Hybrid intelligent wind measuring system of wind generating set

Abstract

The invention discloses a hybrid intelligent wind measurement system of a wind generating set, relates to the technical field of wind power generation, and discloses the hybrid intelligent wind measurement system of the wind generating set, which comprises six core modules: a wind speed data acquisition module acquires the wind speed of front wind of a hub through a laser wind measurement unit, and a mechanical wind measurement unit acquires the wind speed of rear wind of a cabin; the data processing module screens non-abnormal weather data and constructs wind speed correlation models of different unit types; the wind speed conversion module realizes conversion from rear wind to front wind in non-abnormal weather based on a correlation model; the training module adopts a random forest algorithm to train a model, and is combined with cross validation optimization to avoid overfitting; the abnormal weather compensation module corrects back wind data under abnormal weather through three special compensation formulas of rain and snow, freezing and strong wind, and calculates the front wind speed; and the feedback and optimization module adjusts model parameters through real-time data, and carries out regular retraining to adapt to new weather and data changes.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a hybrid intelligent wind measurement system for a wind generator set. Background Art

[0002] The wind measurement system of a wind turbine generator set is an important component of the wind turbine generator set, and the accuracy of the wind measurement system is related to the accuracy of the unit efficiency analysis. Traditional wind turbines mainly rely on mechanical anemometers. Although they are inexpensive, they can only measure the wind speed at the rear of the nacelle and cannot directly obtain the wind speed in front of the hub. In addition, the wind speed data converted by empirical formulas is affected by factors such as wake disturbances and terrain differences, and generally has large deviations, making it difficult to meet the equipment management needs of power generation companies. Although laser wind measurement devices can directly measure the wind speed in front of the hub, they are expensive and are only used in a small number of wind farms. They are also affected by abnormal weather conditions, and the measurement deviation increases significantly, resulting in unreliable wind speed data.

[0003] Furthermore, existing wind speed conversion models are mostly trained based on conventional weather data, failing to fully account for the nonlinear interference of abnormal weather on wind speed measurements. They also lack dynamic optimization mechanisms and are unable to adapt to data characteristic deviations caused by equipment aging or environmental changes. This leads to a continuous decline in prediction accuracy after long-term operation. To address these issues, a hybrid intelligent wind measurement system combining multi-source data, scenario-based modeling, and dynamic optimization is urgently needed to improve wind measurement accuracy and long-term reliability under complex weather conditions, meeting the high-precision equipment management requirements of power generation companies. Summary of the Invention

[0004] The object of the present invention is to provide a hybrid intelligent wind measurement system for a wind turbine generator set to solve the problems raised in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solutions: In a first aspect, the present invention provides a hybrid intelligent wind measurement system for a wind turbine generator set, comprising: Wind speed data acquisition module: includes a laser wind measurement unit and a mechanical wind measurement unit. The laser wind measurement unit is used to measure the wind speed in front of the turbine hub, and the mechanical wind measurement unit is used to measure the wind speed behind the nacelle. Data processing module: includes a data set screening unit and a model building unit. The data set screening unit screens wind speed data under normal weather conditions to obtain a data set for training. The model building unit builds a wind speed data association model based on the data set, and builds different association models according to different wind turbine types. Wind speed conversion module: including a wind speed conversion unit, wherein the wind speed conversion unit is used to convert the wind speed behind the nacelle into the wind speed in front of the hub according to the correlation model under normal weather conditions; Training module: includes algorithm training unit and optimization unit. The algorithm training unit uses random forest algorithm to train the association model, and the optimization unit optimizes the training process through cross-validation method to avoid overfitting. Abnormal weather compensation module: includes rain and snow weather compensation unit, freezing weather compensation unit and strong wind weather compensation unit, which calculates the front wind speed under abnormal weather conditions based on the wind speed data association model and compensation formula. Among them, the rain and snow weather compensation unit corrects the wind speed data according to the rain and snow weather compensation formula and converts the rear wind speed into the front wind speed; the freezing weather compensation unit corrects the wind speed data according to the freezing weather compensation formula and converts the rear wind speed into the front wind speed; the strong wind weather compensation unit corrects the wind speed data according to the strong wind weather compensation formula and converts the rear wind speed into the front wind speed; Feedback and optimization module: includes a feedback mechanism unit and a model retraining unit. The feedback mechanism unit adjusts the model according to real-time data, and the model retraining unit regularly retrains the model with new data to ensure that the model can adapt to new weather and data changes.

[0006] In combination with the first aspect, in a first implementation of the first aspect of the present application, the laser wind measuring unit is used to measure the wind speed in front of the turbine hub, and the mechanical wind measuring unit is used to measure the wind speed behind the nacelle, including: The laser wind measurement unit uses Doppler laser radar technology to measure the three-dimensional wind speed in front of the turbine hub in real time, including longitudinal, lateral and vertical directions. It detects speed changes in the airflow through laser beams, captures wind speed changes, and provides real-time data; The mechanical wind measurement unit includes a mechanical anemometer and a wind vane, which collect data on wind speed, wind direction, and temperature at the rear of the nacelle. The mechanical anemometer senses air velocity through rotating blades, the wind vane provides wind direction information, and the temperature sensor supplements ambient meteorological data to capture the characteristics of the airflow at the rear of the unit. The two sets of data were aligned using GPS timestamps to ensure that the time dimensions of the wind speed data collected by the laser wind measurement unit and the mechanical wind measurement unit were consistent.

