Wind generating set yaw control method based on wind direction prediction
Through multi-dimensional meteorological data prediction and nonlinear optimization models, forward-looking yaw control of wind turbines is achieved, which solves the problems of response lag and mechanical wear, and improves wind energy capture efficiency and equipment life.
Patent Information
- Application Number
- CN202511216859.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-28
- Publication Date
- 2025-10-10
AI Technical Summary
Wind turbines have problems in yaw control such as response lag, accumulated wind deviation, loss of power generation efficiency and increased mechanical wear, and traditional methods lack the means to predict multi-dimensional meteorological data.
By collecting multi-dimensional meteorological data such as wind speed, wind direction, temperature, air pressure, humidity, radiation intensity and visibility, a wind speed and direction mapping model is established, and predictions are made using convolutional neural networks. The optimal cabin position is calculated in combination with a nonlinear optimization model to achieve active yaw control.
Significantly improve wind energy capture efficiency, reduce yaw deviation, extend the life of key components, and enhance the system's operational stability and power generation efficiency under complex meteorological conditions.
Smart Images
Figure CN120759698A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of intelligent control of wind turbines in the wind power industry, and in particular to a yaw control method for wind turbines based on wind direction prediction. BACKGROUND
[0002] As the core equipment of wind power generation, wind turbines play an important role in the transformation of energy structure. The power generation of wind turbines is affected by many factors, among which the yaw-to-wind accuracy is crucial.
[0003] The yaw system of wind turbines mainly relies on precise sensors and advanced control systems to achieve rapid response and adjustment to changes in wind direction, to ensure that the blades are always in the best wind- receiving position. However, in actual operation, due to the instantaneous changes in wind speed and direction in the natural environment, as well as the complexity of the terrain, surrounding obstacles and other factors, wind turbines often cannot always maintain the optimal yaw position in theory, resulting in a certain yaw deviation. This deviation not only reduces the efficiency of converting wind energy into electrical energy and reduces power generation, but also increases the wear and tear of mechanical components due to asymmetric loads, affecting the overall life of the equipment.
[0004] That is, current wind turbines generally rely on real-time wind direction sensors for yaw control, which has significant limitations:
[0005] 1. Response lag: Wind direction changes instantaneously, and adjusting the nacelle position based on current data alone cannot keep up with the rapid changes in wind direction;
[0006] 2. Accumulation of yaw deviation: Frequent small yawing can cause error accumulation, leading to long-term deviation of the nacelle from the optimal yaw direction;
[0007] 3. Loss of power generation efficiency: Inaccurate yawing directly reduces the efficiency of wind energy capture, resulting in a loss of power generation;
[0008] 4. Increased mechanical wear and tear: Continuous yawing deviation causes asymmetric loads, increasing the risk of fatigue damage to key components such as the drive chain;
[0009] 5. Single prediction method: Traditional methods rarely use multi-dimensional meteorological data (such as temperature, humidity, pressure, and cloud conditions) to predict future changes in wind direction to optimize control.
[0010] Therefore, improvement and innovation are imperative. SUMMARY
[0011] In view of the above situation, in order to overcome the shortcomings of the prior art, the purpose of the present application is to provide a yaw control method for wind turbines based on wind direction prediction, which solves the problem of yaw optimization control for wind turbines based on wind direction prediction.
[0012] The technical scheme solved by the present application is:
[0013] A yaw control method for a wind turbine based on wind direction prediction, comprising the following steps:
[0014] Step 1: Collect and establish a data set of continuously changing wind speed, wind direction, air temperature, air pressure, humidity, radiation intensity, visibility, and cloud cover ratio of the wind farm;
[0015] Step 2: Establish a wind speed and wind direction mapping model for a continuous time period before and after yaw action of the wind farm;
[0016] Step 3: Input the data set of wind speed, wind direction, air temperature, air pressure, humidity, radiation intensity, visibility, and cloud cover ratio of a continuous time period before yaw action into the optimal nacelle position direction calculation model to obtain the optimal nacelle position direction;
[0017] Step 5: Calculate the yaw angle of the wind turbine based on the current nacelle position and the optimal nacelle position direction to optimize yaw control.
