Intelligent wind field wake flow loss suppression method based on artificial intelligence

Abstract

The invention relates to the technical field of wind power generation, in particular to an intelligent wind field wake flow loss suppression method based on artificial intelligence, and the method comprises the following steps: collecting the operation state and environment information of each fan in a wind power plant, and carrying out the standardization processing of the collected data, and forming a unified data set; extracting the spatial relationship between the fans, the upstream and downstream position relationship and the data characteristics of the wind field environment from the data set, and encoding the coupling relationship between the fans; and based on the data features, constructing an artificial intelligence prediction model fusing the spatial-temporal features and the fan coupling relationship, and using the artificial intelligence prediction model to output a prediction result including the wind speed, turbulence and power state of the downstream position of each fan in the wind power plant. According to the method, the operation state and the environment information of each fan in the wind power plant are collected, so that the problems of low fan operation efficiency and power generation loss caused by the fact that an experience control method is mostly adopted in traditional wind power plant wake flow loss regulation and control are solved.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, and in particular to a smart wind farm wake loss suppression method based on artificial intelligence. Background Technology

[0002] A wind farm is a renewable energy power generation facility that converts wind energy into electricity by deploying multiple wind turbine generators within a certain area. With the development of wind power generation technology, the scale of wind farms has gradually expanded, with a single wind farm typically containing dozens to hundreds of turbines. The operation of a wind farm involves multiple aspects, including wind energy resource acquisition, turbine layout optimization, generator control, and wind power output management.

[0003] Traditional wind farm wake loss control methods mostly rely on empirical control methods. Because they cannot predict the complex coupling relationships between wind turbines, they result in low turbine operating efficiency and power generation loss. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides an intelligent wind farm wake loss suppression method based on artificial intelligence, which aims to improve the problem that traditional wind farm wake loss control mostly adopts empirical control methods. Since it is impossible to predict the complex coupling relationship between wind turbines, it causes low wind turbine operating efficiency and power generation loss.

[0005] In a first aspect, the present invention provides the following technical solution: a smart wind farm wake loss suppression method based on artificial intelligence, comprising the following steps: Collect the operating status and environmental information of each wind turbine in the wind farm, and standardize the collected data to form a unified dataset; Extract data features of spatial relationships between wind turbines, upstream and downstream location relationships, and wind farm environment from the dataset, and encode the coupling relationships between wind turbines; Based on the data features, an artificial intelligence prediction model is constructed that integrates spatiotemporal features and the coupling relationship between wind turbines. The artificial intelligence prediction model is then used to output prediction results including wind speed, turbulence and power status at the downstream location of each wind turbine in the wind farm. An initial wind turbine collaborative control strategy is generated using a reinforcement learning algorithm, and wind turbine mechanical constraints are introduced during the strategy generation process. The initial wind turbine collaborative control strategy is then iteratively updated using multi-condition training, domain randomization, and strategy regularization. Based on the iteratively updated wind turbine collaborative control strategy, the rotational speed and blade pitch of each wind turbine in the wind farm are dynamically adjusted. Based on the prediction results and the control strategy, the position of the wind turbines in the wind farm is constrained and optimized, and the arrangement order of the wind turbines is adjusted. The control strategy and layout optimization scheme are sent to the wind turbine control system to monitor the wind turbine status and modify the control strategy based on the monitoring data.

[0006] By adopting the above technical solution, the operating status and environmental information of each wind turbine in the wind farm are collected, and the collected data is standardized. Then, based on the extracted spatial relationships, upstream and downstream positional relationships and wind farm environmental characteristics, an artificial intelligence prediction model is constructed to predict the wind speed, turbulence and power status at the downstream position of each wind turbine. This improves the problem that traditional wind farm wake loss control mostly adopts empirical control methods, which cannot predict the complex coupling relationship between wind turbines, resulting in low wind turbine operating efficiency and power generation loss.

[0007] Furthermore, the standardization processing of the collected data includes: Interpolate or fill in missing values; Detect and remove outliers; Normalize or standardize the data for each feature. Align and resample time series data; A unified dataset is constructed for subsequent feature extraction and prediction.

[0008] Furthermore, the extraction of data features related to the spatial relationships between wind turbines, upstream and downstream location relationships, and the wind farm environment includes: Obtain wind turbine location data; Calculate the horizontal distance, vertical distance, and relative azimuth angle between the fans; Determine the upstream and downstream location relationship of the wind turbine based on wind direction information; Extract wind field environmental characteristics, including wind speed, wind direction, turbulence intensity, and other environmental parameters.

