Data-driven linear MPC-based assessment method and apparatus for frequency regulation capability of wind farm, and control method and apparatus

Abstract

Disclosed in the present invention are a data-driven linear MPC-based assessment method and apparatus for a frequency regulation capability of a wind farm, and a control method and apparatus. The assessment method comprises the steps of: offline acquiring historical operation data sets of wind turbines, and training a linear predictive control model, wherein the linear predictive control model is obtained by applying a dimensionality-increasing transformation process to a wind farm frequency modulation dynamic nonlinear model on the basis of a Koopman operator theory; acquiring real-time operation data of the wind turbines, and using the acquired real-time operation data to solve for a linear MPC optimization model, to obtain droop coefficients of the wind farm, wherein the linear MPC optimization model is constructed on the basis of the trained linear predictive control model and by taking the maximum droop coefficient of the wind farm as an objective and setting a safety rotational speed constraint condition; and assessing a frequency regulation capability of the wind farm on the basis of the obtained droop coefficients of all the wind turbines in the wind farm. The present invention has the advantages of ease of implementation, high assessment efficiency and accuracy, and strong scalability, and the like.

Description

A data-driven linear MPC-based method, control method, and device for evaluating the frequency regulation capability of wind farms.

[0001] Cross-reference to related applications

[0002] [Correction 21.05.2025 in accordance with Rule 91] This application is based on and claims priority to Chinese Patent Application No. 2024106736154, filed on May 28, 2024, entitled “Wind Farm Frequency Regulation Capability Evaluation Method, Control Method and Device Based on Data-Driven Linear MPC”, the entire contents of which are incorporated herein by reference as a part of this application. [Technical Field]

[0003] This invention relates to the field of wind farm frequency regulation technology, and in particular to a method, control method and device for evaluating the frequency regulation capability of wind farms based on data-driven linear MPC (Model Predictive Control). [Background Technology]

[0004] In wind power generation, the wind turbine is connected to the grid via a power electronic converter, leading to decoupling between rotor speed and system frequency, making it difficult to respond quickly and effectively to changes in system frequency. Power systems exhibit low inertia in many areas, requiring wind turbines to participate in primary frequency regulation to ensure system frequency stability. Variable-speed wind turbines can capture kinetic energy from flowing air and have a wide speed regulation range. Therefore, the most widely adopted method is to add an additional control module to the power electronic converter. By adjusting the wind turbine's output power, the kinetic energy stored in the turbine blades is used to simulate the sag characteristics of a traditional generator and provide frequency support.

[0005] During primary frequency regulation, to ensure the safe operation of the wind turbine, the rotor speed must be maintained within a safe range. Therefore, before participating in primary frequency regulation, the wind farm first needs to assess the maximum droop coefficient under different wind speed scenarios, i.e., the wind farm's frequency regulation capability. Current technologies typically rely on traditional physical model analysis to realize the frequency regulation capability of a wind farm during primary frequency regulation, which presents the following problems:

[0006] 1. Traditional physical models rely heavily on the accuracy of parameters and the completeness of the model. If the parameters are inaccurate, it is difficult to guarantee the accuracy of the droop coefficient assessment.

[0007] 2. As the system size increases, the solution time for large-scale optimization models will also increase, making them unsuitable for online applications.

[0008] 3. The dynamic model of primary frequency regulation in wind farms exhibits complex nonlinear characteristics. Nonlinear models are difficult to solve, and it is hard to find a suitable solution method.

[0009] Some practitioners have proposed a data-driven method based on dynamic mode decomposition to achieve frequency control during the primary frequency regulation process of wind farms. However, such solutions usually track the power commands issued by the controller to achieve active power control of each wind turbine. The droop control coefficient is set with reference to the external power frequency characteristic curve of the wind farm, which is a fixed value. However, in the actual operation of a wind farm, the operating status of the wind turbines changes with the wind speed, and the frequency regulation capability that the wind farm can provide changes in real time. A fixed droop coefficient is obviously difficult to fully utilize the overall frequency regulation capability of the wind farm. [Summary of the Invention]

[0010] The technical problem to be solved by this invention is: in view of the technical problems existing in the prior art, this invention provides a wind farm frequency regulation capability evaluation method, control method and device based on data-driven linear MPC that is simple to implement, has high evaluation efficiency and accuracy, and good scalability.

[0011] To solve the above-mentioned technical problems, the technical solution proposed by this invention is as follows:

[0012] A method for evaluating the frequency regulation capability of wind farms based on data-driven linear MPC, comprising the following steps:

[0013] Offline acquisition of historical wind turbine operation datasets, and training of linear predictive control models using historical wind turbine operation datasets to obtain trained linear predictive control models. The linear predictive control models are obtained by transforming the dynamic nonlinear model of wind farm frequency regulation based on the Koopman operator theory through a dimensionality-up transformation process.