[0018] Preferably, in step 1, the data set at least includes continuously changing wind speed V=(v1, …, v n}, wind direction D=(d1, …, d n}, air temperature T=(t1, …, t n}, air pressure P={p1, …, p n}, humidity H=(h1, …, h n}, radiation intensity R={r1, …, r n}, visibility L=(l1, …, l n}, and cloud cover ratio C=(c1, …, c n}, and the time interval of each data is not greater than 1 minute.
[0019] Preferably, in step 1, the data set is sorted into model input parameters and model prediction parameters, the input parameters X={x1, …, x n}, and x i =(V i , D i , T i , P i , H i , R i , L i , C i}, V i , D i , T i , P i , H i, R i , L i , C i are respectively the wind speed, wind direction, air temperature, air pressure, humidity, radiation intensity, visibility, cloud cover ratio data within 5 minutes, the prediction parameter Y = {y1, …, y n}, y i = (V ave,i , D ave,i}. V ave，i , D ave，i are the i-th predicted wind speed, wind direction; the sample set is randomly divided into a training set {X train , Y train} and a test set {X test , Y test} according to an 8:2 ratio.
[0020] Preferably, in step 2, the wind speed and wind direction mapping model is a convolutional neural network;
[0021] The model includes a CNN layer, a BN layer, a Flatten layer, a Droupout layer, a pooling layer, and a fully connected layer;
[0022] Model evaluation index:
[0023] Mean square error MSE:
[0024] Determination coefficient R 2 :
[0025] Where y i is the actual observation value, is the model prediction value, is the average value of the actual observation value, and n is the number of prediction samples;
[0026] The model training method is as follows:
[0027] S1: For the training set X train as an input parameter, Y train as a prediction parameter, neural network training is performed, and when the mean square error MSE and the determination coefficient R 2 , MSE is less than 0.2, and R 2 is greater than 0.99, the training is ended;
[0028] S2: The test set X test is used as an input parameter, Y test is used as a prediction parameter, and the neural network is input for calculation, and MSE and R 2 are calculated;
[0029] S3: When the test set MSE is less than 0.3, and R 2When the training is completed, the test set cannot simultaneously satisfy the MSE and R 2 Return S1 to retrain the model when required.
[0030] Preferably, in step 3, the wind speed, wind direction, air temperature, air pressure, humidity, radiation intensity, visibility, and cloud cover data set data in a continuous time period (such as 25 minutes) before the yaw action are input into the wind speed and wind direction mapping model to obtain the wind speed and wind direction prediction data Y = {y1, …, y n} in the time period after the yaw action. i = (V ave，i , D ave，i} ; V ave,i , D ave，i are the ith predicted wind speed and wind direction.
[0031] Preferably, in step 4, the optimal nacelle position direction calculation model is a nonlinear optimization model, and the objective function is:
[0032]
[0033] wherein E is the maximum value of the single-machine integrated wind energy capture when the predicted wind speed and wind direction are fixed, p is the air density, V ave，j is the predicted wind speed at the jth time point, D ave,j is the predicted wind direction at the jth time point, q opt is the optimal nacelle position direction, A is the wind sweeping area of the unit, and C p is the wind energy utilization coefficient of the unit; and the constraint condition is:
[0034] q low ≤ q opt ≤ q high
[0035] wherein q low is the lower limit angle of the nacelle position, and q high is the upper limit angle of the nacelle position.
[0036] Preferably, in step 4, q opt is traversed at an interval of 0.5 degrees, and the nonlinear optimization model is solved to obtain the optimal nacelle position direction q opt when E is the maximum value.
[0037] Preferably, in step 5, the optimal nacelle position direction q opt when E is the maximum value is used to obtain the final target yaw angle as:
[0038] q opt
[0039] Wherein, θ0 is the current nacelle position, when the unit triggers the yaw action, the yaw action is performed according to the yaw angle Δθ of the fan, and the yaw control optimization is performed.
