Method for monitoring state of yaw system of offshore wind generating set

Abstract

The invention discloses a method for monitoring the state of a yaw system of an offshore wind generating set, which relates to the field of wind driven generators and comprises the following steps of: 1, performing initial wind domain division in a working area of a target generating set, deploying wind power acquisition components based on vertical height distribution, constructing distributed acquisition point locations, and constructing a wind power acquisition point location; acquiring wind speed, wind direction and turbulence intensity data of each wind domain in real time; 2, based on the initial wind domain division result, combining wind domains with the correlation degree higher than a preset threshold value through correlation analysis, independently storing low-correlation-degree wind domains, and forming a multi-height-layer wind domain model; the problem of yaw lag or over-regulation caused by wind shear or turbulence is effectively avoided, the response precision and the operation stability of a yaw system are further improved, the power generation efficiency loss caused by yaw errors is reduced, it is ensured that a fan is always located at the optimal windward angle, the service life of core components such as a yaw gearbox and a motor is prolonged, and the power generation efficiency is improved. And the shutdown maintenance cost caused by sudden failure is also reduced.

Description

Technical Field

[0001] This invention relates to the field of wind turbine technology, specifically to a method for monitoring the yaw system status of an offshore wind turbine. Background Technology

[0002] In recent years, offshore wind power installed capacity has grown rapidly, becoming an important development direction in the renewable energy sector. Compared to onshore wind power, offshore wind power has advantages such as higher wind speeds, greater wind energy density, and less land occupation. However, the marine environment is complex and changeable, and wind turbines are exposed to harsh conditions such as high salt spray, high humidity, and strong winds and waves for extended periods, leading to numerous challenges in their operation and maintenance. Traditional single-point wind measurement on the top of the nacelle cannot accurately reflect the three-dimensional wind field distribution within the swept surface of the wind turbine, resulting in a mismatch between the yaw control strategy and the actual wind conditions. Offshore wind farms often encounter extreme wind conditions such as typhoons, strong wind shear, and turbulence, which cause the unit to bear unbalanced loads, affecting the stability and lifespan of the yaw system. Frequent yaw adjustments may lead to increased vibration of key components such as gearboxes and bearings, and increased oil temperature, thereby affecting the unit's operating efficiency and safety. Traditional methods rely solely on sensors atop the nacelle, making it difficult to capture vertical wind zone differences, such as low-altitude strong wind shear or high-altitude turbulence. This leads to a mismatch between yaw strategies and the actual wind field. Existing wind zone models are mostly fixed-height stratified, failing to consider the dynamic correlation of wind conditions, resulting in data redundancy or omission of key wind layers. Current systems mostly trigger alarms after vibration or temperature exceedances, lacking prediction of yaw adjustments and failing to prevent potential failures. They typically analyze yaw errors, vibration, or temperature independently, without establishing a model of the impact of wind data and mechanical conditions, making it difficult to comprehensively optimize yaw strategies. Summary of the Invention

[0003] (a) Technical problems to be solved

[0004] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a method for monitoring the yaw system status of offshore wind turbine generators, which can effectively solve the problems of the existing technology.

