Expansion optimization method of distributed wind power plant control system based on Bayesian optimization

Abstract

The invention discloses an expansion optimization method of a distributed wind power plant control system based on Bayesian optimization, and belongs to the field of intelligent control of a wind power electric field. According to the method, in a distributed framework, modeling is carried out on a black-box objective function of each wind power vortex machine through Bayesian optimization, and an acquisition function based on priori knowledge is utilized to acquire a black-box objective function of each wind power vortex machine; according to the method, data point collection is carried out on all wind power vortex machines, information interaction and collaborative optimization are carried out among different wind power vortex machines through a strong connection network, then an agent model is continuously updated, an optimization strategy is adjusted, an optimal solution of a target function of each wind power vortex machine is obtained, and then an optimal parameter combination of the wind power vortex machines is determined. Experiments show that the method can fit the black box objective function more quickly, the Bayesian optimization evaluation cost is reduced, the power generation efficiency of the wind power plant is improved, and efficient utilization of wind energy is achieved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control of wind farms, and in particular to an extended optimization method of a distributed wind farm control system based on Bayesian optimization. Background Art

[0002] In recent years, artificial intelligence has developed rapidly and has been widely used in power systems. Artificial intelligence optimizes the operation and maintenance of wind farms through data analysis and prediction, improves power generation efficiency and reduces equipment failures. At the same time, machine learning algorithms are used to predict wind speed and optimize wind turbine scheduling, thereby improving the overall benefits of wind power.

[0003] A wind farm is a place where wind resources are used to generate electricity. Using wind turbines to convert wind energy into electrical energy is the main way of generating wind power in my country. A wind turbine is a new type of wind turbine that improves power generation efficiency and stability through rotating vortex technology. The daily operation of a wind farm system generates a large amount of power data. With the development of edge computing technology, multiple wind turbines perform data calculation, storage, and network information exchange to achieve stable and efficient operation of the wind farm. However, how to reasonably dispatch information exchange between wind turbines in a wind farm to match the power generation of the wind farm with the power demand proposed by the grid operator is a key factor in solving the problem of active adaptation of wind farms.

[0004] In the wind farm scenario, since the power generation method is affected by climate change and geographical location, the operation of the wind turbine in the wind farm is affected by factors such as wind speed, wind direction, and atmospheric pressure, and its objective function is usually a non-convex black box function. Therefore, traditional optimization methods may find it difficult to effectively handle these complex situations. Bayesian Optimization, as an efficient global optimization method, can efficiently explore the parameter space with fewer samples by constructing a black box objective function probabilistic proxy model without fully understanding the characteristics of wind power and complex environmental conditions. It is particularly suitable for dealing with high-dimensional, nonlinear and noisy situations in wind farms.

[0005] In a distributed framework, a wind farm is divided into multiple sub-areas. The wind turbines in each sub-area are managed and controlled as independent nodes. The wind turbines can cooperate with each other and autonomously adjust the operating parameters under different wind speed conditions to improve the power generation efficiency and system reliability of the entire wind farm. However, in the process of Bayesian optimization of distributed wind farms, since all wind turbines need to model their black box objective functions, the cost of function evaluation is very expensive. Therefore, it is of great significance to study how to determine the optimal parameter combination in the wind farm power control optimization model faster and more efficiently while reducing the cost of Bayesian optimization evaluation, so as to improve the autonomous dispatching capability and adaptability of the power system. Summary of the invention

[0006] The purpose of the present invention is to provide an extended optimization method for a distributed wind farm control system based on Bayesian optimization, which can fit the black box objective function more quickly, reduce the Bayesian optimization evaluation cost, improve the system's autonomous scheduling capability and adaptability, and achieve efficient utilization of wind energy in wind farms.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is:

[0008] An extended optimization method for a distributed wind farm control system based on Bayesian optimization comprises the following steps:

[0009] Step S1, establishing a local proxy model, a local data set and a black box objective function for each wind turbine in the distributed wind farm, and initializing the data;

[0010] Step S2, using the acquisition function based on prior knowledge, collects data points for all wind turbines, and combines the data collected by all wind turbines in this round. Different wind turbines are connected through a strong network. Conduct information interaction and collaborative optimization to obtain the collaborative consensus point of all wind turbines in this round, where K = [1, 2, ..., K] represents the set of wind turbines, ε = K × K represents the edge of the network, and W represents the K-order weighted adjacency matrix;

[0011] Step S3, updating the local proxy model and local data set of each wind turbine, and then performing a new round of data collection and optimization;

[0012] Step S4, as the number of iterations increases, the weighted adjacency matrix is ​​transformed into a unit matrix, and then the optimal solution of the objective function of each wind turbine is calculated to determine the optimal parameter combination of the wind turbine.