[0040] The application can predict the wind speed and wind direction in real time through real-time meteorological data in the wind farm area, and the predicted wind speed and wind direction at the position of the unit are used to realize active yaw control through multi-parameter wind direction prediction and nonlinear optimization, so that the wind power unit maximizes the use of wind resources. The application innovatively integrates 8-dimensional meteorological data such as wind speed, wind direction, air temperature, air pressure, humidity, radiation intensity, visibility, and cloud coverage ratio as input to build a more comprehensive short-term wind speed and wind direction prediction model; it is a breakthrough to use the predicted future wind speed and wind direction sequence as input to build a nonlinear optimization model to solve the optimal nacelle direction corresponding to the maximum comprehensive wind energy capture in the future period, and to realize the yaw control based on the prediction; the theoretical maximum wind energy capture E calculated from the predicted wind speed and wind direction sequence is clearly defined as the optimization objective function, and its mathematical expression clearly reflects the physical relationship between wind energy capture efficiency, nacelle direction, and predicted wind direction; through accurate prediction of future wind conditions and active adjustment of the optimal nacelle direction, the deviation of the yaw-to-wind is significantly reduced, and the single-machine wind energy capture efficiency is maximized; unnecessary yaw actions and uneven loads caused by inaccurate yaw-to-wind are reduced, effectively prolonging the service life of key components such as yaw bearings and gearboxes; the use of multi-dimensional meteorological data prediction makes the system more adaptable to complex weather conditions (such as sudden wind, shear, and cloud changes), improving the operation stability in complex terrain or weather-varying areas; based on the optimal comprehensive energy capture in the future period, the short-sightedness and frequent small adjustments caused by responding to the instantaneous wind direction are avoided, making the yaw control more forward-looking and globally optimal; that is, the application provides a data-driven and prediction-oriented wind turbine yaw control method, which realizes the forward-looking optimization and adjustment of the nacelle direction by deeply integrating multi-source meteorological data, training a high-precision CNN prediction model, building and solving a wind energy capture maximization nonlinear optimization problem, and finally achieving the core goal of improving power generation efficiency and reducing equipment wear. The technical solution is specific, implementable, and effective, and is an innovation in the field of wind turbine yaw control methods. BRIEF DESCRIPTION OF DRAWINGS
[0041] Figure 1 The method flowchart of the application is shown. DETAILED DESCRIPTION
[0042] The specific implementation of the application is further described in detail below in combination with the drawings and examples.
[0043] As Figure 1 shown is a wind turbine yaw control method based on wind direction prediction, the steps include:
[0044] S1: Collect and establish the wind speed, wind direction, air temperature, air pressure, humidity, radiation intensity, visibility, cloud cover ratio data set of the wind farm continuously changing.
[0045] Step 2: Establish the wind speed, wind direction, air temperature, air pressure, humidity, radiation intensity, visibility, cloud cover ratio and wind speed wind direction mapping model within 25 minutes before the yaw action of the wind farm and 15 minutes after the yaw action;
[0046] Step 3: Input the yaw action yaw action 25 minutes before the yaw action wind speed, wind direction, air temperature, air pressure, humidity, radiation intensity, visibility, cloud cover ratio data set into the wind speed wind direction mapping model to obtain the wind speed wind direction prediction data within the time period after the yaw action;
[0047] S 4: Input the predicted wind speed wind direction data into the optimal nacelle position direction calculation model to obtain the optimal nacelle position direction.
[0048] S 5: Calculate the yaw angle of the wind turbine according to the current nacelle position and the optimal nacelle position direction to perform yaw control optimization.
[0049] The following is a specific embodiment for predicting wind speed and wind direction in the short term using a convolutional neural network method, and optimizing yaw control of the wind farm based on the predicted wind direction and wind speed:
[0050] Step 1.1: Collect wind speed V = {v1, …, v n}, wind direction D = {d1, …, d n}, air temperature T = {t1, …, t n}, humidity H = {h1, …, h n}, wind farm air pressure P = {p1, …, p n}, wind farm radiation intensity R = {r1, …, r n}, wind farm visibility L = {l1, …, l n}, and cloud cover ratio C = {c1, …, c n} for 10 minutes continuously through wind turbine SCADA data.
[0051] Step 1.2: Construct a wind speed and wind direction prediction data set, including input parameters X = {x1, …, x n}, x i = {V i , D i , T i , P i , H i , R i , Li , C i}, V i , D i , T i , P i , H i , R i , L i , C i are respectively the continuous wind speed, wind direction, air temperature, air pressure, humidity, radiation intensity, visibility, cloud cover ratio data within 5 minutes; the prediction parameter Y = {y1, …, y n}, y i = {V ave，i , D ave，i} ; V ave，i , D ave，i are the i-th predicted wind speed, wind direction; the sample set is randomly divided into a training set {X train , Y train} and a test set {X test , Y test} according to an 8:2 ratio.