[0007] In combination with the first aspect, in a second implementation of the first aspect of the present application, the data set screening unit screens wind speed data under non-abnormal weather conditions to obtain a data set for training, including: The data screening unit analyzes the collected wind speed data and removes abnormal data caused by equipment failure or abnormal weather. It also identifies and removes abnormal fluctuations in wind speed data by analyzing the standard deviation of each set of data. The filtered data set includes wind speed data under different weather conditions. Time series analysis is performed on the data to maintain the time series consistency of the wind speed data. During the processing, the data is formatted, standardized and normalized to ensure consistency when the data is input into the model.

[0008] In combination with the first aspect, in a third implementation of the first aspect of the present application, the model building unit builds a wind speed data association model based on the data set, and builds different association models according to different wind turbine types, including: Based on the filtered data set, the correlation model construction unit constructs a correlation model between the wind speed data, establishing a mathematical model of the relationship between the wind speed behind the nacelle and the wind speed in front of the hub. Based on different types of wind turbines, including those on land plains, mountainous areas, and offshore, a correlation model is selected for each type of wind turbine to adapt to different wind speed variation patterns. Under normal weather conditions, a linear regression model is used to capture the linear relationship between the wind speed behind the nacelle and the wind speed in front of the hub. The relationship between the wind speed in front and behind is calculated by linear fitting of the wind speed data. Under abnormal weather conditions, the support vector machine model is used to process the nonlinear wind speed data relationship. By establishing a hyperplane in high-dimensional space, the nonlinear characteristics under abnormal weather conditions are captured and the front wind speed under abnormal weather conditions is calculated.

[0009] In combination with the first aspect, in a fourth implementation of the first aspect of the present application, the wind speed conversion unit is configured to convert the wind speed behind the nacelle into the wind speed in front of the hub according to the correlation model under non-abnormal weather conditions, including: Under non-abnormal weather conditions, the correlation model between wind speed data obtained from the dataset is The calculation formula for converting the wind speed behind the nacelle to the wind speed in front of the hub is obtained. The specific formula is as follows: ; Among them, V front is the estimated front wind speed of the hub, V rear is the measured wind speed behind the cabin, and β is the conversion coefficient under normal weather conditions, which represents the conversion relationship from rear wind to front wind and is calculated through data set analysis and actual wind speed data.

[0010] In combination with the first aspect, in a fifth implementation of the first aspect of the present application, the algorithm training unit uses a random forest algorithm to train the association model, and the optimization unit optimizes the training process through a cross-validation method to avoid overfitting, including: The training process uses supervised learning methods to input the filtered data set to train the model; uses the random forest algorithm for optimization, and adopts the cross-validation method to avoid overfitting to ensure the stability and generalization ability of the model; uses the mean square error function to measure the difference between the predicted wind speed and the actual wind speed, and adjusts the model parameters; and further improves the prediction accuracy of the model under different wind speed data by adjusting the model's hyperparameters.

[0011] In combination with the first aspect, in a sixth implementation of the first aspect of the present application, the rain and snow weather compensation unit corrects the wind speed data according to the rain and snow weather compensation formula to convert the rear wind speed into the front wind speed, including: Under rainy or snowy weather conditions, a rain or snow compensation coefficient is established based on the wind speed model related to rain or snow, targeting the impact of rain or snow on the wind speed measuring equipment, to correct the measurement errors caused by moisture, water accumulation, and snow accumulation. The wind speed relationship model during rain or snow is used to calculate the actual wind speed in front of the hub, using the following formula: ; Among them, V front is the calibrated hub front wind speed, V rear is the measured wind speed behind the cabin, α is the compensation coefficient for rain or snow weather, which represents the effect of rain or snow on wind speed measurement. Based on the analysis of the relationship between wind speed measurement data and rainfall or snow depth under rain or snow conditions, it is obtained from historical data through data fitting algorithm. P is the rainfall or snow depth, and P max is the maximum value of rainfall or snow, normalized compensation coefficient.

[0012] In combination with the first aspect, in a seventh implementation of the first aspect of the present application, the freezing weather compensation unit corrects the wind speed data according to the freezing weather compensation formula to convert the rear wind speed into the front wind speed, including: In freezing weather conditions, wind speed measurement errors may occur due to ice accumulation on the wind turbine blades. This error needs to be corrected using a formula to deduce the actual wind speed in front of the turbine hub. The specific formula is as follows: ; Among them, V front is the calibrated hub front wind speed, V rear is the measured wind speed behind the nacelle, γ is the freezing weather compensation coefficient, which represents the effect of blade ice accumulation on wind speed measurement. It is obtained from historical freezing weather data using a regression analysis algorithm based on the relationship between wind speed and blade obstruction effect under freezing conditions. I is the degree of freezing icing, and the parameter is expressed as the thickness of ice on the blade surface or the amount of ice accumulation.

