Over-speed suppression method based on wind turbine generator performance boundary

Abstract

An overspeed suppression method based on a wind turbine generator performance boundary comprises the following steps: step 1, establishing a dynamic model of a fan for simulating fan response under different wind conditions; 2, evaluating the mechanical stress and fatigue life of the fan under various working conditions; step 3, taking maximization of the service life of the wind turbine generator as a target function, and preliminarily setting a performance boundary; 4, a redundant sensor is installed for the wind turbine generator, and data fusion is carried out; 5, performing anomaly detection and fault prediction on the wind turbine generator by using a big data analysis tool; step 6, learning a normal operation mode from the historical data by using a deep learning model, and identifying an abnormal behavior; step 7, automatically selecting proper suppression measures according to different levels of overspeed conditions; and step 8, repeatedly adjusting the performance boundary to form a final overspeed suppression strategy. The performance boundary of the wind turbine generator can be scientifically analyzed and set, and safe operation of the wind turbine generator in various complex environments can be ensured.

Description

Technical Field

[0001] The invention belongs to the field of overspeed suppression methods for wind turbine generator sets, and in particular relates to an overspeed suppression method based on a performance boundary of a wind turbine generator set. Background Art

[0002] With the continuous development of wind power generation technology, the safety and reliability of wind turbines have become key issues in wind farm operations. Especially in complex and changeable natural environments, wind turbines may face various extreme conditions, such as strong winds and turbulence, which may cause wind turbines to overspeed, thus causing mechanical failures, electrical failures and even serious safety accidents. Therefore, it is particularly important to develop an effective overspeed suppression method that can not only ensure the safe operation of wind turbines in various complex environments, but also continuously improve their operating efficiency and lifespan through continuous optimization.

[0003] At present, traditional overspeed suppression methods are usually based on simple threshold judgments and fixed control strategies, lacking in-depth analysis of wind turbine performance boundaries and dynamic adjustment capabilities. This method is often not flexible enough when dealing with complex and changeable wind conditions, and is prone to the risk of premature shutdown or failure to respond in time. In addition, with the development of big data and artificial intelligence technologies, modern wind farms have accumulated a large amount of operating data, but how to effectively use this data to optimize overspeed suppression strategies remains a problem that needs to be solved.

[0004] Based on the above analysis, the technical problem to be solved by the present invention is that the traditional overspeed suppression method is usually based on simple threshold judgment and fixed control strategy, lacking in-depth analysis of the fan performance boundary. It fails to fully consider the fan response characteristics under different wind conditions, environmental conditions and operating conditions, resulting in unreasonable performance boundary settings, and the risk of premature shutdown or failure to respond in time. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an overspeed suppression method based on the performance boundary of a wind turbine set, which can scientifically analyze and set the performance boundary of a wind turbine set, and develop an effective overspeed suppression strategy on this basis; it can not only ensure the safe operation of the wind turbine in various complex environments, but also continuously improve its operating efficiency and lifespan through continuous optimization.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: An overspeed suppression method based on the performance boundary of a wind turbine generator system comprises the following steps: Step 1: Establish a dynamic model of the wind turbine to simulate the response of the wind turbine under different wind conditions. The specific method is as follows: Use simulation software (such as Bladed, FAST) to build a dynamic model of the wind turbine. These software are based on the principles of aerodynamics, structural mechanics, etc., and can accurately simulate the operating status of the wind turbine under different wind conditions.

[0007] Simulate the speed under different wind conditions (Unit: rpm), power output (Unit: kW), vibration displacement (Unit: mm) and other parameters. By measuring wind parameters such as wind speed (Unit: m / s), wind direction (Unit: degree) and other inputs, combined with the structural parameters and aerodynamic characteristics of the fan, the following dynamic equations are used for simulation: ,in For air power, is the lift coefficient, is the air density, is the blade swept area, is the resistance coefficient.

[0008] ,in is the aerodynamic torque, is the blade radius.

[0009] ,in is the moment of inertia, is the speed change rate, is the generator torque, is the friction torque.

[0010] Through dynamic modeling, the response characteristics of the wind turbine under various working conditions are obtained, including the curve of speed change with wind speed, the relationship between power output and wind speed, etc.

[0011] Step 2: Evaluate the mechanical stress and fatigue life of the fan under various working conditions; the specific method is: Finite element analysis (FEA) and multi-body dynamics analysis methods are used. Finite element analysis discretizes the key components of the wind turbine (such as blades, gearboxes, and generators) into a finite number of units, and calculates the stress and strain of each unit by solving the mechanical equilibrium equation. Multi-body dynamics analysis considers the connection and motion relationship between the components to more accurately simulate the overall dynamic response of the wind turbine.

[0012] Evaluate the stress distribution and fatigue life of key components under different working conditions. For stress analysis, the stress formula can be used ,in is the stress (unit: Pascal), is the force (unit: Newton), is the stress-bearing area (unit: square meters). For fatigue life assessment, fatigue life prediction models can be used, such as Miner's law: ,in For the stress level The number of cycles below, For the stress level Fatigue life under is the damage accumulation parameter.

[0013] Step 3: Taking the maximization of the service life of the wind turbine as the objective function, preliminarily set the performance boundary; the specific method is: Based on theoretical analysis and simulation calculation results, the performance parameter boundaries under different wind conditions are determined. Theoretical analysis can combine the design parameters of the fan and the principles of aerodynamics to derive the theoretical calculation formula for the performance boundary. Simulation calculations simulate the operation of the fan under different wind conditions many times and obtain the reasonable range of performance parameters through statistical analysis.

[0014] Performance parameters include but are not limited to upper speed limit , power output limit , Pitch angle range , Temperature Limit Etc. The objective function can be expressed as ,in is the service life of the wind turbine (unit: hours). Through optimization algorithms, such as genetic algorithm, particle swarm optimization algorithm, etc., under the premise of satisfying various constraints, the performance boundary parameter combination that maximizes the objective function is found.

[0015] Step 4: Install redundant sensors for wind turbines and perform data fusion; the specific method is as follows: Install multiple redundant sensors for key parameters such as speed, power output and temperature. For example, for the speed sensor, it is possible to install multiple sensors of different types, such as photoelectric encoders, Hall sensors, etc.

[0016] Use multi-sensor data fusion technology (such as Kalman filter or multi-source information fusion algorithm) to improve data consistency and reliability. Kalman filter is an optimal estimation method based on the linear system state equation. Its basic formula is: ,in for The estimated state value at time for The estimated state value at time for The Kalman gain at time t, for The measured value at the moment, is the measurement matrix.

[0017] Establish a multi-level alarm system. When the first level threshold is reached When the alarm is triggered, it is close to the higher level threshold , The alarm system can set different thresholds according to different parameters. For example, for the speed, the first threshold can be set as the upper limit of the speed. The second level threshold is the upper speed limit , the third level threshold is the upper speed limit, etc.

