the velocity deficit, turbulence enhancement, and overlapping areas of the wakes of all upstream turbine units are superimposed to form a high-resolution wake coverage matrix of the entire wind farm, which reflects the spatial distribution and dynamic evolution of the wake in real time.

[0128] S4.3. Based on the real-time wake field, identify downstream units located in areas covered by strong wakes, and identify the set of main upstream interference source units that significantly affect their power. Specifically, spatially map the real-time reconstructed high-resolution wake field with the hub positions of downstream units, calculate the wake velocity deficit and turbulence enhancement at each downstream unit, and set a threshold to determine areas covered by strong wakes. Subsequently, for downstream units located in areas with strong wakes, trace their wake sources, and by calculating the contribution ratio of the wakes of each upstream unit to the velocity attenuation at that location, screen out the set of main upstream interference source units that significantly affect power. Finally, record the correspondence between each affected downstream unit and its main upstream interference source units as structured data for use in upstream-downstream coordination and power recovery calculations in the yaw optimization strategy.

[0129] S4.4. Conduct a cost assessment for each feasible yaw angle, and jointly quantify the upstream wake deflection benefits and the yaw loss of the unit itself to form a comprehensive benefit function for yaw actions.

[0130] The process involves evaluating the costs of each feasible yaw angle, quantifying the upstream wake deflection benefits and the unit's own yaw losses together to form a comprehensive benefit function for yaw actions, including the following steps:

[0131] S4.41. For each upstream unit, define its permissible yaw angle range, and select the distance range based on unit design limitations and safety requirements. Construct a set of feasible yaw angles;

[0132] S4.42, For each candidate yaw angle The wake deflection and wake spread width are calculated based on the wake propagation model.

[0133] Among them, wake deflection for:

[0134] ;

[0135] In the formula, This refers to the yaw angle of the upstream unit; The yaw effect coefficient (obtained empirically or through calibration, dimensionless); This indicates the deflection angle of the wake centerline relative to the incoming wind direction;

[0136] Wake spread width for

[0137] ;

[0138] In the formula, The rotor radius of the unit. downstream distance, is the wake diffusion coefficient, which is dimensionless;

[0139] S4.43. Based on the real-time wake field, assess the power changes of downstream affected units after yaw. This refers to the power recovery caused by wake deflection. Specifically, it involves mapping the wake deflection, diffusion radius, and velocity distribution calculated by the upstream unit at the candidate yaw angle to the hub position of the downstream unit, determining the wake velocity deficit and turbulence enhancement experienced by each downstream unit. Subsequently, combining the downstream unit power curves or empirical power-wind speed relationships, the local wind speed attenuation caused by the wake is converted into power change. ;

[0140] S4.44, for each candidate yaw angle According to the unit's power curve or calibration data, the power loss caused by the unit's own yaw is calculated. Specifically, this involves obtaining the reference power output at a normal yaw angle (usually zero yaw) based on the unit's design calibration data or power curve. Subsequently, for candidate yaw angles, the factors such as changes in frontal area, angle of attack shift, and aerodynamic efficiency reduction caused by the yaw angle are converted into power attenuation through a power reduction model or empirical correction function. ,in, Indicates that the unit is Actual output power at yaw angle;

[0141] S4.45. Quantify the upstream wake revenue and its own loss together to form a comprehensive revenue function. ;

[0142] In the formula, As a weight for downstream power recovery, The weights for the yaw loss are dimensionless positive real numbers, and their values ​​are determined through optimization using historical wind farm operating data or heuristic rules based on the relative position of the turbines and the wind direction distribution, satisfying the following conditions: Normalization constraints;

[0143] S4.5. Establish a downstream power recovery model based on the downstream unit power loss curve, predict the downstream power recovery amount under different upstream yaw angles, and optimize it in conjunction with the upstream benefit function. Specifically, based on the historical power loss curves of downstream units under different wake disturbance conditions, use a regression model or response surface methodology to fit the power recovery function, and correlate the downstream power loss amount with the upstream unit yaw angle, wake intensity, and coverage ratio. Subsequently, for each candidate upstream yaw angle, use the model to predict the downstream unit power recovery amount and calculate its contribution to the overall wind farm power. Finally, link the predicted downstream power recovery amount with the wake deflection benefit of the upstream unit, and integrate them through weighted or optimization algorithms to form a total comprehensive benefit index, which is used to select the optimal yaw angle combination to achieve coordinated power recovery and wake optimization between upstream and downstream units.