[0009] Furthermore, encoding the coupling relationship between wind turbines includes: Construct a wind turbine coupling matrix to represent the spatial and upstream / downstream relationships between wind turbines; The status information of each wind turbine is combined with the coupling matrix to form a feature vector; Integrate all wind turbine feature vectors to form a unified data representation; The data is represented as the input format for an artificial intelligence prediction model.

[0010] Furthermore, the construction of the artificial intelligence prediction model that integrates spatiotemporal features and the coupling relationship between wind turbines includes: Input the wind turbine characteristics and coupling relationships into the artificial intelligence model; An artificial intelligence model architecture that integrates spatiotemporal features and wind turbine coupling relationships is selected. The architecture is a fusion architecture that uses convolutional neural networks to extract spatial features, long short-term memory networks to capture time series patterns, and graph neural networks to encode the coupling relationships between wind turbines. The model is trained to output the wind speed, turbulence, and power status at the downstream location of the wind turbine. The consistency between the model output and historical data is verified by the root mean square error or mean absolute error, and the training parameters are saved.

[0011] Furthermore, the initial wind turbine collaborative control strategy includes: Initial wind turbine collaborative control strategy is generated using reinforcement learning algorithms; Mechanical constraints on the wind turbines are introduced during the initial generation of the wind turbine collaborative control strategy. The initial wind turbine collaborative control strategy is iteratively updated using multi-condition training, domain randomization, and strategy regularization. The multi-condition training includes data augmentation training that combines simulation data generated by computational fluid dynamics or large eddy simulation with actual wind field data. The output can be sent to the wind turbine for execution of control strategies.

[0012] Furthermore, the dynamic adjustment of the rotational speed and blade pitch of each wind turbine within the wind farm includes: The generated control strategy is then sent to the wind turbine control system. Adjust the fan speed according to the strategy instructions; Adjust the blade pitch according to strategy instructions; Record the adjustment results to form wind turbine operating status data.

[0013] Furthermore, the constraint optimization of the wind turbine's position in the wind farm includes: Generate constraints based on the prediction results and control strategies; A genetic algorithm is used to calculate and optimize the arrangement order and relative positions of the wind turbines; Generate wind turbine layout schemes that meet the constraints; The fan position adjustment plan is output to the control system.

[0014] Furthermore, the step of modifying the control strategy based on monitoring data includes: Input the monitoring data into the policy update module; Adjust the parameters of the reinforcement learning control strategy based on the wind turbine power output deviation data; The updated control strategy is output and sent to the wind turbine for execution.

[0015] Secondly, the present invention provides the following technical solution: an intelligent wind farm wake loss suppression system based on artificial intelligence, the system comprising: The data acquisition module is used to collect the operating status and environmental information of each wind turbine in the wind farm, and to standardize the collected data to form a unified dataset. The feature extraction module is used to extract data features of the spatial relationship between wind turbines, the upstream and downstream positional relationship, and the wind farm environment from the dataset, and to encode the coupling relationship between wind turbines; The wake prediction module is used to construct an artificial intelligence prediction model based on the data features, which integrates spatiotemporal features and the coupling relationship between wind turbines, and to use the artificial intelligence prediction model to output prediction results including wind speed, turbulence and power status at the downstream location of each wind turbine in the wind farm. The intelligent control module is used to generate an initial wind turbine collaborative control strategy using a reinforcement learning algorithm, and to introduce wind turbine mechanical constraints during the strategy generation process; the initial wind turbine collaborative control strategy is iteratively updated through multi-condition training, domain randomization, and strategy regularization. The strategy update module is used to dynamically adjust the speed and blade pitch of each wind turbine in the wind farm according to the iteratively updated wind turbine collaborative control strategy. The layout optimization module is used to constrain and optimize the position of the wind turbines in the wind farm based on the prediction results and the control strategy, and to adjust the arrangement order of the wind turbines. The execution and monitoring module is used to send the control strategy and layout optimization scheme to the wind turbine control system, monitor the wind turbine status, and modify the control strategy based on the monitoring data.

[0016] The present invention has the following beneficial effects: 1. In this invention, by collecting the operating status and environmental information of each wind turbine in the wind farm and standardizing the collected data, an artificial intelligence prediction model is constructed based on the extracted spatial relationships, upstream and downstream positional relationships and wind farm environmental characteristics to predict the wind speed, turbulence and power status at the downstream position of each wind turbine. This improves the problem that traditional wind farm wake loss control mostly adopts empirical control methods, which cannot predict the complex coupling relationship between wind turbines, resulting in low wind turbine operating efficiency and power generation loss.

