A wind farm unit hybrid control method and device suitable for wake effect

Abstract

The application discloses a wind farm unit hybrid control method and device suitable for wake effect, comprehensively considers the influence of the wake effect in the wind farm, constructs a data-model hybrid control method, combines the advantages of model driving and data driving, optimizes control of the pitch angle of a wind turbine unit group and the torque of a generator, realizes reduction of the fatigue load of the unit group under the wake effect of the wind farm, and provides technical support for optimized operation of the wind farm.

Description

Technical Field

[0001] This invention relates to the field of wind power technology, and specifically to a hybrid control method and device for wind farm units that utilizes wake effects. Background Technology

[0002] As a crucial support for the global green energy transition strategy, wind power has become one of the world's fastest-growing energy sources, especially in the field of newly installed renewable energy. To improve land utilization and maximize power generation by installing as many wind turbines as possible on limited land, the optimal distance between turbines must be carefully considered. The wake effect is a critical factor determining this distance. With advancements in the refined management of wind turbines, the wake effect has attracted increasing attention from researchers due to its potential to increase turbulence intensity and energy loss within wind farms. Turbulence can significantly increase fatigue loads on wind turbines, leading to a higher probability of damage or failure of critical components, resulting in unplanned downtime and maintenance. This not only increases maintenance costs but also ultimately reduces the turbine's lifespan. The energy loss caused by the wake effect has a significant impact on the economics of wind farms. Therefore, mitigating the negative impacts of the wake effect within wind farms and reducing turbine loads has become a critical issue that urgently needs to be addressed in the optimized design of wind farm wake effects. Summary of the Invention

[0003] To address the shortcomings and problems of existing wind farm optimization wake effect models, such as poor interpretability, lack of highly accurate physical models and data quality, and difficulty in handling measurement noise, the present invention aims to provide a hybrid control method and device for wind farm units applicable to wake effects.

[0004] To solve the above-mentioned technical problems, the technical solution provided by the present invention is as follows:

[0005] A hybrid control method for wind farm turbines applicable to wake effects, the method comprising:

[0006] S1. Establish a wind farm wake optimization control model based on model predictive control, where the inputs are wind speed, engine speed, pitch angle, and torque from the collected data, and the output is the pitch angle of the turbine group. and generator torque ;

[0007] Step S1 includes establishing a wake effect calculation model;

[0008] Step S1 includes obtaining an incremental expression with unit speed, pitch angle, and torque as independent variables;

[0009] Step S1 includes combining the wake model to... Ft , T s For fatigue load targets, F t For thrust; T s For shaft torque; establish the controller objective function;

[0010] S2, Establish a data-driven strategy, and obtain wind turbine group data and turbine pitch angle through encoders and graph neural network decoders. and generator torque The complex mapping relationship; among which, the wind turbine group data includes wind speed, wind direction, speed, and pitch angle;

[0011] S3, based on neural networks, mixes the output results of data-driven strategies and model predictive control-based wind farm wake optimization control models to obtain the optimal output;

[0012] Step S3 includes dynamically adjusting weights based on a neural network. The weight parameters are based on the input and labels, and the weights are adjusted according to the input and labels through supervised learning.

[0013] Optionally, the wake effect calculation model established in step S1 is as follows:

[0014]

[0015] in, i, j Represents a positive integer variable; This represents the total number of units within the wake region. λ Indicates the tip speed ratio. β Indicates the pitch angle; C t.j It is the first j Upstream thrust coefficient of each wind turbine unit; v ∞ Indicates the initial wind speed; v i [ t 0] and v ∞ [ t 0] are respectively the first i The single equivalent wind speed of each wind turbine unit and t Initial inflow velocity at time 0; v j [ t 0- ]and C t.j [ t 0- [These are the first] times. j Each wind turbine unit t 0- Wind speed and thrust coefficient at any given time, Indicates the first i Taiwan and the j The distance between typhoon turbine units.

