An adaptive cooperative yaw optimization control system for wind farm wake effect

Abstract

This invention discloses an adaptive cooperative yaw optimization control system for wind farm wake effects, relating to the field of intelligent control technology for wind farms. The system includes: predicting the spatiotemporal trajectory of the wake using a wake prediction model and simultaneously outputting a prediction uncertainty tensor; optimizing the wake's spatiotemporal trajectory using an online controller, aiming to maximize the total power generation of the entire wind farm while embedding unit operation safety constraints; dynamically adjusting the unit operation safety constraints based on the prediction uncertainty tensor, and solving for the optimal cooperative yaw angle command sequence based on a rolling time-domain optimization mechanism; dynamically adjusting the multi-dimensional dynamic safety boundary in conjunction with the prediction uncertainty tensor before issuing the optimal cooperative yaw angle command sequence; comparing and verifying the command sequence with the multi-dimensional dynamic safety boundary and performing soft constraint contraction; obtaining the verified command sequence and issuing it to the upstream units for execution; thus achieving the cooperative maximization of power generation efficiency of the wind farm cluster in a complex flow field.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for wind power plants, and in particular to an adaptive cooperative yaw optimization control system for wind farm wake effect. Background Technology

[0002] Coordinated yaw control technology for wind farms, as a key approach to improving the overall efficiency of wind farm clusters, has evolved from setting static yaw angles to dynamic optimization considering the spatiotemporal evolution of the wake. Existing technologies typically rely on fluid dynamics models or data-driven methods, acquiring flow field information through sensors deployed throughout the wind farm area, and using this information to solve for the optimal yaw strategy under specific operating conditions. For example, a steady-state wake model can be constructed through lidar feedforward measurements or SCADA data fusion from the turbines, thereby generating a yaw command set aimed at improving the output of downstream turbines. These methods have achieved coordination between turbines to some extent, providing a preliminary solution for active wind farm control.

[0003] However, the core limitation of existing yaw optimization techniques lies in the fact that the unit operation safety constraints in the optimization framework they rely on are mostly pre-set fixed thresholds, lacking the ability to adaptively and elastically adjust based on the uncertainty of real-time flow field predictions. This leads to the cooperative yaw strategy often becoming overly conservative under conditions of drastic changes in wind speed and direction or low confidence in flow field predictions, unnecessarily sacrificing power generation potential. Conversely, under steady-state conditions with high prediction confidence, the fixed constraint boundaries cannot provide sufficient optimization space for yaw optimization, making it difficult to achieve a dynamic optimal trade-off between overall power generation efficiency and operational safety, resulting in wasted power generation potential. Furthermore, the cooperative yaw angle command sequence obtained from the optimization solution lacks a final safety verification and smoothing correction mechanism based on uncertainty quantification before final execution, failing to proactively suppress potential over-limit yaw commands, ultimately hindering the wind farm from achieving global optimization of safety and efficiency across all operating conditions. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an adaptive collaborative yaw optimization control system for wind farm wake effect to solve the problem of wasted power generation potential caused by fixed safety constraints under uncertain operating conditions.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an adaptive cooperative yaw optimization control system for wind farm wake effect, comprising: The flow field module constructs a real-time reliable flow field map based on the lidar data of the core unit and the operating and environmental data of the radiation unit. The prediction module, based on real-time reliability flow field maps, predicts the spatiotemporal trajectory of the wake using a wake prediction model and simultaneously outputs the prediction uncertainty tensor. The decision-making module, based on the wake spatiotemporal trajectory, optimizes the controller online to maximize the total power generation of the entire site and embeds unit operation safety constraints; it dynamically adjusts the unit operation safety constraints according to the prediction uncertainty tensor and solves the optimal cooperative yaw angle command sequence based on the rolling time domain optimization mechanism. Before issuing the optimal coordinated yaw angle command sequence, the verification module dynamically adjusts the multi-dimensional dynamic safety boundary by combining the prediction uncertainty tensor; it compares and verifies the command sequence with the multi-dimensional dynamic safety boundary and performs soft constraint contraction; and it obtains the verified command sequence and issues it to the upstream unit for execution. The evolution module acquires wind farm operation data after execution and performs joint adaptive optimization on the wake prediction model and online optimization controller based on the deviation between the wind farm operation data and the expected results.

[0007] Furthermore: the method for constructing the real-time reliability flow field map includes: The lidar sensing data of the core unit and the operation and environmental data of the radiating units are fused together; and a dynamic propagation operator is calculated based on the wind farm's operating status and spatial topology characteristics to generate the propagation weight matrix of the incoming flow from the core unit to the radiating units. Based on the mapping relationship between the incoming flow propagation weight matrix and the unit operating characteristics, the incoming flow status in front of the radiating unit is calculated and physical consistency calibration is performed to obtain virtual incoming flow information; and it is then fused with the real-time feedforward measurement data of the core unit lidar according to the data confidence level and subjected to physical constraint filtering to obtain the fused wind measurement data field. The fused wind measurement data field is input into the three-dimensional interpolation model to reconstruct the volumetric flow field, perform uncertainty quantification and confidence labeling, and output a real-time confidence flow field map.

[0008] Furthermore: the method for predicting the wake spatiotemporal trajectory and simultaneously outputting the prediction uncertainty tensor includes: The real-time confidence flow field map is input into the wake prediction model for graph structure transformation. Each wind turbine is used as a node of the graph to construct a directed graph, and the incoming flow propagation weight matrix is ​​used as the initial directed edge weights for initialization. The confidence of the core unit lidar measurement points is mapped to the edge weight attenuation factor in the graph to suppress the propagation of low confidence information in the graph structure. Message-passing reasoning with embedded aerodynamic priors is performed on a directed graph to obtain the wake spacetime trajectory in the future rolling time domain; where the aerodynamic priors include the velocity deficit propagation law, wake diffusion characteristics and wake deflection relationship caused by yaw; While generating the wake spacetime trajectory, a prediction uncertainty assessment is performed, and the prediction uncertainty tensor corresponding to the wake spacetime trajectory is output.

