Wind power plant unit operation optimization method and system based on artificial intelligence

Abstract

The invention relates to a wind power plant unit operation optimization method and system based on artificial intelligence, and the method comprises the following steps: S1, data collection and processing: collecting the output power data, vibration data, temperature data, rotating speed data, yaw angle data and blade root strain data of each unit, meanwhile, acquiring upwind speed, wind direction, wind shear and turbulence intensity by using a laser radar at the top of a cabin, and performing time synchronization and preliminary noise filtering on the data; according to the invention, unit output power data, vibration data, temperature data, rotating speed data, yaw angle data and blade root strain data are continuously acquired through a data acquisition module, upwind speed, wind direction, wind shear and turbulence intensity are acquired by a laser radar at the top of a cabin, and time synchronization and preliminary noise filtering processing are combined to obtain a wind speed and a wind direction; discrete noisy points generated by wind speed pulsation, equipment jitter and environment interference can be effectively eliminated, so that the acquired data is more stable and reliable.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, and in particular to a method and system for optimizing the operation of wind farm units based on artificial intelligence. Background Technology

[0003] Wind power, as a clean, low-carbon, safe, efficient, and large-scale energy source, is an important choice for promoting green development, establishing and improving a green, low-carbon, and circular economic system, and strengthening the clean energy industry. However, its level of intelligence is relatively low compared to hydropower and thermal power. With the continuous development of digital, information, and intelligent technologies, it is imperative to improve the intelligence level of the wind power industry by utilizing advanced means such as information and communication technology, big data, cloud computing, the Internet of Things, mobile applications, and sensing technology.

[0004] In practice, some problems still exist: With the rapid development of wind power generation technology, the operating efficiency and reliability of wind farm units have become increasingly important for improving wind energy utilization. However, existing wind farm units generally have the following problems in actual operation: On the one hand, due to the instantaneous changes in environmental factors such as wind speed, wind direction and turbulence intensity, the unit output power fluctuates greatly. Traditional experience-based control methods are difficult to adjust the unit attitude and pitch angle in real time and accurately, resulting in effective wind energy loss. On the other hand, the turbine blades and towers bear complex loads during long-term operation. Existing monitoring methods mostly rely on a single sensor or periodic inspections, which cannot fully reflect the fatigue state of the blades and the changes in the bending moment of the tower. This can easily lead to the accumulation of structural fatigue and reduce the life of the turbine. In addition, there is wake interference between the turbines. Downwind turbines are affected by the wake of upwind turbines, resulting in a significant decrease in power output. However, existing power allocation strategies lack the ability to dynamically optimize wake characteristics and the overall layout of the wind farm, resulting in low power generation efficiency of the wind farm cluster. Summary of the Invention

[0006] (a) Technical problems to be solved To address the problems of existing technologies, this invention provides a wind farm turbine operation optimization method and system based on artificial intelligence, thereby resolving the issues raised in the background section.

[0007] (II) Technical Solution To achieve the objective, the main technical solution adopted by this invention is as follows: A method and system for optimizing the operation of wind farm turbines based on artificial intelligence, comprising the following steps: S1: Data acquisition and processing. Collects output power data, vibration data, temperature data, speed data, yaw angle data, and blade root strain data for each unit. At the same time, it uses the lidar on the top of the nacelle to obtain upwind wind speed, wind direction, wind shear, and turbulence intensity, and performs time synchronization and preliminary noise filtering on the data. S2: Digital twin modeling. Based on the whole machine dynamics model, a digital twin model of the wind turbine is constructed. At the same time, a finite element model of the blade is constructed based on the blade main beam structure model, web structure model and shell structure model. A tower structure model is constructed based on the tower equivalent stiffness parameters. The model is mapped in real time to the unit load, unit deformation and unit power response. S3: Artificial intelligence state recognition, which uses convolutional neural network to extract spatial features from real-time data of the unit and long short-term memory network to extract time series features, and generates a state vector composed of yaw error quantification index, wake intensity index, blade health decay index, tower bending moment change index and power decay trend index. S4: Intelligent yaw correction, based on the state vector, calculates the difference between the real-time power curve and the theoretical maximum power curve, establishes the power-to-yaw angle curve through a scatter fitting algorithm, and determines the yaw compensation angle based on the first derivative of the yaw angle curve, and performs automatic yaw correction on the unit. S5: Wake simulation and power distribution. Calculate the wake distribution of the wind farm using a digital twin model. Based on the relative position between the upwind and downwind turbines and the wind direction shift trend, generate the pitch angle adjustment and power limit value of the upwind turbines. S6: Control parameter optimization based on genetic algorithm. A multi-objective function is established with the optimization objectives of power generation improvement rate, structural load reduction rate, reduction in yaw action number and health improvement. Pitch angle, control gain, power limit value and yaw compensation amount are used as optimization variables. The global optimal solution of the optimization variables is sought through adaptive genetic algorithm and sent to the unit actuator.