[0016] Optionally, the incremental expression obtained in step S1 with unit speed, pitch angle, and torque as independent variables is:

[0017]

[0018]

[0019]

[0020]

[0021] in Indicates increment, It is the gearbox speed ratio. It is aerodynamic torque; It is air density; It is the tip speed ratio; It is the generator speed; It is the speed of the filter generator; J t It is equivalent quality; It refers to the generator torque; It is the propeller pitch angle; This is a reference value for the generator torque. t 1, t 2 and t 2 indicates the time parameter;

[0022] and The calculation is as follows:

[0023]

[0024]

[0025] Among them, utilizing λ 0 and β 0 can be used to obtain the power coefficient at the operating point. C p0 ( λ 0, β 0), and The partial derivative is the sensitivity coefficient; R It is the blade radius. v 0 represents the wind speed measurement value. These are the measured values ​​of the wind energy capture factor and the generator speed; based on the above incremental expression, we can obtain:

[0026] ,

[0027] , ,

[0028] , , , , , ,

[0029] , ,

[0030] Subscript i Indicates which wind turbine unit. This represents the set of all increments of independent variables in a wind farm. This represents the set of all control variable increments for a wind farm. This represents the set of wind speed variables from multiple turbines in a wind farm, where k = 1, 2, 3, ..., n represents the number of prediction steps, and n is a positive integer. , , The wind farm coefficient matrix contains the coefficient matrix of multiple turbine units. ， ， For the first i The coefficient matrix of the unit. This is a measurement of mechanical torque. This is the measured value of the generator torque.

[0031] Optionally, in step S1, F t , T s The fatigue load target can be expressed by the following formula:

[0032]

[0033]

[0034] Wherein represents v Wind speed, F t It is thrust; T s It is shaft torque; P avi It is usable wind energy; , , , and It is the sensitivity coefficient;

[0035] The controller objective function can be expressed as:

[0036]

[0037]

[0038] in Ts This represents the equivalent low-speed shaft torque of the gearbox. N c It controls the number of prediction steps. Q 1， Q 2 is a constant weight coefficient matrix. and These are the minimum and maximum torque values ​​of the generator, respectively. and These represent the minimum and maximum pitch angles, respectively.

[0039] Optionally, the data input in step S2 is processed by the autoencoder as follows:

[0040]

[0041]

[0042] in, It consists of standardized historical wake-related data, including active power, wind speed, pitch angle, thrust, torque, and generator speed. n Indicates the dimension of the input data. s express x The length of , where x It can be represented as x = [ x 1, x 2, x 3, x 4, x [5] In the encoding stage of the autoencoder, the input data x is linearly transformed and an activation function (Rectified Linear Unit, ReLU) is applied to map it to the feature representation of the hidden layer. z middle, It is the input weight matrix. It is the input bias vector, and the feature representation of the hidden layer. ,in e This represents the number of neurons in the hidden layer; in the decoding stage of the autoencoder, the features of the hidden layer are also represented by linear transformation and the ReLU activation function. z Convert to reconstructed output ,in It is the weight matrix from the hidden layer to the output layer. The corresponding bias vector is the final output of the autoencoder. ,in d Indicates the output dimension.

[0043] Optionally, step S2 includes inputting the encoder-processed data into the decoder of the graph neural network, which outputs an optimized active power command for the generator group, as shown in the figure. Each node and Undirected edge, k This represents the number of nodes connected by each edge. The calculation process for undirected edges is as follows:

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052]

[0053] in This indicates the number of nodes, i.e., the total number of wind turbines. It is the degree matrix of nodes in the graph that contain self-connections. A Represents the original adjacency matrix. This shows the sum of adjacency matrices with self-connections. This represents an adjacency matrix with self-connections. Indicates the first i Taiwan unit and the first j Adjacency matrix of the kiosks; Indicates the first l Learnable parameters of the layer Indicates the first l +1 layer node feature matrix Represents the identity matrix. It is an activation function. Indicates the first l The node feature matrix of the layer; For nodes The aggregation characteristics, Represents a node The set of adjacent nodes, and They represent the first l Layer and first l Adjacent nodes in layer -1 The feature representation can effectively capture local graph structure information by calculating the average value of the features of neighboring nodes; Indicates the first l The layer's weight matrix is ​​used here to perform a linear transformation on the features of neighboring nodes. Indicates from the previous level to the [number]th level l The weight matrix of the node feature representation of the layer is used to preserve the features of the node itself. The nonlinearity is introduced by the ReLU function so that the model can capture the feature patterns between wind turbine units. Characteristics representing real nodes, Represents the node characteristics of the reconstruction; express and The probability that an edge exists between them. ) indicates nodes in different views The feature cosine similarity. Describes the edge reconstruction loss function. This represents the contrastive learning loss function, used to enhance the model's ability to discriminate the features of wind turbine nodes. and These represent the unit's wind speed characteristics and the enhanced characteristics after random perturbation, respectively.