[0014] When it is necessary to evaluate the frequency regulation capability of a wind farm, real-time operating data of the wind turbines is acquired, and the acquired real-time operating data is used to solve the linear MPC optimization model to obtain the droop coefficient of the wind farm. The linear MPC optimization model is constructed based on the trained linear predictive control model with the objective of maximizing the droop coefficient of the wind farm and setting safe speed constraints. The frequency regulation capability of the wind farm is evaluated based on the obtained droop coefficients of all wind turbines in the wind farm.

[0015] In some embodiments of the present invention, the wind farm frequency regulation dynamic nonlinear model is to represent the wind farm frequency regulation dynamic process as the current wind turbine speed, system frequency, and control variables are transformed by a nonlinear function to determine the next wind turbine speed and power system frequency. The control variables include the wind turbine droop coefficient and wind turbine wind speed.

[0016] In some embodiments of the present invention, the expression of the linear predictive control model is: zk+1 =A·z k +B·u k K = [AB]

[0017] Where K represents the finite-dimensional approximation matrix of the infinite-dimensional Koopman operator, and A represents the corresponding N. lift ×N lift Dimension coefficient matrix, B represents u k The corresponding N lift The coefficient matrix has dimensions ×h, where h represents the control vector u. k The dimension of z k U represents the augmented observation state vector. k =[K f,i,k ,v w,i,k ] T K represents the control vector. f,i,k K represents the sag coefficient of the k-th wind turbine at step k. i,k K f,i The constructed control vector, v w,i,k This represents the control vector formed by the wind speed of the i-th wind turbine at step k.

[0018] In some embodiments of the present invention, during the process of training the linear predictive control model using the historical operating dataset of the wind turbine, the least squares method is used to solve for the coefficient matrices A and B in the linear predictive control model.

[0019] In some embodiments of the present invention, the coefficient matrix obtained by solving the coefficient matrices A and B in the linear predictive control model using the least squares method is as follows:

[0020] in, This represents the pseudo-inverse operation of a matrix, X = [z k,1 ,…,z k,L ] represents the input variable dataset, Y = [z k,+11 ,…,z k+1,L ] represents the output variable dataset, U = [u k,1 ,…,u k,L [This is the dataset for control variables.]

[0021] In some embodiments of the present invention, the linear MPC optimization model is expressed as:

[0022] Where K represents the finite-dimensional approximation matrix of the infinite-dimensional Koopman operator, and A represents the corresponding N. lift ×N lift Dimension coefficient matrix, B represents u k The corresponding N lift The coefficient matrix has dimensions ×h, where h represents the control vector u. kThe dimension of z k N represents lift The augmented observation state vector of dimension z0 represents z k The initial value, u k =[K f,i,k ,v w,i,k ] T K represents the control vector. f,i,k K represents the sag coefficient of the i-th wind turbine at step k. f,i The constructed control vector, v w,i,k ω represents the control vector formed by the wind speed of the i-th wind turbine at step k. max ω min These represent the upper and lower limits of the safe operating speed of the fan.

[0023] In some embodiments of the present invention, the method further includes acquiring real-time wind speed data of the wind turbine, real-time rotational speed data of the wind turbine, and real-time frequency data of the power system as initialization data to determine an initial value z0: z0=[ω0,f0,ξ(ω0,f0)]

[0024] Where ω0 is the initial data of the wind turbine speed, f0 is the initial data of the power system, and ξ(ω0,f0) is the nonlinear increasing-dimensional function ξ(ω k ,f k The initial value of ξ(ω) k ,f k ) represents the nonlinear dimensionality-increasing function of the initial state at step k.

[0025] In some embodiments of the present invention, the nonlinear increasing-dimensional function ξ(ω) k ,f k The calculation expression for ) is:

[0026] Where, ξ N (ω k ,f k ) represents ξ(ω) k ,f k The Nth component of ) Represents the Euclidean norm, r k Represents Euclidean distance, x k Let represent the initial state variable at step k, and c represent the basis vector.

[0027] In some embodiments of the present invention, the step of evaluating the frequency regulation capability of a wind farm based on the obtained droop coefficients of all wind turbines in the wind farm includes: summing the droop coefficients of each wind turbine to obtain the overall droop coefficient of the wind farm, and evaluating the frequency regulation capability of the wind farm based on the obtained overall droop coefficient of the wind farm.

[0028] A wind farm frequency regulation control method based on data-driven linear model predictive control, comprising the following steps:

[0029] The frequency regulation capability of the wind farm is evaluated according to the above-mentioned wind farm frequency regulation capability evaluation method, and the droop coefficient of each wind turbine is obtained.