[0005] (II) Technical Solution

[0006] To achieve the above objectives, the present invention provides the following technical solution: This invention discloses a method for monitoring the yaw system status of an offshore wind turbine generator set, applicable to offshore wind turbine generator sets equipped with a yaw control system, comprising the following steps: Step 1: Perform initial wind zone division in the working area of ​​the target generator set, deploy wind acquisition components based on vertical height distribution, construct distributed acquisition points, and acquire wind speed, wind direction, and turbulence intensity data of each wind zone in real time; when there is a conflict between the nacelle sensor and the distributed wind zone data, the wind zone data of the vertically distributed acquisition points shall prevail; periodically check the calibration status of the acquisition components, and switch to the backup sensor when abnormal; Step 2: Based on the initial wind zone division results, merge wind zones with a correlation degree higher than the preset threshold through correlation analysis, and independently save wind zones with low correlation degree to form a multi-height layer wind zone model; Step 3: Real-time acquisition of generator set yaw angle, speed, braking status, and ambient wind direction fed back by the nacelle top wind sensor; calculation of the deviation between ambient wind direction and nacelle position; identification of yaw error magnitude. Step 4: Synchronously monitor the vibration spectrum of the yaw gearbox, the current fluctuation of the motor load, and the gearbox oil temperature and motor winding temperature in the yaw control system, and extract vibration, current, and temperature characteristic parameters; collect gearbox vibration characteristics through an accelerometer, and the vibration characteristic parameters include the proportion of high-frequency vibration energy of the gearbox and the amplitude of the bearing characteristic frequency, and extract them through fast Fourier transform; temperature characteristics are collected through thermocouples or infrared sensors, and the temperature characteristics include the rate of change of oil temperature and the winding temperature rise gradient; Step 5: Construct an impact prediction model using deep learning algorithms. Generate an initial yaw adjustment scheme based on feedback data from wind sensors pre-installed on the top of the cabin. Combined with wind data from various wind regions, the impact prediction model predicts the impact coefficients of the scheme on yaw error, vibration, temperature, and current characteristics. Step 6: Determine whether each influence coefficient meets the preset safety threshold. If it does, execute the adjustment plan; otherwise, dynamically correct the yaw adjustment plan based on the influence coefficient and related parameters. Step 7: Output the revised yaw adjustment scheme to the generator set's yaw control system and back it up to the repository.

[0007] Furthermore, the wind data acquisition components in step 1 include lidar, ultrasonic anemometer, and cup anemometer, and the data acquisition points are distributed in a vertical gradient, with a height difference of 5-15 meters between adjacent points, covering a range from the top of the cabin to 150 meters above sea level.

[0008] Furthermore, the association analysis process in step 2 includes the following steps: Step 21: Perform time synchronization processing on the wind speed, wind direction and turbulence intensity data collected from each wind area, and standardize the data of different dimensions to make the parameters comparable; Step 22: Calculate the linear correlation between wind speed sequences in different wind regions based on the Pearson correlation coefficient. If the wind speed changes in the two wind regions are highly synchronized, they are considered to have a strong correlation. Calculate the similarity of turbulence intensity based on Euclidean distance. If the turbulence intensity change patterns in the two wind regions are similar, it further supports the merging decision. Step 23: If the wind speed correlation between two wind regions exceeds a preset value, they are determined to belong to the same wind condition layer. If the difference in turbulence intensity between the two wind regions is lower than a certain limit, their turbulence characteristics are determined to be similar. When the wind speed correlation and turbulence intensity similarity both meet the threshold conditions, the two wind regions are marked as highly correlated wind regions. Step 24: Merge highly correlated wind regions into the same wind region layer. Take the weighted average of the merged wind speed and direction, with the weight being the proportion of data collection time for each wind region. Take the maximum value of the merged wind region for turbulence intensity. Save the data of low-correlation wind regions independently and mark their vertical height range. After merging wind regions, the system reconstructs the multi-height-layer wind region model and continuously monitors the newly divided wind region data. If subsequent monitoring finds that changes in wind conditions cause the original division to no longer be applicable, such as a decrease in wind speed correlation or an increase in turbulence differences, the correlation analysis is re-executed.

[0009] Furthermore, the calculation logic for the linear correlation of wind speed sequences between different wind regions in step 22 is as follows: Time alignment processing is performed on the wind speed sequences of the two wind regions within the same time period; Obtain the average value of all data points for the wind speed sequence in each wind region; Obtain the deviation of each data point within each wind zone from its average value; Multiply the deviations at corresponding time points in the two wind regions and sum them up to obtain the cumulative covariance. Calculate the sum of squares of the deviations of all data points in the two wind regions and take the square root to obtain the product of the standard deviations of the two wind regions. The correlation coefficient is obtained by dividing the cumulative covariance by the product of the standard deviations.

