Fast wind farm power optimization method and system based on graph neural network

Abstract

The application provides a fast wind farm power optimization method and system based on a graph neural network. The method comprises the following steps: step a, constructing a graph neural network model; step b, fast power calculation; step c, combining a genetic algorithm to optimize a yaw angle; step d, optimizing the power output of a wind turbine; and step e, real-time data updating and model adaptability. The system comprises a data input module (1), a graph neural network model (2), a genetic algorithm optimization module (3), a control system (4) and a feedback module (5). The application is applied to the technical field of wind power.

Description

Technical Field

[0001] The present invention relates to the field of wind power technology, and in particular to a method and system for rapid wind farm power optimization based on graph neural networks. Background Art

[0002] Field-level control methods for wind farms have become a research hotspot in recent years. As wind farms expand in size and the number of wind turbines increases, effectively coordinating the operation of multiple wind turbines to optimize overall power generation efficiency and minimize the inter-turbine wake effect has become a core challenge. Traditional control methods mostly rely on centralized or distributed control strategies, primarily based on physical models or local measurement data to achieve coordinated control of wind turbines. However, due to the complex and variable factors within a wind farm, such as wind speed, wind direction, and terrain, these traditional methods have limitations in optimizing large-scale wind farms.

[0003] Existing technologies include wind farm control methods based on physical models. The current mainstream approach utilizes wake models and wind turbine performance curves to optimize overall wind farm performance by adjusting wind turbine power output and unit angles. For example, tools such as FLORIS are used to simulate flow field variations within a wind farm, combined with traditional optimization algorithms for control adjustments. These methods are effective for small-scale wind farms, but as wind farms scale, computational complexity increases significantly and accuracy decreases.

[0004] There are also control strategies based on machine learning. In this approach, some research has begun to explore the application of machine learning techniques to wind farm control, particularly using reinforcement learning to optimize wind turbine control strategies. While these methods can achieve good optimization results under specific wind farm conditions, they typically rely on large amounts of historical data for training and have limited ability to model the complex fluid dynamics within wind farms and the interactions between wind turbines.

[0005] From the above analysis, it can be seen that the physical model-based approach relies on detailed wake and flow field simulations, which has high computational complexity. Especially in large-scale wind farms, the response speed is slow in real-time application and it is difficult to cope with complex dynamic environments. Traditional control strategies based on physical models or machine learning have limited performance in dealing with nonlinear interactions between wind turbines. In particular, the wake effects between different wind turbines in a wind farm are difficult to accurately model, resulting in poor overall performance optimization. Existing technologies often focus on optimizing single machines or local areas, lacking attention to the global optimality of the entire wind farm, resulting in a failure to maximize overall efficiency. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology and provide a rapid wind farm power optimization method and system based on graph neural networks, aiming to solve the problems of computational complexity, poor real-time performance, and insufficient global optimization capabilities in the existing technology in dealing with wind farm field-level optimization; by introducing the advanced modeling and optimization capabilities of graph neural networks, the overall power generation efficiency of the wind farm can be significantly improved, and the impact of wake between wind turbines can be reduced.

[0007] The technical solution adopted by the rapid wind farm power optimization method based on graph neural network described in the present invention is a rapid wind farm power optimization method based solely on graph neural network, which includes the following steps:

[0008] Step a: Construct a graph neural network model: Use a graph neural network to model the relationship between wind turbines in a wind farm. Nodes represent wind turbines and their states, and edges represent interactions between wind turbines. Input historical data on wind turbine coordinates, wind speed, wind direction, and yaw angle into the graph neural network model. The model is trained by inputting historical data to obtain a graph neural network model for wind farm power.

[0009] Step b, fast power calculation: obtaining real-time data on the coordinates, wind speed, wind direction, and yaw angle of each wind turbine and inputting them into the graph neural network model. The graph neural network model captures the dynamic influence between wind turbines and outputs the power prediction value of each wind turbine;

[0010] Step c, combining genetic algorithm to optimize yaw angle: Using genetic algorithm under fixed coordinate and wind speed conditions, through multiple iterations, the power prediction value output by the graph neural network model is used as the fitness function of the genetic algorithm. Through selection, crossover, and mutation operations, the optimal yaw angle under fixed wind speed and direction conditions is solved;

[0011] Step d, optimizing the power output of the wind turbine: adjusting the yaw angle of the corresponding wind turbine according to the optimal yaw angle obtained by the genetic algorithm, and achieving precise control of the wind turbine through the wind turbine's electronic control system to achieve optimal operating conditions;

[0012] Step e. Real-time data update and model adaptability: Feedback the real-time measured information to the graph neural network model to promote online learning and model update, and maintain the adaptability of the model.