[0017] 2. In this invention, an initial wind turbine collaborative control strategy is generated by using a reinforcement learning algorithm. During the strategy generation process, wind turbine mechanical constraints, multi-condition training, domain randomization, and strategy regularization are introduced for iterative updates. This dynamically adjusts the wind turbine speed and blade pitch, thereby improving the problem that traditional wind farm control mostly adopts fixed speed or single-machine optimization strategies, which ignores wake interference between wind turbines and the influence of multiple operating conditions, resulting in low overall power generation efficiency of wind turbines.

[0018] 3. In this invention, the position of the wind turbine in the wind farm is constrained and optimized based on the prediction results and control strategy, and the arrangement order of the wind turbines is adjusted. Then, the control strategy and layout optimization scheme are sent to the wind turbine control system for status monitoring and strategy correction. This improves the problem that traditional wind farm layout optimization mostly adopts static planning methods, which lacks integration with real-time control strategies, resulting in the inability to effectively suppress wake interference. Attached Figure Description

[0019] Figure 1 This is a flowchart of the intelligent wind farm wake loss suppression method based on artificial intelligence proposed in this invention. Figure 2 This is a system architecture diagram of the intelligent wind farm wake loss suppression system based on artificial intelligence proposed in this invention. Detailed Implementation

[0020] The technical solutions in 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 embodiments of the present invention, and not all embodiments. 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.

[0021] Example 1 In a first embodiment of the present invention, the present invention provides a smart wind farm wake loss suppression method based on artificial intelligence, such as... Figure 1 As shown, the process includes the following steps: collecting the operating status and environmental information of each wind turbine in the wind farm, and standardizing the collected data to form a unified dataset; Furthermore, the standardization process for the collected data includes: Interpolate or fill in missing values; Detect and remove outliers; Normalize or standardize the data for each feature. Align and resample time series data; A unified dataset is constructed for subsequent feature extraction and prediction.

[0022] Specifically, the basic data preparation for predictive model construction and wake loss suppression methods is achieved by cleaning and processing the operating status and environmental information of each wind turbine in the wind farm. This includes interpolating or supplementing missing values ​​in the original data and assigning the wind turbine number as... The time series index is Missing values ​​can be imputed using linear interpolation, expressed as follows: ;in Indicates the first Typhoon machine in time The measured values ​​are derived from real-time wind speed, wind direction, power output, blade pitch, and rotational speed parameters collected by the wind turbine's SCADA system. Outliers are detected and eliminated by calculating the characteristic value deviation at each time point. ;in and The first The mean and standard deviation of typhoon characteristics within the training window, if If a value is found to be outlier, it is either removed or replaced with a nearby valid data point. (Threshold) This can be obtained from historical data statistics, and each characteristic data point can be normalized or standardized using formulas. Data of different dimensions are unified to a zero-mean, unit-variance form, and time series data alignment and resampling use a uniform time step. Data from different wind turbine acquisition frequencies are synchronized into a standard time series using linear interpolation or resampling methods to form a matrix. As a unified dataset, the input matrix The data is subsequently used for wind turbine feature extraction and coupling relationship encoding, serving as the training input for an artificial intelligence prediction model. The model output consists of predicted values ​​of zone velocity, turbulence, and power state at downstream locations of each wind turbine, which are used to generate wind turbine collaborative control strategies and layout optimization, thereby achieving the goal of suppressing wake loss.

[0023] By standardizing the processing of the operating status and environmental information of each wind turbine in a wind farm and constructing a unified dataset, a consistent and processable data foundation can be provided for subsequent wind turbine feature extraction, coupling relationship encoding, and artificial intelligence prediction model training, thereby improving the availability of data and the reliability of model input.

[0024] Extract data features of spatial relationships between wind turbines, upstream and downstream location relationships, and wind farm environment from the dataset, and encode the coupling relationships between wind turbines; Furthermore, the data features extracted include the spatial relationships between wind turbines, the upstream and downstream location relationships, and the wind farm environment: Obtain wind turbine location data; Calculate the horizontal distance, vertical distance, and relative azimuth angle between the fans; Determine the upstream and downstream location relationship of the wind turbine based on wind direction information; Extract wind field environmental characteristics, including wind speed, wind direction, turbulence intensity, and other environmental parameters.