[0013] Furthermore, the proxy model adopts a Gaussian process proxy model, and the acquisition function includes an expected lift acquisition function and an upper confidence bound acquisition function.

[0014] Furthermore, the expected improvement acquisition function is:

[0015] Where μ(x) and σ(x) are the Gaussian prediction mean and variance, respectively. is the black box objective function f k The optimal function value found so far, φ is the probability density function of the standard normal distribution, and Φ is the cumulative distribution function of the standard normal distribution.

[0016] Furthermore, the upper confidence bound acquisition function is:

[0017] In the formula, μ t-1 (x),σ t-1 (x) are the Gaussian prediction mean and variance, β t is a constant.

[0018] Furthermore, the acquisition function based on prior knowledge is:

[0019] Where α(x,D) represents the acquisition function of the data point x, D represents the local data set of the wind turbine, and π(x) represents the prior knowledge.

[0020] Furthermore, the step S2 specifically includes: using the acquisition function based on prior knowledge to collect information of this round on the objective functions of all wind turbines, splicing all the collected information, and then performing collaborative optimization between the wind turbines, which is expressed as:

[0021] In the formula, [·] k Represents the kth wind turbine at the current collection point is the splicing data of K wind turbines in this round, is the objective function f after collaborative optimization k New data from I d is a d-order identity matrix, Represents the Kronea operation, W t is the weighted adjacency matrix of round t.

[0022] Furthermore, in the collaborative optimization process, the focus of collaborative optimization of multiple wind turbines gradually shifts from collaborative optimization between wind turbines to each wind turbine performing its own matrix optimization, while ensuring that the weighted adjacency matrix W has double randomness.

[0023] Furthermore, each round of collaborative optimization increases the diagonal elements of the weighted adjacency matrix W by The weights of the weight adjacency matrix W are reduced by With the increase of the number of iterations, the weighted adjacency matrix W gradually turns into a unit matrix, where K is the number of wind turbines and T is the number of iterations.

[0024] Further, initialize all elements of the weighted adjacency matrix W to

[0025] Furthermore, in step S3, before updating the local proxy model and the local data set, the function value of the wind turbine is evaluated, and after the evaluation, new data about the black box objective function of all wind turbines is generated, and each new data is sent to the corresponding wind turbine.

[0026] By adopting the above technical solution, the present invention can achieve the following beneficial effects:

[0027] (1) The Gaussian process is used to model the optimization parameters of the black box objective function of the wind turbine in each sub-region, and the acquisition function is used to balance exploration and utilization between the global and local levels. The wind turbines in the wind farm are collaboratively optimized in a distributed framework, so as to quickly solve the optimal parameter combination of the wind turbine under the conditions of high power generation and low power transmission loss, thereby realizing high-speed operation of the wind farm.

[0028] (2) Bayesian optimization optimizes wind farms in a distributed framework and can effectively handle complex multi-parameter optimization problems, make full use of limited computing resources and time, and find the optimal wind turbine configuration solution.

[0029] (3) By guiding the optimization through prior knowledge, the spatial search direction of the wind turbine black box objective function can be more effectively guided in the early stage of optimization, which can effectively avoid evaluation in areas far away from the optimal value of the objective function, reduce the optimization cost of the black box objective function, further improve the evaluation efficiency of the black box objective function, and reduce the evaluation cost of Bayesian optimization.

[0030] (4) The present invention can not only significantly improve the overall power generation efficiency of the wind farm, but also optimize the layout of the wind turbines under complex terrain and changeable wind conditions to maximize energy output, and can dynamically adapt to real-time data and environmental changes to ensure that the wind farm can maintain efficient operation at different operating stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is the overall framework diagram of the present invention;

[0032] Figure 2 It is a flowchart of the optimization algorithm in the embodiment of the present invention;

[0033] Figure 3 This is a comparison diagram of average power generation under condition A1 in an embodiment of the present invention;

[0034] Figure 4 This is a comparison diagram of average power generation under the condition A2 in the embodiment of the present invention;

