A method for hybrid control of output power of offshore wind farm under wake effect

Abstract

This invention discloses a hybrid control method for the output power of an offshore wind farm under wake influence, comprising: S1: acquiring the operating parameters V0 and P of the offshore wind farm in real time; S2: inputting V0 and P into the controller, which outputs the axial induction factor vector a of the wind turbines in the entire offshore wind farm; S3: adjusting each wind turbine in the offshore wind farm according to vector a; repeating S1-S3. This invention proposes a hybrid optimization method oriented towards axial induction factor control, which can significantly improve the output power in the actual operation of the wind farm. Under dynamic wind conditions and uncertain wake conditions, it can realize real-time optimization and updating of the wind turbine control variables, enhance the adaptability of the control strategy to environmental disturbances, and effectively improve the overall power generation efficiency and operational robustness of the wind farm.

Description

Technical Field

[0001] This invention relates to the field of wind farms, and more particularly to a method for hybrid control of output power of offshore wind farms under the influence of wake. Background Technology

[0002] With the development of wind power technology and the continuous transformation of the global energy structure, the scale of offshore wind farms is constantly expanding, and their proportion in the total installed capacity of new energy is increasing. However, the large-scale integration of wind farms has also brought many challenges to operation and control. First, the dense layout of wind turbines generates a significant wake effect, leading to deterioration of wind conditions for downstream turbines. This not only reduces wind energy utilization efficiency but also exacerbates the accumulation of fatigue loads on the turbines, shortening equipment lifespan. Second, the actual wind speed has high volatility and measurement uncertainty, further increasing the unpredictability of wind farm operation, making it imperative for control methods to possess strong robustness and adaptability. To address these issues, in recent years, researchers have gradually focused their control objectives on adjusting the axial induction factor of each turbine in the wind farm. By rationally allocating the axial induction factor to each turbine, the overall power generation performance, load distribution, and system economy can be improved under conditions of wake interference and inaccurate wind conditions, becoming a key path to promote the efficient operation of offshore wind farms.

[0003] Currently, to improve the power generation capacity of offshore wind farms, axial induction factor control methods can be mainly divided into three categories: model-based methods, data-based methods, and hybrid control methods that combine both. Model-based methods rely on accurate modeling of the wind farm's wake propagation and power output mechanisms. By introducing system structural characteristics during the controller design phase, good convergence speed and optimization performance can be achieved. Some studies have utilized the alternating direction multiplier method to achieve distributed optimization control of the axial induction factor, while others have used collaborative optimization or model predictive control methods to improve overall power generation and alleviate unit fatigue loads. However, model-based methods are highly sensitive to wind condition parameters; modeling errors or inaccurate sensor measurements can lead to deviations in system behavior, limiting their adaptability in real-world operating scenarios.

[0004] To overcome the aforementioned limitations, some research has shifted towards exploring data-driven axial induction factor (AIN) control methods. These methods do not rely on prior wind farm models but instead use mechanisms such as reinforcement learning, Markov decision processes, or machine learning to update strategies based on online wind turbine operation data, exhibiting a degree of robustness and exploratory capability. Some studies utilize deep reinforcement learning to obtain the optimal AIN setting strategy or employ genetic algorithms to achieve a trade-off between power generation efficiency and equipment health. However, these methods typically require a large amount of data and a long exploration period, and face challenges such as slow convergence speed and large fluctuations in control performance.

[0005] To balance convergence speed and environmental adaptability, hybrid control methods have become a research hotspot in recent years. These methods attempt to embed data-driven policy search into a model-driven optimization framework, improving policy robustness while maintaining optimization efficiency. Some studies propose first obtaining a high-quality initial policy through free exploration using a model, and then using an accurate model to optimize wind farm output power. Although hybrid methods possess good overall performance, most current research only verifies their effectiveness under static or ideal wind conditions, failing to fully consider the impact of wind speed measurement errors or wind fluctuations on control performance in actual operation. This results in insufficient applicability of the methods in real wind farms.

[0006] In summary, the existing technology has the following problems:

[0007] 1) Existing axial induction factor control methods for offshore wind farms, while exhibiting good performance under ideal wind conditions, are susceptible to model errors and prone to getting trapped in local optima when faced with real-world operating scenarios such as wind speed measurement errors and wake uncertainties. While pure data-driven strategies offer some robustness, they suffer from slow convergence and low optimization efficiency. These existing methods have poor adaptability, easily leading to system optimization failures or performance instability, making it difficult to achieve efficient wind farm operation.