[0010] Furthermore, the calculation process for the turbulence intensity similarity in step 22 is as follows: ; In the formula, The Euclidean distance representing the turbulence intensity between two wind regions A and B. Represents the number of sampling points. Representing Wind Domain A Turbulence intensity values ​​at each time point Representing Wind Domain B Turbulence intensity values ​​at each time point.

[0011] Furthermore, the calculation process for the yaw error magnitude in step 3 includes: ; In the formula, This represents the corrected yaw angle error. Represents the angle of environmental wind direction. Represents the cabin's orientation angle. This represents an empirical coefficient used to adjust the weighting of turbulence in correcting for errors. It represents the intensity of turbulence, which is the ratio of the standard deviation of wind speed to the average wind speed.

[0012] Furthermore, the process of constructing the prediction model in step 5 is as follows: Historical wind force data from multiple wind regions, records of yaw adjustment schemes, and measured changes in the corresponding yaw error amplitude, vibration spectrum characteristics, temperature characteristics, and current fluctuation characteristics are obtained to construct a training dataset. The training data were preprocessed to normalize the wind speed, turbulence intensity, and temperature data, and the vibration spectrum was transformed by time and frequency to extract frequency domain features. A deep neural network architecture is constructed, with its input layer including wind area division data, cabin sensor data and current yaw status, and its output layer being the predicted values ​​of the influence coefficients of each feature parameter. Long Short-Term Memory (LSTM) networks are used to process time-series wind data, and an attention mechanism is introduced to dynamically weight the influence of wind regions at different altitude levels. The model is trained using the mean squared error loss function and backpropagation algorithm, and an early stopping method is used to prevent overfitting. The model parameters with the best performance on the validation set are saved. The trained impact prediction model is integrated into the yaw control system to receive online data in real time and output the impact coefficient prediction results.

[0013] Furthermore, the dynamic correction process of the yaw adjustment scheme in step 6 is as follows: The influence coefficients are classified and determined according to yaw error, vibration, temperature and current fluctuation, and the main types of parameters exceeding the limits are identified. If the yaw error influence coefficient exceeds the limit, prioritize increasing the yaw angle adjustment or shortening the adjustment interval according to the preset scale; if the vibration influence coefficient exceeds the limit, reduce the yaw motor speed or perform yaw action in stages according to the preset scale; if the temperature influence coefficient exceeds the limit, trigger forced cooling of the gearbox or suspend yaw until the temperature drops to a safe threshold; if the current fluctuation influence coefficient exceeds the limit, limit the instantaneous power of the motor and smooth the load change curve. When multiple influence coefficients exceed the limit simultaneously, a weighted adjustment is performed based on a preset priority rule, where the weight is dynamically calculated by the proportion of each coefficient exceeding the limit. Input the revised adjustment parameters into the impact prediction model and recalculate the impact coefficients until all coefficients meet the safety threshold.

[0014] Furthermore, the dynamic calculation logic for the over-limit ratio is as follows: For each impact coefficient exceeding the limit, calculate the percentage deviation between its actual value and the safety threshold; this percentage is the proportion of the coefficient exceeding the limit. Based on the preset priority rules, a basic weight is assigned to each type of coefficient, and a weighted adjustment is made in combination with its over-limit ratio, so that the coefficient with a more severe over-limit degree has a greater influence weight in the final correction strategy.

[0015] Furthermore, in step 7, the database receives the implementation parameters, actual impact coefficients, and corresponding multi-wind-area wind force data of each yaw adjustment scheme, extracts wind force characteristics from vibration and temperature exceedance cases, and updates the training dataset of the step impact prediction model.

[0016] (III) Beneficial Effects

[0017] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: 1. By dividing wind zones into multiple height layers and performing dynamic correlation analysis, the system can collect vertical wind speed gradients, turbulence intensity changes, and wind direction differences in a distributed manner. Based on the yaw error amplitude and combined with real-time monitoring of multi-dimensional characteristic parameters such as vibration, temperature, and current, the system can predict the impact. This can effectively avoid yaw lag or over-adjustment problems caused by wind shear or turbulence, thereby improving the response accuracy and operational stability of the yaw system, reducing power generation efficiency losses caused by yaw errors, and ensuring that the wind turbine is always at the optimal windward angle, thus improving wind energy capture efficiency.