[0018] Step 5: Use big data analysis tools to detect anomalies and predict faults of wind turbines; the specific method is as follows: Utilize big data analysis tools on cloud computing platforms (such as Apache Hadoop or Spark). These tools can process large-scale wind turbine operation data and provide efficient data storage and computing capabilities.

[0019] Combine machine learning algorithms (such as random forest and XGBoost) for anomaly detection and fault prediction. Random forest is an ensemble learning algorithm consisting of multiple decision trees. Its prediction formula is: ,in is the predicted value, is the number of decision trees, For the The predicted value of a decision tree. XGBoost is a gradient boosting tree algorithm that improves prediction accuracy by continuously optimizing the loss function.

[0020] Realize real-time monitoring and early warning of the fan operation status. By analyzing real-time data, abnormal situations can be discovered in time and alarms can be issued.

[0021] Step 6: Use deep learning models to learn normal operating patterns from historical data and identify abnormal behaviors; the specific methods are: High-precision speed sensors, vibration sensors, temperature sensors, power sensors, etc. are installed on wind turbines.

[0022] Continuously collect various sensor data, including but not limited to speed , vibration displacement ,temperature ,power wait.

[0023] Introduce deep learning models (such as convolutional neural networks (CNN) or long short-term memory networks (LSTM)) to learn normal operating modes from a large amount of historical data and identify abnormal behaviors. CNN is mainly used to process image data. For example, the vibration signal of the fan can be converted into an image for processing. LSTM is suitable for processing time series data. The operation data of the fan can be input into LSTM in chronological order for training. For example, the formula of LSTM is: , , , , , in , , are the outputs of the input gate, forget gate, and output gate respectively. is the cell state, is the hidden state, For input data, is the weight matrix, is the bias term, is the sigmoid function, is element-wise multiplication.

[0024] Step 7: Automatically select appropriate suppression measures according to different levels of speeding; the specific method is: Develop a graded response algorithm to automatically select appropriate suppression measures based on different levels of speeding (minor, moderate, severe). Ability to define speeding severity indicators ,in is the current speed, is the rated speed, The upper speed limit.

[0025] Minor speeding ( ): Solved by fine-tuning the pitch angle. The pitch angle adjustment formula can be ,in is the current pitch angle, is the adjustment amount, which can be determined according to an empirical formula or an optimization algorithm.

[0026] Moderate speeding ( ): Reduce power output. The power output adjustment formula can be ,in is the current power output, To reduce the amount, it can be determined according to the overspeed degree and fan characteristics.

[0027] Severe speeding ( ): Initiate emergency braking. Emergency braking can be achieved by triggering a mechanical brake device or an electrical brake system.

[0028] Step 8: Repeatedly adjust the performance boundary to form the final overspeed suppression strategy. The specific method is: Step 8.1. Estimation of mechanical stress and fatigue life based on actual measured sensor parameters 1. Data collection and processing: Continuously collect data from sensors installed at key parts of the wind turbine, including but not limited to speed sensors, vibration sensors, temperature sensors, strain sensors, etc. Ensure the accuracy and real-time nature of the data, and use data filtering technology to remove noise interference.

[0029] Preprocess the collected data, such as segmenting and smoothing the time series data for subsequent analysis.

[0030] 2. Mechanical stress calculation: Based on strain sensor data and material mechanics principles, calculate the stress distribution of key components (such as blades, towers, gearboxes, etc.) under actual operating conditions. Ability to use Hooke's law ,in represents stress, represents the elastic modulus, Indicates strain.

[0031] Considering the combined effects of wind load, gravity, inertia force and other loads, the stress state of complex structures is accurately calculated using the finite element analysis method.

[0032] 3. Fatigue life assessment: Use fatigue life prediction models, such as the SN curve method or Miner linear cumulative damage theory, to calculate the fatigue damage of key components based on the number of stress cycles and stress amplitude.

[0033] Consider the impact of random loads and environmental factors (such as temperature and humidity) on fatigue life during actual operation, and make corrections and adjustments.

[0034] Step 8.2: Compare predictions with actual results 1. Establish a comparative indicator system: Determine key performance indicators, such as maximum stress value, average stress value, fatigue life, speed fluctuation range, etc., for comparison between predicted results and actual measurement results.

[0035] Set a reasonable error range, such as controlling the stress error within a certain percentage and the fatigue life error within a certain number of hours.

[0036] 2. Data analysis and comparison: Compare the actual measured performance indicators with the predicted results based on theoretical analysis and simulation calculations. The differences can be intuitively demonstrated using methods such as charts and statistical analysis.

[0037] The reasons for the analysis of the differences may include model errors, sensor accuracy issues, and insufficient consideration of environmental changes.

[0038] Step 8.3: Repeatedly adjust performance boundaries 1. Optimization algorithm selection: Simulated annealing algorithm: simulates the annealing process of solid materials and searches for the global optimal solution by gradually reducing the temperature. When adjusting the performance boundary, the objective function is defined as minimizing the difference between the actual and predicted performance indicators, randomly generating the initial solution (performance boundary parameter combination), and then continuously iterating and updating the solution, accepting the worse solution with a certain probability to avoid falling into the local optimum.

[0039] Ant Colony Algorithm: simulates the behavior of ants in the process of searching for food, and guides the search for the optimal solution through the accumulation and update of pheromones. The performance boundary parameter combination is regarded as the path of the ants, and the pheromone intensity is defined to be related to the objective function value. The ants select the next parameter combination for exploration based on the pheromone concentration.

[0040] 2. Iterative adjustment process: Initialize the parameters of the optimization algorithm, such as the initial temperature and cooling rate of the simulated annealing algorithm, the number of ants and pheromone volatility coefficient of the ant colony algorithm, etc.

[0041] A set of initial performance boundary parameter combinations is randomly generated as the starting point of the algorithm.

[0042] In each iteration, a new parameter combination is generated according to the rules of the optimization algorithm, and the corresponding objective function value is calculated. If the new parameter combination makes the objective function value better, then the parameter combination is accepted; otherwise, the worse parameter combination is accepted with a certain probability to increase the algorithm's exploration ability.

[0043] The iterative process is repeated until the stopping condition is met, such as reaching the maximum number of iterations or the objective function value is less than a certain threshold.

[0044] 3. Verification and confirmation: After each adjustment of the performance boundaries, the mechanical stress and fatigue life evaluation under the actual operating conditions is re-performed and compared with the predicted results.

[0045] Check whether the adjusted performance boundary meets the safety requirements, reliability requirements, and operating efficiency requirements. If not, continue to adjust; if so, further verify its stability and robustness.

[0046] Step 8.4: Form the final overspeed suppression strategy 1. Integrate optimization results: The optimal performance boundary parameter combination obtained through repeated adjustments is integrated into the overspeed suppression strategy. Key parameters such as upper speed limit, power output limit, pitch angle range, and temperature limit are determined.

[0047] Consider the adaptability under different wind conditions, environmental conditions and operating states, and formulate corresponding adjustment rules and control strategies.