[0008] 2) Traditional control targets are singular, focusing only on maximizing power generation without considering the structural fatigue problems that may be caused by frequent adjustments of axial induction factors, making it difficult to ensure the healthy operation of wind turbines while increasing power.

[0009] Therefore, those skilled in the art are dedicated to developing a hybrid control method for the output power of offshore wind farms under the influence of wake. Summary of the Invention

[0010] To achieve the above objectives, the present invention provides a method for hybrid control of the output power of an offshore wind farm under the influence of wake, comprising the following steps:

[0011] S1: Real-time acquisition of offshore wind farm operating parameters and ,in It is the free-flow wind speed of the wind field. It is the active power vector of all wind turbines obtained from the SCADA system;

[0012] S2: Will and The input controller outputs the axial induction factor vector of the wind turbines throughout the entire offshore wind farm. ;

[0013] S3: Based on vector Adjusting each wind turbine within the offshore wind farm;

[0014] Repeat S1-S3.

[0015] Furthermore, the objective function of the controller can specifically be:

[0016]

[0017] In the formula, the wind turbines in the entire offshore wind farm are collectively referred to as... , This represents the axial induction factor vector for wind turbines throughout the entire offshore wind farm. The initial axial induction factor vector set for the wind farm control center. To adjust the cost coefficient, This refers to the electricity price for wind power generation. This represents the active power vector of all wind turbines in the field. This represents the feasible region for the axial induction factor.

[0018] Furthermore, the controller's control strategy includes a gradient estimation stage and a momentum update stage.

[0019] Furthermore, the gradient estimation stage includes joint estimation from both model-driven and data-driven perspectives.

[0020] Furthermore, the specific iterative process of the gradient estimation stage is as follows:

[0021]

[0022] In the formula, k This indicates the k-th iteration. It is a vector composed of the axial induction factors of the wind turbines in the entire offshore wind farm at the k-th iteration. It is the power vector of all wind turbines in the k-th iteration. The driving direction of the model is obtained based on the power sensitivity matrix. For data-driven approaches based on residual feedback, For the first k The sensitivity matrix is ​​calculated from the current wind speed and the induction factor in the next iteration. Let be a random vector uniformly sampled on a unit sphere, satisfying , This represents the weighting coefficients for model-driven and data-driven approaches.

[0023] Furthermore, the model-driven direction has a higher weight in the initial stage of iteration, while the data-driven direction has a gradually stronger weight in the later stage of iteration.

[0024] Furthermore, the momentum update phase employs the heavy ball method for momentum update.

[0025] Furthermore, the specific process of the heavy ball method is as follows:

[0026]

[0027]

[0028] In the formula, Step size, Momentum factor To explore the disturbance,

[0029] First, update the candidate solutions. Then, using the integrated gradient direction intermediate variables Update and refine the candidate solution trajectory using the momentum term. To avoid the solution getting trapped in a local optimum, a certain exploration perturbation is added, and finally the candidate solution is projected onto the feasible region to obtain a new axial induction factor value. .

[0030] Secondly, the present invention also discloses a power output control device for an offshore wind farm under wake influence, the device being used to implement the above-mentioned control method, comprising:

[0031] Data acquisition module: used to acquire real-time operating parameters of offshore wind farms. and ;

[0032] Control module: Used to input operating parameters of offshore wind farms and It also outputs the axial induction factor vector of the wind turbines throughout the entire offshore wind farm. ;

[0033] Execution module: Used to calculate the axial induction factor vector of the wind turbines throughout the entire offshore wind farm based on the output of the controller module. Adjust each fan.

[0034] Thirdly, the present invention also discloses a computer device, comprising:

[0035] Memory;

[0036] One or more processors are coupled to the memory;

[0037] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by one or more processors, the one or more applications being configured to perform the control method described above.

