A wind turbine clearance prediction control method

Abstract

The application provides a wind turbine clearance prediction control method, which first collects historical clearance data in a recent period of time, including clearance data cl h , a pitch angle alpha and an impeller rotating speed omega, then obtains a clearance prediction result through a machine learning method or a bin query prediction method, and finally takes a unit protection action control according to the clearance prediction result. The wind turbine clearance prediction control method provided by the application can predict the clearance value based on short-term clearance data and working condition data, so as to realize timely control intervention in an extremely severe working condition, prevent the tower scanning risk caused by continuous clearance decline, realize early intervention of the unit, and guarantee safe and stable operation of the unit.

Description

Technical Field

[0001] This invention relates to the field of wind turbine airspace monitoring technology, and in particular to a wind turbine airspace prediction and control method. Background Technology

[0002] With advancements in wind power technology, the size of wind turbine blades is continuously increasing. Larger blades can capture more wind energy, improving power generation efficiency. From blades that were tens of meters long to nearly a hundred meters long, blades are also trending towards lighter and more flexible designs. Simultaneously, to obtain higher wind speeds and more stable wind energy resources, the height of wind turbine towers is gradually increasing. Higher towers can improve power generation efficiency and allow the turbines to operate at higher altitudes. All these trends pose challenges to the design and safety margins of wind turbine clearance. To address the potential for low clearance or tower sweeping, major turbine manufacturers have developed and implemented clearance monitoring methods to provide early warnings and protect the turbines.

[0003] Current methods for monitoring air clearance are all based on real-time measurements. On the one hand, due to the measurement, calculation, and communication of the system, there is generally a delay of several hundred milliseconds. On the other hand, when encountering extreme and severe operating conditions, the air clearance drops significantly. When the wind turbine main control system receives a dangerous air clearance signal, the system may have just taken countermeasures, and the blade air clearance may not be suppressed in a timely and effective manner, resulting in a tower sweeping accident. This lag is a shortcoming of the current unit protection function based on real-time air clearance.

[0004] Some studies have attempted to predict wind headroom based on environmental data such as wind speed and direction. Historical data is input into relatively mature models such as neural networks to calculate and predict wind turbine headroom, and then control interventions are implemented based on these results. However, this approach also has several problems: First, current wind speed prediction methods rely on nacelle-mounted wind-measuring radar to accurately measure wind speeds more than 100 meters away, enabling feedforward control of wind turbines. However, wind speed, direction, and shear patterns are poorly defined, making accurate measurement and the summarization of precise patterns difficult. It is challenging to combine these patterns with turbine status information for accurate prediction, even with integrated blade load sensors. Second, neural network models often require incorporating too much uncertain information and employ coarse-grained control based on average values. Therefore, training a neural network model with so many variables easily leads to overfitting, resulting in decreased generalization performance on new data. Third, during the prediction process, the neural network model needs to perform a series of matrix operations and activation function calculations on the input data. These calculation operations consume a lot of computing resources and are not convenient to be implemented directly in the wind turbine PLC. They require additional hardware, which increases the cost of the system and the difficulty of integration with the wind turbine PLC. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a wind turbine airspace prediction and control method that can predict airspace values ​​based on short-term airspace data and operating condition data. This enables timely control and intervention under extreme and severe operating conditions, preventing tower sweep risks caused by continuous airspace decline, allowing for early intervention of the unit, and ensuring the safe and stable operation of the unit.

[0006] The technical solution adopted by this invention to solve its technical problem is: a wind turbine headroom prediction control method, comprising the following steps:

[0007] S1. Collect historical airspace data for a recent period, including airspace data cl h The pitch angle α and the impeller speed ω;

[0008] S2. Obtain net short forecast results through machine learning methods or warehouse query forecasting methods;

[0009] S3. Take protective action control measures for the unit based on the airspace forecast results.

[0010] Step S1 Clearance Data cl h The pitch angle α and impeller speed ω were acquired by the headroom measurement system from the wind turbine main control system via the industrial bus.