[0025] Specifically, the input data includes the geographical coordinates of the wind turbine. Wind angle Wind speed turbulence intensity and environmental meteorological characteristic parameter set ;in Indicates temperature. Indicates air pressure. Humidity data is collected through wind farm monitoring systems, weather towers, and wind-measuring lidar. Next, calculations are performed on the wind turbines. With wind turbine When considering spatial relationships, the horizontal distance is important. Through the formula: ; Calculate the vertical distance Relative azimuth When determining the upstream and downstream relationship, it is achieved by judging the positional relationship of the wind turbine relative to the wind direction. If the conditions are met... Then determine the fan Located in the wind turbine Downstream, next, when extracting wind field environmental characteristics, the collected wind speed... ,wind direction turbulence intensity and meteorological parameters As input feature vector The spatial characteristic matrix between the wind turbines can be obtained through the above calculations. and upstream and downstream relationship matrix ;in Indicates wind turbine Located in the wind turbine The downstream of, otherwise 0, will eventually , With environmental feature vectors Combining to form a set of wind turbine features This result serves as input to the subsequent wind turbine coupling relationship encoding module, used to construct the spatial dependence and dynamic influence expression between wind turbines in the artificial intelligence model, providing an input basis for predicting wind speed, turbulence, and power state.

[0026] By extracting the spatial relationships, upstream and downstream location relationships, and wind field environmental characteristics of wind turbines, and encoding the coupling relationships between wind turbines, structured input data can be provided for artificial intelligence prediction models. This supports the prediction of downstream wind speed, turbulence, and power status of wind turbines, and provides a foundation for wake loss suppression and collaborative control strategies.

[0027] Furthermore, encoding the coupling relationships between wind turbines includes: Construct a wind turbine coupling matrix to represent the spatial and upstream / downstream relationships between wind turbines; The status information of each wind turbine is combined with the coupling matrix to form a feature vector; Integrate all wind turbine feature vectors to form a unified data representation; Use data representation as the input format for artificial intelligence prediction models.

[0028] Specifically, the input data includes the spatial feature matrix of the wind turbine. Upstream and downstream relationship matrix and the set of operating status parameters for each wind turbine. ;in For the output power of the fan, For fan speed, The blade pitch angle, For the inflow wind speed, To determine the local turbulence intensity, these data are collected in real time by the wind turbine control system and monitoring sensors. First, the wind turbine coupling matrix is ​​constructed. This matrix simultaneously reflects the spatial distance between wind turbines and the upstream and downstream influence relationships. The calculation formula is as follows: ;in The distance attenuation coefficient describes the attenuation trend of the wind turbine wake effect with distance. This matrix is ​​used to quantify the spatial dependence between wind turbines. Then, the state vector of each wind turbine is... With coupling matrix Combined to form local feature vectors The calculation formula is: ;in The total number of wind turbines represents the number of turbines in the series. The comprehensive feature representation of a typhoon turbine after being affected by other wind turbines is then used to set the local feature vectors of all wind turbines. Compared with the original state information The data is then concatenated to form a unified data representation matrix. The matrix The output is used as input to an artificial intelligence prediction model, which then processes the data. The learning process captures the wake effect, spatial coupling effect, and power distribution patterns between wind turbines, thereby enabling the generation of prediction and control strategies for wind speed, turbulence, and power states in subsequent steps.

[0029] By constructing a wind turbine coupling matrix and combining the states of each wind turbine with their coupling relationships to form a unified data representation, a structured input can be provided for artificial intelligence prediction models. This supports the prediction of downstream wind speed, turbulence, and power states, and provides basic data for wake loss suppression and collaborative control strategies.

[0030] An artificial intelligence prediction model is constructed based on data features, which integrates spatiotemporal features with the coupling relationship of wind turbines. The artificial intelligence prediction model is used to output prediction results including wind speed, turbulence and power status at the downstream location of each wind turbine in the wind farm. Furthermore, constructing an artificial intelligence prediction model that integrates spatiotemporal characteristics with the coupling relationship of wind turbines includes: Input the wind turbine characteristics and coupling relationships into the artificial intelligence model; We selected an artificial intelligence model architecture that integrates spatiotemporal features with the coupling relationship between wind turbines. The architecture is a fusion architecture that uses convolutional neural networks to extract spatial features, long short-term memory networks to capture time series patterns, and graph neural networks to encode the coupling relationship between wind turbines. The model is trained to output the wind speed, turbulence, and power status at the downstream location of the wind turbine. The consistency between the model output and historical data is verified by the root mean square error or mean absolute error, and the training parameters are saved.