[0018] 2. By integrating the vibration spectrum of the gearbox, the current fluctuation of the motor, and the temperature characteristics, potential mechanical or electrical faults in the yaw system can be identified. Based on the impact coefficient assessment of the impact prediction model, the negative impact on key components can be predicted before yaw adjustment, proactively avoiding the risk of exceeding limits, extending the service life of core components such as the yaw gearbox and motor, and reducing the downtime maintenance costs caused by sudden failures.

[0019] 3. By dynamically merging highly correlated wind regions, the amount of redundant data calculation can be reduced, the load on edge computing nodes can be lowered, and the system operating efficiency can be improved while ensuring monitoring accuracy. The data accumulated over a long period of operation can also be used to formulate yaw adjustment strategies and further tap the power generation potential. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0021] Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a flowchart illustrating the correlation analysis process in this invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0023] The present invention will be further described below with reference to embodiments.

[0024] This embodiment provides a method for monitoring the yaw system status of an offshore wind turbine, such as... Figure 1 As shown, it includes the following steps: Step 1: Initially divide the working area of ​​the target generator set into wind zones, deploy wind data acquisition components based on vertical height distribution, construct distributed acquisition points, and acquire wind speed, wind direction, and turbulence intensity data for each wind zone in real time; when there is a conflict between the nacelle sensor and the distributed wind zone data, the wind zone data from the vertically distributed acquisition points shall prevail; periodically check the calibration status of the acquisition components, and switch to backup sensors when abnormal; the wind data acquisition components include lidar, ultrasonic anemometer, and cup anemometer, and each acquisition point is distributed according to a vertical gradient, with a height difference of 10 meters between adjacent points, covering a range from the top of the nacelle to 150 meters above sea level; accurately capture wind shear and turbulence stratification phenomena, avoiding the limitations of single-height measurement.

[0025] Step 2: Based on the initial wind area division results, merge wind areas with a correlation degree higher than the preset threshold through correlation analysis, and independently save low correlation wind areas to form a dynamically updated multi-height layer wind area model. Merge similar wind areas through correlation analysis, and independently save areas with significant differences to form a dynamically updated wind area model that better reflects actual wind condition changes. Merging highly correlated wind areas can reduce redundant data processing and improve system real-time performance. Step 3: Real-time acquisition of generator set yaw angle, speed, braking status, and ambient wind direction feedback from the nacelle top wind sensor; calculation of the deviation between the ambient wind direction and the nacelle position; identification of yaw error magnitude; the calculation process for yaw error magnitude includes: ; In the formula, This represents the corrected yaw angle error. Represents the angle of environmental wind direction. Represents the cabin's orientation angle. This represents an empirical coefficient used to adjust the weighting of turbulence in correcting for errors. It represents the intensity of turbulence, which is the ratio of the standard deviation of wind speed to the average wind speed.

[0026] Step 4: Synchronously monitor the vibration spectrum of the yaw gearbox, the current fluctuation of the motor load, and the gearbox oil temperature and motor winding temperature in the yaw control system, and extract vibration, current, and temperature characteristic parameters; collect gearbox vibration characteristics through an accelerometer, including the proportion of high-frequency vibration energy in the gearbox and the amplitude of the bearing characteristic frequency, and extract them through fast Fourier transform; collect temperature characteristics through thermocouples or infrared sensors, including the rate of change of oil temperature and the temperature rise gradient of the windings; this can identify problems such as gearbox wear, motor overload, or lubrication failure at an early stage, capture minor damage to the gearbox, and avoid sudden failures.