[0035] Figure 5 This is a comparison diagram of average power generation under the condition A3 in the embodiment of the present invention;

[0036] Figure 6 This is a comparison diagram of average power generation under the condition A4 in the embodiment of the present invention. DETAILED DESCRIPTION

[0037] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] In order to reduce the expensive evaluation cost in the Bayesian optimization algorithm, improve the autonomous dispatching capability and adaptability of the power system, and realize the efficient utilization of wind energy in the wind farm, the present invention provides an extended optimization method of a distributed wind farm control system based on Bayesian optimization, comprising the following steps:

[0039] Step S1, establishing a local proxy model, a local data set and a black box objective function for each wind turbine in the distributed wind farm, and initializing the data;

[0040] Step S2, using the acquisition function based on prior knowledge, collects data points for all wind turbines, and combines the data collected by all wind turbines in this round. Different wind turbines are connected through a strong network. Conduct information interaction and collaborative optimization to obtain the collaborative consensus point of all wind turbines in this round, where K = [1, 2, ..., K] represents the set of wind turbines, ε = K × K represents the edge of the network, and W represents the K-order weighted adjacency matrix;

[0041] Step S3, updating the local proxy model and local data set of each wind turbine, and then performing a new round of data collection and optimization;

[0042] Step S4, as the number of iterations increases, the weighted adjacency matrix is ​​transformed into a unit matrix, and then the optimal solution of the objective function of each wind turbine is calculated to determine the optimal parameter combination of the wind turbine.

[0043] The present invention enables multiple wind turbines in a wind farm to exchange information efficiently in a distributed framework, and at the same time guides and optimizes the black box objective function of the wind turbine in combination with prior knowledge in the wind farm field, thereby reducing optimization costs and improving the power generation efficiency of the wind farm.

[0044] The system consists of K wind turbines, each of which has its own local agent model and local datasets As the optimization proceeds, each wind turbine continuously updates its local data set and local proxy model. At the same time, different wind turbines exchange information with each other to fit the black box objective function more quickly, thereby achieving efficient utilization of wind energy in the wind farm.

[0045] The present invention is specifically divided into two parts: First, the black box function f of the wind turbine k (k∈K) is optimized by using the Bayesian optimization method. k First, optimization is performed; secondly, in a distributed framework, collaborative optimization of multiple wind turbines in a wind farm is performed.

[0046] In this distributed framework, the wind farm contains K wind turbines, each of which has its corresponding black box objective function The output power of the wind turbine can be calculated through the black box objective function. Initial local agent model and the initial local dataset The specific technical solutions are as follows:

[0047] Bayesian optimization based on prior knowledge:

[0048] Bayesian optimization is a black-box optimization A powerful optimization method whose main goal is to find the black box function f k In real number space The optimal solution in It is expressed as:

[0049] Where d is the spatial dimension. Bayesian optimization also considers the influence of noise during the optimization process, so the objective function disturbed by noise is expressed as: y k =f(x k )+∈ k (2)

[0050] ∈ k ~N(0,σ 2 ) is independent and identically distributed additive Gaussian noise, σ 2 is the noise variance. Bayesian optimization is mainly divided into two parts: agent model and acquisition function.

[0051] The implementation process of the proxy model is as follows:

[0052] The agent models the black box objective function through the data obtained by the acquisition function, and gradually deepens the understanding of the objective function as the amount of data increases. The Gaussian process agent model is widely used in Bayesian optimization due to its flexible function fitting ability, efficient uncertainty measurement ability, and easy Bayesian inference. The Gaussian process is represented by GP(μ(·), k(·, ·)), where is the mean function, Represents a positive definite kernel function. The present invention uses a zero-mean Gaussian process to calculate the objective function of the wind turbine Modeling is performed, expressed as: f k~GP(0,k(x,x′)) (3)

[0053] Assuming that after T rounds of optimization, the local data set corresponding to the wind turbine k (k∈K) is expressed as The parameter update process of the Gaussian process in the Tth round can be expressed as:

[0054] In the formula, y T =[y1,…,y T ] is the function value disturbed by noise in T rounds of optimization.

[0055] By continuously collecting and analyzing meteorological data such as wind speed and direction, as well as the operating status and maintenance records of wind turbines, Bayesian optimization can continuously update the probabilistic agent model, adjust the optimization strategy, and achieve adaptive optimization. This continuous optimization capability not only reduces operating and maintenance costs, but also improves the reliability and economic benefits of wind farms.