[0048] 2. Strategy verification and improvement: The developed overspeed suppression strategy is tested in an actual wind farm. The effectiveness and reliability of the strategy are observed by simulating different overspeed scenarios.

[0049] Based on the test results, the overspeed suppression strategy is further improved, such as adjusting the response time, optimizing the control algorithm, adding redundant protection measures, etc.

[0050] The final overspeed suppression strategy document based on the performance boundary of the wind turbine is formed, including the principle of the strategy, parameter settings, implementation steps and precautions, etc., to facilitate practical application and maintenance.

[0051] The present invention can achieve the following beneficial effects: 1. The present invention establishes a dynamic model of the fan to simulate the fan response under different wind conditions, and can more accurately determine the changing rules of parameters such as speed, power output, and vibration. Finite element analysis and multi-body dynamics analysis methods are used to evaluate the mechanical stress and fatigue life of key components under different working conditions to ensure the scientificity and rationality of the performance boundary.

[0052] 2. The present invention installs multiple redundant sensors for key parameters such as speed, power output and temperature to ensure the reliability and accuracy of data. Multi-sensor data fusion technology is used to improve the consistency and reliability of data and reduce the impact of single sensor failure.

[0053] 3. The present invention uses big data analysis tools on the cloud computing platform to process large-scale wind turbine operation data and provide efficient data storage and computing capabilities. It combines machine learning algorithms such as random forest and XGBoost to perform anomaly detection and fault prediction, realize real-time monitoring and early warning of wind turbine operation status, and improve the response speed and accuracy of the system.

[0054] 4. The present invention dynamically adjusts and optimizes according to actual operation data, combines actual working conditions and theoretical analysis results, and repeatedly adjusts the performance boundary to achieve the best overspeed suppression effect. By continuously collecting and analyzing actual operation data, the overspeed suppression strategy is continuously optimized to ensure that it is always in the best state.

[0055] 5. Through finite element analysis and fatigue life prediction models, the mechanical stress and fatigue life of key components under actual operating conditions are evaluated in detail. The combined effects of multiple loads and the influence of environmental factors are considered to ensure that the setting of performance boundaries is more scientific and accurate.

[0056] 6. Introduce deep learning models such as convolutional neural networks and long short-term memory networks to learn normal operating modes from a large amount of historical data and identify abnormal behaviors. Through the optimization algorithm, automatically find the performance boundary parameter combination that maximizes the objective function, reduce manual intervention and improve efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0058] Embodiment 1: Taking a large wind turbine as an example, a large wind farm is located in an area with complex terrain and changeable wind conditions. The wind farm has multiple wind turbines. Due to the complexity and uncertainty of wind conditions, wind turbines often face the risk of overspeed during operation, which not only affects the safe operation of wind turbines, but also may reduce their service life and power generation efficiency. In order to solve this problem, the present invention provides an overspeed suppression method based on the performance boundary of wind turbines, the steps are: Step 1: Establish a dynamic model of the wind turbine to simulate the response of the wind turbine under different wind conditions. The specific method is as follows: Use professional simulation software (such as Bladed, FAST) to build a dynamic model of the wind turbine. These software are based on the principles of aerodynamics, structural mechanics, etc., and can accurately simulate the operating status of the wind turbine under different wind conditions.

[0059] Simulate the speed n (unit: rpm), power output P (unit: kW), vibration displacement d (unit: mm) and other parameters under different wind conditions. By inputting wind parameters such as wind speed v (unit: m / s), wind direction θ (unit: degree), etc., combined with the structural parameters and aerodynamic characteristics of the fan, the following dynamic equations are used for simulation: ,in For air power, is the lift coefficient, is the air density, is the blade swept area, is the drag coefficient. The determination of lift coefficient and drag coefficient is a complex process that requires consideration of factors such as blade shape, angle of attack, and Reynolds number. For example, for blades of different shapes, at different angles of attack, the lift coefficient and drag coefficient will be different. The lift coefficient and drag coefficient can be determined by wind tunnel tests, numerical simulations, and other methods. Air density is affected by factors such as altitude, temperature, and humidity, and air density varies in different geographical locations and climatic conditions. The blade swept area can be calculated from the length and width of the blade, and the blade swept area will also be different for different models of fans.

[0060] ,in is the aerodynamic torque, is the blade radius. The size of the blade radius directly affects the size of the aerodynamic moment. The larger the radius, the larger the aerodynamic moment. In practical applications, the selection of blade radius needs to consider factors such as the power output and speed of the fan to ensure that the fan can operate normally under different wind conditions.

[0061] ,in is the moment of inertia, is the speed change rate, is the generator torque, is the friction torque. The moment of inertia is a physical quantity that describes the rotational inertia of an object. For a wind turbine, the moment of inertia is related to the structure and mass distribution of the wind turbine. The generator torque depends on the working state and load conditions of the generator, and the friction torque includes the torque generated by various friction factors such as bearing friction and air friction. In practical applications, it is necessary to determine the values ​​of these parameters through measurement and calculation in order to more accurately simulate the operating state of the wind turbine.

[0062] Through dynamic modeling, the response characteristics of the fan under various working conditions are obtained, including the speed change curve with wind speed, the relationship between power output and wind speed, etc. For example, at low wind speeds, the speed of the fan may increase slowly; as the wind speed increases, the speed may increase approximately linearly; but when the wind speed exceeds a certain critical value, due to the effect of the fan control system, the speed growth rate will gradually slow down or even stabilize. The relationship between power output and wind speed is also very important. Generally, within a certain wind speed range, power output is proportional to the cube of wind speed. Through a deep understanding of these relationships, we can better grasp the operating characteristics of the fan and provide a solid foundation for subsequent performance analysis and overspeed suppression strategy formulation.

[0063] Step 2: Evaluate the mechanical stress and fatigue life of the fan under various working conditions; the specific method is: Finite element analysis (FEA) and multi-body dynamics analysis methods are used. Finite element analysis discretizes the key components of the wind turbine (such as blades, gearboxes, and generators) into a finite number of units, and calculates the stress and strain of each unit by solving the mechanical equilibrium equation. For example, for blades, the blades can be discretized into a large number of small units, and the stress and strain of each unit can be obtained by solving the mechanical equilibrium equation. At the root of the blade, the stress is usually large due to the large bending moment and torque; at the tip of the blade, due to the high wind speed and large aerodynamic force, large stress will also be generated. Through finite element analysis, the stress distribution of the blade under different working conditions can be accurately determined, providing an important basis for evaluating the strength and fatigue life of the blade.

[0064] Multibody dynamics analysis considers the connection and motion relationship between the various components to more accurately simulate the overall dynamic response of the wind turbine. For example, the connecting shaft between the gearbox and the generator is subject to torque and bending forces during operation. Through multibody dynamics analysis, the stress and deformation of the connecting shaft can be calculated, while the vibration of the wind turbine under different wind conditions and the impact of vibration on each component can also be considered. This analysis method can more accurately simulate the overall dynamic response of the wind turbine and provide a more comprehensive perspective for evaluating the mechanical stress and fatigue life of the wind turbine.