[0054] Through the self-supervised learning tasks described above, the model can not only learn features from labeled data, but also extract correlation information from unlabeled tail data, thereby improving the robustness and generalization ability of the model.

[0055] Optionally, the process of adjusting the weight matrix in step S3 is represented as follows:

[0056]

[0057]

[0058] in, and The outputs of the model-driven and data-driven control methods are respectively the pitch angle. and generator torque The dataset; It is a connection and The input after; ; n is the data dimension; It is the weight matrix of the input layer; It is the bias of the input layer; It is the activation function of the input layer; It is the weight matrix of the output layer; Indicates the bias of the output layer; It is the activation function of the output layer. This represents the first hidden layer function. This indicates the final control command.

[0059] This invention comprehensively considers the impact of wake effects in wind farms and constructs a data-model hybrid control method, combining the advantages of model-driven and data-driven approaches to control the pitch angle of wind turbine clusters. and generator torque Optimized control was implemented to reduce the fatigue load of the wind farm unit group under the wake effect, providing technical support for the optimized operation of the wind farm. Attached Figure Description

[0060] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 A flowchart illustrating the structure of a convolutional neural network according to an embodiment of the present invention;

[0062] Figure 2 A flowchart illustrating the decoder module of a graph neural network according to an embodiment of the present invention;

[0063] Figure 3 A flowchart illustrating the data-model hybrid control method according to an embodiment of the present invention;

[0064] Figure 4 A structural diagram illustrating a wind farm device according to an embodiment of the present invention;

[0065] Figure 5 This diagram illustrates the active power and pitch angle output of the wind turbine generator set according to an embodiment of the present invention.

[0066] Figure 6 This diagram illustrates the thrust and shaft torque of the wind turbine generator set according to an embodiment of the present invention.

[0067] Figure 7 This diagram illustrates the active power resistance to measurement noise in an embodiment of the present invention. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0070] In various embodiments of the present invention, it should be understood that the sequence number of each process described below does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0071] In addition, the terms "system" and "network" are often used interchangeably in this embodiment of the invention.

[0072] In the embodiments provided in this application, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0073] Furthermore, in this embodiment of the invention, the term "and / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0074] In this embodiment of the invention, the term "multiple" refers to two or more, and other quantifiers are similar.

[0075] The hybrid control method for wind farm units applicable to wake effect of this invention includes:

[0076] S1. Establish a wind farm wake optimization control model based on model predictive control, where the inputs are wind speed, engine speed, pitch angle, and torque from the collected data, and the output is the pitch angle of the turbine group. and generator torque ;

[0077] Step S1 includes establishing a wake effect calculation model;

[0078] Step S1 includes obtaining an incremental expression with unit speed, pitch angle, and torque as independent variables;

[0079] Step S1 includes combining the wake model to... F t , T s For fatigue load targets, F t For thrust; T s For shaft torque; establish the controller objective function;

[0080] S2, Establish a data-driven strategy, and obtain wind turbine group data and turbine pitch angle through encoders and graph neural network decoders. and generator torque The complex mapping relationship; among which, the wind turbine group data includes wind speed, wind direction, speed, and pitch angle;

[0081] S3, based on neural networks, mixes the output results of data-driven strategies and model predictive control-based wind farm wake optimization control models to obtain the optimal output;

[0082] Step S3 includes dynamically adjusting weights based on a neural network. The weight parameters are based on the input and labels, and the weights are adjusted according to the input and labels through supervised learning.