[0009] Furthermore: the method for solving the optimal cooperative yaw angle command sequence includes: Based on the wake spatiotemporal trajectory, an adaptive cooperative yaw roll optimization problem is established through an online optimization controller. This adaptive cooperative yaw roll optimization problem aims to maximize the total power generation of the entire plant and incorporates unit operation safety constraints. Based on the prediction uncertainty tensor, the unit operation safety constraints in the adaptive cooperative yaw roll optimization problem are dynamically adjusted; the adjusted unit operation safety constraints are used to replace the original unit operation safety constraints to form the adjusted adaptive cooperative yaw roll optimization problem. The online optimization controller is based on the rolling time-domain optimization mechanism to solve the adjusted adaptive cooperative yaw rolling optimization problem and obtain the optimal cooperative yaw angle command sequence.

[0010] Furthermore: the method for establishing the adaptive cooperative yaw roll optimization problem includes: The online optimization controller uses the wake spatiotemporal trajectory as the external prediction input; and extracts the flow field evolution information in the future rolling time domain based on the wake spatiotemporal trajectory to construct the predicted state sequence for optimization solution; Based on the predicted state sequence, and according to the impact of future flow field evolution information on the overall power generation efficiency of the wind farm, the future power generation changes corresponding to different cooperative yaw angle strategies are calculated; with the goal of maximizing the total power generation of the entire field, the expected power generation at each time step in the future rolling time domain is accumulated to form the objective function of the optimization problem. The unit operation safety constraints are transformed into mathematical constraints related to the coordinated yaw angle command sequence, and the activation state and boundary value of each mathematical constraint are dynamically determined based on the future flow field evolution information. By combining the objective function and mathematical constraints, an adaptive cooperative yaw rolling optimization problem is established.

[0011] Furthermore: the method for dynamically adjusting the safety constraints of the operating unit includes: The online optimization controller converts the wake prediction error, characterized by the prediction uncertainty tensor, into the constraint confidence level of the unit operation safety constraints in the future rolling time domain; and converts the constraint confidence level into the dynamic relaxation factor and tightening factor of the unit operation safety constraints. By using relaxation and tightening factors, the boundary values ​​of unit operation safety constraints are corrected in real time, and the adjusted unit operation safety constraints are generated.

[0012] Furthermore: the method for obtaining the verified instruction sequence and sending it to the upstream unit for execution includes: Based on the prediction uncertainty tensor, uncertainty propagation analysis is performed on the current optimal cooperative yaw angle command sequence to obtain the trigger probability of the multi-dimensional dynamic safety boundary in the future execution time domain. The trigger probability is then converted into the corresponding boundary confidence level, and the multi-dimensional dynamic safety boundary is dynamically adjusted based on the boundary confidence level. The optimal cooperative yaw angle command sequence is compared and verified with the adjusted multi-dimensional dynamic safety boundary. If the optimal cooperative yaw angle command sequence is found to have a risk of exceeding the boundary, soft constraint contraction is performed on the optimal cooperative yaw angle command sequence to ensure that the safety execution requirements are met. Obtain the verified optimal coordinated yaw angle command sequence and send it to the upstream unit for execution.

[0013] Furthermore: the method for dynamically adjusting multi-dimensional dynamic security boundaries includes: Compare the trigger probability of the multi-dimensional dynamic security boundary with the preset security threshold and the aggressive threshold; When the risk probability is higher than the safety threshold, the corresponding dimension's safety boundary is tightened exponentially according to the extent of the excess; when the risk probability falls between the safety threshold and the aggressive threshold, the corresponding dimension's safety boundary is tightened moderately in a linear manner according to its value to balance safety and efficiency; when the risk probability is lower than the aggressive threshold, the corresponding dimension's safety boundary is relaxed linearly according to the extent of the lower risk.

[0014] Furthermore: the execution of soft-constraint contraction includes: Select the portion of the optimal cooperative yaw angle command sequence whose distance from the multi-dimensional dynamic safety boundary is less than a preset safety buffer threshold; For the portion where the distance is less than the preset safety buffer threshold, the soft constraint shrinkage is calculated based on its distance from the multi-dimensional dynamic safety boundary and the predicted uncertainty tensor. Apply the soft constraint contraction amount to the optimal cooperative yaw angle command sequence, correct the optimal cooperative yaw angle command value, and generate the optimal cooperative yaw angle command sequence after soft constraint contraction, which is used as the optimal cooperative yaw angle command sequence that has passed the verification.

[0015] Furthermore: the method for performing joint adaptive optimization includes: Receive the wind farm operation data after execution, compare it with the expected results, and obtain the deviation between the wind farm operation data and the expected results; Using this deviation as a driving signal, the parameter adjustments for the wake prediction model and the online optimization controller used to minimize the deviation are calculated, and the parameters of both are synchronously corrected and updated accordingly.

[0016] The beneficial effects of this invention are as follows: By quantifying the uncertainty of wake prediction, a dual adaptive optimization mechanism for coordinated yaw across the entire wind farm is driven, ultimately improving the overall power generation efficiency. At the collaborative decision-making level, based on the uncertainty of global wake prediction, the operational safety constraints of each unit are dynamically adjusted. This ensures that the yaw operation of upstream units, while fully considering its impact on downstream units and corresponding uncertainties, can actively guide the wake to enhance the wind energy captured by downstream units, while avoiding sacrificing its own power generation potential due to rigid constraints, thus achieving dynamic optimization of the total power generation across the entire wind farm. At the collaborative execution level, uncertainty is again used to uniformly verify and smoothly fine-tune the yaw command sequence, ensuring that the yaw actions of each unit are coordinated and consistent within a safe range, avoiding cascading performance losses caused by single-unit over-limits. By embedding uncertainty management into the entire decision-making and execution process of collaborative control, the collaborative maximization of power generation efficiency of the wind farm cluster in complex flow fields is achieved. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the adaptive cooperative yaw optimization control system for wind farm wake effect in this invention.

[0019] Figure 2 This is a flowchart of adjusting the unit's operational safety constraints in this invention.

[0020] Figure 3 This is a flowchart illustrating the process of obtaining the optimal cooperative yaw command sequence in this invention.