[0031] Specifically, the aforementioned unified data representation As input, where Indicates the first Typhoon state vector With the row of the coupling matrix Combined with the resulting eigenvectors, For wind speed, For wind direction, For turbulence intensity, For power output, For rotational speed, For blade pitch, these parameters are acquired by the wind turbine's SCADA system and environmental measurement equipment. The input data is processed through an architecture that integrates convolutional neural networks, long short-term memory networks, and graph neural networks. Feature extraction and spatiotemporal correlation encoding are performed, where convolutional neural networks are used for spatial feature extraction, long short-term memory networks are used for capturing time series patterns, and graph neural networks are used for encoding wind turbine coupling relationships. The model output is... This represents the predicted wind speed, turbulence, and power state at the downstream location of each wind turbine. The model is trained by minimizing the root mean square error. or mean absolute error To verify the consistency between the output and historical data and save the training parameters. Output results As the basis for subsequent wind turbine collaborative control strategy generation and wind turbine layout optimization, it is used to guide the dynamic adjustment of wind turbine speed and blade pitch to reduce wake loss and improve the overall power generation efficiency of wind farm.

[0032] By constructing an artificial intelligence prediction model that integrates spatiotemporal features with the coupling relationship of wind turbines, it is possible to predict the wind speed, turbulence, and power status at the downstream locations of each wind turbine in a wind farm, providing data support for the generation of wind turbine collaborative control strategies and the optimization of wake loss.

[0033] An initial wind turbine collaborative control strategy is generated using a reinforcement learning algorithm, and wind turbine mechanical constraints are introduced during the strategy generation process. The initial wind turbine collaborative control strategy is then iteratively updated through multi-condition training, domain randomization, and strategy regularization. Furthermore, the initial wind turbine collaborative control strategy includes: Initial wind turbine collaborative control strategy is generated using reinforcement learning algorithms; Mechanical constraints on the wind turbines are introduced during the initial generation of the wind turbine collaborative control strategy. The initial wind turbine collaborative control strategy is iteratively updated using multi-condition training, domain randomization, and strategy regularization. Multi-condition training includes data augmentation training that combines simulation data generated by computational fluid dynamics or large eddy simulation with actual wind field data. The output can be sent to the wind turbine for execution of control strategies.

[0034] Specifically, the downstream wind speed, turbulence, and power state prediction results output by the aforementioned artificial intelligence prediction model will be used. As status input Give a reinforcement learning agent, where This represents the environment and operating status of each wind turbine at the current time step, and defines the action vector. For the speed of each fan and blade pitch The agent receives the adjustment instructions through the policy function. Output actions, introducing mechanical constraints during strategy generation. To ensure the safe operation of wind turbines, multi-condition training enhances generalization ability by training strategies under different wind speeds, directions, and turbulence conditions. This is combined with domain randomization techniques on simulation data. and actual wind field data Data augmentation training, utilizing policy regularization The policy update magnitude is controlled to stabilize training; the agent's optimization objective is to maximize the cumulative reward function. Among the rewards The strategy can be calculated comprehensively based on wind turbine power output, wake loss, and control constraints, and output after training. The strategy is then implemented on the wind turbines. The next step is to apply this strategy to real-time turbine regulation to dynamically adjust the turbine speed and pitch. The results are used to suppress wake loss, improve the overall power generation efficiency of the wind farm, and provide a control reference for turbine arrangement optimization. All parameters are as follows: All data are collected or output from the wind turbine SCADA system and environmental measurement equipment or prediction models.

[0035] By leveraging reinforcement learning to generate wind turbine collaborative control strategies and combining them with mechanical constraints, multi-condition training, and regularized iterative updates of the strategies, executable dynamic adjustment commands can be provided for wind turbines, supporting wake loss suppression and wind farm power generation performance optimization.

[0036] Based on the iteratively updated wind turbine collaborative control strategy, the rotational speed and blade pitch of each wind turbine in the wind farm are dynamically adjusted. Furthermore, dynamically adjusting the speed and blade pitch of each wind turbine within the wind farm includes: The generated control strategy is then sent to the wind turbine control system. Adjust the fan speed according to the strategy instructions; Adjust the blade pitch according to strategy instructions; Record the adjustment results to form wind turbine operating status data.