[0065] Evaluate the stress distribution and fatigue life of key components under different working conditions. For stress analysis, the stress formula can be used ,in is the stress (unit: Pascal), is the force (unit: Newton), is the stress area (unit: square meters). For example, for a specific part of the blade, the stress at that part can be calculated by measuring the force applied to that part and determining the stress area at that part. For fatigue life assessment, a fatigue life prediction model can be used, such as Miner's law: ,in For the stress level The number of cycles below, For the stress level Fatigue life under is the damage accumulation parameter. In practical applications, it is necessary to count the number of cycles under different stress levels based on the actual operating data of the fan. Then, the fatigue life under different stress levels is determined based on the fatigue characteristic curve of the material. Finally, the fatigue damage of key components is calculated using Miner's law. At the same time, it is also necessary to consider the influence of random loads and environmental factors (such as temperature and humidity) on fatigue life in actual operation, and make corrections and adjustments. For example, in a high temperature and high humidity environment, the fatigue performance of the material may decrease, and the fatigue life needs to be corrected accordingly.

[0066] Step 3: Taking the maximization of the service life of the wind turbine as the objective function, preliminarily set the performance boundary; the specific method is: Based on theoretical analysis and simulation calculation results, the performance parameter boundaries under different wind conditions are determined. Theoretical analysis can combine the design parameters of the wind turbine with the principles of aerodynamics to derive theoretical calculation formulas for performance boundaries. For example, based on the wind turbine's rated power, speed, blade length and other design parameters, as well as the principles of lift and drag in aerodynamics, theoretical calculation formulas for performance parameters such as the maximum power output limit, speed limit, and pitch angle range under different wind conditions can be derived.

[0067] The simulation calculation simulates the operation of the fan under different wind conditions many times, and obtains the reasonable range of performance parameters through statistical analysis. The fan can be simulated a lot by changing wind parameters such as wind speed, wind direction, temperature, humidity, etc. Record the speed, power output, pitch angle and other parameters of the fan under different wind conditions, and then perform statistical analysis on these data. For example, if it is found that the speed is between 1000-1800 rpm and the power output is between 500-1200 kW in the majority, then it can be preliminarily determined that the upper limit of the speed is 1800 rpm and the power output limit is 1200 kW.

[0068] Performance parameters include but are not limited to upper speed limit , power output limit , Pitch angle range , Temperature Limit Etc. The objective function can be expressed as ,in is the service life of the wind turbine (unit: hours). Through optimization algorithms, such as genetic algorithms, particle swarm optimization algorithms, etc., under the premise of satisfying various constraints, the performance boundary parameter combination that maximizes the objective function is found. Taking the genetic algorithm as an example, the performance boundary parameter combination is regarded as a chromosome, and new parameter combinations are generated through genetic operations such as crossover and mutation. In each generation, the objective function value corresponding to each parameter combination is calculated, that is, the service life of the wind turbine. Select excellent parameter combinations to enter the next generation, and after multiple generations of evolution, finally find the performance boundary parameter combination that maximizes the objective function.

[0069] Step 4: Install redundant sensors for wind turbines and perform data fusion; the specific method is as follows: Install multiple redundant sensors for key parameters such as speed, power output, and temperature. For example, for speed sensors, you can install multiple different types of sensors, such as photoelectric encoders, Hall sensors, etc. Photoelectric encoders determine speed by measuring pulses of light, with high accuracy and fast response. It can be installed on the main shaft of the fan and calculate the speed by measuring the rotation angle of the main shaft. Hall sensors determine speed by detecting changes in the magnetic field, with high reliability and low cost. It can be installed on the stator of the motor and calculate the speed by measuring the frequency of changes in the magnetic field. Installing multiple different types of speed sensors can ensure that when one sensor fails, other sensors can continue to provide accurate speed data, improving system reliability.

[0070] Use multi-sensor data fusion technology (such as Kalman filter or multi-source information fusion algorithm) to improve data consistency and reliability. Kalman filter is an optimal estimation method based on the linear system state equation. Its basic formula is: ,in for The estimated state value at time for The estimated state value at time for The Kalman gain at time t, for The measured value at the moment, is the measurement matrix. In practical applications, the Kalman filter can fuse the measurements of multiple sensors to obtain a more accurate state estimate. For example, assuming that the actual speed of the fan is 1500 rpm, the speed measured by the photoelectric encoder is 1510 rpm, and the speed measured by the Hall sensor is 1490 rpm. Through the Kalman filter, these two measurements can be fused to obtain a more accurate speed estimate, such as 1502 rpm.

[0071] Establish a multi-level alarm system. When the first level threshold is reached When the alarm is triggered, it is close to the higher level threshold , The response measures are upgraded step by step. The alarm system can set different thresholds according to different parameters. For example, for the speed, the first-level threshold can be set to 90% of the upper speed limit. When the speed reaches this threshold, the system sends a warning signal to remind the operator to pay attention. The second-level threshold can be set to 95% of the upper speed limit. When the speed approaches this threshold, the system automatically adjusts the control parameters of the fan to reduce the speed. The third-level threshold can be set to the upper speed limit. When the speed reaches this threshold, the system activates the emergency brake device to ensure the safe shutdown of the fan.

[0072] Step 5: Use big data analysis tools to detect anomalies and predict faults of wind turbines; the specific method is as follows: Utilize big data analysis tools on cloud computing platforms (such as Apache Hadoop or Spark). These tools can process large-scale wind turbine operation data and provide efficient data storage and computing capabilities. Apache Hadoop is an open source distributed computing platform that can store large-scale wind turbine operation data in a distributed manner on multiple nodes and improve data processing speed through parallel computing. For example, the historical operation data of the wind turbine can be stored in Hadoop's distributed file system (HDFS), and then the data can be analyzed and processed using the MapReduce programming model. Spark is a fast in-memory computing framework that can perform fast data processing in memory and is suitable for anomaly detection and fault prediction tasks with high real-time requirements. For example, Spark Streaming can be used to process the sensor data of the wind turbine in real time to detect abnormal situations in a timely manner.

[0073] Combine machine learning algorithms (such as random forest and XGBoost) for anomaly detection and fault prediction. Random forest is an ensemble learning algorithm consisting of multiple decision trees. Its prediction formula is: ,in is the predicted value, is the number of decision trees, For the In practical applications, the random forest algorithm can be used to analyze the vibration data of the fan, and multiple decision trees can be used to vote to determine whether there is an abnormality. If most decision trees believe that the vibration data is abnormal, the system will issue an alarm. XGBoost is a gradient boosting tree algorithm that improves prediction accuracy by continuously optimizing the loss function. For example, when predicting the time of a fan failure, the model parameters can be continuously adjusted based on historical data to improve the accuracy of the prediction.