[0008] Preferably, the wind speed and wind direction data output by the lidar in S1 are processed by Kalman filtering before entering the digital twin model. Kalman filtering preserves the stable trend and removes discrete noise caused by sudden changes in wind speed, so that the yaw feedforward control quantity remains stable.

[0009] Preferably, in the digital twin model, the whole-machine dynamics model is constructed using FAST, the blade finite element model is constructed using shell elements, and the tower structure model is constructed using equivalent beam elements. The model is used to predict blade stress, main shaft bending moment, tower bending moment, and cumulative fatigue damage in real time.

[0010] Preferably, the leaf health degradation index output by the artificial intelligence model in S3 includes the leaf root strain change rate and the leaf bending moment root mean square error, which are used to identify the decline in leaf fatigue performance.

[0011] Preferably, the yaw compensation angle in S4 is determined by the following steps: Extract the power difference between real-time power and theoretical maximum power; The power-yaw angle function is generated based on the least squares principle; Calculate the first derivative of the yaw angle function with respect to the yaw angle, and take the position where the derivative is zero as the yaw compensation angle.

[0012] Preferably, the wake simulation in S5 adopts an empirical wake model, which uses the wake diffusion coefficient and attenuation coefficient to describe the wake diffusion characteristics, so that the wake simulation remains real-time.

[0013] Preferably, before generating the power allocation strategy in S5, the units are divided into multiple groups based on their relative orientation and the dominant wind direction. The groups are divided using the K-means algorithm so that units in the same wake chain belong to the same group.

[0014] Preferably, the genetic algorithm in S6 uses a fixed crossover probability and a fixed mutation probability, where the crossover probability is 0.7 and the mutation probability is 0.2, and obtains the optimal solution that satisfies the convergence condition through 50 generations of iteration.

[0015] Preferably, after the optimized control is executed, a blade tip double winglet is set for the blade. The blade tip double winglet includes a base section and an extension section. The base section is fixed to the blade trailing edge structure reinforcement position by bolts. The extension section is arranged along the blade span and is set with a fixed dihedral angle, which increases the blade lift coefficient and reduces the blade tip vortex intensity.

[0016] Preferably, an artificial intelligence-based wind farm turbine operation optimization system includes: a data acquisition module for acquiring turbine output power data, vibration data, temperature data, yaw angle data, blade strain data, and lidar wind speed and direction data; The digital twin model module constructs and updates the overall dynamics model, blade finite element model, and tower structure model in real time. The artificial intelligence analysis module extracts time-series features from multi-source data and outputs state vectors; The yaw correction module calculates the yaw compensation angle based on the state vector and executes yaw control. The wake calculation module generates wake distribution and power allocation strategies based on real-time wind conditions. The cluster scheduling module groups the generating units and executes power optimization strategies within each group. The execution module is used to perform pitch angle adjustment, power limit adjustment, and yaw execution operations.