[0052] Step 2.1: Establish a neural network model as:
[0053] Layer 1: Conv1D layer, number of convolution kernels 4, convolution kernel size 5, step size 5, activation function ReLU, input shape (8, 120);
[0054] Layer 2: BatchNormalization layer;
[0055] Layer 3: MaxPooling1D layer;
[0056] Layer 4: CNN layer, number of convolution kernels 8, convolution kernel size 5, step size 5, activation function ReLU;
[0057] Layer 5: BatchNormalization layer;
[0058] Layer 6: MaxPooling1D layer;
[0059] Layer 7: CNN layer, number of convolution kernels 16, convolution kernel size 4, step size 2, activation function ReLU;
[0060] Layer 8: BatchNormalization layer;
[0061] Layer 9: MaxPooling1D layer;
[0062] Layer 10: Dropout layer, dropout = 0.3;
[0063] Layer 11: Dense layer, number of neurons: 64, activation function: ReLu;
[0064] Layer 12: Dense layer, number of neurons: 2, activation function: ReLu;
[0065] Step 2.2: Model training, for training set X train As input parameters, Y train As a prediction parameter, the neural network model is trained for 300 epochs. If the accuracy is higher than 0.97, it meets the requirements, otherwise repeat the training.
[0066] Step 2.3: Model verification, test set X test As input parameters, Y test As a prediction parameter, input the trained neural network model for calculation. Compare the model output results. If the accuracy is greater than 0.95, the model meets the requirements, otherwise repeat step 2.2.
[0067] Step 3: Input the continuous data of real-time wind speed, wind direction, air temperature, air pressure, humidity, radiation intensity, visibility, and cloud coverage ratio of the wind farm into the trained model to obtain short-term predicted wind speed and wind direction.
[0068] Step 4: Establish a nonlinear optimization model, whose objective function is:
[0069]
[0070] Where E is the maximum value of wind energy capture of a single machine when the predicted wind speed and wind direction are fixed at the machine cabin position. The constraint conditions are:
[0071] θ low ≤θ opt ≤θ high
[0072] Where θ low is the lower limit angle of the machine cabin position, and θ high is the upper limit angle of the machine cabin position.
[0073] Solve the nonlinear optimization model using the Monte Carlo method to obtain the optimal machine cabin position direction θ opt .
[0074] Step 5: Calculate the final target yaw angle based on the optimal machine cabin position direction and the current machine cabin position, and execute the yaw action to complete the yaw control optimization.
[0075] Example 2: The following is a specific embodiment of neural network training for historical meteorological data and SCADA data of a certain wind farm in Anyang, Henan:
[0076] Step 1: Collect the wind farm's 12-month meteorological data and wind speed and direction data. Randomly divide the first 6 months of data into a training set and a test set in an 8:2 ratio.
[0077] Step 2: Train the neural network model using the training set and test the model using the test set. The test set R2 = 0.988, MSE = 0.23, and the model training is qualified.
[0078] Step 3: Divide the remaining six months of data into 1-minute intervals, input them into the neural network model, and optimize the yaw angle.
[0079] Step 4: Calculate the theoretical power generation after yawing based on the optimized yaw angle. Compared with the actual power generation, the theoretical power generation is on average increased by 0.23%.
Claims
1. A yaw control method for a wind turbine generator set based on wind direction prediction, characterized in that: The following steps are involved: Step 1: Collect and establish a dataset of continuously changing wind speed, wind direction, temperature, air pressure, humidity, radiation intensity, visibility, and cloud cover ratio for the wind farm; Step 2: Establish a mapping model between the wind speed, wind direction, temperature, air pressure, humidity, radiation intensity, visibility, and cloud cover ratio in a continuous time period before the yaw action of the wind farm and the wind speed and direction in a continuous time period after the yaw action; Step 3: Input the wind speed, wind direction, temperature, air pressure, humidity, radiation intensity, visibility, and cloud cover ratio data set for the continuous time period before the yaw action into the wind speed and direction mapping model to obtain the wind speed and direction prediction data for the time period after the yaw action; Step 4: Input the predicted wind speed and direction data into the optimal cabin position and direction calculation model to obtain the optimal cabin position and direction; Step 5: Calculate the yaw angle of the wind turbine based on the current nacelle position and the optimal nacelle position direction, and perform yaw control optimization.