[0056] The implementation process of the acquisition function is as follows:

[0057] Collection function It is a collection strategy of function data point x, which guides data collection in the function space of the objective function. The present invention mainly uses the upper confidence bound (UCB) collection function and the expected improvement (EI) collection function to collect black box objective functions of all wind turbines. Collect new data points x. The expected improvement acquisition function mathematical expression is:

[0058] Where μ(x) and σ(x) are the Gaussian prediction mean and variance, respectively. is the black box objective function f k The optimal function value found so far, φ is the probability density function of the standard normal distribution, and Φ is the cumulative distribution function of the standard normal distribution.

[0059] The mathematical expression of the upper confidence bound acquisition function is:

[0060] In the formula, μ t-1 (x),σ t-1 (x) are the Gaussian prediction mean and variance, β t is a constant. t Used to decide whether to explore in unknown exploration space areas or explore near "high-value" areas that have been explored.

[0061] Since the objective function of the wind turbine is a black box, the objective function f k The evaluation is very expensive. In order to reduce the cost of optimization, the present invention also considers using prior knowledge to guide the optimization and avoid excessive collection of "low-value" areas by the acquisition function. These prior knowledge include parameters such as the daily power generation of wind turbines in different climates and the maximum power generation. We use π k (x) is used to represent the prior knowledge. Combining the above knowledge, the acquisition function based on prior knowledge is expressed as:

[0062] The use of prior knowledge for Bayesian optimization can more effectively guide the search direction of the wind turbine black box function space in the early stage of optimization, thereby reducing the need for expensive function evaluation in high-dimensional space, accelerating convergence and improving optimization efficiency.

[0063] The distributed collaborative optimization process of multiple wind turbines is as follows:

[0064] In the distributed wind farm framework, each wind turbine k (k∈K) is connected through a strongly connected network Information exchange is carried out, where K = [1, 2, ..., K] is the set of wind turbines, ε = K × K represents the edge of the network, and W is the K-order weighted adjacency matrix. Wind turbines can exchange information with each other. Since little is known about the black box objective function of wind turbines in the initial stage of optimization, it is necessary to actively promote information exchange between wind turbines. However, as the optimization proceeds, the proxy model of each wind turbine Both of them have black box objective function The modeling is continuously refined. Considering the privacy of different wind turbines and the efficiency of optimization, the amount of data exchanged between wind turbines is reduced. This requires a detailed design of the weighted adjacency matrix W. First, all elements of the weighted adjacency matrix W are initialized as W 0 It is expressed as:

[0065] Then, as the optimization proceeds, the weighted adjacency matrix W is continuously updated to guide the wind turbines to perform collaborative optimization in the distributed framework. Assuming that the current number of optimizations is t, the acquisition function based on prior knowledge is used The objective function for all wind turbines is Carry out this round of information collection, splice all the collected information, and then perform collaborative optimization, which is expressed as:

[0066] In the formula, [·]k Represents the kth wind turbine at the current collection point It is the splicing data of K wind turbines in this round. is the objective function f after collaborative optimization k New data. d is an identity matrix of order d. Representatives of the Cronea operation, A new matrix of order (D×K) will be generated. t is the weighted adjacency matrix of the tth round, expressed as:

[0067] And W t It is a doubly random matrix, that is, the sum of all elements in each row and column is 1, expressed as: ∑ l ω kj =∑ j ω jk =1,j,k∈K (14)

[0068] Next, the weighted adjacency matrix W is updated to protect the privacy data of different wind turbines and achieve efficient collaborative optimization. For the update of the weighted adjacency matrix W, we consider that as the optimization proceeds, the optimization focus gradually shifts from collaborative optimization between wind turbines to matrix optimization of each wind turbine, while ensuring the double randomness of the weighted adjacency matrix W. Therefore, in each round, the diagonal elements of the weighted adjacency matrix W are increased by The weights of the matrix are reduced by The weight of the weight adjacency matrix W from the tth round to the t+1th round element optimization is expressed as:

[0069] In the formula, represents the off-diagonal elements of the matrix, represents the diagonal elements of the matrix. Therefore, the update process of the weighted adjacency matrix W from round t to round t+1 is expressed as:

[0070] As the number of iterations t increases, W gradually transforms into a unit matrix, so that each wind turbine finds its own black box objective function Therefore, we collaboratively optimize the wind turbines in the wind farm in a distributed framework, effectively model the black box objective function of each wind turbine through Bayesian optimization, and use wind farm data to find the best parameter combination that enables the wind turbine to generate high power and low power transmission loss, so as to achieve high-speed operation of the wind farm. In addition, we use the prior knowledge of the wind farm to further reduce the optimization cost of the black box function.