[0017] (III) Beneficial Effects The beneficial effects of this invention are: 1. In this invention, the data acquisition module continuously collects unit output power data, vibration data, temperature data, speed data, yaw angle data, and blade root strain data. The upwind wind speed, wind direction, wind shear, and turbulence intensity are obtained by the lidar on the top of the nacelle. Combined with time synchronization and preliminary noise filtering, discrete noise caused by wind speed fluctuations, equipment vibration, and environmental interference can be effectively eliminated, making the collected data more stable and reliable. This mechanism provides high-quality input data for digital twin modeling and artificial intelligence analysis, ensuring the accuracy of subsequent unit status identification, yaw correction, and wake simulation, thereby reducing control errors, improving the reliability and safety of unit operation, and providing a solid data foundation for overall optimization.

[0018] 2. In this invention, the digital twin model module constructs a digital twin model of the wind turbine through the whole-machine dynamics model, and constructs a finite element model of the blade by combining the blade main beam, web and shell structure models. At the same time, a tower structure model is established based on the equivalent stiffness parameters of the tower, realizing real-time mapping of unit load, deformation and power response. This module can accurately reflect the structural and dynamic state of the unit under different wind conditions, providing a reliable basis for artificial intelligence state recognition and yaw control, making the control strategy more targeted and forward-looking, thereby reducing the fatigue accumulation of blades and tower, improving the structural safety and power generation efficiency of the unit, and providing a guarantee for the long-term stable operation of the wind farm.

[0019] 3. In this invention, the artificial intelligence analysis module extracts spatial features through a convolutional neural network and extracts time series features through a long short-term memory network. It outputs a state vector composed of yaw error quantification index, wake intensity index, blade health decay index, tower bending moment change index, and power decay trend index. The yaw correction module calculates the difference between the real-time power and the theoretical maximum power based on the state vector and generates a yaw compensation angle, driving the unit to automatically perform yaw actions. This control mechanism can effectively reduce wind energy loss caused by yaw error, reduce structural fatigue caused by frequent yaw system actions, and ensure that the unit continues to face upwind, improving wind energy capture rate and power generation stability, and realizing intelligent and precise unit operation optimization.

[0020] 4. The wake calculation module uses a digital twin model to simulate the wake distribution of the wind farm and groups and allocates power according to the relative orientation of the units and the dominant wind direction. At the same time, the genetic algorithm module optimizes the pitch angle, control gain, power limit value and yaw compensation. The execution module implements the optimization strategy to the unit operation, including pitch angle adjustment, power limit adjustment and yaw control. When the unit is equipped with blade tip winglets, the lift coefficient is further improved and the blade tip vortex intensity is reduced by combining its aerodynamic characteristics, so as to achieve better airflow conditions for downwind units. This comprehensive mechanism can significantly improve the power generation of the wind farm, reduce structural load and fatigue accumulation, improve the overall life of the units and economic benefits, and realize the intelligent and optimized operation of the wind farm. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the method of the present invention; Figure 2 This is a flowchart illustrating the optimized system portion of the present invention. Detailed Implementation