2. The yaw control method for a wind turbine generator set based on wind direction prediction according to claim 1, characterized in that: In step 1, the data set includes at least the continuously changing wind speed V={v1, ..., v n }, wind direction D = {d1,…,d n }, temperature T={t1,…,t n }, air pressure P={p1,…,p n }, humidity H={h1,…,h n }, radiation intensity R={r1,…,r n }, visibility L = {l1,…,l n }, cloud cover ratio C={c1,…,c n }, the time interval of each data is no more than 1 minute.
3. The yaw control method for a wind turbine generator set based on wind direction prediction according to claim 2, characterized in that: In step 1, the data set is organized into model input parameters and model prediction parameters. Input parameter X = {x1, ..., x n }, x i ={V i , D i , T i , P i , H i , R i , L i , C i }, V i , D i , T i , P i , H i ,R i , L i , C i They are the continuous wind speed, wind direction, temperature, air pressure, humidity, radiation intensity, visibility, and cloud cover ratio data within 5 minutes, and the prediction parameters Y = {y1, ..., y n },y i ={V ave,i , D ave，i }. V ave，i , D ave，i The wind speed and direction of the ith prediction are randomly divided into training set {X train , Y train } and the test set {X test , Y test }.
4. The yaw control method for a wind turbine generator set based on wind direction prediction according to claim 3, characterized in that: In step 2, the wind speed and direction mapping model is a convolutional neural network; The model training method is as follows: S1: for training set X train As an input parameter, Y train As the prediction parameter, the neural network training is carried out and the root mean square error MSE and the determination coefficient R are calculated. 2 , MSE is less than 0.2, R 2 The training ends when it is greater than 0.99; S2: Test set X test As an input parameter, Y test As the prediction parameters, input the neural network for calculation, and calculate the MSE and R 2 ; S3: When the test set MSE is less than 0.3, R 2 The training is completed when it is greater than 0.
98. When the test set cannot meet the MSE and R 2 Return to S1 to retrain the model when required.
5. The yaw control method for a wind turbine generator set based on wind direction prediction according to claim 4, characterized in that: In step 3, the wind speed, wind direction, temperature, air pressure, humidity, radiation intensity, visibility, and cloud cover ratio data set in the continuous time period before the yaw action are input into the wind speed and direction mapping model to obtain the wind speed and direction prediction data Y={y1, ..., y n },y i =(V ave，i , D ave，i }; V ave，i ,D ave，i is the ith predicted wind speed and direction.
6. The yaw control method for a wind turbine generator set based on wind direction prediction according to claim 5, characterized in that: The optimal cabin position and direction calculation model in step 4 is specifically to establish a nonlinear optimization model, and its objective function is: Where E is the maximum value of wind speed and direction predicted for a single machine when the nacelle position and direction are fixed, ρ is the air density, and V ave，j is the predicted wind speed at the jth time point, D ave,j is the predicted wind direction at the jth time point, θ opt is the optimal cabin position direction, A is the wind sweep area of the unit, C p is the wind energy utilization coefficient of the unit; the constraints are: i low ≤θ opt ≤θ high Among them, θ low is the lower limit angle of the cabin position, θ high is the cabin upper limit angle; 7. The yaw control method for a wind turbine generator set based on wind direction prediction according to claim 6, characterized in that: In step 4, the nonlinear optimization model is solved to find the optimal cabin position direction θ when E is the maximum value. opt .
8. The yaw control method for a wind turbine generator set based on wind direction prediction according to claim 7, characterized in that: The step 5 is specifically as follows: using the optimal cabin position direction q when E is the maximum value opt , the final target yaw angle is obtained as: Δθ=θ0-θ opt Among them, θ0 is the current nacelle position. When the unit triggers the yaw action, the yaw action is performed according to the yaw angle Δθ of the wind turbine to optimize the yaw control.