[0037] Specifically, the control strategies generated by reinforcement learning will be used. The data is sent to the wind turbine control system as input, where the motion vector... Indicates the rotational speed of each fan and blade pitch ; State vector This indicates the predicted downstream wind speed for each wind turbine. turbulence intensity and power output These parameters are obtained by combining the output of an artificial intelligence prediction model with actual operating data collected by the SCADA system, and the wind turbine control system then... Adjust the speed, according to Adjust the blade pitch, and record the adjustment result as a new wind turbine operating state vector. A dataset is generated for subsequent iterative training and strategy updates. The output results are the actual speed and pitch of each wind turbine after execution, which are used to verify the executability of the control strategy and provide real-time operational data support for wake loss suppression and overall power generation optimization.

[0038] By issuing and executing the iteratively updated wind turbine collaborative control strategy, the speed and blade pitch of each wind turbine can be dynamically adjusted to generate real-time operating status data, providing a foundation for suppressing wake loss and optimizing wind farm power generation.

[0039] Based on the prediction results and control strategies, the position of wind turbines in the wind farm is optimized under constraints, and the arrangement order of wind turbines is adjusted. Furthermore, constrained optimization of the wind turbine's location within the wind farm includes: Generate constraints based on the prediction results and control strategies; A genetic algorithm is used to calculate and optimize the arrangement order and relative positions of the wind turbines; Generate wind turbine layout schemes that meet the constraints; The fan position adjustment plan is output to the control system.

[0040] Specifically, the wind speed output by the artificial intelligence prediction model turbulence and power status Compared with the current control strategy Combine to generate a set of constraints Among them, the constraints The algorithm represents the safe distance between wind turbines, the feasible arrangement order, and the power optimization objective. It then uses a genetic algorithm to optimize the arrangement order and relative positions of the wind turbines, defining individual vectors. express The position of the typhoon generator in the planar coordinate system is iteratively updated by a genetic algorithm through selection, crossover, and mutation operations. Maximize the fitness function ;in To predict power output, Penalties for violating the constraints, These parameters, acting as penalty coefficients, are obtained from the prediction model output and wind farm layout constraints. After optimization, a new wind turbine layout scheme is output. The wind turbine control system is used to adjust the position of the wind turbine. The next step is to apply the new layout and control strategy together to the operation of the wind farm to reduce wake interference and improve the overall power generation performance. All coordinates and parameters are provided by wind farm measurement data and prediction models.

[0041] By optimizing the location and arrangement of wind turbines based on prediction results and control strategies, a reasonable wind turbine layout scheme can be formed, which can support the reduction of wake interference and the improvement of the overall power generation efficiency of the wind farm.

[0042] The control strategy and layout optimization scheme are sent to the wind turbine control system to monitor the wind turbine status and modify the control strategy based on the monitoring data. Furthermore, adjusting the control strategy based on monitoring data includes: Input the monitoring data into the policy update module; Adjust the parameters of the reinforcement learning control strategy based on the wind turbine power output deviation data; The updated control strategy is output and sent to the wind turbine for execution.

[0043] Specifically, the power output collected in real time by the wind turbine control system With predicted power Compare the calculated deviations and the deviation vector The input policy update module utilizes reinforcement learning policy parameters. Update rules Adjust the control strategy, among which The loss function representing the control policy measures the difference between the output bias and the expected power; the learning rate... The parameters are set during the training phase, obtained from historical running data and real-time measurements, and the updated strategy is then implemented. The data is output and sent to the wind turbine for execution. The wind turbine then adjusts its speed and pitch according to the new strategy to generate new operating status data. This result is used to continuously optimize wake control and wind farm power generation efficiency. All parameters and status data are provided by wind turbine sensors and predictive models.

[0044] By distributing control strategies and layout optimization schemes and correcting control strategies based on real-time monitoring data, the operating status of wind turbines can be continuously adjusted, enabling wind turbine collaborative control to adapt to changes in the wind field and optimize power generation performance.