[0022] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] Please refer to Figures 1 to 2 As shown, the present invention provides a wind farm turbine operation optimization method and system based on artificial intelligence, comprising the following steps: S1: Data acquisition and processing. Collects output power data, vibration data, temperature data, speed data, yaw angle data, and blade root strain data for each unit. At the same time, it uses the lidar on the top of the nacelle to obtain upwind wind speed, wind direction, wind shear, and turbulence intensity, and performs time synchronization and preliminary noise filtering on the data. S2: Digital twin modeling. Based on the whole machine dynamics model, a digital twin model of the wind turbine is constructed. At the same time, a finite element model of the blade is constructed based on the blade main beam structure model, web structure model and shell structure model. A tower structure model is constructed based on the tower equivalent stiffness parameters. The model is mapped in real time to the unit load, unit deformation and unit power response. S3: Artificial intelligence state recognition, which uses convolutional neural network to extract spatial features from real-time data of the unit and long short-term memory network to extract time series features, and generates a state vector composed of yaw error quantification index, wake intensity index, blade health decay index, tower bending moment change index and power decay trend index. S4: Intelligent yaw correction, which calculates the difference between the real-time power curve and the theoretical maximum power curve based on the state vector, establishes the power-to-yaw angle curve through a scatter fitting algorithm, and determines the yaw compensation angle based on the first derivative of the yaw angle curve, and performs automatic yaw correction on the unit. S5: Wake simulation and power distribution. Calculate the wake distribution of the wind farm using a digital twin model. Based on the relative position between the upwind and downwind turbines and the wind direction shift trend, generate the pitch angle adjustment and power limit value of the upwind turbines. S6: Control parameter optimization based on genetic algorithm. A multi-objective function is established with the optimization objectives of power generation improvement rate, structural load reduction rate, reduction in yaw action number and health improvement. Pitch angle, control gain, power limit value and yaw compensation amount are used as optimization variables. The global optimal solution of the optimization variables is sought through adaptive genetic algorithm and sent to the unit actuator. In actual implementation, the data acquisition module continuously acquires output power data, vibration data, temperature data, speed data, yaw angle data, and blade root strain data for each unit. Simultaneously, it utilizes the lidar on the top of the nacelle to acquire upwind wind speed, wind direction, wind shear, and turbulence intensity. Time synchronization processing is performed within the data acquisition module to initially filter out discrete noise caused by wind speed fluctuations, equipment vibrations, and environmental interference. This enables the subsequent digital twin model module and artificial intelligence analysis module to obtain high-quality, continuous, and stable unit operation information, ensuring the reliability and accuracy of state vector generation. Furthermore, it provides a precise real-time data foundation for the yaw correction module, wake calculation module, and execution module, enabling the entire wind farm unit operation optimization system to achieve efficient and accurate unit operation optimization.

[0024] Optionally, the wind speed and direction data output by the lidar in S1 are processed by Kalman filtering before entering the digital twin model. Kalman filtering preserves the stable trend and removes discrete noise caused by sudden changes in wind speed, thus keeping the yaw feedforward control quantity stable. In actual implementation, the wind speed and direction data output by the lidar are processed by the Kalman filtering module before entering the digital twin model module. Kalman filtering can preserve the stable trend of wind speed and direction, while effectively filtering out discrete noise caused by sudden changes in wind speed, ensuring the stability of the yaw feedforward control quantity, further enhancing the smoothness and accuracy of the yaw correction module when executing yaw actions, enabling the unit to maintain an effective power generation attitude under rapidly changing wind conditions, reducing structural load fluctuations caused by frequent yaw actions, and improving the overall reliability of the unit's operation, providing stable wind field input data for the wake calculation module and the wind farm scheduling module.

[0025] Optionally, the digital twin model uses FAST to construct the overall dynamics model, shell elements to construct the blade finite element model, and equivalent beam elements to construct the tower structure model. The model is used to predict blade stress, main shaft bending moment, tower bending moment, and cumulative fatigue damage in real time. In actual implementation, the digital twin model module achieves real-time mapping of the unit's state through the overall dynamics model, blade finite element model, and tower structure model. The overall dynamics model, constructed using FAST, can simulate main shaft bending moment, blade stress, and tower response. The blade finite element model, constructed using shell elements, can accurately predict blade stress distribution and fatigue accumulation. The tower structure model, constructed using equivalent beam elements, can acquire tower bending moment and structural deformation information in real time. The entire digital twin model module updates the unit load, unit deformation, and power response in real time, providing an accurate and reliable numerical basis for the artificial intelligence analysis module to generate state vectors and the control parameter optimization module, ensuring high-precision decision-making by the wind farm unit operation optimization system under complex wind conditions.