[0013] The system for implementing the above-mentioned rapid wind farm power optimization method based on graph neural network of the present invention includes:

[0014] A data input module for receiving data input;

[0015] A graph neural network model for processing input data from the data input module, establishing mutual influence relationships between wind turbines, and predicting the power output of each wind turbine;

[0016] A genetic algorithm optimization module is used to iteratively optimize the power data output by the graph neural network model using a genetic algorithm to find the optimal yaw angle under fixed wind speed and direction conditions;

[0017] a control system for receiving the optimal yaw angle output by the genetic algorithm optimization module, adjusting the yaw setting of the wind turbine, and achieving precise control of the wind turbine through the wind turbine's electronic control device to ensure optimal operating conditions; and

[0018] A feedback module is used to feed back the real-time data received by the data input module to the graph neural network model, promote online learning and model updating of the graph neural network model, and maintain the adaptability of the model.

[0019] The data input module is a sensor or a meteorological data interface.

[0020] The architecture of the graph neural network model is:

[0021] Input layer: used to accept data input of wind turbine coordinates, wind speed, wind direction and yaw angle;

[0022] Hidden layer: has several graph convolutional layers to capture the relationship between wind turbines; and

[0023] Output layer: used to output the power prediction results of each wind turbine.

[0024] The working process of the genetic algorithm optimization module is as follows:

[0025] I. Initialize the population: randomly generate yaw angle combinations;

[0026] II. Fitness evaluation: Using the power data predicted by the graph neural network model as the fitness function, evaluate the performance of each combination in step I and obtain the fitness value of each combination;

[0027] III. Selection, crossover, and mutation operations: Genetic operations are performed based on the fitness value. After multiple iterative updates, the optimal yaw angle is finally obtained.

[0028] The beneficial effects of the present invention are as follows: in the present invention, due to the use of a power model of a graph neural network, the input data from the data input module is processed, the mutual influence relationship between wind turbines is established, the power output of each wind turbine is predicted, and the genetic algorithm is used under the conditions of fixed coordinates and wind speed. Through multiple iterations, the power prediction value output by the graph neural network model is used as the fitness function of the genetic algorithm, and the optimal yaw angle under the conditions of fixed wind speed and wind direction is solved through the operations of selection, crossover, and mutation; according to the optimal yaw angle obtained by the genetic algorithm, the yaw angle of the corresponding wind turbine is adjusted, and the wind turbine is accurately controlled through the wind turbine's electrical control system to achieve the optimal operating state; this significantly reduces the cost of the calculation. This advantage enables wind farms to quickly respond to environmental changes and optimize operational efficiency. By modeling the interactions between wind turbines through graph neural networks, the accuracy of power prediction is improved. At the same time, by feeding back real-time measured information to the graph neural network model, online learning and model updates are promoted, and the adaptability of the model is maintained, so that the model has the ability to update data in real time, can quickly adapt to changes in wind speed and direction, and ensure long-term operational stability. Combined with the optimization scheme of the genetic algorithm, the optimal yaw angle under specific conditions is quickly solved. Compared with traditional methods, this scheme is more flexible and adaptable, and can improve the energy output of the entire wind farm. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a simplified structural block diagram of the system of the present invention;

[0030] Figure 2 This is a simplified architecture diagram of the graph neural network model;

[0031] Figure 3 It is a simple flow chart of the method of the present invention. DETAILED DESCRIPTION

[0032] In this invention, all modeling and iterative calculation processes are implemented on a computer. The graph neural network model and genetic algorithm used are both mature mathematical computational models. This invention replaces the traditional wake model with a wind farm power model trained using a graph neural network, thereby improving the wind farm power optimization rate.