[0038] Technical effect

[0039] 1) This invention addresses the common problems of wind speed inaccuracy and dynamic wake interference in wind farm operation by proposing a hybrid optimization method for Axial Induced Factor (AIF) control. It integrates three technical mechanisms: first, it utilizes a model-driven stochastic gradient descent method to construct approximate guiding information, providing a basic structure for rapid convergence; second, it introduces a data-driven zero-order residual feedback mechanism to enhance the ability to explore unknown disturbances and local extrema; and third, it employs a momentum-based ball acceleration strategy to effectively improve the algorithm's convergence speed and stability. This invention combines the rapid optimization capability of model-driven approaches with the exploratory capability of data-driven approaches, designing a hybrid control framework with convergence guarantees and introducing a momentum acceleration mechanism for iterative updates, thereby improving the robustness and convergence efficiency of the controller under uncertain wind conditions. This invention can significantly improve output power in actual wind farm operation. Under dynamic wind conditions and uncertain wake conditions, it can achieve real-time optimization and updates of wind turbine control quantities, enhancing the adaptability of the control strategy to environmental disturbances and effectively improving the overall power generation efficiency and operational robustness of the wind farm. Simultaneously, it reduces the computational burden of the algorithm, facilitating engineering deployment and long-term operation.

[0040] 2) This invention designs a wind farm control objective function that balances power generation revenue and mechanical degradation costs, achieving an optimal balance between power generation efficiency and turbine lifespan. This invention explicitly incorporates the fatigue cost of the wind turbine into the control objective function. By balancing power gains and adjustment costs during the optimization process, it guides the rational allocation of axial induction factors, avoiding excessive power pursuit and over-adjustment of local turbines. This invention can extend the service life of wind turbines, reduce maintenance costs, and improve the economic efficiency and safety of wind farms throughout their entire life cycle without significantly affecting power generation levels, demonstrating promising engineering application prospects.

[0041] It is understandable that the beneficial effects of the second and third aspects mentioned above can be found in the relevant descriptions above, and will not be repeated here.

[0042] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0043] Figure 1 This is a flowchart of a preferred embodiment of the invention, showing a method for hybrid control of output power in offshore wind farms under wake effects.

[0044] Figure 2 This is a schematic diagram of a single fan wake model according to a preferred embodiment of the present invention;

[0045] Figure 3 This is a simulation diagram of the wake field of an 8-fan scenario according to a preferred embodiment of the present invention;

[0046] Figure 4This is a convergence effect diagram of the standard wind condition axial induction factor according to a preferred embodiment of the present invention.

[0047] Figure 5 This is a diagram illustrating the improvement effect of the standard wind condition objective function according to a preferred embodiment of the present invention.

[0048] Figure 6 This is a graph showing the improvement effect of the objective function based on actual wind conditions in a preferred embodiment of the present invention. Detailed Implementation

[0049] The following description, with reference to the accompanying drawings, illustrates several preferred embodiments of the present invention to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0050] In the accompanying drawings, components with the same structure are indicated by the same numerical designation, and components with similar structures or functions are indicated by similar numerical designations. The dimensions and thicknesses of each component shown in the drawings are arbitrary, and the present invention does not limit the dimensions and thicknesses of each component. To make the illustrations clearer, the thickness of some components has been appropriately exaggerated in the drawings.

[0051] like Figure 1 As shown, this invention provides a method for hybrid control of the output power of an offshore wind farm under the influence of wake, comprising the following steps:

[0052] S1: Real-time acquisition of offshore wind farm operating parameters and ,in It is the free-flow wind speed of the wind field. It is the active power vector of all wind turbines obtained from the SCADA system;

[0053] S2: Will and The input controller outputs the axial induction factor vector of the wind turbines throughout the entire offshore wind farm. ;

[0054] S3: Based on vector Adjusting each wind turbine within the offshore wind farm;

[0055] Repeat S1-S3.

[0056] Furthermore, the objective function of the controller can specifically be:

[0057]

[0058] In the formula, the wind turbines in the entire offshore wind farm are collectively referred to as... , This represents the axial induction factor vector for wind turbines throughout the entire offshore wind farm. The initial axial induction factor vector set for the wind farm control center. To adjust the cost coefficient, This refers to the electricity price for wind power generation. This represents the active power vector of all wind turbines in the field. This represents the feasible region for the axial induction factor.

[0059] Furthermore, the controller's control strategy includes a gradient estimation stage and a momentum update stage.

[0060] Furthermore, the gradient estimation stage includes joint estimation from both model-driven and data-driven perspectives.

[0061] Furthermore, the specific iterative process of the gradient estimation stage is as follows:

[0062]

[0063] In the formula, k This indicates the k-th iteration. It is a vector composed of the axial induction factors of the wind turbines in the entire offshore wind farm at the k-th iteration. It is the power vector of all wind turbines in the k-th iteration. The driving direction of the model is obtained based on the power sensitivity matrix. For data-driven approaches based on residual feedback, For the first k The sensitivity matrix is ​​calculated from the current wind speed and the induction factor in the next iteration. Let be a random vector uniformly sampled on a unit sphere, satisfying , This represents the weighting coefficients for model-driven and data-driven approaches.