[0144] S4.6. A distributed greedy search strategy is adopted to solve the yaw angle combination for the entire wind farm, so as to optimize the overall power recovery and wake improvement effect. Specifically, the feasible yaw angle set of the units in the wind farm and the corresponding comprehensive benefit function are distributed to the distributed computing units. Each unit independently calculates the gain of its managed unit on downstream power recovery and wake improvement under different yaw angles. Then, in each iteration, each unit selects the yaw angle that maximizes the local comprehensive benefit and updates the selection to the global wake field state. At the same time, the gain information is shared with adjacent units to realize upstream and downstream information interaction. During the iteration process, the yaw angle of each unit is updated in turn, and the power recovery and wake improvement effect of the entire wind farm is accumulated until the yaw angle combination of all units converges or reaches the preset number of iterations (the convergence condition is defined as: in two consecutive iterations, the change of the total power prediction value of the entire field is less than the threshold d (e.g., d is 0.1% of the rated total power), or the maximum number of iterations L is reached (e.g., L is 50 times)). Finally, the optimal yaw angle combination of the entire wind farm is output.

[0145] S5. Based on the yaw optimization strategy, electric field-level yaw coordination control is performed through multi-unit collaborative optimization and closed-loop correction.

[0146] In this embodiment, based on the yaw optimization strategy, electric field-level yaw coordination control is performed through multi-unit collaborative optimization and closed-loop correction, including the following steps:

[0147] S5.1. Input the yaw optimization strategy results generated in step S4 and the lidar measurement data from step S2 into the group control system in real time to form a data input set for multi-unit collaborative control. The input data includes the current yaw angle, power output prediction, correlation weight with neighboring units, and observability indicators of each unit. The group control system refers to a management platform used for centralized or distributed coordinated control of all units in the entire wind farm. Its main functions include real-time acquisition of the operating status of each unit (such as power output, wind speed, yaw angle, speed, etc.), monitoring wind farm environmental data, executing yaw, power, pitch, and other control strategies, and issuing control commands to each unit according to the overall wind farm optimization objectives (such as maximizing total power, minimizing wake loss, or power balance). At the same time, the group control system can perform closed-loop correction of the control feedback of each unit, realize upstream and downstream coordination between units, wake regulation, and global power optimization, and improve the overall power generation efficiency and operational stability of the wind farm.

[0148] S5.2. Integrate the yaw optimization problem of all units into an overall optimization objective, namely, maximizing the total power of the wind farm. Specifically, collect the candidate yaw angle set and corresponding comprehensive benefit function for each unit in the wind farm to form the yaw state space of the entire wind farm. Then, integrate the downstream power recovery and its own power loss of each unit into a local comprehensive benefit through weighting, and establish the total power objective function across the entire wind farm. ,in, Total number of generating units, power of each generating unit Related to all crew yaw angle combinations, For unit indexing;

[0149] S5.3. Error calculation is performed by comparing real-time measurement data with the predicted output of the feature space to form a yaw control error vector. Specifically, real-time observation data such as wind speed, wind direction, yaw angle, and power output of each unit are acquired from lidar measurements, SCADA systems, and other sensors. Subsequently, these real-time measurement data are compared one by one with the expected values ​​output by the wake prediction model or power prediction model constructed based on the feature space to calculate the yaw angle deviation, power deviation, and wake response error of each unit. Finally, the yaw control errors of all units are combined into an error vector according to a unified format.

[0150] S5.4. A gain adjustment algorithm is used to perform closed-loop adjustment of the yaw angle for single-aircraft and multi-aircraft groups, reducing the overall yaw strategy error. Specifically, the yaw control error vector is input into the gain adjustment algorithm, based on the preset single-aircraft gain... and multi-unit coupling gain Calculate the yaw angle correction for each generator unit. ,in, Indicates the relationship with the first A group of generator units that have wake effects or are spatially adjacent to each other. For the first The self-yaw control error of a generator unit is the difference between the unit's current yaw angle or power output and the predicted or expected value in the characteristic space. In order to be with the first The Taiwanese unit has wake effects or spatial proximity. Yaw error of the unit, single unit gain and multi-unit coupling gain The initial value can be set according to the dynamic response characteristics of the unit (e.g.) Typical initial values ​​are 0.05 to 0.2. The value is 0.01~0.05, and it can be adaptively adjusted according to the error convergence speed during closed-loop operation; then, the correction is applied to the current yaw angle of each unit to realize closed-loop adjustment, and the wake field prediction and power output are updated in real time; this process is iterated until the yaw error converges or the optimization target is reached, thereby realizing the full-field yaw optimization control of single unit and multi-unit collaboration;

[0151] S5.5 Output the final yaw angle command for each unit (the final yaw angle command is the current yaw angle plus the correction calculated by closed-loop control). The system obtains data and updates the group control system database to execute control and the next iteration. It also generates global yaw coordination performance indicators, including the wind farm's total power increase, yaw coordination degree, and control convergence, to evaluate the effectiveness of the strategy.