[0074] Real-time monitoring and early warning of the fan operation status. By analyzing real-time data, abnormal situations can be discovered in time and alarms can be issued. For example, changes in parameters such as temperature, speed, and power output of the fan can be monitored in real time. When the temperature rises suddenly, the speed fluctuates abnormally, or the power output decreases, the system can detect abnormal situations in time and issue alarms by monitoring the changes in these parameters in real time. At the same time, the system can also take different response measures according to the severity of the abnormal situation, such as adjusting the control parameters of the fan, notifying maintenance personnel, etc.

[0075] Step 6: Use deep learning models to learn normal operating patterns from historical data and identify abnormal behaviors; the specific methods are: High-precision speed sensors, vibration sensors, temperature sensors, power sensors, etc. are installed on wind turbines. These sensors can monitor the operating status of the wind turbine in real time and provide a large amount of historical data for the deep learning model. For example, the speed sensor can collect data once a second to record the speed change of the wind turbine; the vibration sensor can collect data every 0.1 second to detect the vibration of the wind turbine; the temperature sensor can measure the temperature of each component of the wind turbine, and the power sensor can monitor the power output of the wind turbine.

[0076] Continuously collect various sensor data, including but not limited to speed n, vibration displacement d, temperature T, power P, etc. These data can reflect the operating status and performance of the fan. For example, speed data can reflect the rotation speed of the fan, vibration displacement data can reflect the structural stability of the fan, temperature data can reflect the operating temperature of each component of the fan, and power data can reflect the power generation capacity of the fan.

[0077] Introduce deep learning models (such as convolutional neural networks (CNN) or long short-term memory networks (LSTM)) to learn normal operating modes from a large amount of historical data and identify abnormal behaviors. CNN is mainly used to process image data. For example, the vibration signal of the fan can be converted into an image for processing. For example, the time domain waveform of the vibration signal can be converted into a two-dimensional image, and then CNN can be used for feature extraction and anomaly detection. LSTM is suitable for processing time series data. The operation data of the fan can be input into LSTM in chronological order for training. For example, the time series data of the fan speed, power output, temperature, etc. are input into LSTM to let it learn the normal operation mode. When abnormal data occurs, LSTM can identify abnormal behavior and issue an alarm.

[0078] The formula for LSTM is: , , , , ,in , , are the outputs of the input gate, forget gate, and output gate respectively. is the cell state, is the hidden state, For input data, , , , , , , , is the weight matrix, , , , is the bias term, is the sigmoid function, is element-by-element multiplication. In LSTM, the input gate, forget gate, and output gate control the input, forgetting, and output of information respectively. The input gate determines how much of the current input data can be stored in the cell state, the forget gate determines how much of the cell state at the previous moment can be retained to the current moment, and the output gate determines how much of the cell state at the current moment can be output to the hidden state. The cell state and hidden state store long-term and short-term information respectively, which are used to predict future outputs. The weight matrix and bias term are the parameters of the model. These parameters are adjusted through training so that the model can better fit historical data.

[0079] Step 7: Automatically select appropriate suppression measures according to different levels of speeding; the specific method is: Develop a graded response algorithm to automatically select appropriate suppression measures based on different levels of speeding (minor, moderate, severe). Speeding degree indicators can be defined ,in is the current speed, is the rated speed, For example, when the current speed is 1800 rpm, the rated speed is 1500 rpm, and the speed limit is 2000 rpm, the overspeed indicator is Based on this indicator, the severity of speeding can be judged and corresponding suppressive measures can be taken.

[0080] Minor speeding ( ): Solved by fine-tuning the pitch angle. The pitch angle adjustment formula can be ,in is the current pitch angle, The adjustment amount can be determined by empirical formula or optimization algorithm. For example, the adjustment amount can be determined by empirical formula according to factors such as the fan model and wind conditions. It is also possible to find the optimal adjustment amount by optimization algorithm, such as genetic algorithm, particle swarm optimization algorithm, etc., under certain constraints, so that the fan speed can be restored to the normal range as soon as possible.

[0081] Moderate speeding ( ): Reduce power output. The power output adjustment formula can be ,in is the current power output, The amount of reduction can be determined based on the degree of overspeed and the characteristics of the fan. For example, the proportion of the reduction can be determined based on the degree of overspeed. For moderate overspeed conditions, the power output can be reduced by 20% - 30%.

[0082] Severe speeding ( ): Start the emergency brake. Emergency braking can be achieved by triggering a mechanical brake device or an electrical brake system. For example: when it is detected that the fan is in a serious overspeed state, it is a vital safety measure to immediately start the emergency brake system. Mechanical brake devices usually use the friction between brake pads and brake discs to achieve rapid shutdown. For example, mechanical brakes are installed at key parts such as the main shaft or gearbox of the fan. When the emergency brake is triggered, the brake pads are pressed tightly against the brake disc, and the speed of the fan is quickly reduced by friction. This mechanical brake device has the characteristics of high reliability and obvious braking effect, but it may wear out when used frequently and requires regular maintenance and inspection.

[0083] The electrical braking system can achieve braking by changing the magnetic field or current direction of the motor. For example, the reverse braking method is used to reverse the three-phase power supply of the motor so that the motor generates reverse torque, thereby quickly reducing the speed. Alternatively, the energy consumption braking method is used to convert the kinetic energy of the motor into electrical energy and consume it in the resistor to achieve braking. The electrical braking system has the advantages of fast response speed and precise control of the braking process, but it requires the support of a complex electrical control system, and may generate a lot of heat during the braking process, so the heat dissipation problem needs to be considered.

[0084] Step 8: Repeatedly adjust the performance boundary to form the final overspeed suppression strategy. The specific method is: Step 8.1. Estimation of mechanical stress and fatigue life based on actual measured sensor parameters 1. Data collection and processing: Continuously collect data from sensors installed at key locations of the wind turbine, including but not limited to speed sensors, vibration sensors, temperature sensors, strain sensors, etc. Ensure data accuracy and real-time performance, and use data filtering technology to remove noise interference. For example, a low-pass filter can be used to remove high-frequency noise, or a median filter can be used to remove outliers. At the same time, sensors should be calibrated and maintained regularly to ensure their measurement accuracy.

[0085] Preprocess the collected data, such as segmenting and smoothing the time series data for subsequent analysis. The data can be segmented according to the time window, and then the data in each time period is averaged or weighted averaged to reduce data fluctuations. At the same time, smoothing algorithms such as moving average or exponential smoothing are used to smooth the data to make the data more stable.

[0086] 2. Mechanical stress calculation: Based on strain sensor data and material mechanics principles, the stress distribution of key components (such as blades, towers, gearboxes, etc.) under actual operating conditions is calculated. Hooke's law can be used, where represents stress, represents elastic modulus, and represents strain. For different key components, the elastic modulus may be different and needs to be determined based on the characteristics of the material. For example, blades are usually made of composite materials, and their elastic modulus is very different from that of metal materials. The strain values ​​of key components at different positions are measured by strain sensors, and then the corresponding stress values ​​are calculated in combination with the elastic modulus.