[0083] The wake effect calculation model established in step S1 is as follows:

[0084]

[0085] in, i, j Represents a positive integer variable; This represents the total number of units within the wake region. λ Indicates the tip speed ratio. β Indicates the pitch angle; C t.j It is the first j Upstream thrust coefficient of each wind turbine unit; v ∞ Indicates the initial wind speed; v i [ t 0] and v ∞ [ t 0] are respectively the first i The single equivalent wind speed of each wind turbine unit and t Initial inflow velocity at time 0; v j [ t 0- ]andC t.j [ t 0- [These are the first] times. j Each wind turbine unit t 0- Wind speed and thrust coefficient at any given time, Indicates the first i Taiwan and the j The distance between typhoon turbine units.

[0086] The incremental expression obtained in step S1 with unit speed, pitch angle, and torque as independent variables is:

[0087]

[0088]

[0089]

[0090]

[0091] in Indicates increment, It is the gearbox speed ratio. It is aerodynamic torque; It is air density; It is the tip speed ratio; It is the generator speed; It is the speed of the filter generator; J t It is equivalent quality; It refers to the generator torque; It is the propeller pitch angle; This is a reference value for the generator torque. t 1, t 2 and t 2 indicates the time parameter;

[0092] and The calculation is as follows:

[0093]

[0094]

[0095] Among them, utilizing λ 0 and β 0 can be used to obtain the power coefficient at the operating point. C p0 ( λ 0, β 0), and The partial derivative is the sensitivity coefficient; R It is the blade radius. v 0 represents the wind speed measurement value. These are the measured values ​​of the wind energy capture factor and the generator speed; based on the above incremental expression, we can obtain:

[0096] ,

[0097] , ,

[0098] , , , , , ,

[0099] , ,

[0100] Subscript i Indicates which wind turbine unit. This represents the set of all increments of independent variables in a wind farm. This represents the set of all control variable increments for a wind farm. This represents the set of wind speed variables from multiple turbines in a wind farm, where k = 1, 2, 3, ..., n represents the number of prediction steps, and n is a positive integer. , , The wind farm coefficient matrix contains the coefficient matrix of multiple turbine units. ， ， For the first i The coefficient matrix of the unit. This is a measurement of mechanical torque. This is the measured value of the generator torque.

[0101] In step S1, F t , T s The fatigue load target can be expressed by the following formula:

[0102]

[0103]

[0104] Wherein represents v Wind speed, F t It is thrust; Ts It is shaft torque; P avi It is usable wind energy; , , , and It is the sensitivity coefficient;

[0105] The controller objective function can be expressed as:

[0106]

[0107]

[0108] in Ts This represents the equivalent low-speed shaft torque of the gearbox. N c It controls the number of prediction steps. Q 1， Q 2 is a constant weight coefficient matrix. and These are the minimum and maximum torque values ​​of the generator, respectively. and These represent the minimum and maximum pitch angles, respectively.

[0109] Optionally, the data input in step S2 is processed by the autoencoder as follows:

[0110]

[0111]

[0112] in, It consists of standardized historical wake-related data, including active power, wind speed, pitch angle, thrust, torque, and generator speed. n Indicates the dimension of the input data. s express x The length of , where x It can be represented as x = [ x 1, x 2, x 3, x 4, x [5] In the encoding stage of the autoencoder, the input data x is linearly transformed and an activation function (Rectified Linear Unit, ReLU) is applied to map it to the feature representation of the hidden layer. z middle, It is the input weight matrix. It is the input bias vector, and the feature representation of the hidden layer. ,in e This represents the number of neurons in the hidden layer; in the decoding stage of the autoencoder, the features of the hidden layer are also represented by linear transformation and the ReLU activation function. z Convert to reconstructed output ,in It is the weight matrix from the hidden layer to the output layer. The corresponding bias vector is the final output of the autoencoder. ,in d Indicates the output dimension.