[0064] Furthermore, the model-driven direction has a higher weight in the initial stage of iteration, while the data-driven direction has a gradually stronger weight in the later stage of iteration.

[0065] Furthermore, the momentum update phase employs the heavy ball method for momentum update.

[0066] Furthermore, the specific process of the heavy ball method is as follows:

[0067]

[0068]

[0069] In the formula, Step size, Momentum factor To explore the disturbance,

[0070] First, update the candidate solutions. Then, using the integrated gradient direction intermediate variables Update and refine the candidate solution trajectory using the momentum term. To avoid the solution getting trapped in a local optimum, a certain exploration perturbation is added, and finally the candidate solution is projected onto the feasible region to obtain a new axial induction factor value. .

[0071] (The following is a further description of the specific implementation methods described above)

[0072] 1. Overall description of the control scheme

[0073] The active power output of a wind farm is affected by the operating status of its wind turbines and their wake effects. Especially in densely packed offshore wind farms, the wake effect can worsen wind resource conditions for downstream turbines, leading to reduced overall wind energy utilization and power generation. It also increases mechanical damage to equipment during turbine adjustments. Therefore, wind farm control requires actively regulating the wake effect by coordinating the axial induction factors of each turbine, thereby improving overall power output and reducing the risk of structural degradation.

[0074] Traditional control schemes often rely on model-driven methods to optimize and adjust the axial induction factor of wind turbines. However, these methods often depend on precise modeling and are difficult to adapt to actual operating conditions such as wind speed measurement errors and wake propagation uncertainties. Some studies have attempted to use data-driven methods for policy learning, which, while possessing some robustness, typically suffer from slow convergence speed and unstable control performance, making long-term deployment in actual wind farms difficult. To address these issues, this embodiment proposes a hybrid control framework combining model-driven and data-driven approaches. Specifically, the system estimates the directional gradient based on the wind farm wake-power model and introduces a real-time output feedback correction mechanism to adapt to actual wind disturbances, thereby achieving dynamic adjustment of the wind turbine's axial induction factor. To improve the controller's convergence performance, a heavy ball acceleration mechanism is further introduced, correcting the current search direction based on historical search trajectories in each iteration, thus achieving faster objective function convergence. This control scheme can achieve stable operation and target power enhancement of wind farms even under conditions of inaccurate wind speed measurements and incompletely accurate wake models, demonstrating good engineering adaptability and promotion potential.

[0075] 2. Introduction to Offshore Wind Farm Wake-Power Model

[0076] (1) Wind farm wake model

[0077] like Figure 2 As shown, consider a containing n Offshore wind farms with typhoon-powered turbines consist of a collection of wind turbines. , among which number jThe fan is located at number i Upstream of the wind turbines. To describe the wake interference relationship between wind turbines in an offshore wind farm, this embodiment uses the Park model to model the wake velocity loss. The wake gradually expands downstream in the horizontal plane, and its influence range can be defined by the distance downstream of the wake centerline. x Horizontal distance r A joint decision.

[0078] Fan i For a certain point downstream The resulting speed loss is defined as

[0079]

[0080] in, This is the axial induction factor for the wind turbine. The diameter of the fan rotor. This is the wake radius expansion coefficient. If there are multiple upstream wind turbines in a wind farm, their wake regions may overlap. In this case, the wind turbines... i The wind speed loss at a given location should comprehensively consider the wake contributions of all upstream wind turbines, and is represented by a linear superposition model.

[0081]

[0082] in, Let be the axial induction factor vector of the wind turbines throughout the entire wind farm. At this point, the effective wind speed of the wind turbine is...

[0083]

[0084] in, The free-flow wind speed of the wind field.

[0085] (2) Wind farm power model

[0086] Given the wind speed, the output power of a wind turbine can be calculated from its axial induction factor. Assuming all wind turbines in the wind farm are of the same model, then the... i The output power of a typhoon can be expressed as:

[0087]

[0088] in, air density, The swept area of ​​the wind turbine is calculated from the blade diameter. The power coefficient is given. Based on this, a mapping relationship between the axial induction factor and the wake is established under the condition that the axial induction factor and the power output of the wind turbine are coupled together.