[0087] Considering the combined effects of multiple loads such as wind load, gravity, inertial force, etc., the stress state of complex structures is accurately calculated by finite element analysis. In finite element analysis, key components are discretized into a finite number of units, and the stress and strain of each unit are calculated by solving the mechanical equilibrium equation. For example, for a blade, it can be divided into a large number of small triangular units, and then the forces and moments acting on each unit by loads such as wind load, gravity and inertial force are considered, and the stress distribution of the blade is obtained by calculation through finite element software.

[0088] 3. Fatigue life assessment: Use fatigue life prediction models, such as the SN curve method or Miner linear cumulative damage theory. Calculate the fatigue damage of key components based on the number of stress cycles and stress amplitude. In practical applications, it is necessary to select a suitable fatigue life prediction model based on the material properties and working environment of key components. For example, for key components of metal materials, the SN curve method can be used to predict fatigue life based on the relationship between stress amplitude and number of cycles. For key components of composite materials, a more complex fatigue life prediction model may be required.

[0089] Consider the effects of random loads and environmental factors (such as temperature and humidity) on fatigue life in actual operation, and make corrections and adjustments. For example, in high temperature and high humidity environments, the fatigue performance of materials may decline, and the fatigue life needs to be corrected accordingly. At the same time, the effect of random loads will also accelerate fatigue damage of key components, and the statistical characteristics of random loads need to be considered to adjust the fatigue life.

[0090] Step 8.2: Compare predictions with actual results 1. Establish a comparative indicator system: Determine key performance indicators, such as maximum stress value, average stress value, fatigue life, speed fluctuation range, etc., to compare the predicted results with the actual measurement results. These key performance indicators can reflect the performance and reliability of the fan under different working conditions. For example, the maximum stress value can measure the stress of key components under the worst working conditions, the average stress value can reflect the average stress of key components in long-term operation, the fatigue life can evaluate the service life of key components, and the speed fluctuation range can reflect the stability of fan operation.

[0091] Set a reasonable error range, such as controlling the stress error within a certain percentage and the fatigue life error within a certain number of hours. Determine a reasonable error range based on the design requirements and actual application of the fan. For example, the stress error of key components can be controlled within 10%; the fatigue life error can be controlled within a certain number of hours, such as within 1000 hours.

[0092] 2. Data analysis and comparison: Compare the actual measured performance indicators with the predicted results based on theoretical analysis and simulation calculations. The differences can be intuitively displayed using charts, statistical analysis, and other methods. For example, a comparison curve between the actual measured stress value and the predicted stress value can be drawn, or a bar chart can be used to show the difference between the actual fatigue life and the predicted fatigue life. At the same time, statistical analysis methods such as mean, variance, standard deviation, etc. can be used to quantitatively analyze the size and significance of the difference.

[0093] Analyze the causes of the differences, which may include model errors, sensor accuracy issues, and insufficient consideration of environmental changes. By analyzing the causes of the differences, you can improve the model and algorithm in a targeted manner to improve the accuracy of the prediction. For example, if the model error is large, you can further optimize the model parameters and structure; if it is a sensor accuracy issue, you can replace the sensor with a higher accuracy or calibrate the sensor; if it is due to insufficient consideration of environmental changes, you can add the impact of environmental factors to the model.

[0094] Step 8.3: Repeatedly adjust performance boundaries 1. Optimization algorithm selection: Simulated annealing algorithm: simulates the annealing process of solid materials and searches for the global optimal solution by gradually lowering the temperature. When adjusting the performance boundary, define the objective function as minimizing the difference between the actual and predicted performance indicators, randomly generate the initial solution (performance boundary parameter combination), and then update the solution through continuous iteration, accepting the poor solution with a certain probability to avoid falling into the local optimum. For example, the objective function can be defined as the sum of the square difference between the actual measured stress value and the predicted stress value, or the absolute difference between the actual fatigue life and the predicted fatigue life. Minimize the objective function value by continuously adjusting the performance boundary parameter combination. In the iterative process, accepting the poor solution with a certain probability can increase the algorithm's exploration ability and avoid falling into the local optimal solution.

[0095] Ant Colony Algorithm: simulates the behavior of ants in the process of searching for food, and guides the search for the optimal solution through the accumulation and update of pheromones. The performance boundary parameter combination is regarded as the path of the ants, and the pheromone intensity is defined to be related to the objective function value. The ants select the next parameter combination for exploration based on the pheromone concentration. For example, the pheromone intensity can be defined as the inverse of the objective function value, that is, the smaller the objective function value, the greater the pheromone intensity. When selecting the next parameter combination, the ants make decisions based on the pheromone concentration and heuristic information. Heuristic information can be defined according to the characteristics of the problem, for example, the changing trend of the performance boundary parameters, constraints, etc. can be considered.

[0096] 2. Iterative adjustment process: Initialize the parameters of the optimization algorithm, such as the initial temperature and cooling rate of the simulated annealing algorithm, the number of ants and the pheromone volatility coefficient of the ant colony algorithm. According to the scale and complexity of the problem, the parameters of the optimization algorithm should be reasonably set. For example, for the simulated annealing algorithm, the initial temperature can be set to a higher value, and the cooling rate can be gradually reduced according to the number of iterations. For the ant colony algorithm, the number of ants can be determined according to the dimension of the performance boundary parameter and the size of the search space, and the pheromone volatility coefficient can control the accumulation and update speed of pheromones.

[0097] Randomly generate a set of initial performance boundary parameter combinations as the starting point of the algorithm. A random number generator can be used to generate the initial performance boundary parameter combinations to ensure that the algorithm has a certain diversity in the search space.

[0098] In each iteration, a new parameter combination is generated according to the rules of the optimization algorithm, and the corresponding objective function value is calculated. If the new parameter combination makes the objective function value better, then the parameter combination is accepted; otherwise, the worse parameter combination is accepted with a certain probability to increase the algorithm's exploration ability. For example, in the simulated annealing algorithm, if the objective function value corresponding to the new parameter combination is less than the objective function value of the current solution, then the parameter combination is directly accepted; if the objective function value corresponding to the new parameter combination is greater than the objective function value of the current solution, then the parameter combination is accepted with a certain probability, and the probability can be calculated based on the difference between the temperature and the objective function value. In the ant colony algorithm, ants select the next parameter combination based on the pheromone concentration and heuristic information. If the objective function value corresponding to the new parameter combination is better, the pheromone concentration is updated.

[0099] Repeat the iteration process until the stopping condition is met, such as reaching the maximum number of iterations, the objective function value is less than a certain threshold, etc. Determine a reasonable stopping condition based on the requirements of the problem and the performance of the algorithm. For example, you can set the maximum number of iterations to 1000, or stop the iteration when the objective function value is less than a certain threshold.

[0100] 3. Verification and confirmation: After each adjustment of the performance boundary, re-evaluate the mechanical stress and fatigue life under the actual operating conditions and compare them with the predicted results. Through actual measurement and calculation, verify whether the adjusted performance boundary can accurately reflect the actual operation of the wind turbine. If the difference between the actual measurement results and the predicted results is large, it is necessary to further adjust the performance boundary parameter combination.