[0113] Step S2 includes inputting the encoder-processed data into the decoder of the graph neural network, which outputs the optimized active power command for the generator group. The figure contains... Each node and Undirected edge, k This represents the number of nodes connected by each edge. The calculation process for undirected edges is as follows:

[0114]

[0115]

[0116]

[0117]

[0118]

[0119]

[0120]

[0121]

[0122]

[0123] in This indicates the number of nodes, i.e., the total number of wind turbines. It is the degree matrix of nodes in the graph that contain self-connections. A Represents the original adjacency matrix. This shows the sum of adjacency matrices with self-connections. This represents an adjacency matrix with self-connections. Indicates the first i Taiwan unit and the first j Adjacency matrix of the kiosks; Indicates the first l Learnable parameters of the layer Indicates the first l +1 layer node feature matrix Represents the identity matrix. It is an activation function. Indicates the first l The node feature matrix of the layer; For nodes The aggregation characteristics, Represents a node The set of adjacent nodes, and They represent the first l Layer and first l Adjacent nodes in layer -1 The feature representation can effectively capture local graph structure information by calculating the average value of the features of neighboring nodes; Indicates the first l The layer's weight matrix is ​​used here to perform a linear transformation on the features of neighboring nodes. Indicates from the previous level to the [number]th level l The weight matrix of the node feature representation of the layer is used to preserve the features of the node itself. The nonlinearity is introduced by the ReLU function so that the model can capture the feature patterns between wind turbine units. Characteristics representing real nodes, Represents the node characteristics of the reconstruction; express and The probability that an edge exists between them. ) indicates nodes in different views The feature cosine similarity. Describes the edge reconstruction loss function. This represents the contrastive learning loss function, used to enhance the model's ability to discriminate the features of wind turbine nodes. and These represent the unit's wind speed characteristics and the enhanced characteristics after random perturbation, respectively.

[0124] Through the self-supervised learning tasks described above, the model can not only learn features from labeled data, but also extract correlation information from unlabeled tail data, thereby improving the robustness and generalization ability of the model.

[0125] The process of adjusting the weight matrix in step S3 is represented as follows:

[0126]

[0127]

[0128] in, and The outputs of the model-driven and data-driven control methods are respectively the pitch angle. and generator torque The dataset; It is a connection and The input after; ; n is the data dimension; It is the weight matrix of the input layer; It is the bias of the input layer; It is the activation function of the input layer; It is the weight matrix of the output layer; Indicates the bias of the output layer; It is the activation function of the output layer. This represents the first hidden layer function. This indicates the final control command.

[0129] To verify the effectiveness of the hybrid control method for wind farm units with wake effect applicable to this embodiment, the following section provides a detailed description of the test system setup and various tests.

[0130] 1) Test system setup

[0131] Taking a wind farm layout with a 4x5 configuration and varying degrees of wake overlap downstream of the wind turbines as an example, this layout utilizes a 5-MW wind turbine model developed by the National Renewable Energy Laboratory, with a rotor diameter of 126m. Figure 4 As shown, as a common wind farm layout, it is assumed that the wind turbines are spaced four rotor diameters apart in the row and six rotor diameters apart in the column.

[0132] The simulation duration was set to 800 seconds. From 0 to 100 seconds, all wind turbines experienced uniform wind speeds and were configured for constant power output, simulating the initial wind speed across the entire wind farm. Therefore, the wind speed of each wind turbine was affected by the wake effect from 100 to 800 seconds.

[0133] 2) Model prediction accuracy test

[0134] To verify the prediction accuracy, this method was compared with Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Transformer. Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Coefficient of Determination (R²) were used as quantitative evaluation indicators to assess the accuracy and reliability of each method. The general statistical indicators MAE, RMSE, and R² can be calculated as follows:

[0135]

[0136]

[0137]

[0138] in, These are measured values. It is the average of the measured values. It is the output of the evaluation model.

[0139] Lower MAE and RMSE values ​​indicate smaller prediction errors, suggesting a better model fit. Higher R² values ​​indicate better fit and prediction performance of the parent model.

[0140] The predicted active power and pitch angle of the wind turbine are as follows: Figure 5 As shown in the figure. For ease of interpretation, generator torque is converted into active power. The target curve in the figure reflects the wind turbine output data derived from model predictive control. Compared with LSTM, GRU, and Transformer methods, the curve of the proposed data-driven method is closer to the target, which implies better predictive performance.

[0141] According to Table 1, the active power MAE, RMSE, and R² values ​​of the DDM are 0.0452, 0.0443, and 0.9967, respectively, and the pitch angle MAE, RMSE, and R² values ​​of the DDM are 0.0542, 0.088, and 0.9945, respectively. Clearly, the data-model hybrid control method maintains a higher standard in terms of overall data fitting.