[0027] Step 5: Construct an impact prediction model using deep learning algorithms. Generate an initial yaw adjustment scheme based on feedback data from wind sensors pre-installed on the top of the cabin. Combined with wind data from various wind regions, the impact prediction model predicts the impact coefficients of the scheme on yaw error, vibration, temperature, and current characteristics. The process of constructing the predictive model is as follows: Historical wind force data from multiple wind regions, records of yaw adjustment schemes, and measured changes in the corresponding yaw error amplitude, vibration spectrum characteristics, temperature characteristics, and current fluctuation characteristics are obtained to construct a training dataset. The training data were preprocessed to normalize the wind speed, turbulence intensity, and temperature data, and the vibration spectrum was transformed by time and frequency to extract frequency domain features. A deep neural network architecture is constructed, with its input layer including wind area division data, cabin sensor data and current yaw status, and its output layer being the predicted values ​​of the influence coefficients of each feature parameter. Long Short-Term Memory (LSTM) networks are used to process time-series wind data, and an attention mechanism is introduced to dynamically weight the influence of wind regions at different altitude levels. The model is trained using the mean squared error loss function and backpropagation algorithm, and an early stopping method is used to prevent overfitting. The model parameters with the best performance on the validation set are saved. The trained impact prediction model is integrated into the yaw control system to receive online data in real time and output the impact coefficient prediction results.

[0028] Step 6: Determine whether each influence coefficient meets the preset safety threshold. If it does, execute the adjustment plan; otherwise, dynamically correct the yaw adjustment plan based on the influence coefficient and related parameters.

[0029] Step 7: Output the revised yaw adjustment scheme to the yaw control system of the generator set and back it up to the storage repository; The database receives the implementation parameters, actual impact coefficients and corresponding multi-wind area wind data of each yaw adjustment scheme, extracts the wind characteristics in vibration and temperature exceedance cases, updates the training dataset of the step impact prediction model, and optimizes the system through historical data feedback. It can establish yaw adjustment effect evaluation index, and generate long-term yaw strategy optimization suggestions by combining nacelle structural fatigue data.

[0030] Compared with existing technologies, this method constructs a multi-height-layer wind domain model by using wind acquisition components with vertical gradient distribution. Combined with correlation analysis, it dynamically merges or independently divides wind domains, improving the accuracy of wind field characterization under complex sea conditions and overcoming the yaw error caused by traditional single-point wind measurement. Based on deep learning, the impact prediction model integrates multi-wind domain data and historical cases to predict the impact coefficient of yaw adjustment on vibration, temperature and other characteristics in real time, achieving proactive safety control. By storing actual adjustment effect data and feeding it back into model training, the wind domain division rules and yaw strategies are continuously optimized.

[0031] At other levels, this embodiment provides a process for correlation analysis, such as Figure 2 As shown, it includes the following steps: Step 21: Perform time synchronization processing on the wind speed, wind direction and turbulence intensity data collected from each wind area, and standardize the data of different dimensions to make the parameters comparable.

[0032] Step 22: Calculate the linear correlation between wind speed sequences in different wind regions based on the Pearson correlation coefficient. If the wind speed changes in two wind regions are highly synchronized, they are considered to have a strong correlation. Calculate the similarity of turbulence intensity based on Euclidean distance. If the turbulence intensity change patterns in two wind regions are similar, this further supports the merging decision. The calculation logic for the linear correlation between wind speed sequences in different wind regions is as follows: Time alignment processing is performed on the wind speed sequences of the two wind regions within the same time period; Obtain the average value of all data points for the wind speed sequence in each wind region; Obtain the deviation of each data point within each wind zone from its average value; Multiply the deviations at corresponding time points in the two wind regions and sum them up to obtain the cumulative covariance. Calculate the sum of squares of the deviations of all data points in the two wind regions and take the square root to obtain the product of the standard deviations of the two wind regions. The correlation coefficient is obtained by dividing the cumulative covariance by the product of the standard deviations. If the absolute value of the coefficient is greater than or equal to 0.8, the wind speed changes in the two wind zones are considered to be highly synchronized. The calculation process for turbulence intensity similarity is as follows: ; In the formula, The Euclidean distance representing the turbulence intensity between two wind regions A and B. Represents the number of sampling points. Representing Wind Domain A Turbulence intensity values ​​at each time point, Representing Wind Domain B The turbulence intensity value at each time point, if If the turbulence characteristics of the two wind zones are similar, then it can be determined that they are similar.