[0089] 3. Introduction to Offshore Wind Farm Power Optimization Model

[0090] (1) Introduction to the integrated optimization problem of power generation revenue and adjustment costs

[0091] Since adjusting the axial induction factor causes changes in the wind turbine pitch angle and torque controller, thus affecting the unit's fatigue life, it is necessary to consider the impact on wind turbine operational stability while improving power generation performance. Therefore, this embodiment constructs the following optimization problem.

[0092]

[0093] in, The initial axial induction factor vector set for the wind farm control center. To adjust the cost coefficient, This refers to the electricity price for wind power generation. This represents the power vector of all wind turbines in the field. This represents the feasible region for the axial induction factor. The control objective aims to improve the overall power output of the wind farm without significantly deviating from the original turbine settings or causing structural damage.

[0094] (2) Introduction to the simplified objective function for axial induction factor control

[0095] In actual control, the fan power output It can be expressed as axial induction factor The objective function is a function of the control variable, thus the optimization problem can be equivalently transformed into a constrained optimization problem with respect only to the control variable, i.e., the axial induction factor. The simplified objective function is defined as follows:

[0096]

[0097]

[0098]

[0099]

[0100] This problem has clear engineering and physical significance: under conditions of wake interference and wind uncertainty, adjusting the axial induction factor can improve the economic efficiency and system reliability of wind farms. The simplified objective function will be used in algorithm iteration and property proof.

[0101] 4. Design of a hybrid controller for axial induction factors in wind turbines

[0102] (1) Introduction to the structure of the hybrid controller

[0103] The wind farm controller proposed in this invention is based on a feedback control framework and dynamically adjusts the axial induction factor of each wind turbine through an iterative optimization method, aiming to improve the overall power generation of the wind farm and suppress structural damage.

[0104] The controller design integrates model-driven and data-driven optimization approaches and introduces a momentum mechanism to improve the smoothness of the search trajectory and convergence speed. It is officially named the "Heavy-ball Feedback Optimization (HBFO)" controller. The controller operation consists of two parts: a gradient estimation phase and a momentum update phase. The iterative process is described below.

[0105] 1) Gradient estimation stage

[0106] This stage is used to determine the decreasing direction of the current axial induction factor. To enhance the robustness of the controller to wind condition uncertainties, a joint estimation approach combining model-driven and data-driven methods is introduced. First, based on the axial induction factors of each wind turbine in the current iteration, the wind speed and power generation under the influence of the wake are calculated, followed by the following iterative process:

[0107] .

[0108] in, The model driving direction, obtained based on the power sensitivity matrix, reflects the power change trend under the model prediction and has a good convergence speed. As a data-driven approach based on residual feedback, it estimates the downward trend by utilizing the difference between the current and previous iteration objective function values, thus exhibiting stronger adaptability to model errors. For the first k The sensitivity matrix calculated from the current wind speed and induction factor in the next iteration reflects the impact of changes in the axial induction factor on power output. Let be a random vector uniformly sampled on a unit sphere, satisfying . The weighting coefficients represent the model-driven and data-driven directions. In the initial stage of iteration, the model-driven direction has a higher weight to accelerate convergence; in the later stage, the data-driven direction is gradually strengthened to improve environmental adaptability.

[0109] 2) Momentum renewal phase

[0110] This stage employs the heavy-ball method for momentum updates to avoid getting trapped in local optima and improve iterative stability. The specific process is as follows:

[0111]

[0112]

[0113] In the formula, Step size, Momentum factor To explore the disturbance,

[0114] First, update the candidate solutions. Then, using the integrated gradient direction intermediate variables Update and refine the candidate solution trajectory using the momentum term. To avoid the solution getting trapped in a local optimum, a certain exploration perturbation is added, and finally the candidate solution is projected onto the feasible region to obtain a new axial induction factor value. .

[0115] In summary, in each iteration, the controller updates the axial induction factor of each wind turbine based on the current wind conditions and power generation status, and comprehensively utilizes model sensitivity information and real-time performance feedback, thereby gradually approaching the optimal value of the objective function and achieving a comprehensive improvement in the overall output performance and operational reliability of the wind farm.

[0116] (2) Controller convergence analysis

[0117] To theoretically verify the effectiveness of the Heavy Ball Feedback Optimization (HBFO) controller proposed in this embodiment, the convergence of its simplified objective function is analyzed. Firstly, it can be seen that... Regarding axial induction factor It is M-Lipschitz and L-smooth. At the same time, Throughout the feasible domain It is bounded, meaning there exists a constant. , so that for any They all .