[0045] Example 2: In the actual operation of large-scale coastal wind farms, the variable sea wind direction, tight turbine spacing, and significant wake effect lead to reduced wind speeds and increased turbulence downstream, resulting in decreased overall power generation efficiency and increased equipment fatigue wear. To address these issues, the AI-based intelligent wind farm wake loss suppression system provided in this invention is employed, with the structure as follows: Figure 2 As shown. The specific implementation process of this system is as follows: First, the data acquisition module collects the operating status and environmental information of each wind turbine in the wind farm, including data such as wind speed, wind direction, blade angle and output power, and performs standardization processing on the data to form a unified dataset. This process achieves consistency of wind farm operating data in terms of time series and dimensions, and improves the accuracy of subsequent modeling and analysis. Secondly, the feature extraction module extracts the spatial relationship between wind turbines, the upstream and downstream positional relationship, and the wind field environmental features from the dataset, and encodes the coupling relationship between wind turbines to obtain a structured feature representation that can reflect the mutual influence of wind turbines, so that the model can capture the propagation law of wake effect. Then, the wake prediction module constructs an artificial intelligence prediction model based on the extracted features, which integrates spatiotemporal features and the coupling relationship between wind turbines. It extracts spatial features through convolutional neural networks, captures time series patterns through long short-term memory networks, and encodes wind turbine coupling relationships through graph neural networks. This enables the prediction of wind speed, turbulence, and power status at downstream locations of each wind turbine in the wind farm. This step can identify the impact range of the wake zone in advance and provide a quantitative basis for subsequent control. Next, the intelligent control module uses reinforcement learning algorithms to generate an initial wind turbine collaborative control strategy. During the strategy generation process, wind turbine mechanical constraints are introduced, and iterative updates are performed by combining multi-condition training, domain randomization, and strategy regularization, thereby forming a control scheme that balances power generation efficiency and equipment safety. Furthermore, the strategy update module dynamically adjusts the speed and blade pitch of each wind turbine in the wind farm according to the updated wind turbine collaborative control strategy, and sends the control strategy to the wind turbine control system for execution, so that the operating status of each wind turbine responds to the changes in the wind farm in real time. Through this process, the wake effect is actively weakened. Subsequently, the layout optimization module optimizes the position of the wind turbines in the wind farm based on the prediction results and control strategies. It uses a genetic algorithm to adjust the arrangement order and relative position of the wind turbines and outputs a wind turbine layout scheme that meets the constraints, thereby improving the overall aerodynamic efficiency of the wind farm and reducing wake loss. Finally, the execution and monitoring module sends the control strategy and layout optimization scheme to the wind turbine control system, monitors the wind turbine operating status in real time, and dynamically corrects the control strategy based on the monitoring data to ensure that the system continuously adapts to changes in the wind farm environment and maintains efficient operation. Through the synergistic effect of the above steps, the present invention can significantly suppress wake loss under complex wind farm conditions, realize intelligent coordination and optimal energy efficiency control of the wind turbine group, and thus improve the overall power generation performance and equipment reliability of the wind farm.

[0046] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart wind farm wake loss suppression method based on artificial intelligence, characterized in that, Includes the following steps: Collect the operating status and environmental information of each wind turbine in the wind farm, and standardize the collected data to form a unified dataset; Extract data features of spatial relationships between wind turbines, upstream and downstream location relationships, and wind farm environment from the dataset, and encode the coupling relationships between wind turbines; Based on the data features, an artificial intelligence prediction model is constructed that integrates spatiotemporal features and the coupling relationship between wind turbines. The artificial intelligence prediction model is then used to output prediction results including wind speed, turbulence and power status at the downstream location of each wind turbine in the wind farm. An initial wind turbine collaborative control strategy is generated using a reinforcement learning algorithm, and wind turbine mechanical constraints are introduced during the strategy generation process. The initial wind turbine collaborative control strategy is then iteratively updated using multi-condition training, domain randomization, and strategy regularization. Based on the iteratively updated wind turbine collaborative control strategy, the rotational speed and blade pitch of each wind turbine in the wind farm are dynamically adjusted. Based on the prediction results and the control strategy, the position of the wind turbines in the wind farm is constrained and optimized, and the arrangement order of the wind turbines is adjusted. The control strategy and layout optimization scheme are sent to the wind turbine control system to monitor the wind turbine status and modify the control strategy based on the monitoring data.

2. The method for suppressing wake loss in intelligent wind farms based on artificial intelligence according to claim 1, characterized in that, The standardization process for the collected data includes: Interpolate or fill in missing values; Detect and remove outliers; Normalize or standardize the data for each feature. Align and resample time series data; A unified dataset is constructed for subsequent feature extraction and prediction.

3. The method for suppressing wake loss in intelligent wind farms based on artificial intelligence according to claim 1, characterized in that, The extracted data features of the spatial relationship between wind turbines, the upstream and downstream location relationship, and the wind farm environment include: Obtain wind turbine location data; Calculate the horizontal distance, vertical distance, and relative azimuth angle between the fans; Determine the upstream and downstream location relationship of the wind turbine based on wind direction information; Extract wind field environmental characteristics, including wind speed, wind direction, turbulence intensity, and other environmental parameters.