[0013] In combination with the first aspect, in an eighth implementation of the first aspect of the present application, the strong wind weather compensation unit corrects the wind speed data according to the strong wind weather compensation formula to convert the rear wind speed into the front wind speed, including: In strong winds, the anemometer will experience fluctuations, affecting measurement accuracy. Based on the IEC61400 standard, the wind intensity W is divided into three levels: 18m / s≤W<22m / s is considered mild wind, 22m / s≤W<25m / s is considered moderate wind, and W≥25m / s is considered extreme wind. Based on the strong wind weather data, correction is made using the following formula: ; V front is the calibrated hub front wind speed, V rear is the measured wind speed behind the cabin, δ is the high wind weather compensation coefficient, which represents the impact of high wind on wind speed measurement, W is the high wind intensity, and the parameter is expressed as the instantaneous maximum wind speed or the wind speed fluctuation amplitude. It is obtained from the measurement data during high winds through a data fitting algorithm based on the historical high wind weather data and the response characteristics of the anemometer.

[0014] In conjunction with the first aspect, in a ninth implementation of the first aspect of the present application, the feedback mechanism unit adjusts the model based on real-time data, and the model retraining unit periodically retrains the model using new data to ensure that the model can adapt to new weather changes and data changes, including: Through the new wind speed data collected in real time, the associated model is fed back and adjusted, and the model parameters are optimized to ensure that the model can adapt to data changes under different weather conditions; when there is new collected data, the data is input into the model for retraining to ensure the model's adaptability to new weather conditions; by analyzing the new wind speed data, the associated model is retrained regularly to further optimize the model's prediction accuracy and stability, ensuring the accuracy of wind speed prediction under different weather conditions.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. In the existing technology, traditional wind speed conversion models mostly use a unified empirical formula, which leads to generally high errors in forward wind estimation under normal weather conditions. The present invention trains an association model based on screened non-abnormal weather data and significantly improves the conversion accuracy through a random forest algorithm combined with cross-validation optimization.

[0016] 2. Most existing models are trained based on conventional weather data and cannot handle the nonlinear interference of abnormal weather. The present invention designs special compensation formulas for three types of abnormal weather, combines historical data to fit the correction parameters, and accurately corrects the back wind measurement error under abnormal weather.

[0017] 3. The existing model relies on fixed historical data training and lacks adaptive capabilities. After long-term operation, the error continues to increase. The present invention achieves dynamic optimization through a feedback mechanism, monitors the forward wind prediction error in real time, optimizes the associated model and retrains it, dynamically adjusts the compensation coefficient, and ensures that the system continues to adapt to new weather patterns and changes in equipment status. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a system structure diagram of the hybrid intelligent wind measurement system for wind turbine generator sets of the present invention; Figure 2 This is a schematic diagram of the abnormal weather compensation module steps of the hybrid intelligent wind measurement system for wind turbines of the present invention. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] Example: Figure 1-Figure 2 As shown, the present invention provides a technical solution. like Figure 1 As shown in the system structure diagram of the hybrid intelligent wind measurement system for wind turbine generator sets, the present invention provides a hybrid intelligent wind measurement system for wind turbine generator sets, comprising: Step S100: The laser wind measuring unit is used to measure the wind speed in front of the turbine hub, and the mechanical wind measuring unit is used to measure the wind speed behind the nacelle; Specifically, the laser wind measurement unit uses Doppler laser radar technology to measure the three-dimensional wind speed in front of the turbine hub in real time, including longitudinal, lateral and vertical directions. It detects speed changes in the airflow through a laser beam, captures wind speed changes, and provides real-time data; The mechanical wind measurement unit includes a mechanical anemometer and a wind vane, which collect data on wind speed, wind direction, and temperature at the rear of the nacelle. The mechanical anemometer senses air velocity through rotating blades, the wind vane provides wind direction information, and the temperature sensor supplements ambient meteorological data to capture the characteristics of the airflow at the rear of the unit. The two sets of data were aligned using GPS timestamps to ensure that the time dimensions of the wind speed data collected by the laser wind measurement unit and the mechanical wind measurement unit were consistent.

[0021] In one specific embodiment, a wind turbine generator system in a medium-sized city was selected as the test environment, located at a wind farm on the outskirts of the city. The test environment included five wind turbine generators, each equipped with a laser wind measurement unit and a mechanical wind measurement unit for real-time wind speed monitoring. Each unit was equipped with a laser wind measurement unit and a mechanical wind measurement unit, and data synchronization technology was used to collect and analyze wind speed data from behind the nacelle and in front of the hub.

[0022] Data collection lasted for one month, during which more than 200,000 samples were collected. The data includes wind speed data under different weather conditions, covering both normal and abnormal weather. During the test, the wind speed data from the laser wind measuring unit and the mechanical wind measuring unit were aligned with GPS timestamps to ensure consistency in the time dimension of the two sets of data. At 12:00:00 on April 1, 2025, the laser wind measuring unit measured the wind speed in front of the turbine hub as 8.5 m / s, 8.6 m / s at 12:00:01, and 8.4 m / s at 12:00:02. At the same time, the mechanical wind measuring unit measured the wind speed behind the nacelle as 7.9 m / s, 8.0 m / s, and 7.8 m / s, respectively, with wind directions of 35°, 34°, and 36°, and a temperature of 15°C.

[0023] To ensure the representativeness of the data, the test environment included multiple samples under different wind speed conditions. Through data synchronization and verification, wind speed data was obtained for different time periods and weather conditions, ensuring the comprehensiveness and accuracy of the data.