[0033] like Figures 1 to 3 As shown, a fast wind farm power optimization method based on graph neural network includes the following steps:

[0034] Step a: Construct a graph neural network model: Use a graph neural network to model the relationship between wind turbines in a wind farm. Nodes represent wind turbines and their states, and edges represent interactions between wind turbines. Input historical data on wind turbine coordinates, wind speed, wind direction, and yaw angle into the graph neural network model. The model is trained by inputting historical data to obtain a graph neural network model for wind farm power.

[0035] Step b, fast power calculation: obtaining real-time data on the coordinates, wind speed, wind direction, and yaw angle of each wind turbine and inputting them into the graph neural network model. The graph neural network model captures the dynamic influence between wind turbines and outputs the power prediction value of each wind turbine;

[0036] Step c, combining genetic algorithm to optimize yaw angle: Using genetic algorithm under fixed coordinate and wind speed conditions, through multiple iterations, the power prediction value output by the graph neural network model is used as the fitness function of the genetic algorithm. Through selection, crossover, and mutation operations, the optimal yaw angle under fixed wind speed and direction conditions is solved;

[0037] Step d, optimizing the power output of the wind turbine: adjusting the yaw angle of the corresponding wind turbine according to the optimal yaw angle obtained by the genetic algorithm, and achieving precise control of the wind turbine through the wind turbine's electronic control system to achieve optimal operating conditions;

[0038] Step e. Real-time data update and model adaptability: Feedback the real-time measured information to the graph neural network model to promote online learning and model update, and maintain the adaptability of the model.

[0039] The above method is implemented by a rapid wind farm power optimization system based on a graph neural network, which includes a data input module 1 for receiving data input;

[0040] A graph neural network model 2 for processing input data from the data input module 1, establishing the mutual influence relationship between wind turbines, and predicting the power output of each wind turbine;

[0041] A genetic algorithm optimization module 3 is used to perform iterative optimization using a genetic algorithm based on the power data output by the graph neural network model 2 to solve the optimal yaw angle under fixed wind speed and wind direction conditions;

[0042] a control system 4 for receiving the optimal yaw angle output by the genetic algorithm optimization module 3, adjusting the yaw setting of the wind turbine, and achieving precise control of the wind turbine through the wind turbine's electronic control device to ensure optimal operating conditions; and

[0043] A feedback module 5 is used to feed back the real-time data received by the data input module 1 to the graph neural network model 2, promote the online learning and model update of the graph neural network model 2, and maintain the adaptability of the model.

[0044] Specifically, the data input module 1 is a sensor or a meteorological data interface. Here, the sensor includes various sensors, such as wind speed sensor, temperature sensor, humidity sensor, angle sensor, direction sensor, etc., which respectively detect the wind speed, temperature and humidity, and yaw angle of the fan blades in the environment where the fan is located.

[0045] The architecture of the graph neural network model 2 is:

[0046] Input layer 21: used to receive data input of wind turbine coordinates, wind speed, wind direction and yaw angle;

[0047] Hidden layer 22: has several graph convolutional layers to capture the relationship between wind turbines; and

[0048] Output layer 23: used to output the power prediction results of each wind turbine.

[0049] The working process of the genetic algorithm optimization module 3 is as follows:

[0050] I. Initialize the population: randomly generate yaw angle combinations;

[0051] II. Fitness evaluation: Using the power data predicted by the graph neural network model 2 as the fitness function, evaluate the performance of each combination in step I and obtain the fitness value of each combination;

[0052] III. Selection, crossover, and mutation operations: Genetic operations are performed based on the fitness value. After multiple iterative updates, the optimal yaw angle is finally obtained.

[0053] The technical solution of the present invention is applicable to the intelligent control system of large wind farms. The following are specific examples of the best implementation method:

[0054] 1. Implementation environment: A large offshore wind farm with multiple wind turbines and frequent changes in wind speed and direction;

[0055] 2. Data input: Use a weather station and wind speed sensor to obtain wind speed (12 m / s) and wind direction (45°) in real time. Wind turbine coordinate data is obtained through the GPS system.

[0056] 3. Run the trained graph neural network model 2: Input the data into the graph neural network model 2 to predict the power of each wind turbine. The predicted power of wind turbine A is 1.5 MW, and the power of wind turbine B is 1.2 MW.