[0021] Figure 4 This is a flowchart of the verification module in this invention. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figure 1 , Figure 2 , Figure 3 and Figure 4 This is one embodiment of the present invention, which provides an adaptive cooperative yaw optimization control system for wind farm wake effect, comprising the following steps: Methods for constructing real-time reliability flow field maps include: The system integrates lidar sensing data from the core turbine units with operational and environmental data from the radiating turbine units; and calculates dynamic propagation operators based on the wind farm's operating status and spatial topology characteristics to generate a propagation weight matrix from the core turbine units to the radiating turbine units.

[0026] It should be noted that the upstream wind speed and direction information provided by the lidar sensing data of the core unit should be spatiotemporally aligned with the power and speed information in the operating data of the radiation unit and the environmental wind speed and direction information in the environmental data. The dynamic propagation operator is calculated as follows: The Euclidean distance between the core and radiating units is calculated based on their coordinates. The propagation intensity decreases continuously as this distance increases. Specifically, the propagation intensity is a function with a natural constant as the base and the product of the distance and a negative attenuation coefficient (e.g., -0.01) as the exponent. The angle between the current prevailing wind direction and the line connecting the core and radiating units is calculated. If this angle is less than a specific angle (e.g., 30 degrees), a higher intensity coefficient (e.g., 1.0) is assigned. This intensity coefficient then decreases at a fixed rate as the angle increases until it reaches a lower limit (e.g., 0.2). The elevation difference between the two locations is calculated based on the difference in elevation data. The propagation intensity decreases linearly based on the product of this elevation difference and a reduction coefficient (e.g., 0.005). The three intensity coefficients calculated from the distance, wind direction angle, and elevation difference are multiplied together to obtain the dynamic propagation operator between the units. This product is then arranged with the core unit as the row and the radiating units as the column to form the incoming flow propagation weight matrix.

[0027] Based on the mapping relationship between the incoming flow propagation weight matrix and the unit operating characteristics, the incoming flow status in front of the radiating unit is calculated and physical consistency calibration is performed to obtain virtual incoming flow information; and then, the data is fused with the real-time feedforward measurement data of the core unit's lidar according to the data confidence level and subjected to physical constraint filtering to obtain the fused wind measurement data field.

[0028] It should be noted that, based on the incoming flow propagation weight matrix, the lidar incoming flow information of the core unit is transmitted to each radiating unit according to weights; using the power and rotational speed data of the radiating units, the local incoming flow wind speed of the radiating unit is estimated in reverse through its characteristic mapping relationship with wind speed. The lidar incoming flow information and the local incoming flow wind speed are then weighted and synthesized to obtain the incoming flow status in front of the radiating units; The incoming flow conditions in front of the radiant turbine are constrained within the cut-in and cut-out wind speed range of the wind turbine. The rate of change of wind direction is limited to no more than 10 degrees per minute, in accordance with the physical characteristics of the atmospheric boundary layer. The calibrated incoming flow conditions in front of the radiant turbine are substituted into the power characteristic curve of the wind turbine to calculate the theoretical power, and the deviation between this and the actual power recorded by SCADA is not more than 5% of the rated power, thus completing the physical consistency calibration. The calibrated virtual inflow information from each radiating unit is combined with real-time measurement data from the core unit's lidar. A high confidence weight (e.g., 0.8) is assigned to the lidar data, while a lower weight (e.g., 0.2) is assigned to the virtual inflow information, and a weighted average is calculated. Subsequently, physical constraint filtering is performed to remove data points with negative wind speeds, wind speeds exceeding the cutoff wind speed, or wind direction changes exceeding 45 degrees instantaneously, generating a spatially continuous and physically reasonable fused wind measurement data field.

[0029] The fused wind measurement data field is input into the three-dimensional interpolation model to reconstruct the volumetric flow field, perform uncertainty quantification and confidence labeling, and output a real-time confidence flow field map.

[0030] It should be noted that when inputting the fused wind measurement data field into the three-dimensional interpolation model, the three-dimensional interpolation model first establishes a regular three-dimensional spatial grid within the wind farm area, and layers it by height. Then, based on the wind speed and wind direction values ​​of the lidar measurement points of each core unit, weights are assigned according to the spatial distance between the measurement points and the grid points, and the data confidence of the measurement points themselves is combined to make high-confidence data contribute more to the interpolation process. By weighted summarization of multiple measurement points in the neighborhood, wind speed and wind direction estimates of each height layer and each spatial grid point can be obtained, forming continuous three-dimensional wind field basic data. After obtaining the basic three-dimensional wind field data, in order to further determine the reliability of the wind field, the consistency of the surrounding measurement points is calculated for each three-dimensional spatial grid point, including whether there are abrupt changes in the measured wind speed and direction, and whether the data from different sources at the same location are similar, etc., which serve as the basis for measuring uncertainty. The higher the uncertainty, the lower the reliability of the wind field data at that location. The uncertainty is converted into the corresponding confidence level, and the confidence level information is attached to each grid point of the three-dimensional wind field. The generated real-time reliability flow field map simultaneously includes the wind speed field, wind direction field, and point-by-point confidence level information.

[0031] Methods for predicting wake spatiotemporal trajectories and simultaneously outputting the prediction uncertainty tensor include: The real-time confidence flow field map is input into the wake prediction model for graph structure transformation. Each wind turbine is used as a node of the graph to construct a directed graph, and the incoming flow propagation weight matrix is ​​used as the initial directed edge weights for initialization. The confidence of the core unit lidar measurement points is mapped to the edge weight attenuation factor in the graph to suppress the propagation of low confidence information in the graph structure.

[0032] It should be noted that the wake prediction model can employ a hybrid structure of convolutional neural networks (CNNs) and graph neural networks (GNNs). The CNNs are responsible for extracting spatial wind speed and direction gradient features from the real-time reliability flow field map, while the GNNs are used to describe the wake propagation relationships between wind turbines. After inputting the real-time reliability flow field map into the wake prediction model, each wind turbine within the wind farm is defined as a node in the graph structure based on its spatial location. The incoming flow propagation weight matrix is ​​then used to map its elements to initial directed edge weights, ensuring that the edge weights reflect the influence of the core turbine on the radiating turbines. Simultaneously, the confidence level of the core turbine's lidar measurement points is used as an edge weight attenuation factor to suppress the propagation of low-confidence information within the graph structure, forming a directed graph that combines spatial topological constraints with data reliability adjustment.