[0026] Optionally, the blade health degradation index output by the AI ​​model in S3 includes the blade root strain change rate and the blade bending moment root mean square error, used to identify the decline in blade fatigue performance. In actual implementation, the AI ​​analysis module extracts spatial features from the real-time data of the unit through a convolutional neural network and extracts time series features through a long short-term memory network, generating a state vector composed of yaw error quantification index, wake intensity index, blade health degradation index, tower bending moment change index, and power degradation trend index. Among them, the blade health degradation index includes the blade root strain change rate and the blade bending moment root mean square error, which can identify the decline in blade fatigue performance. This provides blade health status information to the yaw correction module and the control parameter optimization module, ensuring that yaw actions and pitch angle adjustments meet power generation efficiency requirements while reducing blade structural fatigue. It also assists the wake calculation module and the wind farm scheduling module in formulating reasonable power allocation strategies, improving the overall operating life of the unit and the efficiency of the wind farm group.

[0027] Optionally, the yaw compensation angle in S4 is determined by the following steps: Extract the power difference between real-time power and theoretical maximum power; The power-yaw angle function is generated based on the least squares principle; The first derivative of the yaw angle function with respect to the yaw angle is calculated, and the position where the derivative is zero is taken as the yaw compensation angle. In actual implementation, the yaw correction module generates the power difference between the real-time power curve and the theoretical maximum power curve based on the state vector. It generates a power-yaw angle curve through a scatter-point fitting algorithm and calculates the first derivative of this curve with respect to the yaw angle. The position where the derivative is zero is determined as the yaw compensation angle, driving the unit to perform automatic yaw correction, so that the unit is continuously aligned with the upwind direction, improving the effective wind energy capture rate, reducing power generation losses caused by yaw errors, and reducing fatigue of the tower and blade structure caused by frequent yaw actions. It also provides accurate yaw action references for the wake calculation module and the execution module, realizing dynamic and fine control of the wind farm unit operation optimization system.

[0028] Optionally, the wake simulation in S5 adopts an empirical wake model. This model uses the wake diffusion coefficient and attenuation coefficient to describe the wake diffusion characteristics, ensuring real-time wake simulation. In actual implementation, the wake calculation module calculates the wind farm wake distribution using a digital twin model, employing an empirical wake model. This model utilizes the wake diffusion coefficient and attenuation coefficient to describe the wake diffusion characteristics, enabling real-time updates of the wake simulation. Based on the relative positions of the upwind and downwind turbines and the wind direction shift trend, it generates the pitch angle adjustment and power limit values ​​for the upwind turbines. This reduces the impact of the wake on the downwind turbines, improves wind energy capture efficiency, provides power allocation references for the wind farm group scheduling module, and enables the entire wind farm group operation optimization system to achieve inter-unit collaborative optimization, improving total power generation and turbine structural safety.

[0029] Optionally, before generating the power allocation strategy, the S5 module divides the turbines into multiple groups based on their relative orientation and the dominant wind direction. The K-means algorithm is used for grouping, ensuring that turbines within the same wake chain belong to the same group. In actual implementation, before generating the power allocation strategy, the wake calculation module divides the turbines into multiple groups based on their relative orientation and the dominant wind direction. The K-means algorithm is used for grouping, ensuring that turbines within the same wake chain belong to the same group, minimizing wake interference. Simultaneously, an independent power allocation strategy is generated within each group, providing a grouping basis for the execution module to adjust pitch angle and power limits, improving the overall power output balance and stability of the wind farm, and providing grouping information to the control parameter optimization module, thus achieving coordinated operation of optimized control and wake allocation.

[0030] Optionally, the genetic algorithm in S6 uses a fixed crossover probability and a fixed mutation probability, with a crossover probability of 0.7 and a mutation probability of 0.2, and obtains the optimal solution that meets the convergence condition through 50 generations of iteration. In actual implementation, the control parameter optimization module uses a genetic algorithm to perform multi-objective optimization of pitch angle, control gain, power limit value, and yaw compensation amount. The optimization objectives include the rate of increase in power generation, the rate of decrease in structural load, the reduction in the number of yaw actions, and the improvement in health. The genetic algorithm uses a fixed crossover probability of 0.7 and a fixed mutation probability of 0.2, and obtains the optimal solution that meets the convergence condition through 50 generations of iteration. The optimized control parameters are sent to the execution module to achieve precise adjustment of the unit's pitch angle, power limit, and yaw compensation angle, assisting in the aerodynamic optimization of the blade tip winglets, and improving the unit's power generation efficiency and structural safety.