[0071] This embodiment uses a distributed multi-task coordination based on prior knowledge (PMCBO) algorithm to achieve optimal dispatch of wind farm power. Figure 2 The algorithm deploys the information interaction server of the wind farm to decide the wind farm power dispatching plan. When the power generation and energy storage of the wind turbine are affected by factors such as wind speed, wind direction, and atmospheric pressure, the power dispatching time is minimized and the power waste is minimized, thus improving the efficiency of wind farm power dispatching and achieving accurate matching of power supply and demand.

[0072] Specifically, the algorithm flow for achieving optimal dispatch of wind farm power using PMCBO is as follows:

[0073] (1) First, the initial data Assigned to each wind turbine, each wind turbine has its corresponding black box objective function The initial data of any wind turbine k includes data, and the maximum number of optimization times of the algorithm is represented by T.

[0074] (2) The next step is to initialize each wind turbine Local dataset and local proxy models

[0075] (3) Selecting a sampling strategy: The present invention mainly uses the UCB acquisition function and the EI acquisition function to perform sampling on the black box objective function. New data points Collection of.

[0076] The UCB acquisition function is expressed as:

[0077] The EI acquisition function is expressed as:

[0078] (4) The new data collected by all wind turbines in the tth round are combined, expressed as X = [x1, ..., x k ].

[0079] (5) K wind turbines are connected through a strong network Exchange information and calculate the collaborative consensus point of all wind speed vortex machines in round t

[0080] (6) The t-th round function value evaluation of the wind turbine k (k∈K) is expressed as in is an additive Gaussian error that conforms to the standard normal distribution, σ 2 is the noise variance. After function value evaluation, new data about the black box objective function of all wind turbines is generated, expressed as

[0081] (7) Each new data is sent to the corresponding wind turbine k to update its local proxy model and local data set.

[0082] (8) Update the weighted adjacency matrix W of this round.

[0083] After t (t≤T) rounds of optimization, the optimal parameter combination scheme for the power generation, energy storage and transmission of the wind turbine k (k∈K) is determined, and each wind turbine is enabled to exchange information in a distributed framework, reducing the high optimization cost in distributed Bayesian optimization. Therefore, the PMCBO algorithm can achieve the best solution for power dispatch and matching in wind farms while saving resources. Bayesian optimization optimizes wind farms in a distributed framework, can effectively handle complex multi-parameter optimization problems, make full use of limited computing resources and time, and find the optimal wind turbine configuration scheme.

[0084] The idealized large-eddy simulations method is used below to illustrate the effectiveness and superiority of the present invention through experiments. This method has been widely used in wind farm scenarios under neutral, stable and unstable stratified flow conditions. This embodiment uses an offshore wind farm combined with a floating photovoltaic device. The specific wind farm layout is shown in Table 1. The offshore wind farm simulation set includes four cases: A1, A2, A3, and A4. SF is the offshore wind farm layout code combined with floating photovoltaics. represents the driving pressure gradient, N s , l x × y are the number and size of floating PV strips, Ri h / Ri h，ref It indicates that the heterogeneity parameter is normalized, WF indicates the number of rows and columns of DTU 10MW wind turbines included in the wind farm. Under ideal conditions, the total installed capacity of the wind farm is 450MW. Table 1 Wind farm initialization simulation parameters

[0085] By initializing wind farm scenarios with different simulation conditions (wind speed, temperature difference, layout, etc.), taking a time interval of 10 minutes as an iteration number t, it is found that the extended optimization method of the present invention (Proposed Method) increases the average power generation by 10% compared with the traditional wind farm control method (Standard Method). The results are as follows: Figures 3 to 6 As shown, the present invention also performs adaptive adjustment of parameters such as wind speed, wind direction, and pressure in a distributed framework based on prior knowledge, which greatly saves power resources. In general, the present invention significantly improves the power generation efficiency and economic benefits of wind farms through scientific optimization methods, making a huge contribution to the development of renewable energy.