[0030] The droop coefficient of each wind turbine is used as a power adjustment parameter to generate control commands for each wind turbine.

[0031] Control commands for each wind turbine are sent to each wind turbine to control the primary frequency regulation of each wind turbine.

[0032] A wind farm frequency regulation capability assessment device based on data-driven linear MPC, comprising:

[0033] The offline model training module is used to acquire historical operation datasets offline and use the historical operation datasets of wind turbines to train the linear predictive control model, thereby obtaining the trained linear predictive control model. The linear predictive control model is obtained by transforming the dynamic nonlinear model of wind farm frequency regulation through a dimensionality-up transformation process based on the Koopman operator theory.

[0034] The online evaluation module is used to acquire real-time operating data of wind turbines when it is necessary to evaluate the frequency regulation capability of a wind farm. The acquired real-time operating data is used to solve the linear MPC optimization model to obtain the droop coefficient of the wind farm. The linear MPC optimization model is constructed based on the trained linear predictive control model with the objective of maximizing the droop coefficient of the wind farm and setting safe speed constraints. The frequency regulation capability of the wind farm is evaluated based on the obtained droop coefficients of all wind turbines in the wind farm.

[0035] A wind farm frequency regulation capability assessment device based on data-driven linear MPC includes a processor and a memory, wherein the memory is used to store a computer program and the processor is used to execute the computer program to perform the method described above.

[0036] A wind farm frequency regulation control device based on data-driven linear model predictive control includes:

[0037] Such as the wind farm frequency regulation capability assessment device mentioned above;

[0038] The control command generation module is used to generate control commands for each wind turbine by using the droop coefficient of each wind turbine as a power adjustment parameter.

[0039] The instruction sending module is used to send control instructions for each wind turbine to each wind turbine to control the primary frequency regulation of each wind turbine.

[0040] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0041] Compared with the prior art, the advantages of the present invention are as follows:

[0042] 1. This invention transforms the dynamic nonlinear MPC model of a wind farm into a linear MPC model based on the Koopman operator theory. It uses a data-driven approach to train and construct a linear MPC model of the wind farm, enabling the solution of the nonlinear MPC model without relying on the accuracy of the model parameters. This allows for a fast and accurate online evaluation of the frequency regulation capability of the wind farm, solving the problem that traditional models are easily affected by parameter accuracy and model completeness, and also has good scalability.

[0043] 2. This invention achieves online evaluation of the wind farm's frequency regulation capability by constructing a wind power linear MPC model through data-driven training before the wind farm's frequency regulation control. It obtains the droop coefficient of each wind turbine and uses the droop coefficient of each wind turbine as a power regulation parameter to generate corresponding control commands for each wind turbine, which are then sent to each wind turbine to control the primary frequency regulation process of each wind turbine. This fully utilizes the wind farm's real-time frequency regulation capability to dynamically control the frequency regulation process of the wind turbines, thereby maximizing the wind farm's frequency regulation capability. [Attached Image Description]

[0044] Figure 1 is a schematic diagram of the implementation process of the wind farm frequency regulation capability evaluation method based on data-driven linear MPC in Embodiment 1 of the present invention.

[0045] Figure 2 is a detailed flowchart illustrating the method for evaluating the frequency regulation capability of a wind farm in Embodiment 1 of the present invention.

[0046] Figure 3 is a schematic diagram illustrating the principle of wind farm frequency regulation control in Embodiment 2 of the present invention.

Detailed Implementation Methods

[0047] The present invention will be further described below with reference to the accompanying drawings and specific preferred embodiments, but this does not limit the scope of protection of the present invention.

[0048] Example 1:

[0049] As shown in Figures 1 and 2, the steps of the wind farm frequency regulation capability assessment method based on data-driven linear MPC in this embodiment include:

[0050] Step 1: Offline Model Training: Obtain the historical operation dataset of wind turbines offline, and use the historical operation dataset of wind turbines to train the linear predictive control model to obtain the trained linear predictive control model. The linear predictive control model is obtained by transforming the dynamic nonlinear model of wind farm frequency regulation based on the Koopman operator theory through a dimensionality-up transformation process.

[0051] In this embodiment, the historical operating data of the wind turbines includes data such as rotational speed, wind speed, system frequency, and turbine droop coefficient obtained during the primary frequency regulation process at different wind speeds. The Supervisory Control and Data Acquisition (SCADA) system obtains the rotational speed, wind speed, system frequency, and turbine droop coefficient data of all wind turbines in the wind farm during primary frequency regulation, forming a historical operating dataset stored on the grid dispatching server or other designated storage space. After obtaining sufficient data from the grid dispatching server's storage, data-driven training is performed.