[0031] Optionally, after the optimized control is executed, a double tip winglet is installed on the blade. The double tip winglet includes a base section and an extension section. The base section is fixed to the reinforced position of the blade trailing edge structure with bolts. The extension section is arranged along the blade spanwise and has a fixed dihedral angle, which increases the blade lift coefficient and reduces the tip vortex intensity. In actual implementation, after the optimized control is executed, a double tip winglet is installed on the blade. The double tip winglet includes a base section and an extension section. The base section is fixed to the reinforced position of the blade trailing edge structure with bolts. The extension section is arranged along the blade spanwise and has a fixed dihedral angle, which increases the blade lift coefficient, reduces the tip vortex intensity, improves the blade aerodynamic characteristics, reduces local vortex energy loss, improves the overall power generation efficiency of the wind farm unit, and reduces the load on the blade and tower structure. This provides an optimized basis for the operation of the yaw correction module and the execution module, and enables the wind farm group operation optimization system to output stably under different wind conditions.

[0032] Optionally, an artificial intelligence-based wind farm turbine operation optimization system includes: a data acquisition module for acquiring turbine output power data, vibration data, temperature data, yaw angle data, blade strain data, and lidar wind speed and direction data; The digital twin model module constructs and updates the overall dynamics model, blade finite element model, and tower structure model in real time. The artificial intelligence analysis module extracts time-series features from multi-source data and outputs state vectors; The yaw correction module calculates the yaw compensation angle based on the state vector and executes yaw control. The wake calculation module generates wake distribution and power allocation strategies based on real-time wind conditions. The cluster scheduling module groups the generating units and executes power optimization strategies within each group. The execution module is used to implement pitch angle adjustment, power limit adjustment, and yaw execution operations. In actual implementation, the AI-based wind farm turbine operation optimization system includes a data acquisition module, a digital twin model module, an AI analysis module, a yaw correction module, a wake calculation module, a wind farm group scheduling module, and an execution module. The data acquisition module continuously collects turbine output power data, vibration data, temperature data, yaw angle data, blade strain data, and lidar wind speed and direction data. The digital twin model module constructs and updates the whole-machine dynamics model, blade finite element model, and tower structure model in real time. The AI ​​analysis module generates state vectors. The yaw correction module executes yaw control. The wake calculation module generates wake distribution and power strategy. The wind farm group scheduling module optimizes power in groups. The execution module implements pitch angle, power limit, and yaw actions, achieving efficient, coordinated, and intelligent control of the wind farm turbine operation optimization system.