[0011] If step S2 employs a machine learning method, it specifically includes the following process:

[0012] A1. Load the collected historical airspace data and preprocess the historical airspace data, including missing values, outliers and duplicate values ​​detection and correction;

[0013] A2. Group the historical clearance data according to time sequence. Each group of data includes historical information of three blades during the process of the blade rotating two consecutive revolutions. And the clearance value cl after the first blade rotates twice. A3 The first clearance value of the first blade in the historical information, as well as the pitch angle and impeller speed at the same moment, are represented by cl. A1 α1 and ω1 represent the first clearance value of the second blade, as well as the pitch angle and impeller speed at the same time, respectively, denoted by cl. B1 α2 and ω2 represent the first clearance value of the third blade, as well as the pitch angle and impeller speed at the same time, respectively, denoted by cl. c1 α3 and ω3 represent the second clearance value of the first blade, and the pitch angle and impeller speed at the same time are represented by cl. A2 α4 and ω4 represent the second clearance value of the second blade, as well as the pitch angle and impeller speed at the same time, respectively, denoted by cl. B2 α5 and ω5 represent the second clearance value of the third blade, as well as the pitch angle and impeller speed at the same time, respectively denoted by cl. C2α6 and ω6 represent;

[0014] Then a set of data is used express;

[0015] A3. Based on prior knowledge and expert advice, combine existing features to derive new features, and then standardize the feature data.

[0016] A4. Randomly split each group of data into a training set and a test set in a 7:3 ratio;

[0017] A5. Train the multiple linear regression model on the training set, and use the gradient descent algorithm to optimize the parameters of the linear regression model until the loss rate of the model is reduced to the minimum, thus obtaining the initial net air volume prediction model.

[0018] A6. Use test set data to evaluate the initial airspace prediction model. If the performance is insufficient, use feature engineering and parameter optimization methods to improve the performance until the best-performing airspace prediction model is obtained.

[0019] A7. Use the best-performing airspace prediction model to predict airspace value and obtain airspace prediction results.

[0020] In step A6, the initial net airspace prediction model is evaluated, specifically by using mean squared error, absolute mean error, and coefficient of determination to calculate and evaluate the model, plotting the predicted values ​​against the actual values, and observing the model's predictive performance.

[0021] If the warehouse-based query prediction method is used in step S2, it specifically includes the following process:

[0022] B1. Divide the historical clearance data into compartments according to speed and pitch angle, and use the average clearance data in each compartment as the benchmark value to obtain the speed-pitch-clearance benchmark table.

[0023] B2. Group the historical clearance data according to time sequence. Each group of data includes historical information of three blades during the process of the blades rotating two consecutive revolutions. The first clearance value of the first blade, the pitch angle and impeller speed at the same moment are respectively represented by cl. A1 α1 and ω1 represent the first clearance value of the second blade, as well as the pitch angle and impeller speed at the same time, respectively, denoted by cl. B1 α2 and ω2 represent the first clearance value of the third blade, as well as the pitch angle and impeller speed at the same time, respectively, denoted by cl. C1 α3 and ω3 represent the second clearance value of the first blade, and the pitch angle and impeller speed at the same time are represented by cl. A2 α4 and ω4 represent the second clearance value of the second blade, as well as the pitch angle and impeller speed at the same time, respectively, denoted by cl. B2α5 and ω5 represent the second clearance value of the third blade, as well as the pitch angle and impeller speed at the same time, respectively denoted by cl. C2 α6 and ω6 represent,

[0024] Then any set of data is used express;

[0025] B3. For any set of data d, find the corresponding reference net clearance value (i.e., the i-th reference net clearance value cl) by referring to the pitch angle and impeller speed at each moment in the speed-pitch angle-net clearance reference table. bi According to the i-th propeller pitch angle α i and impeller speed ω i The data d is obtained from the speed-pitch-headroom reference table, and then converted into data.

[0026] B4. According to formula cl qi =cl Ai -cl bi Further transform data d2 into data d3 = [cl q1 cl q2 cl q3 cl q4 cl q5 cl q6 ];

[0027] B5. The predicted headroom wind shear component cl is calculated based on the following empirical formula. q7 :

[0028] cl q7 =k1(cl q4 -cl q1 )+k2(cl q5 -cl q2 )+k3(cl q6 -cl q3 )

[0029] Where k1, k2, and k3 are the first, second, and third decreasing coefficients, respectively, and are preset empirical values.

[0030] B6. Calculate the predicted rotational speed ω7 and the predicted pitch angle α7 using the following formulas:

[0031] α7=2α6-α5

[0032] ω7=2ω6-ω5

[0033] B7. Using the predicted rotational speed ω7 and the predicted pitch angle α7, the predicted reference net clearance value cl is obtained by looking up the predicted reference net clearance value cl in the rotational speed / pitch angle-net clearance reference table. b7 ;

[0034] B8. The predicted net airspace value cl is calculated using the following formula. A3 :

[0035] cl A3 =cl b7 +cl q7 .