[0052] Understandably, in practice, other methods can be used to obtain historical operating data of each wind turbine in a wind farm, depending on actual needs.

[0053] In this embodiment, the wind farm frequency regulation dynamic nonlinear model represents the wind farm frequency regulation dynamic process as a transformation of the current wind turbine speed, system frequency, and control variables using a nonlinear function to determine the next wind turbine speed and power system frequency. The control variables include the turbine droop coefficient and the turbine wind speed. In other words, the wind farm frequency regulation dynamic process can be approximately represented as a nonlinear model where the next wind turbine speed and system frequency are determined by the current wind turbine speed, system frequency, and the control variables turbine droop coefficient and turbine wind speed.

[0054] In a specific application embodiment, the dynamic nonlinear model of wind farm frequency regulation can be expressed as: [ω k+1 ,f k+1 ]=δ(ω k ,f k ,u k (1)u k =[K f,i,k ,v w,i,k ] T (2)

[0055] Where δ represents a nonlinear function, k represents the discrete control point corresponding to the k-th step, and ω k f represents the vector formed by the fan speed at step k. k u represents the actual frequency of the power system at step k. k ∈R h The control vector consists of the turbine droop coefficient K and the wind speed v. f,i,k K represents the sag coefficient of the i-th wind turbine at step k. f,i The constructed control vector, v w,i,k v represents the wind speed of the i-th wind turbine at step k. w,i The control vector formed.

[0056] Koopman operator theory can project a nonlinear dynamic model from a low-dimensional nonlinear equation system to a high-dimensional linear equation system with high computational efficiency, meeting the needs of online applications. To construct a linear predictive control model, this embodiment first defines a state-space mapping function with enhanced observation states based on Koopman operator theory, as shown below: φ(ω k ,f k ,u k )=[ω k ,f k ,ξ(ω k ,f k ),u k ] T =[z k ,u k ] T (3)

[0057] Where, ξ(ω) k ,f k ) represents the nonlinear dimensionality-increasing function of the initial state at step k. Represents N lift Dimensionally enhanced observation state vector.

[0058] ξ(ω k ,f k This can be further expressed as:

[0059] Where, ξ N (ω k ,f k ) represents ξ(ω) k ,f k The Nth component of ) is a scalar function.

[0060] In a specific application embodiment, the dimension-upgrading function adopts the form of a thin-plate spline, as shown below:

[0061] Where, ξ N (ω k ,f k ) represents ξ(ω) k ,f k The Nth component of ) is a scalar function.

[0062] In a specific application embodiment, the dimension-upgrading function adopts the form of a thin-plate spline, as shown below:

[0063] in, Represents the Euclidean norm, r k Represents Euclidean distance; x kThis represents the initial state variable at step k, which can be specifically represented by [ω]. k ,f k ] T Let c represent the basis vector, taken as [0, x]. k Random values ​​within the specified interval.

[0064] The wind farm frequency regulation dynamic nonlinear model is further transformed into a linear predictive control model through a dimension-up transformation process:

[0065] First, the nonlinear model shown in equation (1) above is transformed into a linear model as shown below by incomplete dimension increase based on the Koopman operator theory: [ω k+1 ,f k+1 ,ξ(ω k+1 f k+1 )]=Kφ(ω k ,f k u k (6)

[0066] Where K represents the finite-dimensional approximation matrix of the infinite-dimensional Koopman operator, and can be further divided as: K = [AB] (7)

[0067] Where A represents the corresponding N lift ×N lift Dimension coefficient matrix, B represents u k The corresponding N lift The coefficient matrix has dimensions ×h, where h represents the control vector u. k The dimension of.

[0068] Based on equations (3) and (7) above, formula (6) can be transformed into the following linear predictor form: z k+1 =A·z k +B·u k (8)

[0069] This leads to the construction of the required linear predictive control model. It is understandable that the linear predictive control model can also be expressed in other forms based on Koopman operator theory, and other forms of nonlinear up-dimensional functions ξ(ω) can be selected according to actual needs. k ,f k ).

[0070] The key to training a linear predictive control model using historical wind turbine operation datasets is solving for the coefficient matrices A and B. This embodiment transforms the dynamic nonlinear MPC model of a wind farm into a linear MPC model based on Koopman operator theory, thus solving the problem of the difficulty in solving nonlinear MPC models. To achieve rapid solution of the coefficient matrices, this embodiment specifically uses the least squares method to solve for coefficient matrices A and B. The specific steps are as follows:

[0071] The input variable dataset, output variable dataset, and control variable dataset are represented as follows:

[0072] Using the historical dataset as the training set, the solution of coefficient matrices A and B is transformed into a least squares problem, namely:

[0073] The analytical solution to the above equation is as follows:

[0074] in, Representing the pseudo-inverse operation of matrices, matrices G and M can be further specifically represented as:

[0075] That is, the coefficient matrices A and B can be solved according to the above equations (11) and (12) to obtain the trained linear predictive control model.