[0024] Step S200: The data set screening unit screens the wind speed data under normal weather conditions to obtain a data set for training. The model building unit builds a wind speed data association model based on the data set, and builds different association models according to different wind turbine types. The wind speed conversion unit is used to convert the wind speed behind the nacelle into the wind speed in front of the hub according to the association model under normal weather conditions. Specifically, the data screening unit analyzes the collected wind speed data to eliminate abnormal data caused by equipment failure or abnormal weather; and identifies and eliminates abnormal fluctuations in the wind speed data by analyzing the standard deviation of each set of data; The filtered data set includes wind speed data under different weather conditions. Time series analysis is performed on the data to maintain the time series consistency of the wind speed data. During the processing, the data is formatted, standardized and normalized to ensure consistency when the data is input into the model.

[0025] Furthermore, based on the filtered data set, the correlation model construction unit constructs a correlation model between the wind speed data, and establishes a mathematical model of the relationship between the wind speed behind the nacelle and the wind speed in front of the hub. According to different types of wind turbines, including onshore plains, mountainous areas, and offshore areas, a correlation model is selected for each type of wind turbine to adapt to different wind speed variation patterns. Under normal weather conditions, a linear regression model is used to capture the linear relationship between the wind speed behind the nacelle and the wind speed in front of the hub. The relationship between the wind speed in front and behind is calculated by linear fitting of the wind speed data. Under abnormal weather conditions, the support vector machine model is used to process the nonlinear wind speed data relationship. By establishing a hyperplane in high-dimensional space, the nonlinear characteristics under abnormal weather conditions are captured and the front wind speed under abnormal weather conditions is calculated.

[0026] Furthermore, under normal weather conditions, the correlation model between the wind speed data obtained from the dataset is used to obtain a calculation formula for converting the wind speed behind the nacelle into the wind speed in front of the hub. The specific formula is as follows: ; Among them, V front is the estimated front wind speed of the hub, V rear is the measured wind speed behind the cabin, and β is the conversion coefficient under normal weather conditions, which represents the conversion relationship from rear wind to front wind and is calculated through data set analysis and actual wind speed data.

[0027] In one specific embodiment, wind speed data is collected from a wind farm unit. The data is processed through a data screening unit to eliminate abnormal data caused by equipment failure, including sensor readings that suddenly drop to zero or abnormally high values. By analyzing the standard deviation of each data set, abnormal fluctuations in the wind speed data are identified and eliminated. If the standard deviation exceeds a preset threshold of 2 m / s, the data is considered abnormal and eliminated. The data is then formatted, standardized, and normalized. The filtered data set includes wind speed data under different weather conditions. Time series analysis is performed on the data to maintain temporal consistency.

[0028] Under normal weather conditions, the collected regular weather data are as follows: Timestamp: 2025-04-20 12:00:00, Backwind Speed: V rear =7.5m / s, front wind speed: V front =8.0m / s; Timestamp: 2025-04-20 12:00:01, Backwind Speed: V rear =7.6m / s, front wind speed: V front =8.1m / s; Timestamp: 2025-04-20 12:00:02, Backwind Speed: V rear =7.4m / s, front wind speed: V front =7.9m / s; By performing linear regression analysis on these data, the calculation formula for converting the wind speed behind the nacelle to the wind speed in front of the hub was obtained: ; Among them, V front is the estimated front wind speed of the hub, V rear is the measured wind speed behind the cabin, and β = 1.05 is the conversion coefficient under normal weather conditions, which represents the conversion relationship from rear wind to front wind and is calculated through data set analysis and actual wind speed data.

[0029] Step S300: the algorithm training unit uses the random forest algorithm to train the association model, and the optimization unit optimizes the training process through the cross-validation method to avoid overfitting; Specifically, the training process uses the supervised learning method to input the filtered data set to train the model; the random forest algorithm is used for optimization, and the cross-validation method is adopted to avoid overfitting and ensure the stability and generalization ability of the model; the mean square error function is used to measure the difference between the predicted wind speed and the actual wind speed, and the model parameters are adjusted; the model's hyperparameters are adjusted to further improve the model's prediction accuracy under different wind speed data.

[0030] In one specific example, meteorological and wind turbine operation data from a wind farm in northwest my country was selected as training data. The original dataset contained 420,000 records spanning 547 days. Outliers caused by sensor failures, including extreme wind speeds exceeding 35 m / s, were removed. Missing values were filled using a forward-filling method. Features with a correlation greater than 0.3 with the target variable were selected based on the Pearson correlation coefficient. The training and test sets were then divided chronologically to ensure non-overlapping time periods to verify model generalization.

[0031] Use Scikit-Learn's RandomForestRegressor to implement the random forest algorithm. The specific steps are as follows: Initial hyperparameters were set as follows: n_estimators = 100 (the number of trees), max_depth = 8 (the maximum depth of the tree), and min_samples_split = 5 (the minimum number of samples for internal node splitting). The mean squared error (MSE) of the training set was 1.92 (i.e., the mean squared difference between the predicted and actual wind speeds was 1.92 (m / s²). The validation set MSE was 2.15, and the cross-validation MSE standard deviation was 0.12.