[0033] Message-passing reasoning with embedded aerodynamic priors is performed on a directed graph to obtain the wake spacetime trajectory in the future rolling time domain; where the aerodynamic priors include the velocity deficit propagation law, wake diffusion characteristics, and wake deflection relationship caused by yaw.

[0034] It should be noted that when performing message passing reasoning with embedded aerodynamic priors on a directed graph, in each message passing process, the influence intensity of the core unit on the downstream radiating units is controlled according to the initial directed edge weights of the directed graph. In the node update process, the propagation law of velocity loss, wake diffusion characteristics, and wake deflection relationship caused by yaw are explicitly introduced, so that each node follows the above aerodynamic prior constraints when receiving messages from upstream nodes. By iterating the message passing process multiple times within the rolling prediction time window, the node state is continuously updated in the time dimension, thereby forming the wake spatiotemporal trajectory in the future rolling time domain. Furthermore, the velocity loss propagation law refers to the momentum loss caused by upstream wind turbines propagating downstream, causing downstream wind speeds to fall below free wind speeds; the wake diffusion characteristics indicate that the velocity loss region gradually expands laterally and vertically with propagation distance, and the degree of loss gradually weakens; the wake deflection relationship caused by yaw refers to the yaw of upstream wind turbines changing the relative position of the rotor and the incoming flow, causing the wake centerline to deflect laterally and extend with propagation. Aerodynamic priors, based on the momentum influence of the turbines and the characteristics of the atmospheric boundary layer, can serve as physical constraints during node state updates, ensuring that message transmission conforms to both the graph structure propagation path and the evolution of the wake in real space.

[0035] While generating the wake spacetime trajectory, a prediction uncertainty assessment is performed, and the prediction uncertainty tensor corresponding to the wake spacetime trajectory is output.

[0036] It should be noted that when performing prediction uncertainty assessment while generating wake prediction results, a parallel uncertainty estimation branch is set in the output layer of the wake prediction model to quantify the prediction confidence level of the wake prediction model. In this uncertainty estimation branch, the stability of the prediction results is measured based on the magnitude of changes in node states, the consistency of message transmission between nodes, and the degree of satisfaction of aerodynamic prior constraints. By calculating the uncertainty index of each wind turbine in the rolling time domain, a prediction uncertainty tensor that completely corresponds to the wake spatiotemporal trajectory in the node dimension and time dimension can be formed, so that the wake prediction results and the prediction uncertainty tensor are output synchronously.

[0037] Methods for finding the optimal cooperative yaw angle command sequence include: Based on the wake spatiotemporal trajectory, an adaptive cooperative yaw and roll optimization problem is established through an online optimization controller. This adaptive cooperative yaw and roll optimization problem aims to maximize the total power generation of the entire plant and incorporates unit operation safety constraints.

[0038] It should be noted that the online optimization controller adopts a rolling time-domain optimization (Model Predictive Control, MPC) framework. This controller is responsible for using the wake spatiotemporal trajectory and prediction uncertainty tensor as feedforward information to perform rolling optimization in real time. Its core structure includes: a power generation model for the entire power plant, used to predict future power generation based on wake information and yaw angle; a constraint processing module for managing and dynamically adjusting unit operation safety constraints; and an optimization solver for solving the optimal control sequence at each time step. The online optimization controller utilizes flow field evolution information predicted by the wake spatiotemporal trajectory to calculate the expected power generation in the future rolling time domain using a field group power generation model. The optimization objective function aims to maximize the cumulative total power generation of the entire field at each time step within this rolling time domain. The safety constraints for wind turbine operation are designed to ensure that the wind turbine does not exceed its design limits when executing yaw commands. These constraints cover load safety (such as tower foundation bending moment and fatigue load), mechanical safety (such as yaw rate and acceleration), and power generation quality requirements.

[0039] Based on the prediction uncertainty tensor, the unit operation safety constraints in the adaptive cooperative yaw roll optimization problem are dynamically adjusted; the adjusted unit operation safety constraints are used to replace the original unit operation safety constraints, thus forming the adjusted adaptive cooperative yaw roll optimization problem.

[0040] It should be noted that the constraint processing module of the online optimization controller quantifies the wake prediction error represented by the prediction uncertainty tensor and converts it into the constraint confidence level of the unit operation safety constraints in the future rolling time domain. When the prediction uncertainty is high, the system generates a tightening factor to tighten the original constraint boundary, making the yaw command more conservative to ensure safety. When the uncertainty is low, the system generates a relaxation factor to relax the constraint boundary, releasing the optimization space to maximize the power generation potential. Through this uncertainty-driven dynamic relaxation and tightening mechanism, adjusted unit operation safety constraints are formed, replacing the original fixed constraints, thus forming an adjusted adaptive cooperative yaw rolling optimization problem.

[0041] The online optimization controller is based on the rolling time-domain optimization mechanism to solve the adjusted adaptive cooperative yaw rolling optimization problem and obtain the optimal cooperative yaw angle command sequence.

[0042] It should be noted that the online optimization controller adopts the rolling temporal optimization (Model Predictive Control, MPC) mechanism, and uses an optimization solver to solve the adjusted adaptive cooperative yaw rolling optimization problem; In each control cycle (e.g., every 10 seconds), the optimizer, based on the current flow field state and the predicted wake spatiotemporal trajectory, performs a rolling time-domain optimization in the future. Inside Find an optimal cooperative yaw angle command sequence ;in, It is the optimal coordinated yaw angle command sequence. It is the first optimal yaw angle command. It is the second optimal yaw angle command. This is the last optimal yaw angle command. It is the end point of the rolling time domain. It is the starting point of the rolling time domain.

[0043] The optimization solver performs a final safety check on the first optimal yaw angle command in the sequence. In the next control cycle, the optimization solver receives the latest flow field information and unit operating status as feedback, rolls the optimization time domain forward by one time step, and repeats the entire optimization process. This rolling time domain optimization mechanism can continuously cope with the dynamic changes in the flow field and prediction uncertainty, and can find a continuously optimal trade-off between the overall power generation efficiency and the dynamically adjusted operational safety constraints, thereby outputting the latest optimal cooperative yaw angle command sequence in real time.