[0076] This embodiment, through the above steps, can transform the dynamic nonlinear MPC model of a wind farm into a linear MPC based on the Koopman operator theory, and then construct a convex optimization problem. By using a data-driven approach to train and construct a linear model, it is possible to solve the nonlinear MPC model without relying on the accuracy of the model parameters. This allows for a fast and accurate online evaluation of the frequency regulation capability of the wind farm, solving the problem that traditional models are easily affected by parameter accuracy and model completeness, and also has good scalability.

[0077] Step 2: Online Evaluation: When it is necessary to evaluate the frequency regulation capability of a wind farm, real-time operating data of the wind turbines is acquired, and the acquired real-time operating data is used to solve the linear MPC optimization model to obtain the droop coefficient of the wind farm. The linear MPC optimization model is constructed based on the trained linear predictive control model with the objective of maximizing the droop coefficient of the wind farm and setting safe speed constraints. The frequency regulation capability of the wind farm is evaluated based on the obtained droop coefficients of all wind turbines in the wind farm.

[0078] Based on the coefficient matrices A and B obtained in step 1, the trained linear predictive control model is determined. Based on this model, the linear MPC optimization model can be constructed by setting the maximum droop coefficient of the wind farm as the objective and setting the safe speed constraint (i.e., the safe speed constraint).

[0079] In a specific application embodiment, the constructed linear MPC optimization model can be expressed as:

[0080] Where z0 represents z k The initial value, u k =[K f,i,k ,v w,i,k ] T Represents the control vector K f,i,k K represents the sag coefficient of the i-th wind turbine at step k. f,i The constructed control vector, v w,i,k ω represents the control vector formed by the wind speed of the i-th wind turbine at step k. max ω min These represent the upper and lower limits of the safe operating speed of the wind turbine, respectively; l represents the number of wind turbines in the wind farm; ω i,k This represents the rotational speed of the i-th fan at step k.

[0081] Equation (13) above is constructed by adding a safe speed constraint and an objective function (aiming to maximize the droop coefficient of the wind farm) to the linear predictive control model shown in Equation (8). The droop coefficient K of each wind turbine is then solved according to Equation (13). f,i Then, the sag coefficient K of each wind turbine was calculated. f,i Summing these values ​​will yield the overall droop coefficient K of the wind farm. f Understandably, it could also be based on the droop coefficient K of each wind turbine. f,i The overall droop coefficient K of the wind farm was calculated using other methods. f For example, first calculate the sag coefficient K of each wind turbine. f,i The statistical values ​​are then used to determine the overall droop coefficient K of the wind farm. f .

[0082] Understandably, the safe rotational speed constraint can also take other forms depending on actual needs, and the objective function aimed at maximizing the droop coefficient of the wind farm can also be optimized and adjusted accordingly. Specifically, it can be configured based on actual requirements. This is based on the overall droop coefficient K of the wind farm. f The frequency regulation capability of a wind farm can be evaluated online. If the overall droop coefficient K f A higher overall sag coefficient K results in a more sensitive wind turbine response to frequency deviations, leading to stronger frequency regulation capabilities for the wind farm. Conversely, a lower overall sag coefficient K... fIf the frequency regulation value is low, the wind farm's frequency regulation capability is weak, thus allowing for real-time and rapid online assessment of the wind farm's frequency regulation capability.

[0083] In this embodiment, the initial value z0 can be determined as follows:

[0084] The real-time wind speed data, real-time rotational speed data of the wind turbine, and real-time frequency data of the power system are obtained as initialization data to determine the initial value z0: z0=[ω0,f0,ξ(ω0,f0)] (14)

[0085] Where ω0 is the initial data of the wind turbine speed, f0 is the initial data of the power system, and ξ(ω0,f0) is the nonlinear increasing-dimensional function ξ(ω k ,f k The initial value of ξ(ω) k ,f k ) represents the nonlinear dimensionality-increasing function of the initial state at step k.

[0086] The wind speed data of the aforementioned wind turbines can be obtained from the wind measurement tower device, the wind turbine speed data can be obtained from the wind turbine sensor, and the power system frequency data can be obtained from the phase-locked loop. The specific acquisition method can also be selected according to actual needs.