[0032] To further improve accuracy, we used grid search combined with 5-fold cross-validation. The search space was set to: n_estimators: [100, 200, 300], increasing the number of trees to improve model complexity; max_depth: [5, 8, 10], limiting or relaxing the tree depth to balance underfitting and overfitting; and min_samples_leaf: [2, 4], the minimum number of leaf samples, to control tree complexity. After tuning, the optimal hyperparameters were: n_estimators = 200, max_depth = 10, and min_samples_leaf = 2. At this point, the average cross-validation MSE dropped to 1.63, and the validation set MSE also dropped to 1.78.

[0033] The optimal model was used to predict the test set, achieving an MSE of 1.81 and a mean absolute error (MAE) of 0.98 m / s. Using a random forest algorithm combined with cross-validation optimization, the model's MSE on the test set stabilized at around 1.8, with a mean absolute error of less than 1 m / s, meeting the engineering requirements for short-term wind speed forecasting at wind farms. After hyperparameter tuning, the model's prediction accuracy improved by approximately 15%-20% across various wind speed ranges, validating the effectiveness of the method.

[0034] Step S400: Calculating the front wind speed under abnormal weather conditions based on the wind speed data association model and the compensation formula, wherein the rain and snow weather compensation unit corrects the wind speed data according to the rain and snow weather compensation formula and converts the rear wind speed into the front wind speed; the freezing weather compensation unit corrects the wind speed data according to the freezing weather compensation formula and converts the rear wind speed into the front wind speed; and the strong wind weather compensation unit corrects the wind speed data according to the strong wind weather compensation formula and converts the rear wind speed into the front wind speed. Specifically, under rainy or snowy weather conditions, a rain or snow compensation coefficient is established based on the wind speed model related to rain or snow, targeting the impact of rain or snow on the wind speed measuring equipment, to correct the measurement errors caused by moisture, accumulated water, and accumulated snow. The wind speed relationship model during rain or snow is used to infer the actual wind speed in front of the hub, which is calculated using the following formula: ; Among them, V front is the calibrated hub front wind speed, V rear is the measured wind speed behind the cabin, α is the compensation coefficient for rain or snow weather, which represents the effect of rain or snow on wind speed measurement. Based on the analysis of the relationship between wind speed measurement data and rainfall or snow depth under rain or snow conditions, it is obtained from historical data through data fitting algorithm. P is the rainfall or snow depth, and P max is the maximum value of rainfall or snow, normalized compensation coefficient.

[0035] Furthermore, in freezing weather conditions, wind speed measurement errors caused by ice accumulation on the wind turbine blades need to be corrected using a formula to calculate the actual wind speed in front of the turbine hub. The specific formula is as follows: ; Among them, V front is the calibrated hub front wind speed, V rear is the measured wind speed behind the nacelle, γ is the freezing weather compensation coefficient, which represents the effect of blade ice accumulation on wind speed measurement. It is obtained from historical freezing weather data using a regression analysis algorithm based on the relationship between wind speed and blade obstruction effect under freezing conditions. I is the degree of freezing icing, and the parameter is expressed as the thickness of ice on the blade surface or the amount of ice accumulation.

[0036] Furthermore, in strong wind weather conditions, strong winds cause fluctuations in the anemometer, affecting measurement accuracy. Based on the strong wind classification strategy of the IEC61400 standard, the strong wind intensity W is divided into three levels: when 18m / s≤W<22m / s, it is a mild strong wind; when 22m / s≤W<25m / s, it is a moderate strong wind; when W≥25m / s, it is an extreme strong wind; Based on the strong wind weather data, correction is made using the following formula: ; V front is the calibrated hub front wind speed, V rear is the measured wind speed behind the cabin, δ is the high wind weather compensation coefficient, which represents the impact of high wind on wind speed measurement, W is the high wind intensity, and the parameter is expressed as the instantaneous maximum wind speed or the wind speed fluctuation amplitude. It is obtained from the measurement data during high winds through a data fitting algorithm based on the historical high wind weather data and the response characteristics of the anemometer.

[0037] In one specific example, operational data from a mountain wind farm in North my country from winter to spring was selected as a validation scenario. Data was collected once per minute, and analysis was conducted during abnormal weather periods in three typical months: freezing / snowfall in January, rain in March, and high winds in April. The focus was on analyzing the wind speed correction process for the wind farm under these three abnormal weather conditions: rain and snowfall, freezing, and high winds.

[0038] When the wind farm encounters rainy weather, the cumulative rainfall is P = 20 mm, and the maximum rainfall in the history of this field is P max =50mm, select 120 records of continuous data, and the wind speed meter measures V rear The average value is 8.5m / s, but due to rain adhering to the sensor surface, the deviation of the uncorrected front wind speed and rear wind speed in the same period in history reached 12%-15%.

[0039] By analyzing 32 historical rainfall data with P=5-50mm, the wind speed V of each rainfall period was extracted. rear , LiDAR measured front wind speed V front _ true , the compensation coefficient α and P / P are obtained by fitting max The relationship is: α=0.05, fitting R²=0.89, indicating that for every 10% increase in rainfall, the correction coefficient increases by 0.005.

[0040] Take one of the data at a certain moment: V rear =8.8m / s,P=22mm,P max =50mm, substitute into the rainfall compensation formula: ; Finally, we get V front =8.8×(1+0.05×22 / 50)≈8.99m / s.