[0033] Step 23: If the wind speed correlation between two wind regions exceeds a preset value, they are determined to belong to the same wind condition layer. If the difference in turbulence intensity between the two wind regions is lower than a certain limit, their turbulence characteristics are determined to be similar. When the wind speed correlation and turbulence intensity similarity both meet the threshold conditions, the two wind regions are marked as highly correlated wind regions.

[0034] Step 24: Merge highly correlated wind regions into the same wind region layer. Take the weighted average of the merged wind speed and direction, with the weight being the proportion of data collection time for each wind region. Take the maximum value of the merged wind region for turbulence intensity. Save the data of low-correlation wind regions independently and mark their vertical height range. After merging wind regions, the system reconstructs the multi-altitude wind region model and continuously monitors the newly divided wind region data. If subsequent monitoring reveals changes in wind conditions that render the original divisions inapplicable, such as decreased wind speed correlation or increased turbulence differences, the correlation analysis is re-executed to ensure that the wind region divisions always conform to the current environmental conditions.

[0035] Compared with existing technologies, this method is based on the vertically distributed collection of wind speed, wind direction and turbulence intensity data. It dynamically merges highly correlated wind areas using both Pearson correlation coefficient and Euclidean distance criteria, which solves the problem that traditional single-point monitoring cannot reflect the wind shear and turbulence stratification characteristics. It introduces time synchronization and standardization processing and weighted fusion rules to ensure that the merged wind area data is both representative and real-time, and overcomes the lag defect of fixed height layer models under complex sea conditions. By jointly determining the wind speed correlation threshold and the turbulence intensity similarity threshold, and combining them with subsequent continuous monitoring and dynamic reclassification mechanisms, the risk of yaw misjudgment caused by sudden changes in wind conditions is reduced. Compared with existing technologies, invalid yaw actions are reduced. The fast calculation method of the product of cumulative covariance and standard deviation and the simplified formula of Euclidean distance of turbulence intensity are used to reduce the computational load while ensuring the accuracy of analysis, making it suitable for edge computing scenarios of offshore units.

[0036] This embodiment provides a dynamic correction process for a yaw adjustment scheme, specifically as follows: The influence coefficients are classified and determined according to yaw error, vibration, temperature and current fluctuation, and the main types of parameters exceeding the limits are identified. If the yaw error influence coefficient exceeds the limit, increase the yaw angle adjustment amount or shorten the adjustment interval according to the preset scale; if the vibration influence coefficient exceeds the limit, reduce the yaw motor speed or perform yaw action in stages according to the preset scale; if the temperature influence coefficient exceeds the limit, trigger the gearbox to force cooling or suspend yaw until the temperature drops back to the safe threshold; if the current fluctuation influence coefficient exceeds the limit, limit the instantaneous power of the motor and smooth the load change curve. When multiple influence coefficients exceed the limit simultaneously, a weighted adjustment is performed based on a preset priority rule, which defaults to vibration > temperature > current > error. The weights are dynamically calculated by the proportion of each coefficient exceeding the limit. Input the revised adjustment parameters into the impact prediction model and recalculate the impact coefficients until all coefficients meet the safety threshold.

[0037] The dynamic calculation logic for the excess ratio is as follows: For each impact coefficient exceeding the limit, calculate the percentage deviation between its actual value and the safety threshold; this percentage is the proportion of the coefficient exceeding the limit. Based on the preset priority rules, a basic weight is assigned to each type of coefficient, and a weighted adjustment is made in combination with its over-limit ratio, so that the coefficient with a more severe over-limit degree has a greater influence weight in the final correction strategy; by combining the weighted results of all coefficients, an optimization scheme that takes into account both safety and adjustment efficiency is generated, ensuring that the corrected yaw action can effectively mitigate the main risks and avoid secondary impact on other parameters.