[0118] Based on the above properties, we further analyze the sources of error in gradient estimation in the controller and establish its upper bound. The error between the approximate sensitivity matrix and the true sensitivity matrix is ​​defined as follows:

[0119]

[0120] The error between the synthesized descent direction and the true gradient of the system in the controller proposed in this invention satisfies the following limits.

[0121]

[0122] in, This is the overall upper bound for the mixed estimation error. The three terms on the right-hand side of the above equation represent the model error, the model-independent estimation error, and the additional bias caused by the disturbance, respectively. Further, we have the following squared error bound.

[0123]

[0124] in It is a constant. The upper bound of the variance is estimated for the descent direction of the controller. This leads to the following overall convergence conclusion: Assume the controller operates... Iterate in steps, with a step size of

[0125]

[0126] Then it satisfies the following asymptotic convergence property

[0127]

[0128] in, This represents the optimal objective function value. The convergence term on the right indicates the expected value of the squared gradient norm. (1 / The sublinear velocity decays.

[0129] In summary, the controller proposed in this embodiment still possesses asymptotic convergence properties even in the presence of model inaccuracies and disturbances, and can achieve the pre-set control objectives.

[0130] 5. Verify the control method using an actual offshore wind farm system.

[0131] To verify the effectiveness of the proposed control method, this embodiment constructs an 8-turbine system based on the Horns Rev I offshore wind farm layout in the MATLAB platform. The turbines are spaced at 7 times the rotor diameter (7D), and all turbines use the NREL-5MW model. The control variable is the axial induction factor of each turbine, with a value range of [0, 0.5] and a nominal initial value of 1 / 3. This system considers the model uncertainty caused by wake interference and wind speed measurement errors, and the control effect under actual operating conditions is used as the evaluation criterion. The wake field simulation of this verification scenario is shown below. Figure 3 As shown.

[0132] To verify the effectiveness of our proposed method, the following three sets of comparative experiments were conducted:

[0133] (1) Convergence Comparison Experiment of Axial Induced Factor: Under the condition of a sudden wind speed change from an initial wind speed of 8 m / s to 10 m / s at the 1000th iteration, random Gaussian noise was superimposed every 10 steps to simulate the inaccuracy of the wind speed. The convergence performance of the axial induced factor of the three control methods under this environment was compared. The methods used for comparison were: the Heavy Ball Feedback Optimization (HBFO) method proposed in this invention, the standard stochastic gradient descent (SGD) method, and the unaccelerated hybrid control method (Hybrid). To obtain an accurate reference benchmark, the prior results of the ideal model under known wind conditions were used as a control.

[0134] (2) Comparative Experiment on the Improvement Effect of Objective Function: Under the same wind condition fluctuations and inaccuracies as Experiment 1, the improvement effects of different methods on the objective function of the wind farm were evaluated. The objective function comprehensively considers the mechanical degradation cost caused by wind power revenue and axial induction factor adjustment, and uses the optimal solution obtained offline by FLORIS software under known precise wind conditions as the theoretical upper limit benchmark. The methods involved in the comparison include: the HBFO method proposed in this invention, the non-accelerated hybrid method, and the Greedy control strategy;

[0135] (3) Verification experiment on the objective function improvement effect under measured SCADA data: To evaluate the operating performance of the proposed controller in actual wind conditions, 30 sets of continuous wind speed samples were selected from the SCADA data of an offshore wind farm in China as the input scenario. Each set of samples corresponds to a 10-minute time period, covering a total operating cycle of 300 minutes. For each wind condition sample, the objective function was calculated using the following three methods: the HBFO method proposed in this invention, the model-based control method, and the greedy method. The effects of each method during the iteration process were compared.

[0136] Experiments were conducted under three different conditions, and the following experimental results were obtained.

[0137] The results of experiment (1) are as follows Figure 4 As shown in the figure. The results show that the HBFO method can quickly converge to the vicinity of the reference result and maintain strong robustness and exploration ability under the presence of perturbations. In contrast, the traditional hybrid method has a significantly longer convergence time under the influence of perturbations. The stochastic gradient descent method based on the static model has a slow response to AIF adjustment in the wind condition change range and has obvious oscillation phenomenon, making it difficult to converge and lacking adaptability.