[0142] Table 1. Statistical Indicators of Active Power and Pitch Angle

[0143]

[0144] 2) Fatigue load and power generation test

[0145] To verify the performance of fatigue load suppression and power generation increase, the data-model hybrid control method was compared with three other control schemes: Optimal Wake Control (OWC), MPC-based Generator Torque and Pitch Angle (GTPA) control, and MPC-based Active Power Control (APC). Damage Equivalent Load (DEL) quantization was employed. F t and T s The changes.

[0146] Thrust and shaft torque are the main fatigue loads of wind turbines, from Figure 6 As can be seen, compared with the GTPA and APC schemes, the proposed data-model hybrid has a smoother curve, which means it more effectively suppresses wind turbine loads. F t and T s Fluctuations. Five wind turbine units were calculated. F t and T sThe DEL is listed in Table 2.

[0147] According to Table 2, compared with the APC control scheme, the total DEL of Ft and Ts in the data-model hybrid scheme (DMHD) was reduced by 45.27% and 12.11%, respectively. This verifies the effectiveness of the data-model hybrid method in fatigue load suppression. Compared with the OWC method, the data-model hybrid method... F t and T s The total DEL decreased by 1.86% and 1.24%, respectively. This indicates that the data-model hybrid method and the OWC method have better fatigue load suppression performance.

[0148] Table 2. DEL Quantization of Wind Turbines under Different Control Schemes F t and T s Changes

[0149]

[0150] The average active power output of the five wind turbines was calculated, the impact on power generation was quantified, and the results are listed in Table 3.

[0151] According to Table 3, compared with the APC scheme, the average total active power output of the five wind turbines using the OWC scheme increased by 9.09%. Furthermore, when comparing the OWC scheme with the data-model hybrid scheme, the average total active power output of the five wind turbines increased by an additional 1.17%. This indicates that the data-model hybrid scheme demonstrates superior performance in improving power generation efficiency.

[0152] Table 3 shows the average active power output of wind turbine generators from 100s to 800s under different control schemes.

[0153]

[0154] 4) Measurement noise immunity test

[0155] Developing a unified model for noise sources is challenging due to their complexity. To simulate the measured noise encountered during wind turbine operation, white noise with an average value 0.05 times the original value is superimposed on the input data collected using the data-model hybrid and OWC methods. Two wind turbines are selected as exemplary cases for analysis, such as... Figure 7 As shown, after introducing noise, the deviation between the data-model mixture and the target curve is more moderate compared to OWC, indicating that the data-model mixture method exhibits greater resilience to measurement noise.

[0156] The methods in the above embodiments can be implemented by a processor calling a program stored in memory (which can be the device's memory or external memory). That is, the device may include a processor that executes the methods in the above embodiments by calling a program in memory. The processor here can be an integrated circuit with signal processing capabilities, such as a CPU. The device can be implemented by one or more integrated circuits configured to implement the above methods. For example: one or more ASICs, or one or more microprocessors (DSPs), or one or more FPGAs, or a combination of at least two of these integrated circuit forms. Alternatively, a combination of the above implementation methods can be used.

[0157] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0158] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0159] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0160] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0161] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0162] In another embodiment of this application, a computer-readable storage medium is also provided, which stores computer-executable instructions. When at least one processor of the device executes the computer-executable instructions, the device performs the method described in the above-described partial embodiments.

[0163] In another embodiment of this application, a computer program product is also provided, the computer program product including computer-executable instructions stored in a computer-readable storage medium; at least one processor of the device can read the computer-executable instructions from the computer-readable storage medium, and the at least one processor executes the computer-executable instructions to cause the device to perform the methods described in some of the above embodiments.

[0164] The above description and accompanying drawings provide specific example embodiments and implementations. However, the described subject matter can be embodied in a variety of different forms, and therefore, the covered or claimed subject matter is intended to be construed as not being limited to any of the example embodiments set forth herein. A fairly broad scope is intended for the claimed or covered subject matter. Among other things, the subject matter can be embodied as a method, apparatus, component, system, or non-transitory computer-readable medium for storing computer code. Thus, embodiments can take the form of, for example, hardware, software, firmware, storage medium, or any combination thereof. For example, the method embodiments described above can be implemented by a component, apparatus, or system including a memory and a processor by executing computer code stored in said memory.