[0036] Step S3 determines the decline in airspace based on the predicted airspace value. If the predicted airspace value is higher than the upper limit of the danger zone th1, a mild control strategy is adopted. If the predicted airspace value is higher than 70% of the upper limit of the danger zone, i.e., th2, an emergency control strategy is adopted. If 60% of the actual airspace value is in the transition zone [th2, th1], a mild control strategy is adopted.

[0037] The gentle control strategy is to reduce the propeller speed at θ1 to a safe pitch angle α. e If the trigger condition is not met for 30 seconds, the system will switch off; the emergency control strategy is to retract the propeller at a rate of θ2 to the safe pitch angle α. e If the trigger condition is not met for 30 seconds, the process will switch off.

[0038] The beneficial effects of this invention based on its technical solution are as follows:

[0039] This invention provides a wind turbine headroom prediction and control method that utilizes accurate headroom prediction results and reasonable control logic to identify and detect sharp headroom drops caused by negative shear degradation as early and quickly as possible, effectively controlling them and avoiding tower sweep accidents in extreme situations. Compared with schemes and strategies based on measured headroom protection, this method has greater advantages, achieving this with minimal power generation loss while identifying headroom decline trends in low-probability, low-headroom events. Early detection and intervention provide sufficient time for turbine safety protection to withstand the threats posed by the uncertainty of severe wind conditions. Furthermore, this invention uses historical headroom data and the wind turbine's rotational speed and pitch angle to predict headroom without adding other hardware sensors, effectively controlling prediction costs. Attached Figure Description

[0040] Figure 1 This is a diagram of event one.

[0041] Figure 2 This is a diagram of Event Two.

[0042] Figure 3 This is a diagram of event three.

[0043] Figure 4 This is a schematic diagram showing the relationship between airspace, wind conditions, and unit control parameters.

[0044] Figure 5 This is a diagram illustrating the correlation between wind speed and airspace clearance.

[0045] Figure 6 This is a schematic diagram illustrating the correlation between rotational speed and headroom.

[0046] Figure 7 This is a schematic diagram illustrating the correlation between pitch angle and clearance.

[0047] Figure 8 It is a heat map of the control parameter net clearance reference table.

[0048] Figure 9 This is a time sequence diagram of the airspace clearance and control parameters for the first time period.

[0049] Figure 10 This is the time sequence diagram of the airspace and control parameters for the second time period.

[0050] Figure 11 This is the time sequence diagram of the airspace and control parameters for the third time period.

[0051] Figure 12 This is the time sequence diagram of the airspace and control parameters for the fourth time period.

[0052] Figure 13 This is a comparative diagram showing the difference between the predicted net air volume and the actual net air volume using linear regression machine learning.

[0053] Figure 14 This is a schematic diagram illustrating the prediction bias analysis using linear regression machine learning.

[0054] Figure 15 This is a schematic diagram comparing the predicted rotational speed with the actual measurement using the compartmentalized query prediction method.

[0055] Figure 16 This is a schematic diagram comparing the predicted pitch angle with the actual measurement using the compartment query prediction method.

[0056] Figure 17 This is a schematic diagram comparing the predicted and measured net air shear components using the compartmentalized query prediction method.

[0057] Figure 18 This is a schematic diagram of the net air forecast deviation analysis using the warehouse-based query forecasting method. Detailed Implementation

[0058] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0059] Explanation of the principle:

[0060] Statistical analysis of extensive data on minor airspace clearance incidents revealed that these incidents can be categorized into three types, dividing the clearance area into safe zones, transition zones, and danger zones. (Refer to...) Figure 1 Event 1 involves a slow but significant decrease in clearance, with the clearance gradually decreasing from the safe zone to the danger zone; the decrease in clearance is generally less than 2 meters. (Refer to...) Figure 2 Event two involves a rapid and significant drop in air clearance, with the clearance decreasing slowly and rapidly from the safe zone to the danger zone. This type of situation mostly occurs under severe operating conditions, such as a sharp increase in wind speed or severe negative shear. The clearance drop in nearby areas is generally greater than 2.5 meters. (Refer to...) Figure 3 Event three involves a persistently low airspace clearance in a transitional zone, with occasional jumps into the safe zone. Overall, however, the clearance remains in a risky area. Worsening wind conditions could lead to dangerous clearance conditions, requiring preventative measures. Analysis of historical airspace data... h It reflects the changes in wind conditions in a short period of time. The pitch angle α and the rotor speed ω represent the control characteristics of the unit. The increase in the pitch angle leads to a decrease in the load in the forward and backward directions of the blades, and the deformation in the forward and backward directions of the blades is reduced, thus improving the headroom. The change in rotor speed reflects the change in wind energy in the entire rotor plane, and will also affect the deformation of the blades and the headroom value. Therefore, it is reasonable to predict the headroom using these three variables.