[0138] The results of experiment (2) are as follows Figure 5As shown in the figure. Simulation results show that at a wind speed of 8 m / s, the HBFO method is only about 20.02% better than the FLORIS result, significantly outperforming the Hybrid and Greedy methods; while at a wind speed of 10 m / s, this gap further narrows to 12.27%. The proposed HBFO controller can effectively adjust the control strategy in real time under the condition of periodic wind speed disturbances in actual wind data, ensuring that the power generation trajectory of the cluster always approaches the static optimal power level obtained from offline FLORIS optimization. Especially in stages with large wind changes and significant disturbances, the HBFO method exhibits good dynamic response capabilities, quickly adjusting the AIF control strategy and following the trend of optimal power changes, avoiding significant power deviation and lag. In contrast, other control methods consistently show that the average power output level is significantly lower than the offline optimal decision when wind speed disturbances occur. In summary, the HBFO controller possesses excellent optimal approximation capability and energy capture efficiency under dynamic disturbance environments, exhibiting not only good convergence performance but also achieving real-time control results closer to the ideal optimization result.

[0139] The results of experiment (3) are as follows Figure 6 As shown, the HBFO method achieves a higher objective function value in most time periods, and its performance curve is significantly improved compared to the other two methods. This indicates that it has strong environmental adaptability and robustness in actual wind condition fluctuation scenarios, can effectively optimize the overall power generation capacity of wind farms, and has good application potential in improving the power generation efficiency of offshore wind farms.

[0140] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for hybrid control of output power of an offshore wind farm under wake effect, characterized in that, The method comprises the following steps: S1: Real-time acquisition of offshore wind farm operating parameters and wherein is the free stream wind speed of the wind farm, is the full farm wind turbine active power vector obtained from the SCADA system; S2: the and an input controller, the controller outputting a vector of axial induction factors for the wind turbines within the entire offshore wind farm ; S3: According to the vector Adjusting each wind turbine within an offshore wind farm; S1-S3 are repeated; The objective function of the controller is: In the formula, the wind turbine set in the entire offshore wind farm is , is the axial induction factor vector of the wind turbine in the entire offshore wind farm, is the initial axial induction factor vector set by the wind farm control center, is the adjustment cost coefficient, is the wind power generation price, is the active power vector of the wind turbine in the entire offshore wind farm, is the axial induction factor feasible region, The control strategy of the controller comprises a gradient estimation stage and a momentum update stage, The gradient estimation stage comprises joint estimation in two directions of model driving and data driving, The specific iteration process of the gradient estimation stage is: In the formula, k This indicates the k-th iteration. It is a vector composed of the axial induction factors of the wind turbines in the entire offshore wind farm at the k-th iteration. It is the power vector of all wind turbines in the k-th iteration. The driving direction of the model is obtained based on the power sensitivity matrix. For data-driven approaches based on residual feedback, For the first k The sensitivity matrix is ​​calculated from the current wind speed and the induction factor in the next iteration. Let be a random vector uniformly sampled on a unit sphere, satisfying , The weighting coefficients for model-driven and data-driven approaches.

2. The control method according to claim 1, characterized by, In the initial stage of iteration, the weight of the model driving direction is higher, and the weight of the data driving direction is gradually increased in the later stage of iteration.

3. The control method according to claim 1, characterized by, The momentum update stage adopts a heavy ball method for momentum update.

4. The control method according to claim 3, characterized by, The specific process of the heavy ball method is: wherein is a step size, is a momentum factor, is an exploration perturbation, First, update the candidate solutions. Then, using the integrated gradient direction intermediate variables Update and refine the candidate solution trajectory using the momentum term. To avoid the solution getting trapped in a local optimum, a certain exploration perturbation is added, and finally the candidate solution is projected onto the feasible region to obtain a new axial induction factor value. .

5. A device for controlling the output power of an offshore wind farm under wake effect, characterized in that, The device is used to implement the control method according to any one of claims 1-4, comprising: Data acquisition module: for real-time acquisition of offshore wind farm operation parameters , and ; Control module: for inputting offshore wind farm operation parameters , and and outputting axial induction factor vector of wind turbines in the whole offshore wind farm ; Execution module: for adjusting individual wind turbines in accordance with the axial induction factor vector output by the controller module Adjusting individual wind turbines.

6. A computer device, comprising: including: a memory; one or more processors coupled to the memory; one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by one or more processors, and the one or more application programs are configured to execute the control method according to any one of claims 1-4.

Images