[0041] When the wind farm encountered freezing weather, ice accumulation appeared on the wind turbine blades. 120 data points were selected and the wind speed meter measured V rear The average value is 6.2m / s, but due to the airflow disturbance caused by ice accumulation on the blades, the uncorrected front wind speed and rear wind speed deviate by 20%-25%.

[0042] By analyzing the winter freezing data of 18 ice accumulation events with I=0.5-5cm, the posterior wind V rear , LiDAR measured front wind V front _ true , using quadratic regression fitting, we get γ=0.2, and exponential regression R²=0.91, which shows that the correction coefficient decays by about 18% for every 1 cm increase in ice thickness.

[0043] Take the data at a certain moment: V rear =6.5m / s, I=3cm, substitute into the freezing compensation formula: ; Finally, we get V front =6.5×0.5488≈3.57m / s.

[0044] When the wind farm encounters strong winds, the instantaneous maximum wind speed W=25m / s. 60 data are selected. The anemometer then changes its measurement value due to turbulence fluctuations. rear The fluctuation range is 8-12m / s.

[0045] By analyzing the wind data of 21 strong wind events with W=20-30m / s, the 10-minute mean value of the wind V was extracted. rear , LiDAR measured front wind V front _ true, the fitting results show that when W=20m / s, δ=0.05, when W=30m / s, δ=0.04, and the fitting R²=0.87.

[0046] Take the data at a certain moment: the mean value of the wind for 10 minutes V rear =15.5m / s, the instantaneous maximum wind speed in the current 10 minutes is W=25m / s, δ=0.045, substitute into the strong wind compensation formula: ; Finally, we get V front =15.5×1.146≈17.77m / s.

[0047] Step S500: The feedback mechanism unit adjusts the model according to real-time data, and the model retraining unit regularly retrains the model using new data to ensure that the model can adapt to new weather changes and data changes.

[0048] Specifically, through the real-time collection of new wind speed data, the associated model is feedback-adjusted and the model parameters are optimized to ensure that the model can adapt to data changes under different weather conditions; when there is new collected data, the data is input into the model for retraining to ensure the model's adaptability to new weather conditions; by analyzing new wind speed data, the associated model is retrained regularly to further optimize the model's prediction accuracy and stability, ensuring the accuracy of wind speed prediction under different weather conditions.

[0049] In a specific embodiment, data of thunderstorm-prone periods are collected, and the back wind speed V rear The one-hour average is 12-18m / s, and the front wind speed is V front _ true The hourly average is 11-16m / s, the rainfall P is 5-20mm, the instantaneous maximum wind speed W=22-28m / s, and the forward wind speed V is predicted by the formula front _ pred , combined with the rain and snow compensation formula, the absolute error is calculated. If the error is greater than 0.5m / s in two hours, the feedback mechanism is triggered.

[0050] Analyzing the source of the error, it was found that the δ coefficient in the high wind compensation formula did not fully consider the high-frequency fluctuations of thunderstorms, resulting in V front _ pred Overestimation, based on the correlation between real-time error and W, dynamically adjust δ, δ 新 =δδ 旧 -0.001×average error, before adjustment δ=0.05, V front _ pred =15.2m / s, V front _ true =14.5m / s, error 0.7m / s, after adjustment δ=0.04, Vfront _ pred =14.6m / s, V front _ true =14.5m / s, error 0.1m / s.

[0051] Retrain the model, accumulate and clean the data, eliminate sensor fault data, fill in missing values, screen effective features, adjust hyperparameters through grid search, and refit the compensation coefficients.

[0052] like Figure 2 As shown in the schematic diagram of the abnormal weather compensation module steps of the hybrid intelligent wind measurement system for wind turbines, the present invention provides a hybrid intelligent wind measurement system for wind turbines, comprising: Through the coordinated work of three compensation units for rain, snow, ice, and strong winds, and based on the correlation model and compensation formula trained with historical data, the wind speed behind the nacelle is corrected to the actual wind speed in front of the hub, adapting to measurement errors in different abnormal weather conditions. The specific steps are as follows: Input real-time data from the weather station, including rainfall P, temperature T, instantaneous maximum wind speed W, and wind speed V behind the mechanical wind measuring unit rear During execution, different abnormal weather conditions are triggered by the following conditions: rainy and snowy weather when P>0mm or snow depth>0cm; freezing weather when T<0℃ and blade ice thickness I>0cm; strong wind weather when W≥18m / s.

[0053] The goal of the rain and snow weather compensation unit is to correct the attachment error of rain or snow to the back wind anemometer. The input data includes the back wind speed V of the mechanical wind measuring unit. rear , the real-time rainfall at the weather station P = 22mm, the maximum historical rainfall at the wind farm P max =50mm, and the compensation coefficient α=0.05 and R²=0.89 fitted by historical rainfall data. The rain and snow compensation formula is used in the calculation: ; Substitute the data to get V front =8.8×(1+0.05×22 / 50)≈8.99m / s.

[0054] The goal of the freezing weather compensation unit is to correct the airflow disturbance error caused by ice accumulation on the blade, including ice accumulation changing the aerodynamic shape of the blade and the high backwind measurement value. The input data includes the backwind speed V of the mechanical wind measuring unit. rear =6.5m / s, the blade ice thickness measured by the drone is I=3cm, and the compensation coefficient γ=0.2 and R²=0.91 are fitted based on historical freezing data. The freezing compensation formula is used in the calculation: ; Substitute the data to get V front=6.5×e^(-0.2×3)≈3.57m / s.