[0044] Methods for establishing adaptive cooperative yaw roll optimization problems include: The online optimization controller uses the wake spatiotemporal trajectory as the external prediction input; and extracts the flow field evolution information in the future rolling time domain based on the wake spatiotemporal trajectory to construct the predicted state sequence for optimization solution.

[0045] It should be noted that the online optimization controller uses key physical quantities contained in the wake's spatiotemporal trajectory, such as the incoming wind speed, wind direction, wake centerline position, and wake diffusion degree in the future rolling time domain, as flow field evolution information. This information is extracted and formatted into a series of time-series vectors, i.e., a predicted state sequence. This predicted state sequence provides the online optimization controller with accurate environmental input with a time dimension, serving as the basis for subsequent accurate power prediction and load calculation by the wind farm power generation model.

[0046] Based on the predicted state sequence, and according to the impact of future flow field evolution information on the overall power generation efficiency of the wind farm, the future power generation changes corresponding to different cooperative yaw angle strategies are calculated. With the goal of maximizing the total power generation of the entire field, the expected power generation at each time step in the future rolling time domain is accumulated to form the objective function of the optimization problem.

[0047] It should be noted that the online optimization controller uses the predicted state sequence and different cooperative yaw angle strategies as inputs to drive the wind farm cluster power generation model to dynamically model the wake loss of the wind farm in the future rolling time domain. The expected power generation of each wind turbine at each time step is calculated using this cluster power generation model, and the expected power generation of all turbines is accumulated in both spatial and temporal dimensions to obtain the total accumulated power generation of the entire wind farm. This accumulated value is the objective function of the optimization problem. By maximizing this objective function, the optimal overall power generation efficiency of the wind farm in the future rolling time domain is achieved. It should be further explained that the calculation process of the power generation model takes the future flow field evolution information (including upstream wind speed, wake diffusion, and wake deflection of each unit) provided in the predicted state sequence and the cooperative yaw angle strategy to be evaluated as input. The calculation process first determines the velocity loss and wake trajectory deflection caused by the upstream unit to the downstream unit based on the current state of each unit and the cooperative yaw angle command, combined with the embedded wake attenuation and deflection laws; then, using the power characteristic curve of the unit itself, the actual captured wind speed after the wake influence is calculated is substituted to calculate the expected power generation of each unit at the corresponding time step.

[0048] The unit operation safety constraints are transformed into mathematical constraints related to the coordinated yaw angle command sequence, and the activation state and boundary value of each mathematical constraint are dynamically determined based on future flow field evolution information.

[0049] It should be noted that the online optimization controller transforms unit operating safety constraints, such as maximum yaw rate and ultimate blade load, into mathematical constraints (such as inequality constraints) related to the optimal cooperative yaw angle command sequence. Simultaneously, based on future flow field evolution information (i.e., predicted state sequence), it dynamically assesses the impact of the flow field on unit safety at each time step. For example, under conditions of drastic wind speed changes or strong wake, corresponding load and mechanical constraints are activated, and their boundary values ​​are determined in real time based on flow field information to ensure that the optimization process always operates within a dynamically safe range.

[0050] By combining the objective function and mathematical constraints, an adaptive cooperative yaw rolling optimization problem is established.

[0051] Specifically, the cumulative total power generation of the entire field is used as the objective function for optimization, and the mathematical constraints, after dynamically determining the activation state and boundary value processing, are used as the constraints for optimization. These two are combined and input into the optimization solver, thus formally establishing the adaptive cooperative yaw roll optimization problem. This problem forms the mathematical basis for the online optimization controller to solve for the optimal cooperative yaw angle command sequence in each control cycle.

[0052] The dynamic adjustment methods for the safety constraints of operating units include: The online optimization controller converts the wake prediction error, characterized by the prediction uncertainty tensor, into the constraint confidence level of the unit operation safety constraints in the future rolling time domain by quantifying the prediction uncertainty tensor; and converts the constraint confidence level into the dynamic relaxation factor and tightening factor of the unit operation safety constraints.

[0053] It should be noted that the constraint processing module of the online optimization controller receives a prediction uncertainty tensor as input, which corresponds completely to the wake spatiotemporal trajectory in both the node and time dimensions. The constraint processing module calculates the magnitude of this prediction uncertainty tensor and quantizes it. That is, the constraint confidence level. This characterizes the impact of wake prediction errors on the risk of triggering unit operation safety constraints. It is converted into a dynamic relaxation factor for adjusting the constraint boundary. Or contraction factor For example, the calculation of the contraction factor can be based on the constraint confidence level. Determine using the following expression; ; in, It is a preset sensitivity coefficient. The settings are based on wind farm operation experience data and unit load tolerance limits. The exemplary value range is between 0.1 and 0.5. This coefficient is used to adjust the constraint confidence level. When it decreases (i.e., when uncertainty increases), the contraction factor The response strength when the constraint boundary is tightened.

[0054] By using relaxation and tightening factors, the boundary values ​​of unit operation safety constraints are corrected in real time, and the adjusted unit operation safety constraints are generated.

[0055] It should be noted that the calculated dynamic relaxation factor is used. Or contraction factor The boundary values ​​of the original unit operation safety constraints are corrected in real time. If the prediction uncertainty tensor shows a large prediction error (i.e., If the value is low, then use the contraction factor. Tightening the boundary constraints makes the optimization solution more conservative; if the prediction error is small (i.e., If the value is high, then use the relaxation factor. Relaxing the boundary constraints provides a larger optimization space for maximizing the total power generation across the entire field for the optimizer; these boundary values, which are corrected in real time, define the adjusted unit operation safety constraints.

[0056] Methods for obtaining and sending verified instruction sequences to upstream units for execution include: Based on the prediction uncertainty tensor, uncertainty propagation analysis is performed on the current optimal cooperative yaw angle command sequence to obtain the trigger probability of the multi-dimensional dynamic safety boundary in the future execution time domain. The trigger probability is then converted into the corresponding boundary confidence level, and the multi-dimensional dynamic safety boundary is dynamically adjusted based on the boundary confidence level.