[0038] Compared with existing technologies, this technology adopts differentiated preset adjustment strategies for different types of over-limit errors such as yaw error, vibration, temperature, and current fluctuations. This avoids secondary risks caused by single, crude adjustments. By integrating over-limit ratio calculation with preset priorities, it achieves scientific decision-making when multiple parameters conflict. It prioritizes the suppression of high-risk parameters while taking into account the stability of other parameters. The over-limit ratio is dynamically calculated based on the deviation percentage, so that the correction strategy automatically focuses on the most serious problem. This avoids the lag of human experience intervention, improves response efficiency under complex wind conditions, and effectively extends the life of key components such as gearboxes and motors while ensuring power generation efficiency.

[0039] In summary, this invention constructs a multi-height-layer wind domain model by vertically distributed wind acquisition components, and combines real-time wind speed, wind direction and turbulence intensity data to achieve refined wind field perception. Based on dynamic wind domain division and correlation analysis, it improves the accuracy and adaptability of yaw control. By monitoring yaw error, vibration spectrum, current fluctuation and temperature characteristics, and predicting the influence coefficient of yaw adjustment scheme on each parameter, it ensures the safety and reliability of the adjustment scheme. By constructing an impact prediction model, the yaw strategy can be dynamically optimized, thereby effectively reducing mechanical wear and energy loss, improving power generation efficiency, and extending equipment life.

[0040] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring the state of a yaw system of an offshore wind turbine generator system, applied to an offshore wind turbine generator system, wherein a yaw control system is arranged on the offshore wind turbine generator system, and characterized in that, The method comprises the following steps: Step 1: initial wind area division is performed in the working area of the target generator set, wind power collection components are arranged based on vertical height distribution, distributed collection points are constructed, and wind power data of each wind area is obtained in real time; Step 2: based on the initial wind area division result, wind areas with a correlation degree higher than a preset threshold are merged through correlation analysis, and wind areas with a low correlation degree are saved independently to form a multi-height layer wind area model; Step 3: real-time collection of environmental wind direction and operating parameters of the generator set is performed, and yaw error is identified; Step 4: the yaw control system is synchronously monitored, and vibration, current and temperature characteristic parameters are extracted; Step 5: an influence prediction model is constructed through a deep learning algorithm, an initial yaw adjustment scheme is generated based on feedback data of a wind sensor preset on the top of the nacelle, wind power data of each wind area is combined, and the influence prediction model predicts influence coefficients of the scheme on yaw error, vibration, temperature and current characteristics; Step 6: whether each influence coefficient meets a preset safety threshold is judged, if yes, the adjustment scheme is executed, otherwise, the yaw adjustment scheme is dynamically corrected based on the influence coefficients and correlation parameters; Step 7: the corrected yaw adjustment scheme is output to the yaw control system.

2. A method of offshore wind turbine yaw system condition monitoring according to claim 1, characterized in that, The wind power collection components in the step 1 comprise a laser radar, an ultrasonic anemometer and a cup anemometer, and each collection point is distributed according to a vertical gradient, and the height difference between adjacent points is 5-15 meters.

3. A method of offshore wind turbine yaw system condition monitoring according to claim 1, characterized in that, The correlation analysis process in the step 2 comprises the following steps: Step 21: time synchronization processing is performed on wind speed, wind direction and turbulence intensity data collected by each wind area; Step 22: linear correlation of wind speed sequences between different wind areas is calculated based on a Pearson correlation coefficient, and similarity of turbulence intensity is calculated based on an Euclidean distance; Step 23: if the wind speed correlation of two wind areas exceeds a preset value, it is determined that they belong to the same wind condition layer, if the turbulence intensity difference of two wind areas is lower than a certain limit value, it is determined that their turbulence characteristics are similar, and when the wind speed correlation and the turbulence intensity similarity both meet the threshold conditions, the two wind areas are marked as high-correlation wind areas; Step 24: high-correlation wind areas are merged into the same wind area layer, the wind speed and wind direction of the merged wind areas are taken as weighted average values, the turbulence intensity is taken as the maximum value of the merged wind areas, low-correlation wind area data is saved independently, and the vertical height interval thereof is marked.