[0055] The goal of the high wind weather compensation unit is to eliminate the signal jump error of the back wind anemometer caused by strong turbulence, including the fluctuation of the measured value caused by instantaneous high wind. The input data includes the 10-minute average value of the back wind speed V of the mechanical wind measuring unit. rear =15.5m / s, the instantaneous maximum wind speed within 10 minutes W=20m / s, and the compensation coefficient δ=0.05 fitted by historical high wind data. The high wind compensation formula is used during calculation: ; Substitute the data to get V front =15.5×(1+0.045×log(21))≈17.86m / s.

[0056] The abnormal weather compensation module uses a process of weather identification, data input, and formula calculation to correct the wind speed for three types of abnormal weather: rain and snow, freezing, and strong winds, using dedicated compensation formulas. Combined with real-time feedback and regular retraining, it ensures the accuracy of the estimated forward wind speed under abnormal weather conditions, providing reliable wind measurement data support for wind farm power prediction and wind turbine control.

[0057] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A hybrid intelligent wind measurement system for wind turbines, characterized in that: include: Wind speed data acquisition module: includes a laser wind measurement unit and a mechanical wind measurement unit. The laser wind measurement unit is used to measure the wind speed in front of the turbine hub, and the mechanical wind measurement unit is used to measure the wind speed behind the nacelle. Data processing module: includes a data set screening unit and a model building unit. The data set screening unit screens wind speed data under normal weather conditions to obtain a data set for training. The model building unit builds a wind speed data association model based on the data set, and builds different association models according to different wind turbine types. Wind speed conversion module: including a wind speed conversion unit, wherein the wind speed conversion unit is used to convert the wind speed behind the nacelle into the wind speed in front of the hub according to the correlation model under normal weather conditions; Training module: includes algorithm training unit and optimization unit. The algorithm training unit uses random forest algorithm to train the association model, and the optimization unit optimizes the training process through cross-validation method to avoid overfitting. Abnormal weather compensation module: includes rain and snow weather compensation unit, freezing weather compensation unit and strong wind weather compensation unit, which calculates the front wind speed under abnormal weather conditions based on the wind speed data association model and compensation formula. Among them, the rain and snow weather compensation unit corrects the wind speed data according to the rain and snow weather compensation formula and converts the rear wind speed into the front wind speed; the freezing weather compensation unit corrects the wind speed data according to the freezing weather compensation formula and converts the rear wind speed into the front wind speed; the strong wind weather compensation unit corrects the wind speed data according to the strong wind weather compensation formula and converts the rear wind speed into the front wind speed; Feedback and optimization module: includes a feedback mechanism unit and a model retraining unit. The feedback mechanism unit adjusts the model according to real-time data, and the model retraining unit regularly retrains the model with new data to ensure that the model can adapt to new weather and data changes.

2. The hybrid intelligent wind measurement system for wind turbine generator sets according to claim 1, characterized in that: The laser wind measuring unit is used to measure the wind speed in front of the turbine hub, and the mechanical wind measuring unit is used to measure the wind speed behind the nacelle, including: The laser wind measurement unit uses Doppler laser radar technology to measure the three-dimensional wind speed in front of the turbine hub in real time, including longitudinal, lateral and vertical directions. It detects speed changes in the airflow through laser beams, captures wind speed changes, and provides real-time data; The mechanical wind measurement unit includes a mechanical anemometer and a wind vane, which collect data on wind speed, wind direction, and temperature at the rear of the nacelle. The mechanical anemometer senses air velocity through rotating blades, the wind vane provides wind direction information, and the temperature sensor supplements ambient meteorological data to capture the characteristics of the airflow at the rear of the unit. The two sets of data were aligned using GPS timestamps to ensure that the time dimensions of the wind speed data collected by the laser wind measurement unit and the mechanical wind measurement unit were consistent.

3. The hybrid intelligent wind measurement system for wind turbine generator sets according to claim 1, characterized in that: The data set screening unit screens the wind speed data under non-abnormal weather conditions to obtain a data set for training, including: The data screening unit analyzes the collected wind speed data and removes abnormal data caused by equipment failure or abnormal weather. It also identifies and removes abnormal fluctuations in wind speed data by analyzing the standard deviation of each set of data. The filtered data set includes wind speed data under different weather conditions. Time series analysis is performed on the data to maintain the time series consistency of the wind speed data. During the processing, the data is formatted, standardized and normalized to ensure consistency when the data is input into the model.

4. The hybrid intelligent wind measurement system for wind turbine generator sets according to claim 1, characterized in that: The model building unit builds a wind speed data association model based on the data set, and builds different association models according to different wind turbine generator types, including: Based on the filtered data set, the correlation model construction unit constructs a correlation model between the wind speed data, establishing a mathematical model of the relationship between the wind speed behind the nacelle and the wind speed in front of the hub. Based on different types of wind turbines, including those on land plains, mountainous areas, and offshore, a correlation model is selected for each type of wind turbine to adapt to different wind speed variation patterns. Under normal weather conditions, a linear regression model is used to capture the linear relationship between the wind speed behind the nacelle and the wind speed in front of the hub. The relationship between the wind speed in front and behind is calculated by linear fitting of the wind speed data. Under abnormal weather conditions, the support vector machine model is used to process the nonlinear wind speed data relationship. By establishing a hyperplane in high-dimensional space, the nonlinear characteristics under abnormal weather conditions are captured and the front wind speed under abnormal weather conditions is calculated.