[0057] It should be noted that after the online optimization controller outputs the optimal cooperative yaw angle command sequence, the verification module takes the prediction uncertainty tensor as input and performs uncertainty propagation analysis on the expected effect of the optimal cooperative yaw angle command sequence in the future execution time domain. This analysis calculates the probability that various multi-dimensional dynamic safety boundaries (such as maximum yaw rate, load limit, etc.) will be breached after the command is executed, i.e., the trigger probability, by superimposing the prediction uncertainty tensor (representing the prediction error range) onto the expected flow field result of the optimal cooperative yaw angle command sequence. This trigger probability is then processed through a preset mapping function (e.g., The relationship, among which It's the trigger probability. It is the boundary confidence level. This is a mapping relationship, converted into a boundary confidence level. Finally, the verification module dynamically adjusts the multi-dimensional dynamic security boundary based on this boundary confidence level to ensure that the tightness of the security boundary matches the risk of instruction execution.

[0058] The optimal cooperative yaw angle command sequence is compared and verified with the adjusted multi-dimensional dynamic safety boundary. If the optimal cooperative yaw angle command sequence is found to have a risk of exceeding the boundary, soft constraint contraction is performed on the optimal cooperative yaw angle command sequence to ensure that the safety execution requirements are met.

[0059] It should be noted that the verification module compares each command value in the optimal coordinated yaw angle command sequence with the corresponding boundary value of the adjusted multi-dimensional dynamic safety boundary point by point. If the command value exceeds the multi-dimensional dynamic safety boundary (hard boundary crossing) or the distance to the boundary is less than the preset safety buffer threshold (risk of boundary crossing), it is determined that there is a risk of boundary crossing. For commands with a risk of boundary crossing, the verification module initiates soft constraint contraction. This contraction operation calculates a soft constraint contraction amount to ensure that the command sequence can smoothly fall back to the safe range of the adjusted multi-dimensional dynamic safety boundary, thereby ensuring that the safe execution requirements of the crew are met.

[0060] Obtain the verified optimal coordinated yaw angle command sequence and send it to the upstream unit for execution.

[0061] It should be noted that after soft constraint contraction processing, a valid optimal cooperative yaw angle command sequence is obtained. The verification module then extracts the first optimal yaw angle command (i.e., the optimal cooperative yaw angle at the current moment) from this valid optimal cooperative yaw angle command sequence and sends it to the underlying controller of the upstream unit for execution. This first optimal yaw angle command represents the system's optimal trade-off between maximizing total power generation and dynamic safety at the current moment.

[0062] Methods for dynamically adjusting multi-dimensional dynamic security boundaries include: The trigger probability of the multi-dimensional dynamic security boundary is compared with the preset security threshold and the aggressive threshold.

[0063] When the risk probability is higher than the safety threshold, the corresponding dimension's safety boundary is tightened exponentially according to the extent of the excess; when the risk probability falls between the safety threshold and the aggressive threshold, the corresponding dimension's safety boundary is tightened moderately in a linear manner according to its value to balance safety and efficiency; when the risk probability is lower than the aggressive threshold, the corresponding dimension's safety boundary is relaxed linearly according to the extent of the lower risk.

[0064] It should be noted that the preset safety threshold (5% for example) and aggressive threshold (1% for example) are preset according to the wind farm's operation specifications and risk preferences. The safety threshold defines the highest trigger probability of the multi-dimensional dynamic safety boundary that the verification module can tolerate. Exceeding this safety threshold is considered to be too risky, and the boundary needs to be tightened quickly. The aggressive threshold defines the lowest trigger probability at which the boundary can be moderately relaxed in order to pursue higher efficiency. When the trigger probability exceeds the safety threshold, the verification module adjusts the threshold exponentially (e.g., based on the magnitude of the exceedance). ,in For the new safety boundary value, The old safety boundary value, For mathematical constants, As a safety threshold, The corresponding dimensional safety boundary is tightened using an exponential tightening coefficient. When the trigger probability falls between the safety threshold and the aggressive threshold, the boundary is tightened linearly to achieve a moderate balance. This segmented adjustment mechanism ensures that the multi-dimensional dynamic safety boundary can respond accurately and appropriately to risks at different risk levels.

[0065] Performing soft-constraint contraction includes: Select the portion of the optimal cooperative yaw angle command sequence that is less than the preset safety buffer threshold in distance from the multi-dimensional dynamic safety boundary.

[0066] It should be noted that, firstly, the distance between each command value in the optimal cooperative yaw angle command sequence and its corresponding adjusted multi-dimensional dynamic safety boundary is calculated. This distance is determined by calculating the difference between the command value and the boundary value. Then, the entire optimal cooperative yaw angle command sequence is traversed, and all portions with distances less than a preset safety buffer threshold are selected; these portions are considered potential boundary violation points. Preset safety buffer threshold (e.g.) The angle value is used to define the safety buffer area for the instruction sequence. Its setting is based on the dynamic response lag of the unit executing instructions and the measurement error of the sensors. This preset safety buffer threshold ensures that even with minor system uncertainties, the adjusted multi-dimensional dynamic safety boundary will not be immediately reached or exceeded when instructions are actually issued and executed.

[0067] For the portion where the distance is less than the preset safety buffer threshold, the soft constraint shrinkage is calculated based on its distance from the multi-dimensional dynamic safety boundary and the predicted uncertainty tensor.

[0068] Specifically, for each boundary-crossing risk point, the soft constraint shrinkage is calculated jointly based on its distance (the difference from the safety buffer threshold) and the corresponding prediction uncertainty tensor (characterizing the range of prediction deviation); the formula for calculating the soft constraint shrinkage is as follows: ; in, It is the amount of soft constraint contraction. and These are weighting coefficients, set based on the relative importance that wind farms place on safety buffer margins and predicted risks: This is used to measure the importance attached to safety margins (i.e., how close an instruction is to the boundary), with an example value of 0.5-0.8. This value is used to measure the importance attached to the prediction risk (characterized by the prediction uncertainty tensor), with an example value of 0.2-0.5. It is a preset safety buffer threshold. It is the distance between the instruction value and its corresponding adjusted multi-dimensional dynamic security boundary. It is the uncertainty measure extracted from the prediction uncertainty tensor; Apply the soft constraint contraction amount to the optimal cooperative yaw angle command sequence, correct the optimal cooperative yaw angle command value, and generate the optimal cooperative yaw angle command sequence after soft constraint contraction, which is used as the optimal cooperative yaw angle command sequence that has passed the verification.