[0087] As shown in Figure 2, in a specific application embodiment, the present invention implements the wind farm frequency regulation capability in two stages using the above method. The detailed steps are as follows:

[0088] Phase 1: Offline Training Process

[0089] (1) Data on the rotational speed, wind speed, system frequency, and turbine droop coefficient of all wind turbines in the wind farm during primary frequency regulation are obtained through the data acquisition and monitoring control system and stored on the grid dispatch side server;

[0090] (2) Determine whether the data stored on the power grid dispatch side is sufficient. If it is sufficient, proceed to the next step to train the model.

[0091] (3) Determine the nonlinear dimension-upgrading function ξ(ω) k ,f k The formation and construction of z k ;

[0092] (4) Construct a high-dimensional linear Koopman model, which is a linear predictive control model formed by transforming the wind farm frequency regulation dynamic nonlinear model through a dimensionality-up transformation process, as shown in equation (8).

[0093] (5) Estimate the coefficient matrices A and B using the least squares method;

[0094] Phase Two: Online Assessment Process

[0095] (1) Real-time assessment of whether wind farms require evaluation;

[0096] (2) If there is an assessment requirement, first obtain the real-time wind speed data of the wind turbine, the real-time rotational speed data of the wind turbine, and the real-time frequency data of the power system as initialization data to determine the initial value z0.

[0097] (3) Construct a linear MPC optimization model and solve it, as shown in equation (13), to obtain the droop coefficient of each wind turbine:

[0098] (4) The droop coefficient of each wind turbine is summed to obtain the overall droop coefficient of the wind farm, and then sent to the wind turbine to control the primary frequency regulation process, thereby realizing the online evaluation and allocation of the frequency regulation capability of the wind farm.

[0099] This embodiment also provides a wind farm frequency regulation capability evaluation device based on data-driven linear MPC, including:

[0100] The offline model solving module is used to acquire historical operation datasets offline and use the historical operation datasets of wind turbines to solve the coefficient matrix in the linear predictive control model to obtain the solved linear predictive control model. The linear predictive control model is obtained by transforming the dynamic nonlinear model of wind farm frequency regulation based on the Koopman operator theory through a dimensionality-up transformation process.

[0101] The online evaluation module is used to acquire real-time operating data of wind turbines when it is necessary to evaluate the frequency regulation capability of a wind farm. The acquired real-time operating data is used to solve the linear MPC optimization model to obtain the droop coefficient of the wind farm. The linear MPC optimization model is constructed based on the solved linear predictive control model with the objective of maximizing the droop coefficient of the wind farm and setting safe speed constraints. The frequency regulation capability of the wind farm is evaluated based on the obtained droop coefficients of all wind turbines in the wind farm.

[0102] The wind farm frequency regulation capability assessment device based on data-driven linear MPC in this embodiment corresponds one-to-one with the wind farm frequency regulation capability assessment method based on data-driven linear MPC described above, and will not be described in detail here.

[0103] Example 2:

[0104] The steps of the wind farm frequency regulation control method based on data-driven linear model predictive control in this embodiment include:

[0105] The frequency regulation capability of the wind farm is evaluated according to the wind farm frequency regulation capability evaluation method in Example 1, and the droop coefficient of each wind turbine is obtained.

[0106] The droop coefficient of each wind turbine is used as a power adjustment parameter to generate control commands for each wind turbine.

[0107] Control commands for each wind turbine are sent to each wind turbine to control the primary frequency regulation of each wind turbine.

[0108] In this embodiment, before frequency regulation control of the wind farm, the frequency regulation capability of the wind farm is evaluated online to obtain the droop coefficient of each wind turbine. Then, the droop coefficient of each wind turbine is used as the power regulation parameter to generate control commands for each wind turbine and send them to each wind turbine to control the primary frequency regulation process of each wind turbine. This can make full use of the real-time frequency regulation capability of the wind farm to dynamically control the frequency regulation process of the wind turbine, thereby giving full play to the frequency regulation capability of the wind farm.

[0109] As shown in Figure 3, to achieve frequency regulation control of the wind farm, during the offline training phase, the data acquisition and monitoring control system obtains the rotational speed, wind speed, system frequency, and turbine droop coefficient data of all wind turbines in the wind farm during primary frequency regulation. This data is stored in the grid dispatching server. After constructing a high-dimensional linear Koopman model, the model is trained using historical operation datasets to solve for the coefficient matrices A and B. On the wind farm side, it is determined in real time whether the wind farm has an evaluation requirement. If an evaluation requirement is found, the real-time wind speed data, real-time rotational speed data, and real-time frequency data of the power system are obtained as initialization data to determine the initial value z0. A linear MPC optimization model is constructed and solved to obtain the droop coefficient of each wind turbine, realizing online evaluation of the wind farm's frequency regulation capability. The sum of the solved droop coefficients of each wind turbine is used to obtain the overall droop coefficient of the wind farm, which is then sent to the wind turbines to control the primary frequency regulation process. On the wind turbine side, online frequency regulation control of the wind farm is achieved based on the received control parameters.