[0086] Finally, it should be noted that the parts of the present invention that are not described in detail are all prior art. Those of ordinary skill in the art can understand that the above are only preferred examples of the invention and are not intended to limit the invention. Although the invention is described in detail with reference to the aforementioned examples, those of ordinary skill in the art can still modify the technical solutions recorded in the aforementioned examples, or replace some of the technical features therein with equivalents. Any modifications, equivalent replacements, etc. made within the spirit and principles of the invention should be included in the scope of protection of the invention.

Claims

1. An extended optimization method for distributed wind farm control systems based on Bayesian optimization, characterized in that: The following steps are involved: Step S1, establishing a local proxy model, a local data set and a black box objective function for each wind turbine in the distributed wind farm, and initializing the data; Step S2, using the acquisition function based on prior knowledge, collects data points for all wind turbines, and combines the data collected by all wind turbines in this round. Different wind turbines are connected through a strong network. Conduct information interaction and collaborative optimization to obtain the collaborative consensus point of all wind turbines in this round, where K = [12, ..., K] represents the set of wind turbines, ε = K × K represents the edge of the network, and W represents the K-order weighted adjacency matrix; Step S3, updating the local proxy model and local data set of each wind turbine, and then performing a new round of data collection and optimization; Step S4, as the number of iterations increases, the weighted adjacency matrix is ​​transformed into a unit matrix, and then the optimal solution of the objective function of each wind turbine is calculated to determine the optimal parameter combination of the wind turbine.

2. The extended optimization method of distributed wind farm control system based on Bayesian optimization according to claim 1 is characterized in that: The proxy model adopts a Gaussian process proxy model, and the acquisition function includes an expected lift acquisition function and an upper confidence bound acquisition function.

3. The extended optimization method of distributed wind farm control system based on Bayesian optimization according to claim 2 is characterized in that: The expected improvement acquisition function is: Where μ(x) and σ(x) are the Gaussian prediction mean and variance, respectively. is the black box objective function f k The optimal function value found so far, φ is the probability density function of the standard normal distribution, and Φ is the cumulative distribution function of the standard normal distribution.

4. The extended optimization method of distributed wind farm control system based on Bayesian optimization according to claim 2 is characterized in that: The upper confidence bound acquisition function is: In the formula, μ t-1 (x),σ t-1 (x) are the Gaussian prediction mean and variance, β t is a constant.

5. The extended optimization method of distributed wind farm control system based on Bayesian optimization according to claim 1, characterized in that: The acquisition function based on prior knowledge is: Where α(x, D) represents the acquisition function of the data point x, D represents the local data set of the wind turbine, and π(x) represents the prior knowledge.

6. The extended optimization method for distributed wind farm control system based on Bayesian optimization according to claim 1, characterized in that: The step S2 specifically includes: using a collection function based on prior knowledge to collect information of this round on the objective functions of all wind turbines, splicing all collected information, and then performing collaborative optimization between wind turbines, which is expressed as: In the formula, [·] k Represents the kth wind turbine at the current collection point is the splicing data of K wind turbines in this round, is the objective function f after collaborative optimization k New data from I d is a d-order identity matrix, Represents the Kronea operation, W t is the weighted adjacency matrix of round t.

7. The extended optimization method for distributed wind farm control system based on Bayesian optimization according to claim 6, characterized in that: During the collaborative optimization process, the focus of collaborative optimization of multiple wind turbines gradually shifts from collaborative optimization between wind turbines to each wind turbine performing its own matrix optimization, while ensuring that the weighted adjacency matrix W has double randomness.

8. The extended optimization method for distributed wind farm control system based on Bayesian optimization according to claim 6, characterized in that: Each round of collaborative optimization increases the diagonal elements of the weighted adjacency matrix W by The weights of the weight adjacency matrix W are reduced by With the increase of the number of iterations, the weighted adjacency matrix W gradually turns into a unit matrix, where K is the number of wind turbines and T is the number of iterations.

9. The extended optimization method of distributed wind farm control system based on Bayesian optimization according to claim 6, characterized in that: Initialize all elements of the weighted adjacency matrix W to 10. The extended optimization method of distributed wind farm control system based on Bayesian optimization according to claim 1, characterized in that: In step S3, before updating the local proxy model and the local data set, the function value of the wind turbine is evaluated, and after the evaluation, new data about the black box objective function of all wind turbines is generated, and each new data is sent to the corresponding wind turbine.

Images