[0101] Check whether the adjusted performance boundary meets the safety requirements, reliability requirements, and operating efficiency requirements. If not, continue to adjust; if satisfied, further verify its stability and robustness. For example, by simulating different wind conditions and working conditions, check whether the adjusted performance boundary can ensure the safe operation and efficient power generation of the wind turbine under various conditions. At the same time, methods such as sensitivity analysis can be used to verify the impact of changes in performance boundary parameters on wind turbine performance and ensure the stability and robustness of the performance boundary.

[0102] Step 8.4: Form the final overspeed suppression strategy 1. Integrate optimization results: The optimal performance boundary parameter combination obtained through repeated adjustments is integrated into the overspeed suppression strategy. Key parameters such as the upper speed limit, power output limit, pitch angle range, and temperature limit are determined. Based on the optimal performance boundary parameter combination, the safe operating range and control strategy of the wind turbine under different working conditions are determined. For example, the upper speed limit is determined to be 1800 rpm, the power output limit is 1200 kW, the pitch angle range is [0°, 30°], and the temperature limit is 80°C.

[0103] Consider the adaptability under different wind conditions, environmental conditions and operating states, and formulate corresponding adjustment rules and control strategies. For example, when the wind speed is high, the power output can be appropriately reduced and the pitch angle can be adjusted to avoid overspeed of the wind turbine; when the ambient temperature is high, the heat dissipation measures of the wind turbine can be strengthened to ensure that the temperature of key components is within the allowable range. At the same time, control strategies under different operating states, such as startup, shutdown, fault handling, etc., can be formulated to ensure the safe operation and efficient power generation of the wind turbine.

[0104] 2. Strategy verification and improvement: The formed overspeed suppression strategy is verified and tested in an actual wind farm. By simulating different overspeed scenarios, the effectiveness and reliability of the strategy can be observed. For example, different wind speeds and load conditions can be set in the wind farm to simulate the overspeed of the wind turbine, and then observe whether the overspeed suppression strategy can effectively control the speed of the wind turbine in a timely manner to avoid safety accidents.

[0105] According to the test results, the overspeed suppression strategy can be further improved, such as adjusting the response time, optimizing the control algorithm, adding redundant protection measures, etc. If the test results show that the overspeed suppression strategy has problems such as too long response time and unsatisfactory control effect, the parameters and algorithms of the strategy can be further adjusted to improve the performance of the strategy. At the same time, it is possible to consider adding redundant protection measures, such as backup sensors and backup controllers, to improve the reliability and safety of the system.

[0106] The final overspeed suppression strategy document based on the performance boundary of wind turbines is formed, including the principle of the strategy, parameter settings, implementation steps and precautions, etc., to facilitate practical application and maintenance. The document should describe in detail the principle and basis of the overspeed suppression strategy, clarify the setting method and adjustment rules of the performance boundary parameters, and explain in detail the implementation steps and operation process of the strategy. At the same time, it lists precautions and solutions to common problems to provide guidance for practical application and maintenance.

[0107] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. An overspeed suppression method based on the performance boundary of a wind turbine generator set, characterized in that The following steps are involved: Step 1: Establish a dynamic model of the wind turbine to simulate the response of the wind turbine under different wind conditions; Step 2: Evaluate the mechanical stress and fatigue life of the fan under various working conditions; Step 3: Taking the maximization of the service life of the wind turbine as the objective function, preliminarily set the performance boundary; Step 4: Install redundant sensors for wind turbines and perform data fusion; Step 5: Use big data analysis tools to detect anomalies and predict faults of wind turbines; Step 6: Use deep learning models to learn normal operating patterns from historical data and identify abnormal behaviors; Step 7: Automatically select appropriate suppression measures according to different levels of speeding; Step 8: Repeatedly adjust the performance boundary to form the final overspeed suppression strategy.

2. The overspeed suppression method based on the performance boundary of a wind turbine generator set according to claim 1 is characterized in that: In step 1, the method of establishing a dynamic model of the wind turbine for simulating the response of the wind turbine under different wind conditions is: Use simulation software to build a dynamic model of the fan; Simulate the speed under different wind conditions , Power Output , vibration displacement parameter; By analyzing wind parameters including wind speed ,wind direction The following dynamic equations are used to simulate the fan, combined with the fan's structural parameters and aerodynamic characteristics: ， in For air power, is the lift coefficient, is the air density, is the blade swept area, is the drag coefficient; ， in is the aerodynamic torque, is the blade radius; ， in is the moment of inertia, is the speed change rate, is the generator torque, is the friction torque; Through dynamic modeling, the response characteristics of the wind turbine under various working conditions are obtained, including the curve of speed change with wind speed and the relationship between power output and wind speed.

3. The overspeed suppression method based on the performance boundary of a wind turbine generator set according to claim 1 is characterized in that: In step 2, the method for evaluating the mechanical stress and fatigue life of the fan under various working conditions is: Finite element analysis and multi-body dynamics analysis methods are used to evaluate the stress distribution and fatigue life of key components under different working conditions; the stress formula is ,in is stress, is the force, is the stress area; the fatigue life assessment adopts Miner's law: ,in For the stress level The number of cycles under For the stress level Fatigue life under is the damage accumulation parameter; Finite element analysis discretizes the key components of the wind turbine into a finite number of units, and calculates the stress and strain of each unit by solving the mechanical equilibrium equation; the key components include blades, gearboxes and generators; Multibody dynamics analysis is used to consider the connection and motion relationship between various components and more accurately simulate the overall dynamic response of the wind turbine.

4. The overspeed suppression method based on the performance boundary of a wind turbine generator set according to claim 1 is characterized in that: In step 3, taking the maximization of the service life of the wind turbine as the objective function, the method for preliminarily setting the performance boundary is: Based on theoretical analysis and simulation results, determine the performance parameter boundaries under different wind conditions; Performance parameters include upper speed limit , power output limit , Pitch angle range , Temperature Limit ; The objective function is , in The service life of the wind turbine generator set; find the performance boundary parameter combination that maximizes the objective function through the optimization algorithm; Based on theoretical analysis and simulation calculation results, the performance parameter boundaries under different wind conditions are determined; theoretical analysis can combine the design parameters of the fan and aerodynamic principles to derive the theoretical calculation formula for the performance boundary; simulation calculation simulates the operation of the fan under different wind conditions many times, and obtains the reasonable range of performance parameters through statistical analysis.

5. The overspeed suppression method based on the performance boundary of a wind turbine generator set according to claim 1 is characterized in that: In step 4, the method for installing redundant sensors for wind turbines and performing data fusion is as follows: Install multiple redundant sensors for key parameters; key parameters include speed, power output and temperature; Use Kalman filter to improve the consistency and reliability of data; the basic formula of Kalman filter is: ， in for The estimated state value at time for The Kalman gain at time t, for The measured value at the moment, For the measurement matrix; establish a multi-level alarm system and set different thresholds; Establish a multi-level alarm system. When the first level threshold is reached When the alarm is triggered, it is close to the higher level threshold , The response measures are upgraded step by step in a timely manner; the alarm system can set different thresholds according to different parameters.