[0152] 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 yaw optimization method suitable for densely packed wind farms, characterized in that, Includes the following steps: S1. Collect wind farm operation information, layout parameters, environmental data and three-dimensional time-series operation data of the units, and construct a digital flow field description system for the wind farm to form a data feature space for yaw decision-making; S2. Based on the data feature space, a virtual lidar measurement model is introduced to perform observability calculations, and the optimal installation location of the lidar is selected and determined. S3. The optimal installation location is fused with the three-dimensional time-series operation data of the unit, and a mapping relationship is established between the incoming flow characteristics, wake characteristics and the power change of the unit, in order to identify the yaw effect on the power generation of the downstream unit. S4. Based on the yaw effect law, a yaw optimization strategy including real-time wake reconstruction, upstream yaw compensation and downstream power recovery is constructed to set the unit yaw angle under different wind direction and wind speed conditions. S5. Based on the yaw optimization strategy, yaw coordination control is carried out through multi-unit collaborative optimization and closed-loop correction.

2. The yaw optimization method suitable for densely packed wind farms according to claim 1, characterized in that: In step S1, the construction of a digital flow field description system for wind farms includes the following steps: S1.1 Collect wind farm operation information, unit layout parameters and environmental data, and establish a unified data input interface; The wind farm's operational information must include at least wind speed, wind direction, power, and yaw angle; the turbine layout parameters must include at least relative position, spacing, and azimuth angle; and the environmental data must include at least meteorological tower data and altitude. S1.2, and preprocess the wind farm's operation information, layout parameters and environmental data; S1.

3. Generate a spatial connection matrix based on the relative positions and spacing of the units, and extract spatial features based on the spatial connection matrix; S1.4 Utilize historical operational data to construct a time series window, extract operational trend features that change over time, and form a dynamic expression in the time dimension; S1.5 Integrate environmental parameters to construct environmental sensitivity features that describe the changes in wind field operation characteristics under different climatic and surface conditions; S1.

6. The spatial characteristics, operational trend characteristics, and environmental sensitivity characteristics are weighted and integrated to form the data feature space of the digital flow field of the wind farm.

3. The yaw optimization method suitable for densely packed wind farms according to claim 1, characterized in that: In step S2, a virtual lidar measurement model is introduced based on the data feature space to perform observability calculations, and the optimal installation location of the lidar is selected and determined, including the following steps: S2.1 Introduce a virtual lidar measurement model in the data feature space to digitally simulate the observation capabilities of different installation points in the feature space; S2.

2. Generate a set of candidate installation locations based on the relative positions and spacing of the units, and map all candidate installation location points into the feature space to form an observation node set; S2.3 Construct the observability coverage vector for each candidate installation location and form an observability matrix; S2.4 Calculate the comprehensive observability score of candidate points using the observability matrix; S2.

5. Based on the comprehensive observability score, a greedy algorithm is used to select a subset of 16 points from all candidate installation locations that maximizes overall observability. S2.6 Output the final 16 optimal installation positions and generate the corresponding global observability results as input data for yaw decision.

4. The yaw optimization method suitable for densely packed wind farms according to claim 3, characterized in that: In step S2.1, a virtual lidar measurement model is introduced into the data feature space to digitally simulate the observation capabilities of different installation points in the feature space, including the following steps: S2.

11. Establish a virtual measurement model based on the measurement method of real lidar; S2.12 Unify the coordinates of the scanning geometry and data feature space of the virtual lidar; S2.13, Constructing a virtual measurement response function in feature space to describe the variation in sensitivity of the lidar to different spatial regions, and incorporating an observed sensitivity function in the construction process to describe range attenuation, viewshed occlusion, and wake effects; S2.

14. Discretize the target area according to the scanning path of the virtual lidar, and map the observation results of each sampling point into a measurement vector in the feature space to form the digital measurement output corresponding to the installation location point; S2.

15. Repeatedly generate digital measurement outputs for all candidate installation locations.

5. The yaw optimization method suitable for densely packed wind farms according to claim 4, characterized in that: In S2.13, a virtual measurement response function is constructed in the feature space to describe the sensitivity variation of the laser radar to different spatial regions, and an observation sensitivity function is introduced in the construction process , involving the following specific steps: The characteristic space of the wind farm is discretized into a series of observation nodes; For each candidate installation point, define the scanning parameters of the virtual LiDAR and map the LiDAR scanning point to the feature space coordinate system; combining the virtual measurement response with the nodal eigenvalues to construct a virtual measurement response function including an observation sensitivity function ; Virtual measurement response values ​​are generated based on the virtual measurement response function.

6. The yaw optimization method suitable for densely packed wind farms according to claim 3, characterized in that: In step S2.5, based on the comprehensive observability score, a greedy algorithm is used to select a subset of 16 points from all candidate installation locations that maximize overall observability. This includes the following steps: S2.