[0110] This embodiment also provides a wind farm frequency regulation control device based on data-driven linear model predictive control, including:

[0111] Such as the wind farm frequency regulation capability assessment device mentioned above;

[0112] The control command generation module is used to generate control commands for each wind turbine by using the droop coefficient of each wind turbine as a power adjustment parameter.

[0113] The instruction sending module is used to send control instructions for each wind turbine to each wind turbine to control the primary frequency regulation of each wind turbine.

[0114] The wind farm frequency regulation control device of data-driven linear model predictive control in this embodiment corresponds one-to-one with the wind farm frequency regulation control method of data-driven linear model predictive control described above, and will not be described in detail here.

[0115] In another embodiment, the wind farm frequency regulation capability assessment device based on data-driven linear MPC of the present invention may further include a processor and a memory, wherein the memory is used to store computer programs and the processor is used to execute the computer programs to perform the methods as described in Embodiments 1 and 2.

[0116] It is understood that the method described in this embodiment can be executed by a single device, such as a computer or server, or it can be applied to a distributed scenario where multiple devices cooperate to complete the task. In a distributed scenario, one of the multiple devices may execute only one or more steps of the method described in this embodiment, and the multiple devices interact to complete the method. The processor can be implemented using a general-purpose CPU, microprocessor, application-specific integrated circuit, or one or more integrated circuits, and is used to execute relevant programs to implement the method described in this embodiment. The memory can be implemented using read-only memory (ROM), random access memory (RAM), static storage devices, and dynamic storage devices. The memory can store the operating system and other applications. When the method described in this embodiment is implemented through software or firmware, the relevant program code is stored in the memory and called and executed by the processor.

[0117] This embodiment further provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0118] Those skilled in the art will understand that the above embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0119] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should fall within the protection scope of the present invention.

Claims

1. A method for evaluating the frequency regulation capability of wind farms based on data-driven linear MPC, characterized by the following steps: include: Offline acquisition of historical wind turbine operation datasets, and training of linear predictive control models using historical wind turbine operation datasets to obtain trained linear predictive control models. The linear predictive control models are obtained by transforming the dynamic nonlinear model of wind farm frequency regulation based on the Koopman operator theory through a dimensionality-up transformation process. When it is necessary to evaluate the frequency regulation capability of a wind farm, real-time operating data of the wind turbines is acquired, and the acquired real-time operating data is used to solve the linear MPC optimization model to obtain the droop coefficient of the wind farm. The linear MPC optimization model is constructed based on the trained linear predictive control model with the objective of maximizing the droop coefficient of the wind farm and setting safe speed constraints. The frequency regulation capability of the wind farm is evaluated based on the obtained droop coefficients of all wind turbines in the wind farm.

2. The wind farm frequency regulation capability evaluation method based on data-driven linear MPC according to claim 1, characterized in that, The wind farm frequency regulation dynamic nonlinear model represents the wind farm frequency regulation dynamic process as the current wind turbine speed, system frequency, and control variables are transformed by a nonlinear function to determine the next wind turbine speed and power system frequency. The control variables include the wind turbine droop coefficient and wind turbine speed.

3. The wind farm frequency regulation capability evaluation method based on data-driven linear MPC according to claim 1, characterized in that, The expression of the linear predictive control model is: z k+1 = A · z k + B · u k K = [A B] Wherein, K represents the finite-dimensional approximation matrix of infinite-dimensional Koopman operator, A represents the corresponding N lift ×N lift dimension coefficient matrix, B represents the u k corresponding N lift ×h dimension coefficient matrix, h represents the dimension of the control vector u k , z k represents the N lift dimensional enhanced observation state vector, u k =[K f,i,k ,v w,i,k ] T represents the control vector, K f,i,k represents the kth step of the first wind turbine droop coefficient K i,k K f,i control vector, v w,i,k represents the kth step of the first i wind turbine speed control vector.

4. The wind farm frequency regulation capability evaluation method based on data-driven linear MPC according to claim 3, characterized in that, In the process of training the linear predictive control model using the historical operation dataset of wind turbines, the least squares method is used to solve for the coefficient matrices A and B in the linear predictive control model.