6. The overspeed suppression method based on the performance boundary of a wind turbine generator set according to claim 1, characterized in that: In step 5, the method of using big data analysis tools to perform abnormality detection and fault prediction on wind turbines is as follows: Utilize big data analysis tools on the cloud computing platform; combine machine learning algorithms for anomaly detection and fault prediction; random forest is an integrated learning algorithm consisting of multiple decision trees; its prediction formula is: ， in is the predicted value, is the number of decision trees, For the The predicted value of a decision tree; Realize real-time monitoring and early warning of the fan operating status; timely detect abnormal situations and issue alarms through analysis of real-time data.

7. The overspeed suppression method based on the performance boundary of a wind turbine generator set according to claim 1, characterized in that: In step 6, the method of using the deep learning model to learn the normal operation mode from historical data and identify abnormal behavior is as follows: Install high-precision speed sensors, vibration sensors, temperature sensors, and power sensors on wind turbines; continuously collect data from various sensors, including but not limited to speed , vibration displacement ,temperature ,power ; Introducing deep learning models, including convolutional neural networks (CNN) or long short-term memory networks (LSTM), to learn normal operating modes from a large amount of historical data and identify abnormal behaviors. Among them, the long short-term memory network (LSTM) is suitable for processing time series data. The operation data of the wind turbine can be input into the long short-term memory network (LSTM) in chronological order for training; The formula for LSTM is: ， ， ， ， ， in , , are the outputs of the input gate, forget gate, and output gate respectively. is the cell state, is the hidden state, For input data, is the weight matrix, is the bias term, is the sigmoid function, is element-wise multiplication.

8. The overspeed suppression method based on the performance boundary of a wind turbine generator set according to claim 1, characterized in that: In step 7, the method of automatically selecting appropriate suppression measures according to different levels of speeding conditions is: Develop a graded response algorithm to automatically select appropriate suppression measures based on different levels of speeding; define speeding degree indicators ,in is the current speed, is the rated speed, The upper speed limit; Slight speeding :Solved by fine-tuning the pitch angle; the pitch angle adjustment formula can be ,in is the current pitch angle, is the adjustment amount, which can be determined according to an empirical formula or an optimization algorithm; Moderate speeding : Reduce power output; the power output adjustment formula can be ,in is the current power output, To reduce the amount, it can be determined according to the degree of overspeed and fan characteristics; Severe speeding :Initiate emergency braking; emergency braking can be achieved by triggering a mechanical brake device or an electrical brake system.

9. The overspeed suppression method based on the performance boundary of a wind turbine generator set according to claim 1, characterized in that: In step 8, the method of repeatedly adjusting the performance boundary to form the final overspeed suppression strategy is: Step 8.1, evaluating mechanical stress and fatigue life based on the actual measured sensor parameters; Step 8.2: Compare the predictions with the actual results; Step 8.3, repeatedly adjust the performance boundary; Step 8.4: Form a final overspeed suppression strategy.

10. The overspeed suppression method based on the performance boundary of a wind turbine generator set according to claim 1, characterized in that: Step 8.

1. Estimation of mechanical stress and fatigue life based on actual measured sensor parameters: (1) Data collection and processing: Continuously collect data from sensors installed at key locations of the wind turbine, including speed sensors, vibration sensors, temperature sensors, and strain sensors; ensure data accuracy and real-time performance, and use data filtering technology to remove noise interference; Preprocess the collected data; (2) Mechanical stress calculation: Based on strain sensor data and material mechanics principles, calculate the stress distribution of key components under actual operating conditions; be able to use Hooke's law ,in represents stress, represents the elastic modulus, Indicates strain; (3) Fatigue life assessment: Using the fatigue life prediction model, the fatigue damage of key components is calculated based on the number of stress cycles and stress amplitude; Step 8.2: Compare the predictions with the actual results; (1) Establish a comparison index system: determine the key performance indicators, including maximum stress value, average stress value, fatigue life, and speed fluctuation range, to compare the predicted results with the actual measurement results; set a reasonable error range; (2) Data analysis and comparison: Compare the actual measured performance indicators with the predicted results based on theoretical analysis and simulation calculations; Step 8.3, repeatedly adjust the performance boundary; (1) Selection of optimization algorithm: Simulated annealing algorithm: simulates the annealing process of solid materials and searches for the global optimal solution by gradually lowering the temperature. When adjusting the performance boundary, the objective function is defined as minimizing the difference between the actual and predicted performance indicators. The initial solution is randomly generated, and then the solution is updated through continuous iteration. The poor solution is accepted with a certain probability to avoid falling into the local optimum. Ant Colony Algorithm: simulates the behavior of ants in the process of searching for food, and guides the search for the optimal solution through the accumulation and update of pheromones; regards the performance boundary parameter combination as the path of the ants, defines the relationship between the pheromone intensity and the objective function value, and the ants select the next parameter combination for exploration based on the pheromone concentration; (2) Iterative adjustment process: Initialize the optimization algorithm parameters, including the initial temperature and cooling rate of the simulated annealing algorithm, the number of ants and the pheromone volatility coefficient of the ant colony algorithm; Randomly generate a set of initial performance boundary parameter combinations as the starting point of the algorithm; In each iteration, a new parameter combination is generated according to the rules of the optimization algorithm, and the corresponding objective function value is calculated; if the new parameter combination makes the objective function value better, then the parameter combination is accepted; otherwise, the worse parameter combination is accepted with a certain probability to increase the exploration ability of the algorithm; Repeat the iterative process until the stopping condition is met; (3) Verification and confirmation: After each adjustment of the performance boundary, re-evaluate the mechanical stress and fatigue life under the actual operating conditions and compare them with the predicted results; Check whether the adjusted performance boundary meets the safety requirements, reliability requirements and operating efficiency requirements; if not, continue to adjust; if so, further verify its stability and robustness; Step 8.4, forming a final overspeed suppression strategy; (1) Integration and optimization results: Integrate the optimal performance boundary parameter combination obtained through repeated adjustments into the overspeed suppression strategy; determine key parameters including upper speed limit, power output limit, pitch angle range, and temperature limit; Consider the adaptability under different wind conditions, environmental conditions and operating states, and formulate corresponding adjustment rules and control strategies; (2) Strategy verification and improvement: The formed overspeed suppression strategy is verified and tested in an actual wind farm; the effectiveness and reliability of the strategy are observed by simulating different overspeed scenarios; based on the test results, the overspeed suppression strategy is further improved to form a final overspeed suppression strategy document based on the performance boundary of the wind turbine, including the principle of the strategy, parameter settings, implementation steps and precautions, to facilitate practical application and maintenance.

Images