4. The method for suppressing wake loss in intelligent wind farms based on artificial intelligence according to claim 1, characterized in that, The encoding of the coupling relationship between wind turbines includes: Construct a wind turbine coupling matrix to represent the spatial and upstream / downstream relationships between wind turbines; The status information of each wind turbine is combined with the coupling matrix to form a feature vector; Integrate all wind turbine feature vectors to form a unified data representation; The data is represented as the input format for an artificial intelligence prediction model.

5. The method for suppressing wake loss in intelligent wind farms based on artificial intelligence according to claim 1, characterized in that, The artificial intelligence prediction model that integrates spatiotemporal features with the coupling relationship between wind turbines includes: Input the wind turbine characteristics and coupling relationships into the artificial intelligence model; An artificial intelligence model architecture that integrates spatiotemporal features and wind turbine coupling relationships is selected. The architecture is a fusion architecture that uses convolutional neural networks to extract spatial features, long short-term memory networks to capture time series patterns, and graph neural networks to encode the coupling relationships between wind turbines. The model is trained to output the wind speed, turbulence, and power status at the downstream location of the wind turbine. The consistency between the model output and historical data is verified by the root mean square error or mean absolute error, and the training parameters are saved.

6. The method for suppressing wake loss in intelligent wind farms based on artificial intelligence according to claim 1, characterized in that, The initial wind turbine collaborative control strategy includes: Initial wind turbine collaborative control strategy is generated using reinforcement learning algorithms; Mechanical constraints on the wind turbines are introduced during the initial generation of the wind turbine collaborative control strategy. The initial wind turbine collaborative control strategy is iteratively updated using multi-condition training, domain randomization, and strategy regularization. The multi-condition training includes data augmentation training that combines simulation data generated by computational fluid dynamics or large eddy simulation with actual wind field data. The output can be sent to the wind turbine for execution of control strategies.

7. The method for suppressing wake loss in intelligent wind farms based on artificial intelligence according to claim 1, characterized in that, The dynamic adjustment of the rotational speed and blade pitch of each wind turbine in the wind farm includes: The generated control strategy is then sent to the wind turbine control system. Adjust the fan speed according to the strategy instructions; Adjust the blade pitch according to strategy instructions; Record the adjustment results to form wind turbine operating status data.

8. The method for suppressing wake loss in intelligent wind farms based on artificial intelligence according to claim 1, characterized in that, The constraint optimization of the wind turbine's location in the wind farm includes: Generate constraints based on the prediction results and control strategies; A genetic algorithm is used to calculate and optimize the arrangement order and relative positions of the wind turbines; Generate wind turbine layout schemes that meet the constraints; The fan position adjustment plan is output to the control system.

9. The method for suppressing wake loss in intelligent wind farms based on artificial intelligence according to claim 1, characterized in that, The step of modifying the control strategy based on monitoring data includes: Input the monitoring data into the policy update module; Adjust the parameters of the reinforcement learning control strategy based on the wind turbine power output deviation data; The updated control strategy is output and sent to the wind turbine for execution.

10. An intelligent wind farm wake loss suppression system based on artificial intelligence, characterized in that, The system for the AI-based smart wind farm wake loss suppression method according to any one of claims 1-9 comprises: The data acquisition module is used to collect the operating status and environmental information of each wind turbine in the wind farm, and to standardize the collected data to form a unified dataset. The feature extraction module is used to extract data features of the spatial relationship between wind turbines, the upstream and downstream positional relationship, and the wind farm environment from the dataset, and to encode the coupling relationship between wind turbines; The wake prediction module is used to construct an artificial intelligence prediction model based on the data features, which integrates spatiotemporal features and the coupling relationship between wind turbines, and to use the artificial intelligence prediction model to output prediction results including wind speed, turbulence and power status at the downstream location of each wind turbine in the wind farm. The intelligent control module is used to generate an initial wind turbine collaborative control strategy using a reinforcement learning algorithm, and to introduce wind turbine mechanical constraints during the strategy generation process; the initial wind turbine collaborative control strategy is iteratively updated through multi-condition training, domain randomization, and strategy regularization. The strategy update module is used to dynamically adjust the speed and blade pitch of each wind turbine in the wind farm according to the iteratively updated wind turbine collaborative control strategy. The layout optimization module is used to constrain and optimize the position of the wind turbines in the wind farm based on the prediction results and the control strategy, and to adjust the arrangement order of the wind turbines. The execution and monitoring module is used to send the control strategy and layout optimization scheme to the wind turbine control system, monitor the wind turbine status, and modify the control strategy based on the monitoring data.

Images