[0061] Based on the above principles, this invention provides a wind turbine headroom prediction and control method, comprising the following steps:

[0062] S1. Collect historical airspace data for a recent period, including airspace data cl h The pitch angle α and impeller speed ω. The net clearance data cl h The pitch angle α and impeller speed ω were acquired by the headroom measurement system from the wind turbine main control system via the industrial bus.

[0063] In practical implementation, it is recommended to use 2 rings of 6 clearances as the historical clearance. This ensures the clearance quantity and links each blade, while also shortening the length of the historical clearance data. Excessively long historical clearances are not conducive to accurately predicting the current clearance, or in other words, have a smaller impact. A wind turbine has 3 blades, which can be understood as ABCABC→A.

[0064] S2. Obtain net airspace prediction results through machine learning methods or sub-warehouse parameter query methods.

[0065] If a machine learning method is used, the specific process includes the following:

[0066] A1. Load the collected historical airspace data into an appropriate data structure and preprocess the historical airspace data, including missing values, outliers and duplicate values ​​detection and correction, thereby improving data quality and ensuring data consistency, accuracy and timeliness.

[0067] A2. Group the historical clearance data according to time sequence. Each group of data includes historical information of three blades during the process of the blade rotating two consecutive revolutions. And the clearance value cl after the first blade rotates twice. A3The first clearance value of the first blade in the historical information, as well as the pitch angle and impeller speed at the same moment, are represented by cl. A1 α1 and ω1 represent the first clearance value of the second blade, as well as the pitch angle and impeller speed at the same time, respectively, denoted by cl. B1 α2 and ω2 represent the first clearance value of the third blade, as well as the pitch angle and impeller speed at the same time, respectively, denoted by cl. C1 α3 and ω3 represent the second clearance value of the first blade, and the pitch angle and impeller speed at the same time are represented by cl. A2 α4 and ω4 represent the second clearance value of the second blade, as well as the pitch angle and impeller speed at the same time, respectively, denoted by cl. B2 α5 and ω5 represent the second clearance value of the third blade, as well as the pitch angle and impeller speed at the same time, respectively denoted by cl. C2 α6 and ω6 represent;

[0068] Then a set of data is used express.

[0069] A3. Based on prior knowledge and expert advice, new features are derived by combining existing features. If the features have different dimensions, feature data standardization can be performed to increase model accuracy and iteration speed.

[0070] A4. Randomly split each group of data into a training set (Atr) and a test set (Ate) in a 7:3 ratio. The training set is used for model training, and the test set is used for model validation and evaluation.

[0071] AS trains a multiple linear regression model on the training set, and uses the gradient descent algorithm to optimize the parameters of the linear regression model until the loss rate of the model is reduced to the minimum, thus obtaining the initial net air volume prediction model.

[0072] The linear regression model is: y i =β0+β1cl A1 +β2cl B1 +…+β 18 ω6+∈ i In the formula y i Let β0 be the target variable for the i-th sample, and β0 be the target variable for the i-th sample. 18 These are the feature weights, i.e., coefficients, ∈ i It is the error term of the model.

[0073] A6. Evaluate the initial net airspace prediction model using test set data. If the performance is insufficient, perform feature engineering and parameter optimization to improve performance until the best-performing net airspace prediction model is obtained. Specifically, mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²) can be used. 2 And so on, and plot the predicted values ​​against the actual values ​​to observe the predictive effect of the model.

[0074] In this embodiment, to minimize the sum of squared residuals, the mean squared error (MSE) is used as the loss function, expressed as:

[0075]

[0076] In the formula, m is the length of the training set Atr. It is the model's prediction for the i-th sample.