5. The hybrid intelligent wind measurement system for wind turbine generator sets according to claim 1, characterized in that: The wind speed conversion unit is used to convert the wind speed behind the nacelle into the wind speed in front of the hub according to the correlation model under normal weather conditions, including: Under non-abnormal weather conditions, the correlation model between wind speed data obtained from the dataset is The calculation formula for converting the wind speed behind the nacelle to the wind speed in front of the hub is obtained. The specific formula is as follows: ； Among them, V front is the estimated front wind speed of the hub, V rear is the measured wind speed behind the cabin, and β is the conversion coefficient under normal weather conditions, which represents the conversion relationship from rear wind to front wind and is calculated through data set analysis and actual wind speed data.

6. The hybrid intelligent wind measurement system for wind turbine generator sets according to claim 1, characterized in that: The algorithm training unit uses the random forest algorithm to train the association model, and the optimization unit optimizes the training process through a cross-validation method to avoid overfitting, including: The training process uses supervised learning methods to input the filtered data set to train the model; uses the random forest algorithm for optimization, and adopts the cross-validation method to avoid overfitting to ensure the stability and generalization ability of the model; uses the mean square error function to measure the difference between the predicted wind speed and the actual wind speed, and adjusts the model parameters; and further improves the prediction accuracy of the model under different wind speed data by adjusting the model's hyperparameters.

7. The hybrid intelligent wind measurement system for wind turbine generator sets according to claim 1, characterized in that: The rain and snow weather compensation unit corrects the wind speed data according to the rain and snow weather compensation formula to convert the rear wind speed into the front wind speed, including: Under rainy or snowy weather conditions, a rain or snow compensation coefficient is established based on the wind speed model related to rain or snow, targeting the impact of rain or snow on the wind speed measuring equipment, to correct the measurement errors caused by moisture, water accumulation, and snow accumulation. The wind speed relationship model during rain or snow is used to calculate the actual wind speed in front of the hub, using the following formula: ； Among them, V front is the calibrated hub front wind speed, V rear is the measured wind speed behind the cabin, α is the compensation coefficient for rain or snow weather, which represents the effect of rain or snow on wind speed measurement. Based on the analysis of the relationship between wind speed measurement data and rainfall or snow depth under rain or snow conditions, it is obtained from historical data through data fitting algorithm. P is the rainfall or snow depth, and P max is the maximum value of rainfall or snow, normalized compensation coefficient.

8. The hybrid intelligent wind measurement system for wind turbine generator sets according to claim 1, characterized in that: The freezing weather compensation unit corrects the wind speed data according to the freezing weather compensation formula to convert the rear wind speed into the front wind speed, including: In freezing weather conditions, wind speed measurement errors may occur due to ice accumulation on the wind turbine blades. This error needs to be corrected using a formula to deduce the actual wind speed in front of the turbine hub. The specific formula is as follows: ； Among them, V front is the calibrated hub front wind speed, V rear is the measured wind speed behind the nacelle, γ is the freezing weather compensation coefficient, which represents the effect of blade ice accumulation on wind speed measurement. It is obtained from historical freezing weather data using a regression analysis algorithm based on the relationship between wind speed and blade obstruction effect under freezing conditions. I is the degree of freezing icing, and the parameter is expressed as the thickness of ice on the blade surface or the amount of ice accumulation.

9. The hybrid intelligent wind measurement system for wind turbine generator sets according to claim 1, characterized in that: The high wind weather compensation unit corrects the wind speed data according to the high wind weather compensation formula to convert the rear wind speed into the front wind speed, including: In strong winds, the anemometer will experience fluctuations, affecting measurement accuracy. Based on the IEC61400 standard, the wind intensity W is divided into three levels: 18m / s≤W<22m / s is considered mild wind, 22m / s≤W<25m / s is considered moderate wind, and W≥25m / s is considered extreme wind. Based on the strong wind weather data, correction is made using the following formula: ； V front is the calibrated hub front wind speed, V rear is the measured wind speed behind the cabin, δ is the high wind weather compensation coefficient, which represents the impact of high wind on wind speed measurement, W is the high wind intensity, and the parameter is expressed as the instantaneous maximum wind speed or the wind speed fluctuation amplitude. It is obtained from the measurement data during high winds through a data fitting algorithm based on the historical high wind weather data and the response characteristics of the anemometer.

10. The hybrid intelligent wind measurement system for wind turbine generator sets according to claim 1, characterized in that: The feedback mechanism unit adjusts the model based on real-time data, and the model retraining unit regularly retrains the model using new data to ensure that the model can adapt to new weather and data changes, including: Through the new wind speed data collected in real time, the associated model is fed back and adjusted, and the model parameters are optimized to ensure that the model can adapt to data changes under different weather conditions; when there is new collected data, the data is input into the model for retraining to ensure the model's adaptability to new weather conditions; by analyzing the new wind speed data, the associated model is retrained regularly to further optimize the model's prediction accuracy and stability, ensuring the accuracy of wind speed prediction under different weather conditions.

Images