[0165] References to features, advantages, or similar language anywhere in this specification do not imply that all features and advantages achievable with this technical solution are included or should be included in any single implementation thereof. Rather, language relating to features and advantages is to be understood as meaning that a particular feature, advantage, or characteristic described in connection with an embodiment is included in at least one embodiment of this technical solution. Therefore, discussions of features and advantages, and similar language anywhere in this specification may, but do not necessarily, refer to the same embodiments.

[0166] Furthermore, the features, advantages, and characteristics described in this technical solution can be combined in any suitable manner in one or more embodiments. In view of the description herein, those skilled in the art will recognize that this technical solution can be practiced without one or more specific features or advantages of a particular embodiment. In other instances, additional features and advantages that may not be present in all embodiments of this technical solution may be found in certain embodiments.

Claims

1. A hybrid control method for wind farm turbines applicable to wake effects, characterized in that, The method includes: S1. Establish a wind farm wake optimization control model based on model predictive control, where the inputs are wind speed, engine speed, pitch angle, and torque from the collected data, and the output is the pitch angle of the turbine group. and generator torque ; Step S1 includes establishing a wake effect calculation model; Step S1 includes obtaining an incremental expression with unit speed, pitch angle, and torque as independent variables; Step S1 includes combining the wake model to... F t , T s For fatigue load targets, F t For thrust; T s For shaft torque; establish the controller objective function; S2, Establish a data-driven strategy, and obtain wind turbine group data and turbine pitch angle through encoders and graph neural network decoders. and generator torque The complex mapping relationship; among which, the wind turbine group data includes wind speed, wind direction, speed, and pitch angle; S3, based on neural networks, mixes the output results of data-driven strategies and model predictive control-based wind farm wake optimization control models to obtain the optimal output; Step S3 includes dynamically adjusting weights based on a neural network. The weight parameters are based on the input and labels, and the weights are adjusted according to the input and labels through supervised learning.

2. The method according to claim 1, characterized in that, The wake effect calculation model established in step S1 is as follows: ； in, i, j Represents a positive integer variable; This represents the total number of units within the wake region. λ Indicates the tip speed ratio, β Indicates the pitch angle; C t.j It is the first j Upstream thrust coefficient of each wind turbine unit; v ∞ Indicates the initial wind speed; v i [ t 0] and v ∞ [ t 0] are respectively the first i The single equivalent wind speed of each wind turbine unit and t Initial inflow velocity at time 0; v j [ t 0- ]and C t.j [ t 0- [These are the first] times. j Each wind turbine unit t 0- Wind speed and thrust coefficient at any given time, Indicates the first i Taiwan and the j The distance between typhoon turbine units.

3. The method according to claim 2, characterized in that, The incremental expression obtained in step S1 with unit speed, pitch angle, and torque as independent variables is: ； ； ； ； in Indicates increment, It is the gearbox speed ratio. It is aerodynamic torque; It is air density; It is the tip speed ratio; It is the generator speed; It is the speed of the filter generator; J t It is equivalent quality; It is the generator torque; It is the propeller pitch angle; This is a reference value for the generator torque. t 1, t 2 and t 2 indicates the time parameter; and The calculation is as follows: ； ； Among them, utilizing λ 0 and β 0 can be used to obtain the power coefficient at the operating point. C p0 ( λ 0, β 0), and The partial derivative is the sensitivity coefficient; R It is the blade radius. v 0 represents the wind speed measurement value. These are the measured values ​​of the wind energy capture factor and the generator speed; based on the above incremental expression, we can obtain... ， , , , , , , , , , , ； Subscript i Indicates which wind turbine unit. This represents the set of all increments of independent variables in a wind farm. This represents the set of all control variable increments for a wind farm. This represents the set of wind speed variables from multiple turbines in a wind farm, where k = 1, 2, 3, ..., n represents the number of prediction steps, and n is a positive integer. , , The wind farm coefficient matrix contains the coefficient matrix of multiple turbine units. ， ， For the first i The coefficient matrix of the unit. This is a measurement of mechanical torque. This is the measured value of the generator torque.