51. Identify all candidate installation location points and obtain the observability matrix corresponding to each candidate point; S2.

52. Initialize the set of selected installation locations to be empty; S2.

53. Filter installation location points through iterative operations; S2.

54. Repeat the iterative operation until the number of selected installation points reaches the preset 16; S2.55 After the iteration is completed, the selected set is the optimal subset of installation points.

7. The yaw optimization method suitable for densely packed wind farms according to claim 1, characterized in that: In step S3, the optimal installation location is fused with the unit's three-dimensional time-series operating data, and a mapping relationship is established between the incoming flow characteristics, wake characteristics, and unit power changes. This includes the following steps: S3.

1. Based on the optimal installation location, generate incoming flow characteristics by calling the measured lidar measurement results; S3.

2. Time synchronization and alignment of lidar measurement data and unit three-dimensional time-series operation data; S3.

3. By combining the unit's spatial coordinate system and wind direction sequence with the yaw angle, wake deflection angle and wake diffusion model of the unit ahead, the dynamic wake coverage characteristics of each unit at each moment are generated, including wake velocity deficit, turbulence enhancement degree and wake overlap ratio. S3.

4. The incoming flow characteristics and dynamic wake coverage characteristics are jointly encoded according to the unit location order to generate a joint feature vector of incoming flow and wake. S3.5 Decompose the power change sequence of downstream units and extract the characteristics of downstream power change, including power loss, wake disturbance response time and dynamic impact of yaw state switching on power. S3.

6. The future flow characteristics, dynamic wake coverage characteristics and downstream power changes are input into a multiple linear regression model, that is, a mapping relationship is constructed, and the degree of yaw impact is output.

8. The yaw optimization method suitable for densely packed wind farms according to claim 7, characterized in that: In step S4, based on the yaw effect law, a yaw optimization strategy is constructed, which includes real-time wake reconstruction, upstream yaw compensation, and downstream power recovery. The strategy includes the following steps: S4.1 Based on the mapping relationship in step S3.6, calculate in real time the wake deflection, wake spread range and wake intensity distribution formed by the current upstream unit under the existing yaw angle conditions; S4.2 Based on the wake propagation model and the spatial geometric relationship of the units, the wake coverage state of all units in the wind farm is reconstructed in real time to form a high-resolution wake field; S4.3 Identify downstream units located in the strong wake coverage area based on the real-time wake field, and identify the main upstream interference source units. S4.

4. Conduct a cost assessment for each feasible yaw angle, and jointly quantify the upstream wake deflection benefits and the yaw loss of the unit itself to form a comprehensive benefit function for yaw actions. S4.5 Establish a downstream power recovery model based on the downstream unit power loss curve, predict the downstream power recovery amount under different upstream yaw angles, and optimize it in conjunction with the upstream revenue function; S4.

6. A distributed greedy search strategy is adopted to solve the yaw angle combination of the entire wind field, so as to achieve the best overall power recovery and wake improvement effect.

9. The yaw optimization method suitable for densely packed wind farms of claim 8, wherein: In S4.4, the cost of each feasible yaw angle is evaluated, and the upstream wake deflection benefit and the yaw loss of the unit itself are jointly quantified to form a comprehensive benefit function for yaw action, including the following steps: S4.

41. For each upstream unit, define its permissible yaw angle range, and select the distance range based on unit design limitations and safety requirements. Construct a set of feasible yaw angles; S4.42, for each candidate yaw angle calculating a wake deflection and a wake spread width according to a wake propagation model; S4.43, based on the real-time wake field, assess the power change of the downstream affected aircraft after yawing ; S4.44, for each candidate yaw angle , calculate the power loss due to yaw of the engine itself from the engine power curve or calibration data ; S4.

45. Quantify the upstream wake revenue and its own loss together to form a comprehensive revenue function.

10. The yaw optimization method suitable for densely packed wind farms of claim 1, wherein: In step S5, based on the yaw optimization strategy, electric field-level yaw coordination control is performed through multi-unit collaborative optimization and closed-loop correction, including the following steps: S5.1 Input the yaw optimization strategy results generated in step S4 and the lidar measurement data from step S2 into the group control system in real time to form a data input set for multi-unit collaborative control. S5.2 Integrate the yaw optimization problem of all units into an overall optimization objective, namely, maximize the total power of the wind farm; S5.

3. Error calculation is performed using real-time measurement data and feature space prediction output to form a yaw control error vector; S5.

4. Use the gain adjustment algorithm to perform closed-loop adjustment of the yaw angle for single and multi-unit aircraft. S5.5 Outputs the final yaw angle command for each unit, and generates a global yaw coordination effect index.

Images