[0077] Least squares estimation: Find the model parameters that minimize MSE by solving the following formula.

[0078]

[0079] In the formula, Let the parameter estimates be a vector, in the form of X is a matrix containing the features of all training samples (one sample per row, one feature per column, an m*(n+1) dimensional matrix). The model training result is achieved by solving the above model.

[0080] Using the trained model, run the test set Ate to evaluate the deviation between the predicted and measured results. Evaluate whether the model meets the requirements using metrics such as the mean deviation μ and the root mean square error σ. A probability of μ±2σ within ±0.6m of 95% or higher is considered acceptable. If the model's performance is insufficient, feature engineering and model parameter optimization methods can be tried to improve it.

[0081] A7. Use the best-performing airspace prediction model to predict airspace value and obtain airspace prediction results.

[0082] For wind turbines of different models, tower heights, and rotor diameters, the above process can be used to establish a headroom prediction model.

[0083] Reference Figure 4 and Figure 5 The headroom of a wind turbine is affected by a combination of wind conditions and turbine control parameters. Generally, before reaching the rated wind speed, as wind speed increases, the rotor speed increases, the thrust on the rotor increases, and the headroom decreases. After the wind speed reaches the rated speed, the rotor speed also reaches the rated speed. As the wind speed further increases, the pitch angle increases, the control speed remains constant near the rated speed, and the load on the blades in the forward and backward directions decreases, resulting in an upward trend in headroom.

[0084] Reference Figure 6 and Figure 7By examining the correlation between rotational speed and headroom, it can be seen that the headroom change trend is closely related to rotational speed and pitch angle. As mentioned above, an increase in rotational speed actually indicates an increase in equivalent wind speed. The different headroom ranges corresponding to the same rotational speed and pitch angle, which are represented by a bandwidth, are the wind shear effects under the action of equivalent wind speed. Positive shear leads to an increase in headroom, while negative shear leads to a decrease in headroom. The rapid decrease in unit headroom and tower sweeping are important reasons for the strong negative shear effect.

[0085] Based on this, a method can also be adopted to analyze and predict the trend of net clearance change caused by wind shear based on the net clearance corresponding to the unit control parameters, and then predict the net clearance result. This is the compartment query prediction method, which specifically includes the following process:

[0086] B1. Divide historical clearance data into compartments based on engine speed and pitch angle. Use the average clearance data within each compartment as the baseline value to obtain the engine speed / pitch angle-clearance baseline table. (Refer to...) Figure 8 This is to illustrate the effect of displaying the speed, pitch angle, and headroom reference table using a heatmap.

[0087] B2. Group the historical clearance data according to time sequence. Each group of data includes historical information of three blades during the process of the blades rotating two consecutive revolutions. The first clearance value of the first blade, the pitch angle and impeller speed at the same moment are respectively represented by cl. A1 α1 and ω1 represent the first clearance value of the second blade, as well as the pitch angle and impeller speed at the same time, respectively, denoted by cl. B1 α2 and ω2 represent the first clearance value of the third blade, as well as the pitch angle and impeller speed at the same time, respectively, denoted by cl. C1 α3 and ω3 represent the second clearance value of the first blade, and the pitch angle and impeller speed at the same time are represented by cl. A2 α4 and ω4 represent the second clearance value of the second blade, as well as the pitch angle and impeller speed at the same time, respectively, denoted by cl. B2 α5 and ω5 represent the second clearance value of the third blade, as well as the pitch angle and impeller speed at the same time, respectively denoted by cl. C2 α6 and ω6 represent,

[0088] Then any set of data is used express;

[0089] B3. For any set of data d, find the corresponding reference net clearance value (i.e., the i-th reference net clearance value cl) by referring to the pitch angle and impeller speed at each moment in the speed-pitch angle-net clearance reference table. bi According to the i-th propeller pitch angle α i and impeller speed ω i The data d is obtained from the speed-pitch-headroom reference table, and then converted into data.

[0090] B4. According to formula cl qi =cl Ai -cl bi Further transform data d2 into data d3 = [cl q1 cl q2 cl q3 cl q4 cl q5 cl q6 ];

[0091] B5. The predicted headroom wind shear component cl is calculated based on the following empirical formula. q7 :

[0092] cl q7 =k1(cl q4 -cl q1 )+k2(cl q5 -cl q2 )+k3(cl q6 -cl q3 )

[0093] Where k1, k2, and k3 are the first, second, and third decreasing coefficients, respectively, and are preset empirical values.