4. A method of offshore wind turbine yaw system condition monitoring according to claim 1, characterized in that, The wind power data in the step 1 comprises wind speed, wind direction and turbulence intensity.

5. A method of offshore wind turbine yaw system condition monitoring according to claim 3, characterized in that, The calculation process of the turbulence intensity similarity in the step 22 is as follows: ; wherein, represents the Euclidean distance of the turbulence intensity of the two wind zones A and B, represents the number of sampling points, represents the turbulence intensity value of the wind zone A at the time point, represents the turbulence intensity value of the wind zone B at the time point.

6. A method of offshore wind turbine yaw system condition monitoring according to claim 1, characterized in that, The calculation process of the yaw error amplitude in the step 3 comprises: ; In the formula, represents the corrected yaw angle error, represents the environmental wind direction angle, represents the cabin heading angle, represents the experience coefficient, represents the turbulence intensity.

7. A method of offshore wind turbine yaw system condition monitoring according to claim 1, characterized in that, The construction process of the influence prediction model in the step 5 is as follows: Historical multi-wind-area wind power data, yaw adjustment scheme records, and corresponding yaw error amplitudes, vibration frequency spectrum characteristics, temperature characteristics and current fluctuation characteristics are obtained to construct a training data set; The training data is preprocessed, wind speed, turbulence intensity and temperature data are normalized, and vibration frequency spectrum is subjected to time-frequency transformation to extract frequency domain characteristics; A deep neural network architecture is constructed, the input layer of which comprises wind area division data, nacelle sensor data and current yaw state, and the output layer is a prediction value of the influence coefficient of each characteristic parameter; The long short-term memory network is used to process the time-series wind condition data, and the model is trained through the mean square error loss function and the back propagation algorithm. The trained influence prediction model is integrated into the yaw control system to receive online data and output the prediction results of the influence coefficients in real time.

8. A method of offshore wind turbine yaw system condition monitoring according to claim 1, characterized in that, The dynamic correction process of the yaw adjustment scheme in step 6 is as follows: The influence coefficient classification determination is performed, and the influence coefficients exceeding the limit are classified according to the yaw error, vibration, temperature, and current fluctuation to determine the main parameter type exceeding the limit. If the yaw error influence coefficient exceeds the limit, the yaw angle adjustment amount is preferentially increased or the adjustment interval is shortened according to the preset scale; if the vibration influence coefficient exceeds the limit, the yaw motor speed is reduced or the yaw action is executed in stages according to the preset scale; if the temperature influence coefficient exceeds the limit, the gear box forced cooling is triggered or the yaw is suspended until the temperature falls below the safety threshold; if the current fluctuation influence coefficient exceeds the limit, the motor instantaneous power is limited and the load change curve is smoothed. When multiple types of influence coefficients exceed the limit simultaneously, weighted adjustment is performed based on the preset priority rules, and the weight is dynamically calculated based on the exceeding limit proportion of each coefficient. The corrected adjustment parameters are input into the influence prediction model to recalculate the influence coefficients until all coefficients meet the safety threshold.

9. A method of offshore wind turbine yaw system condition monitoring according to claim 8, characterized in that, The dynamic calculation logic of the exceeding limit proportion is as follows: For each influence coefficient exceeding the limit, the deviation percentage between the actual value and the safety threshold is calculated, which is the exceeding limit proportion of the coefficient. According to the preset priority rules, each type of coefficient is assigned a basic weight, and weighted adjustment is performed based on the exceeding limit proportion, so that the influence weight of the coefficient with a more serious exceeding limit in the final correction strategy is greater.

10. A method of offshore wind turbine yaw system condition monitoring according to claim 1, characterized in that, In step 7, the database receives the implementation parameters of each yaw adjustment scheme, the actual influence coefficients, and the corresponding multi-wind-region wind data, extracts the wind characteristics in the vibration and temperature exceeding limit cases, and updates the training data set of the influence prediction model in step 6.

Images