5. The wind farm frequency regulation capability evaluation method based on data-driven linear MPC according to claim 4, characterized in that, The coefficient matrix obtained by solving for the coefficient matrices A and B in the linear predictive control model using the least squares method is: in, This represents the pseudo-inverse operation of a matrix, X = [z k,1 ,…,z k,L ] represents the input variable dataset, Y = [z k,+11 ,…,z k+1,L ] represents the output variable dataset, U = [u k,1 ,…,u k,L [This is the dataset for control variables.] 6. The wind farm frequency regulation capability evaluation method based on data-driven linear MPC according to any one of claims 1 to 5, characterized in that, The linear MPC optimization model is expressed as follows: stz k+1 =A·z k +B·u k z0 = [x0, ξ(x0)] T ω min ≤ω i,k ≤ω max , i∈l Where K represents the finite-dimensional approximation matrix of the infinite-dimensional Koopman operator, and A represents the corresponding N. lift ×N lift Dimension coefficient matrix, B represents u k The corresponding N lift The coefficient matrix has dimensions ×h, where h represents the control vector u. k The dimension of z k N represents lift The augmented observation state vector of dimension z0 represents z k The initial value, u k =[K f,i,k ,v w,i,k ] T K represents the control vector. f,i,k K represents the sag coefficient of the i-th wind turbine at step k. f,i The constructed control vector, v w,i,k ω represents the control vector formed by the wind speed of the i-th wind turbine at step k. max ω min These represent the upper and lower limits of the safe operating speed of the fan.

7. The wind farm frequency regulation capability evaluation method based on data-driven linear MPC according to claim 6, characterized in that, It also includes acquiring real-time wind speed data, real-time rotational speed data of the wind turbine, and real-time frequency data of the power system as initialization data to determine the initial value z0: z0=[ω0,f0,ξ(ω0,f0)] Where ω0 is the initial data of the wind turbine speed, f0 is the initial data of the power system, and ξ(ω0,f0) is the nonlinear increasing-dimensional function ξ(ω k ,f k The initial value of ξ(ω) k ,f k ) represents the nonlinear dimensionality-increasing function of the initial state at step k.

8. The wind farm frequency regulation capability evaluation method based on data-driven linear MPC according to claim 7, characterized in that, Nonlinear dimensionality-increasing function ξ(ω) k ,f k The calculation expression for ) is: Where, ξ N (ω k ,f k ) represents ξ(ω) k ,f k The Nth component of ) Represents the Euclidean norm, r k Represents Euclidean distance, x k Let represent the initial state variable at step k, and c represent the basis vector.

9. The wind farm frequency regulation capability evaluation method based on data-driven linear MPC according to any one of claims 1 to 5, characterized in that, The step of evaluating the frequency regulation capability of a wind farm based on the obtained droop coefficients of all wind turbines includes: summing the droop coefficients of each wind turbine to obtain the overall droop coefficient of the wind farm, and evaluating the frequency regulation capability of the wind farm based on the obtained overall droop coefficient of the wind farm.

10. A wind farm frequency regulation control method based on data-driven linear model predictive control, characterized in that the steps are as follows: include: The wind farm frequency regulation capability is evaluated according to the wind farm frequency regulation capability evaluation method according to any one of claims 1 to 9, and the droop coefficient of each wind turbine is obtained. The droop coefficient of each wind turbine is used as a power adjustment parameter to generate control commands for each wind turbine. Control commands for each wind turbine are sent to each wind turbine to control the primary frequency regulation of each wind turbine.

11. A wind farm frequency regulation capability evaluation device based on data-driven linear MPC, characterized in that, include: The offline model training module is used to acquire historical operation datasets offline and use the historical operation datasets of wind turbines to train the linear predictive control model, thereby obtaining the trained linear predictive control model. The linear predictive control model is obtained by transforming the dynamic nonlinear model of wind farm frequency regulation through a dimensionality-up transformation process based on the Koopman operator theory. The online evaluation module is used to acquire real-time operating data of wind turbines when it is necessary to evaluate the frequency regulation capability of a wind farm. The acquired real-time operating data is used to solve the linear MPC optimization model to obtain the droop coefficient of the wind farm. The linear MPC optimization model is constructed based on the trained linear predictive control model with the objective of maximizing the droop coefficient of the wind farm and setting safe speed constraints. The frequency regulation capability of the wind farm is evaluated based on the obtained droop coefficients of all wind turbines in the wind farm.

12. A wind farm frequency regulation capability evaluation device based on data-driven linear MPC, comprising a processor and a memory, wherein the memory is used to store computer programs, characterized in that, The processor is used to execute the computer program to perform the method as described in any one of claims 1 to 9.

13. A wind farm frequency regulation control device based on data-driven linear model predictive control, characterized in that, include: The wind farm frequency regulation capability assessment device as described in claim 11 or 12; The control command generation module is used to generate control commands for each wind turbine by using the droop coefficient of each wind turbine as a power adjustment parameter. The instruction sending module is used to send control instructions for each wind turbine to each wind turbine to control the primary frequency regulation of each wind turbine.

14. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method as described in any one of claims 1 to 10.

Images