[0094] B6. Calculate the predicted rotational speed ω7 and the predicted pitch angle α7 using the following formulas:

[0095] α7=2α6-α5

[0096] ω7=2ω6-ω5

[0097] B7. Using the predicted rotational speed ω7 and the predicted pitch angle α7, the predicted reference net clearance value cl is obtained by looking up the predicted reference net clearance value cl in the rotational speed / pitch angle-net clearance reference table. b7 ;

[0098] B8. The predicted net airspace value cl is calculated using the following formula. A3 :

[0099] cl A3 =cl b7 +cl q7 .

[0100] S3. Take protective action control measures for the unit based on the airspace forecast results.

[0101] The airspace decline is assessed based on the predicted airspace value. If the predicted airspace value is higher than the upper limit of the danger zone (th1), a mild control strategy is adopted. If the predicted airspace value is higher than 70% of the upper limit of the danger zone (th2), an emergency control strategy is adopted. If 60% of the actual airspace value is within the transition zone [th2, th1], a mild control strategy is adopted. th1 can be determined based on acceptable simulation values, or the 0.1% to 0.5% percentile of the historical airspace of similar units. The lower limit is selected for ordinary units, and the upper limit can be selected for dangerous units.

[0102] The gentle control strategy is to reduce the propeller speed at θ1 to a safe pitch angle α. e If the trigger condition is not met for 30 seconds, the system will switch off; the emergency control strategy is to retract the propeller at a rate of θ2 to the safe pitch angle α. e If the trigger condition is not met for 30 seconds, the switch will be initiated. θ1 can be set with reference to the high-speed pitch control, θ2 can be set with reference to the emergency pitch control, and α... e A 5° pitch angle can be selected based on experience.

[0103] Figures 9 to 12 The analysis results of the airspace prediction model established using a linear regression machine learning method are presented across several time periods. It is evident that the predicted airspace can accurately predict downward trends and is sensitive to changes in the pitch angle, thus predicting upward trends. The model comprehensively considers the influence of airspace trends and control parameters. (Refer to...) Figure 13 and Figure 14 This is a comparative analysis of predicted and measured air clearance. It can be seen that through comparison of numerous air clearance values, the results are similar, with the deviation generally controlled within 0.5m, and the average deviation close to 0. This demonstrates that the prediction method provided by this invention has a high and good prediction effect and accuracy for predicting the air clearance of this unit.

[0104] Figures 15 to 18 The results demonstrate the effectiveness of using a compartmentalized query forecasting method. The mean deviation μ is 2.4 cm, the standard deviation σ is 0.28, and the probability of μ ± 2σ being within ± 0.6 m is 97.2%, meeting the 95% expectation requirement.