[0033] Working principle: First, the data acquisition module continuously acquires the unit's output power data, vibration data, temperature data, speed data, yaw angle data, and blade root strain data. At the same time, the lidar deployed on the top of the nacelle acquires the upwind wind speed, wind direction, wind shear, and turbulence intensity. All the acquired data is synchronized in time within the module to match various signals on a unified time axis. A preliminary noise filtering algorithm is used to remove discrete noise caused by wind speed fluctuations, equipment vibration, and environmental interference. The preprocessed data serves as the input basis for subsequent analysis and optimization control. Subsequently, the digital twin model module constructs a digital twin model of the wind turbine based on the whole-machine dynamics model, and constructs a blade finite element model based on the blade main beam structure model, web structure model and shell structure model. It also constructs a tower structure model based on the tower equivalent stiffness parameters. The above models are updated synchronously with the unit load, unit deformation and unit power response under real-time data drive, so that the digital twin model can accurately reflect the dynamic state and structural state of the actual unit under different wind conditions, providing a reliable basis for control strategy. While the digital twin model is continuously updated, the artificial intelligence analysis module extracts spatial features from the real-time data of the unit through convolutional neural networks and extracts time series features through long short-term memory networks. Finally, it generates a state vector composed of yaw error quantification index, wake intensity index, blade health decay index, tower bending moment change index, and power decay trend index. This state vector comprehensively reflects the unit's operating deviation, structural fatigue degree, and power generation performance decline trend, and is the core input for subsequent intelligent control. Once the state vector is generated, the yaw correction module establishes a power-to-yaw angle curve based on the difference between the real-time power curve and the theoretical maximum power curve using a scatter-point fitting algorithm. It then calculates the first derivative of this curve and uses the position where the first derivative is zero as the yaw compensation angle, thereby driving the unit to perform yaw actions. This mechanism keeps the unit aligned with the upwind direction, reduces the effective wind energy loss caused by yaw errors, and reduces structural fatigue problems caused by frequent yaw system actions. In addition to yaw correction, the wake calculation module uses a digital twin model to calculate the wake distribution of the wind farm. Based on the relative position between the upwind and downwind turbines and the wind direction shift trend, this module dynamically generates the pitch angle adjustment and power limit value of the upwind turbines, so that the downwind turbines are in the region with weaker wake turbulence. By adjusting the operating mode of the upwind turbines, the quality of the airflow entering the downwind turbines is improved, thereby enhancing the overall energy capture capability and increasing the power generation of the wind farm. After obtaining wake information, the cluster scheduling module divides the units into multiple groups based on their relative positions and the prevailing wind direction. The groups are divided using the K-means algorithm to ensure that units in the same wake chain belong to the same group. This module generates an independent power allocation strategy for each group to ensure minimal mutual interference between units while keeping the overall output close to the optimal level. Ultimately, the execution module implements pitch angle adjustment, power limit adjustment, and yaw execution actions based on the optimization strategy, realizing the application of control parameters on the physical unit. When the unit is equipped with blade tip biplanes, the execution module further combines the aerodynamic characteristics of the blade tip biplanes to achieve higher wind energy utilization, enabling the unit to maintain stable output under different wind conditions.

[0034] The above description shows and illustrates the basic principles, main features, and advantages of the present invention. Standard parts used in the present invention can be purchased from the market, and irregular parts can be customized according to the description and drawings. The specific connection methods of each part adopt conventional methods such as bolts, rivets, and welding that are mature in the prior art. The machinery, parts, and equipment adopt conventional models in the prior art, and the circuit connection adopts conventional connection methods in the prior art, which will not be described in detail here.

[0035] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A wind farm turbine operation optimization method based on artificial intelligence, characterized in that, Includes the following steps: S1: Data acquisition and processing. Collects output power data, vibration data, temperature data, speed data, yaw angle data, and blade root strain data for each unit. At the same time, it uses the lidar on the top of the nacelle to obtain upwind wind speed, wind direction, wind shear, and turbulence intensity, and performs time synchronization and preliminary noise filtering on the data. S2: Digital twin modeling. Based on the whole machine dynamics model, a digital twin model of the wind turbine is constructed. At the same time, a finite element model of the blade is constructed based on the blade main beam structure model, web structure model and shell structure model. A tower structure model is constructed based on the tower equivalent stiffness parameters. The model is mapped in real time to the unit load, unit deformation and unit power response. S3: Artificial intelligence state recognition, which uses convolutional neural network to extract spatial features from real-time data of the unit and long short-term memory network to extract time series features, and generates a state vector composed of yaw error quantification index, wake intensity index, blade health decay index, tower bending moment change index and power decay trend index. S4: Intelligent yaw correction, based on the state vector, calculates the difference between the real-time power curve and the theoretical maximum power curve, establishes the power-to-yaw angle curve through a scatter fitting algorithm, and determines the yaw compensation angle based on the first derivative of the yaw angle curve, and performs automatic yaw correction on the unit. S5: Wake simulation and power distribution. Calculate the wake distribution of the wind farm using a digital twin model. Based on the relative position between the upwind and downwind turbines and the wind direction shift trend, generate the pitch angle adjustment and power limit value of the upwind turbines. S6: Control parameter optimization based on genetic algorithm. A multi-objective function is established with the optimization objectives of power generation improvement rate, structural load reduction rate, reduction in yaw action number and health improvement. Pitch angle, control gain, power limit value and yaw compensation amount are used as optimization variables. The global optimal solution of the optimization variables is sought through adaptive genetic algorithm and sent to the unit actuator.