[0069] It should be noted that the verification module applies the calculated soft constraint contraction amount to the corresponding out-of-bounds risk points in the optimal cooperative yaw angle command sequence, performing a correction operation. The corrected command value will be limited to the safety range formed by the adjusted multi-dimensional dynamic safety boundary and the preset safety buffer threshold. After all corrections, the optimal cooperative yaw angle command sequence is generated as the optimal cooperative yaw angle command sequence after soft constraint contraction, and is identified as the optimal cooperative yaw angle command sequence that has passed verification, ready to be issued for execution.

[0070] Methods for joint adaptive optimization include: Receive the wind farm operation data after execution, compare it with the expected results, and obtain the deviation between the wind farm operation data and the expected results.

[0071] Specifically, based on the wind farm operation data (including actual power generation, actual turbine load, and actual wind speed) after the validated optimal coordinated yaw angle command sequence is executed on the upstream turbines, the wind farm cluster power generation model and wake prediction model are invoked to calculate the expected results (i.e., expected power generation and load, etc.) based on the optimal coordinated yaw angle command and the actual incoming flow. By aligning and comparing the wind farm operation data with this expected result in time, the difference between the two is calculated, thereby obtaining the deviation between the wind farm operation data and the expected result. This deviation (e.g., the difference between actual power generation and expected power generation) quantifies the degree of inaccuracy of the system's prediction and optimization strategies.

[0072] Using this deviation as a driving signal, the parameter adjustments for the wake prediction model and the online optimization controller used to minimize the deviation are calculated, and the parameters of both are synchronously corrected and updated accordingly.

[0073] Specifically, the deviation between the acquired wind farm operation data and the expected results is used as a driving signal. An algorithm based on gradient descent or similar iterative optimization is used to calculate the parameter adjustment of the wake prediction model and the parameter adjustment of the online optimization controller. For wake prediction models (such as hybrid structures of convolutional neural networks and graph neural networks), the calculated parameter adjustments are used to correct the model's weights and biases to improve its accuracy in predicting wake spatiotemporal trajectories and prediction uncertainty tensors. For online optimization controllers, the calculated parameter adjustments are used to correct key physical parameters in their internal field-group power generation model, such as empirical correction factors related to wake attenuation and deflection (e.g., wake diffusion coefficient, momentum loss coefficient, etc.) and uncertainty response coefficients (sensitivity coefficients) in the constraint handling module. ); These calculated parameter adjustments are applied to the parameter sets of the wake prediction model and the online optimization controller, respectively, to achieve synchronous correction and updates, thereby minimizing the deviation and enabling it to have continuous joint adaptive optimization capabilities.

[0074] In summary, this invention achieves improved overall power generation efficiency by: quantifying the uncertainty of wake prediction to drive a dual adaptive optimization mechanism for coordinated yaw across the entire wind farm; dynamically adjusting the operational safety constraints of each unit at the collaborative decision-making level based on the uncertainty of global wake prediction, ensuring that the yaw operation of upstream units, while fully considering their impact on downstream units and corresponding uncertainties, can both actively guide the wake to enhance wind energy capture by downstream units and avoid sacrificing their own power generation potential due to rigid constraints, thus achieving dynamic optimization of the total power generation of the entire wind farm; and again using uncertainty at the collaborative execution level to uniformly verify and smooth the yaw command sequence, ensuring that the yaw actions of each unit are coordinated and consistent within a safe range, avoiding cascading performance losses caused by single unit exceeding limits. By embedding uncertainty management into the entire decision-making and execution process of collaborative control, the collaborative maximization of power generation efficiency of the wind farm cluster in complex flow fields is achieved.

[0075] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A wind farm wake effect adaptive cooperative yaw optimization control system, characterized in that, include: The flow field module constructs a real-time reliable flow field map based on the lidar data of the core unit and the operating and environmental data of the radiation unit. The prediction module, based on real-time reliability flow field maps, predicts the spatiotemporal trajectory of the wake using a wake prediction model and simultaneously outputs the prediction uncertainty tensor. The decision-making module, based on the wake spatiotemporal trajectory, optimizes the controller online to maximize the total power generation of the entire site and embeds unit operation safety constraints; it dynamically adjusts the unit operation safety constraints according to the prediction uncertainty tensor and solves the optimal cooperative yaw angle command sequence based on the rolling time domain optimization mechanism. Before issuing the optimal coordinated yaw angle command sequence, the verification module dynamically adjusts the multi-dimensional dynamic safety boundary by combining the prediction uncertainty tensor; it compares and verifies the command sequence with the multi-dimensional dynamic safety boundary and performs soft constraint contraction; and it obtains the verified command sequence and issues it to the upstream unit for execution. The evolution module acquires wind farm operation data after execution and performs joint adaptive optimization on the wake prediction model and online optimization controller based on the deviation between the wind farm operation data and the expected results.

2. The wind farm wake effect adaptive cooperative yaw optimization control system as described in claim 1, characterized in that, The method for constructing real-time reliability flow field maps includes: The lidar sensing data of the core unit and the operation and environmental data of the radiating units are fused together; and a dynamic propagation operator is calculated based on the wind farm's operating status and spatial topology characteristics to generate the propagation weight matrix of the incoming flow from the core unit to the radiating units. Based on the mapping relationship between the incoming flow propagation weight matrix and the unit operating characteristics, the incoming flow status in front of the radiating unit is calculated and physical consistency calibration is performed to obtain virtual incoming flow information; and it is then fused with the real-time feedforward measurement data of the core unit lidar according to the data confidence level and subjected to physical constraint filtering to obtain the fused wind measurement data field. The fused wind measurement data field is input into the three-dimensional interpolation model to reconstruct the volumetric flow field, perform uncertainty quantification and confidence labeling, and output a real-time confidence flow field map.