[0057] 4. Optimization process of genetic algorithm optimization module 3: The yaw angle is optimized using the genetic algorithm optimization module. After several generations of iteration, the optimal yaw angle is finally obtained (the optimal yaw angle of wind turbine A is 30°, and that of wind turbine B is 25°);

[0058] 5. Control system adjustment: The control system receives the optimal yaw angle command and adjusts the yaw setting of the wind turbine;

[0059] 6. Real-time feedback and model updates: The system feeds new wind speed and direction data back to the graph neural network model 2 for online learning to ensure that the model adapts to future changes.

[0060] Through this technical solution, the present invention can significantly improve the power output and operating efficiency of wind farms, adapt to complex meteorological conditions, and provide technical support for the efficient use of renewable energy.

[0061] Finally, it should be emphasized that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A fast wind farm power optimization method based on graph neural network, characterized in that: The method comprises the following steps: Step a: Construct a graph neural network model: Use a graph neural network to model the relationship between wind turbines in a wind farm. Nodes represent wind turbines and their states, and edges represent interactions between wind turbines. Input historical data on wind turbine coordinates, wind speed, wind direction, and yaw angle into the graph neural network model. The model is trained by inputting historical data to obtain a graph neural network model for wind farm power. Step b, fast power calculation: obtaining real-time data on the coordinates, wind speed, wind direction, and yaw angle of each wind turbine and inputting them into the graph neural network model. The graph neural network model captures the dynamic influence between wind turbines and outputs the power prediction value of each wind turbine; Step c, combining genetic algorithm to optimize yaw angle: Using genetic algorithm under fixed coordinate and wind speed conditions, through multiple iterations, the power prediction value output by the graph neural network model is used as the fitness function of the genetic algorithm. Through selection, crossover, and mutation operations, the optimal yaw angle under fixed wind speed and direction conditions is solved; Step d, optimizing the power output of the wind turbine: adjusting the yaw angle of the corresponding wind turbine according to the optimal yaw angle obtained by the genetic algorithm, and achieving precise control of the wind turbine through the wind turbine's electronic control system to achieve optimal operating conditions; Step e. Real-time data update and model adaptability: Feedback the real-time measured information to the graph neural network model to promote online learning and model update, and maintain the adaptability of the model.

2. A system for implementing the method for rapid wind farm power optimization based on graph neural network according to claim 1, characterized in that: The system includes A data input module (1) for receiving data input; A graph neural network model (2) for processing input data from the data input module (1), establishing a mutual influence relationship between wind turbines, and predicting the power output of each wind turbine; A genetic algorithm optimization module (3) for iteratively optimizing the power data output by the graph neural network model (2) using a genetic algorithm to find the optimal yaw angle under fixed wind speed and wind direction conditions; A control system (4) for receiving the optimal yaw angle output by the genetic algorithm optimization module (3), adjusting the yaw setting of the wind turbine, and achieving precise control of the wind turbine through the wind turbine's electronic control device to ensure optimal operating conditions; as well as A feedback module (5) is used to feed back the real-time data received by the data input module (1) to the graph neural network model (2), promote online learning and model updating of the graph neural network model (2), and maintain the adaptability of the model.

3. The rapid wind farm power optimization system based on graph neural network according to claim 2 is characterized in that: The data input module (1) is a sensor or a meteorological data interface.

4. The rapid wind farm power optimization system based on graph neural network according to claim 2 is characterized in that: The architecture of the graph neural network model (2) is: Input layer (21): used to receive data input of wind turbine coordinates, wind speed, wind direction and yaw angle; Hidden layer (22): has several graph convolutional layers to capture the relationship between wind turbines; and Output layer (23): used to output the power prediction results of each wind turbine.

5. The rapid wind farm power optimization system based on graph neural network according to claim 2 is characterized in that: The working process of the genetic algorithm optimization module (3) is as follows: I. Initialize the population: randomly generate yaw angle combinations; II. Fitness evaluation: Using the power data predicted by the graph neural network model (2) as the fitness function, evaluate the performance of each combination in step I and obtain the fitness value of each combination; III. Selection, crossover, and mutation operations: Genetic operations are performed based on the fitness value. After multiple iterative updates, the optimal yaw angle is finally obtained.

Images