2. The wind farm turbine operation optimization method based on artificial intelligence according to claim 1, characterized in that: Before the wind speed and wind direction data output by the lidar in S1 enter the digital twin model, they are processed by Kalman filtering. Kalman filtering retains the stable trend and removes the discrete noise caused by sudden changes in wind speed, so that the yaw feedforward control quantity remains stable.

3. The method for optimizing wind farm turbine operation based on artificial intelligence according to claim 1, characterized in that: In the digital twin model, the overall dynamic model is constructed using FAST, the blade finite element model is constructed using shell elements, and the tower structure model is constructed using equivalent beam elements. The model is used to predict blade stress, main shaft bending moment, tower bending moment, and cumulative fatigue damage in real time.

4. The wind farm turbine operation optimization method based on artificial intelligence according to claim 1, characterized in that: The blade health degradation index output by the artificial intelligence model in S3 includes the blade root strain change rate and the root mean square error of the blade bending moment, which are used to identify the decline in blade fatigue performance.

5. The wind farm turbine operation optimization method based on artificial intelligence according to claim 1, characterized in that: The yaw compensation angle in S4 is determined by the following steps: Extract the power difference between real-time power and theoretical maximum power; The power-yaw angle function is generated based on the least squares principle; Calculate the first derivative of the yaw angle function with respect to the yaw angle, and take the position where the derivative is zero as the yaw compensation angle.

6. The wind farm turbine operation optimization method based on artificial intelligence according to claim 1, characterized in that: The wake simulation in S5 adopts an empirical wake model, which uses the wake diffusion coefficient and attenuation coefficient to describe the wake diffusion characteristics, thus ensuring the wake simulation is real-time.

7. The wind farm turbine operation optimization method based on artificial intelligence according to claim 1, characterized in that: Before the S5 power generation allocation strategy, the units are divided into multiple groups based on their relative orientation and the dominant wind direction. The groups are divided using the K-means algorithm so that units in the same wake chain belong to the same group.

8. The wind farm turbine operation optimization method based on artificial intelligence according to claim 1, characterized in that: The genetic algorithm in S6 uses a fixed crossover probability and a fixed mutation probability, with a crossover probability of 0.7 and a mutation probability of 0.2, and obtains the optimal solution that meets the convergence condition through 50 generations of iteration.

9. The wind farm turbine operation optimization method based on artificial intelligence according to claim 1, characterized in that: After the optimized control is executed, a double tip winglet is set for the blade. The double tip winglet includes a base section and an extension section. The base section is fixed to the reinforcement position of the trailing edge structure of the blade by bolts. The extension section is arranged along the span of the blade and is set with a fixed dihedral angle, which increases the lift coefficient of the blade and reduces the tip vortex intensity.

10. A wind farm turbine operation optimization system based on artificial intelligence, characterized in that: The optimization method applied to any one of claims 1-9 includes: a data acquisition module for acquiring unit output power data, vibration data, temperature data, yaw angle data, blade strain data, and lidar wind speed and direction data; The digital twin model module constructs and updates the overall dynamics model, blade finite element model, and tower structure model in real time. The artificial intelligence analysis module extracts time-series features from multi-source data and outputs state vectors; The yaw correction module calculates the yaw compensation angle based on the state vector and executes yaw control. The wake calculation module generates wake distribution and power allocation strategies based on real-time wind conditions. The cluster scheduling module groups the generating units and executes power optimization strategies within each group. The execution module is used to perform pitch angle adjustment, power limit adjustment, and yaw execution operations.

Images