3. The wind farm wake effect adaptive cooperative yaw optimization control system as described in claim 2, characterized in that, The method for predicting the wake spatiotemporal trajectory and simultaneously outputting the prediction uncertainty tensor includes: The real-time confidence flow field map is input into the wake prediction model for graph structure transformation. Each wind turbine is used as a node of the graph to construct a directed graph, and the incoming flow propagation weight matrix is ​​used as the initial directed edge weights for initialization. The confidence of the core unit lidar measurement points is mapped to the edge weight attenuation factor in the graph to suppress the propagation of low confidence information in the graph structure. Message-passing reasoning with embedded aerodynamic priors is performed on a directed graph to obtain the wake spacetime trajectory in the future rolling time domain; where the aerodynamic priors include the velocity deficit propagation law, wake diffusion characteristics and wake deflection relationship caused by yaw; While generating the wake spacetime trajectory, a prediction uncertainty assessment is performed, and the prediction uncertainty tensor corresponding to the wake spacetime trajectory is output.

4. The wind farm wake effect adaptive cooperative yaw optimization control system as described in claim 3, characterized in that, The method for solving the optimal cooperative yaw angle command sequence includes: Based on the wake spatiotemporal trajectory, an adaptive cooperative yaw roll optimization problem is established through an online optimization controller. This adaptive cooperative yaw roll optimization problem aims to maximize the total power generation of the entire plant and incorporates unit operation safety constraints. Based on the prediction uncertainty tensor, the unit operation safety constraints in the adaptive cooperative yaw roll optimization problem are dynamically adjusted; the adjusted unit operation safety constraints are used to replace the original unit operation safety constraints to form the adjusted adaptive cooperative yaw roll optimization problem. The online optimization controller is based on the rolling time-domain optimization mechanism to solve the adjusted adaptive cooperative yaw rolling optimization problem and obtain the optimal cooperative yaw angle command sequence.

5. The wind farm wake effect adaptive cooperative yaw optimization control system as described in claim 4, characterized in that, The method for establishing the adaptive cooperative yaw roll optimization problem includes: The online optimization controller uses the wake spatiotemporal trajectory as the external prediction input; and extracts the flow field evolution information in the future rolling time domain based on the wake spatiotemporal trajectory to construct the predicted state sequence for optimization solution; Based on the predicted state sequence, and according to the impact of future flow field evolution information on the overall power generation efficiency of the wind farm, the future power generation changes corresponding to different cooperative yaw angle strategies are calculated; with the goal of maximizing the total power generation of the entire field, the expected power generation at each time step in the future rolling time domain is accumulated to form the objective function of the optimization problem. The unit operation safety constraints are transformed into mathematical constraints related to the coordinated yaw angle command sequence, and the activation state and boundary value of each mathematical constraint are dynamically determined based on the future flow field evolution information. By combining the objective function and mathematical constraints, an adaptive cooperative yaw rolling optimization problem is established.

6. The wind farm wake effect adaptive cooperative yaw optimization control system as described in claim 4, characterized in that, The dynamic adjustment method for the safety constraints of the operating units includes: The online optimization controller converts the wake prediction error, characterized by the prediction uncertainty tensor, into the constraint confidence level of the unit operation safety constraints in the future rolling time domain; and converts the constraint confidence level into the dynamic relaxation factor and tightening factor of the unit operation safety constraints. By using relaxation and tightening factors, the boundary values ​​of unit operation safety constraints are corrected in real time, and the adjusted unit operation safety constraints are generated.

7. The wind farm wake effect adaptive cooperative yaw optimization control system as described in claim 1, characterized in that, The method for obtaining the verified instruction sequence and sending it to the upstream unit for execution includes: Based on the prediction uncertainty tensor, uncertainty propagation analysis is performed on the current optimal cooperative yaw angle command sequence to obtain the trigger probability of the multi-dimensional dynamic safety boundary in the future execution time domain. The trigger probability is then converted into the corresponding boundary confidence level, and the multi-dimensional dynamic safety boundary is dynamically adjusted based on the boundary confidence level. The optimal cooperative yaw angle command sequence is compared and verified with the adjusted multi-dimensional dynamic safety boundary. If the optimal cooperative yaw angle command sequence is found to have a risk of exceeding the boundary, soft constraint contraction is performed on the optimal cooperative yaw angle command sequence to ensure that the safety execution requirements are met. Obtain the verified optimal coordinated yaw angle command sequence and send it to the upstream unit for execution.

8. The wind farm wake effect adaptive cooperative yaw optimization control system as described in claim 7, characterized in that, The method for dynamically adjusting multi-dimensional dynamic security boundaries includes: Compare the trigger probability of the multi-dimensional dynamic security boundary with the preset security threshold and the aggressive threshold; When the risk probability is higher than the safety threshold, the corresponding dimension's safety boundary is tightened exponentially according to the extent of the excess; when the risk probability falls between the safety threshold and the aggressive threshold, the corresponding dimension's safety boundary is tightened moderately in a linear manner according to its value to balance safety and efficiency; when the risk probability is lower than the aggressive threshold, the corresponding dimension's safety boundary is relaxed linearly according to the extent of the lower risk.

9. The wind farm wake effect adaptive cooperative yaw optimization control system as described in claim 7, characterized in that, The execution of soft-constraint contraction includes: Select the portion of the optimal cooperative yaw angle command sequence whose distance from the multi-dimensional dynamic safety boundary is less than a preset safety buffer threshold; For the portion where the distance is less than the preset safety buffer threshold, the soft constraint shrinkage is calculated based on its distance from the multi-dimensional dynamic safety boundary and the predicted uncertainty tensor. Apply the soft constraint contraction amount to the optimal cooperative yaw angle command sequence, correct the optimal cooperative yaw angle command value, and generate the optimal cooperative yaw angle command sequence after soft constraint contraction, which is used as the optimal cooperative yaw angle command sequence that has passed the verification.

10. The wind farm wake effect adaptive cooperative yaw optimization control system as described in claim 1, characterized in that, The method for performing joint adaptive optimization includes: Receive the wind farm operation data after execution, compare it with the expected results, and obtain the deviation between the wind farm operation data and the expected results; Using this deviation as a driving signal, the parameter adjustments for the wake prediction model and the online optimization controller used to minimize the deviation are calculated, and the parameters of both are synchronously corrected and updated accordingly.

Images