Claims

1. A wind turbine headroom prediction and control method, characterized in that... Includes the following steps: S1. Collect historical airspace data for a recent period, including airspace data. cl h Pitch angle α and impeller speed ω ; S2. Obtain net short forecast results through machine learning methods or warehouse query forecasting methods; If a machine learning method is used, the specific process includes the following: A1. Load the collected historical airspace data and preprocess the historical airspace data, including missing values, outliers and duplicate values ​​detection and correction; A2. Group the historical clearance data according to time sequence. Each group of data includes historical information of three blades during the process of the blade rotating two consecutive revolutions. And the clearance value of the first blade after two rotations. cl A3 The first clearance value of the first blade in the historical information, as well as the pitch angle and impeller speed at the same moment, are respectively represented by... cl A1 , α 1 and ω 1 indicates that the first clearance value of the second blade, as well as the pitch angle and impeller speed at the same moment, are respectively represented by... cl B1 , α 2 and ω 2 indicates that the first clearance value of the third blade, as well as the pitch angle and impeller speed at the same time, are respectively represented by... cl C1 , α 3 and ω 3 indicates that the second clearance value of the first blade, as well as the pitch angle and impeller speed at the same time, are respectively represented by... cl A2 , α 4 and ω 4 indicates that the second clearance value of the second blade, as well as the pitch angle and impeller speed at the same time, are respectively represented by... cl B2 , α 5 and ω 5 indicates that the second clearance value of the third blade, as well as the pitch angle and impeller speed at the same time, are respectively represented by... cl C2 , α 6 and ω 6 represents; Then a set of data is used express; A3. Based on prior knowledge and expert advice, combine existing features to derive new features, and then standardize the feature data. A4. Randomly split each group of data into a training set and a test set in a 7:3 ratio; A5. Train the multiple linear regression model on the training set, and use the gradient descent algorithm to optimize the parameters of the linear regression model until the loss rate of the model is reduced to the minimum, thus obtaining the initial net air volume prediction model. A6. Use test set data to evaluate the initial airspace prediction model. If the performance is insufficient, use feature engineering and parameter optimization methods to improve the performance until the best-performing airspace prediction model is obtained. A7. Use the best-performing airspace prediction model to predict airspace value and obtain airspace prediction results. If the warehouse-based query prediction method is used, the specific process includes the following: B1. Divide the historical clearance data into compartments according to speed and pitch angle, and use the average clearance data in each compartment as the benchmark value to obtain the speed-pitch-clearance benchmark table. B2. Group the historical clearance data according to time sequence. Each group of data includes historical information of three blades during the process of the blade rotating two consecutive revolutions. The first clearance value of the first blade, the pitch angle and impeller speed at the same moment are respectively represented by... cl A1 , α 1 and ω 1 indicates that the first clearance value of the second blade, as well as the pitch angle and impeller speed at the same moment, are respectively represented by... cl B1 , α 2 and ω 2 indicates that the first clearance value of the third blade, as well as the pitch angle and impeller speed at the same time, are respectively represented by... cl C1 , α 3 and ω 3 indicates that the second clearance value of the first blade, as well as the pitch angle and impeller speed at the same time, are respectively represented by... cl A2 , α 4 and ω 4 indicates that the second clearance value of the second blade, as well as the pitch angle and impeller speed at the same time, are respectively represented by... cl B2 , α 5 and ω 5 indicates that the second clearance value of the third blade, as well as the pitch angle and impeller speed at the same time, are respectively represented by... cl C2 , α 6 and ω 6 represents, Then any set of data is used express; B3. For any set of data d, find the corresponding reference clearance value by looking up the pitch angle and impeller speed at each moment in the speed-pitch angle-clearance reference table, i.e., the first... i Benchmark Net Value cl bi According to the i Pitch angle α i and impeller speed ω i The data d is obtained from the speed-pitch-headroom reference table, and then converted into data. ； B4. According to the formula Further transform the data d2 into data ； B5. The predicted headwind shear component is calculated using the following empirical formula. cl q7 : ， in k 1. k 2 and k 3 represents the first, second, and third decreasing coefficients, respectively, using preset empirical values; B6. Calculate the predicted rotational speed using the following formula. ω 7 and predicted pitch angle α 7: ， ， B7. Utilizing Predicted Rotation Speed ω 7 and predicted pitch angle α 7. Find the predicted reference net clearance value in the speed-pitch-headroom reference table. cl b7 ; B8. The predicted net airspace value is calculated using the following formula. cl A3 : ； S3. Take protective action control measures for the unit based on the airspace forecast results.

2. The wind turbine headroom prediction and control method according to claim 1, characterized in that: Step S1 Clearance Data cl h Pitch angle obtained from the clearance measurement system α and impeller speed ω All data were collected from the wind turbine main control system via an industrial bus.

3. The wind turbine headroom prediction and control method according to claim 1, characterized in that: In step A6, the initial net airspace prediction model is evaluated, specifically by calculating and evaluating the mean square error, absolute mean error, and coefficient of determination, and plotting the predicted values ​​against the actual values ​​to observe the model's predictive performance.

4. The wind turbine headroom prediction and control method according to claim 1, characterized in that: Step S3 determines the decline in airspace based on the predicted airspace value. If the predicted airspace value is higher than the upper limit of the danger zone th1, a mild control strategy is adopted. If the predicted airspace value is higher than 70% of the upper limit of the danger zone, i.e., th2, an emergency control strategy is adopted. If 60% of the actual airspace value is in the transition zone [th2, th1], a mild control strategy is adopted.

5. The wind turbine headroom prediction control method according to claim 4, characterized in that: The gentle control strategy is to reduce the propeller speed to a safe pitch angle at a rate of θ1. α e If the trigger condition is not met for 30 seconds, the switch will be initiated; the emergency control strategy is to retract the propeller at a rate of θ2 to a safe pitch angle. α e If the trigger condition is not met for 30 seconds, the process will switch off.

Images