4. The method according to claim 3, characterized in that, In step S1, F t , T s The fatigue load target can be expressed by the following formula: ； ； Wherein represents v Wind speed, F t It is thrust; T s It is shaft torque; P avi It is usable wind energy; , , , and It is the sensitivity coefficient; The controller objective function can be expressed as: ； ； in Ts This represents the equivalent low-speed shaft torque of the gearbox. N c It controls the number of prediction steps. Q 1， Q 2 is a constant weight coefficient matrix. and These are the minimum and maximum torque values ​​of the generator, respectively. and These represent the minimum and maximum pitch angles, respectively.

5. The method according to claim 4, characterized in that, The process of data input being processed by the autoencoder in step S2 is as follows: ； ； in, It consists of standardized historical wake-related data, including active power, wind speed, pitch angle, thrust, torque, and generator speed. n Indicates the dimension of the input data. s express x The length of , where x It can be represented as x = [ x 1, x 2, x 3, x 4, x [5] In the encoding stage of the autoencoder, the input data x is linearly transformed and an activation function (Rectified Linear Unit, ReLU) is applied to map it to the feature representation of the hidden layer. z middle, It is the input weight matrix. It is the input bias vector, and the feature representation of the hidden layer. ,in e This represents the number of neurons in the hidden layer; in the decoding stage of the autoencoder, the features of the hidden layer are also represented by linear transformation and the ReLU activation function. z Convert to reconstructed output ,in It is the weight matrix from the hidden layer to the output layer. The corresponding bias vector is the final output of the autoencoder. ,in d Indicates the output dimension.

6. The method according to claim 5, characterized in that, Step S2 includes inputting the encoder-processed data into the decoder of the graph neural network, which outputs the optimized active power command for the generator group. The figure contains... Each node and Undirected edge, k This represents the number of nodes connected by each edge. The calculation process for undirected edges is as follows: ； ； ； ； ； ； ； ； ； in This indicates the number of nodes, i.e., the total number of wind turbines. It is the degree matrix of nodes in the graph that contain self-connections. A Represents the original adjacency matrix. This shows the sum of adjacency matrices with self-connections. This represents an adjacency matrix with self-connections. Indicates the first i Taiwan unit and the first j Adjacency matrix of the kiosks; Indicates the first l Learnable parameters of the layer Indicates the first l +1 layer node feature matrix Represents the identity matrix. It is an activation function. Indicates the first l The node feature matrix of the layer; For nodes The aggregation characteristics, Represents a node The set of adjacent nodes, and They represent the first l Layer and first l Adjacent nodes in layer -1 The feature representation can effectively capture local graph structure information by calculating the average value of the features of neighboring nodes; Indicates the first l The layer's weight matrix is ​​used here to perform a linear transformation on the features of neighboring nodes. Indicates from the previous level to the [number]th level l The weight matrix of the node feature representation of the layer is used to preserve the features of the node itself. The nonlinearity is introduced by the ReLU function so that the model can capture the feature patterns between wind turbine units. Characteristics representing real nodes, Represents the node characteristics of the reconstruction; express and The probability that an edge exists between them. ) indicates nodes in different views The feature cosine similarity. Describes the edge reconstruction loss function. This represents the contrastive learning loss function, used to enhance the model's ability to discriminate the features of wind turbine nodes. and These represent the unit's wind speed characteristics and the enhanced characteristics after random perturbation, respectively.

7. The method according to claim 6, characterized in that, The process of adjusting the weight matrix in step S3 is represented as follows: ； ； in, and The outputs of the model-driven and data-driven control methods are respectively the pitch angle. and generator torque The dataset; It is a connection and The input after; ; n is the data dimension; It is the weight matrix of the input layer; It is the bias of the input layer; It is the activation function of the input layer; It is the weight matrix of the output layer; Indicates the bias of the output layer; It is the activation function of the output layer. This represents the first hidden layer function. This indicates the final control command.

8. A hybrid control device for wind farm turbines applicable to wake effects, characterized in that, include: A processor coupled to a memory for storing programs or instructions that, when executed by the processor, cause the apparatus to perform the method as described in any one of claims 1 to 7.

9. A chip system, characterized in that, It includes at least one processor and an interface for receiving data and / or signals, wherein the at least one processor is configured to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 7.

Images