A Pitch Angle Dynamic Compensation Method Based on Multi-Radar Spatiotemporal Synchronization

Abstract

The present invention discloses a pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization, which relates to the technical field of wind power generation. Aiming at the problem that the traditional single-point wind speed measurement cannot capture the spatio-temporal characteristics of the three-dimensional wind field, resulting in insufficient pitch angle control accuracy, the present invention synchronously constructs a three-dimensional spatio-temporal velocity field of the wind turbine plane through a horizontal-vertical radar array, combines a flow field-structure coupling model to calculate the aerodynamic load distribution of the blade segments in real time, generates a reference pitch command by mapping the dynamic load deviation through a neural network, and performs pre-compensation by integrating the tower shadow interference eigenvector; adopts a quadratic programming algorithm to counteract saturation optimization, so that the pitch rate strictly follows the dynamic constraint curve of the hydraulic system. The present invention effectively solves the problem of aerodynamic load phase delay, reduces the periodic load fluctuation caused by the tower shadow effect, and provides technical guarantee for the stable operation of large-scale wind turbines under complex wind conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and particularly to a pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization. Background Art

[0002] As an important part of clean energy, wind power generation is playing an increasingly important role in the global energy structure transformation. However, wind turbines face complex and variable wind conditions, which pose great challenges to pitch angle control. The traditional control system of wind turbines mainly relies on single-point wind speed measurement. This method is difficult to accurately capture the spatio-temporal variation characteristics of the wind field under complex wind conditions, resulting in limitations in the accuracy and response speed of pitch angle control.

[0003] Currently, the pitch angle control systems of wind turbines generally have the following problems: First, single-point wind speed measurement cannot comprehensively reflect the wind speed distribution in the rotor plane, resulting in uneven forces on the blades; Second, traditional control algorithms are difficult to cope with rapidly changing wind conditions, causing phase delays in the dynamic response of aerodynamic loads and affecting power generation efficiency and equipment life. Summary of the Invention

[0004] The present invention provides a pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization to solve the problems in the background art.

[0005] The present application provides a pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization. The technical solution is as follows: By synchronously measuring data from a horizontal scanning radar array and a vertical profile radar array, a three-dimensional spatio-temporal velocity field of the rotor plane is constructed to comprehensively capture the wind speed changes around the rotor; The velocity field data is input into a fluid-structure coupling model to calculate the dynamic aerodynamic load distribution of each blade segment in real time, reflecting the force conditions of the blades under different wind conditions; Based on the deviation between the aerodynamic load distribution data and the target load, an initial pitch compensation instruction set is generated; The initial instruction set is input into an actuator dynamic constraint model for anti-saturation optimization, and the final pitch angle control instruction is output and synchronized to the pitch actuator.

[0006] Furthermore, the present application also proposes that the construction process of the three-dimensional spatio-temporal velocity field includes: receiving the original point cloud data stream of multiple radar nodes, eliminating the coordinate offset caused by tower sway through a spatio-temporal registration algorithm; mapping the registered point cloud data to a rotating coordinate system to generate a three-dimensional grid velocity field with the impeller center as the origin; Based on the difference in velocity field data between adjacent time steps, the spatial distribution of the flow field acceleration field is calculated.

[0007] Further, the present application also proposes that the generation of the initial pitch compensation instruction set includes: inputting the dynamic load deviation vector into a pre-trained neural network model to map and generate the reference pitch angle increment of each blade; superimposing the tower shadow interference feature vector constructed based on historical operation data to generate an initial pitch instruction with pre-compensation; and fusing the initial instruction set with the current pitch angle state data to generate a time-continuous pitch instruction sequence.

[0008] Further, the present application also proposes that the generation of the tower shadow interference feature vector includes: obtaining the historical wind direction data stream from the SCADA system to construct a time-space distribution map of the tower wake interference pattern; extracting the near-tower area velocity profile in the radar data in real time and performing convolution similarity matching with the distribution map; and extracting the wake vortex intensity coefficient and phase delay parameter according to the matching result to construct a time-varying interference feature vector.

[0009] Further, the present application also proposes that the anti-saturation optimization data processing includes: receiving the real-time state data stream of the pitch actuator, including the hydraulic oil pressure, actuator displacement, and temperature monitoring values; inputting the state data into the actuator dynamic model to predict the pitch rate limit curve within a future time window; and re-planning the initial instruction set using a constrained optimization algorithm to make the pitch trajectory fall within the range of the rate limit curve.

[0010] Further, the present application also proposes that the update of the actuator dynamic model includes: injecting a swept-frequency test signal during the idle period of the pitch system to collect the frequency response data of the actuator; identifying the damping coefficient and stiffness parameter in the dynamic model based on the response data; and updating the weight matrix of the dynamic model when the parameter change amount exceeds the adaptive threshold.

[0011] Further, the present application also proposes that the method further includes prospective flow field compensation: performing proper orthogonal decomposition on the radar time-series velocity field data to extract the spatio-temporal evolution coefficients of the dominant flow modes; inputting the evolution coefficients into a pre-trained LSTM network to predict the flow field evolution trend in multiple future rotation periods; and injecting the predicted flow field data into the flow field-structure coupling model in advance to generate an advanced compensation instruction component.

[0012] Further, the present application also proposes that the fusion of the advanced compensation instructions includes: calculating the real-time error amount between the actual flow field data and the predicted data to construct a prediction credibility weight factor; using a weighted fusion algorithm to mix the advanced compensation instructions and the real-time compensation instructions; and gradually reducing the weight ratio of the advanced compensation instructions until they are closed when the prediction error continues to increase.

[0013] As described above, a pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization provided by the present application constructs a three-dimensional spatio-temporal velocity field of the wind turbine plane through the synchronized measurement data of the horizontal scanning radar array and the vertical profile radar array; inputs the velocity field data into a flow field-structure coupling model to calculate the dynamic aerodynamic load distribution of each blade segment in real time; generates an initial pitch compensation instruction set based on the deviation between the aerodynamic load distribution data and the target load; inputs the initial instruction set into an actuator dynamic constraint model for anti-saturation optimization, outputs the final pitch angle control instruction and synchronizes it to the pitch actuator, thereby achieving precise capture and compensation of complex wind conditions, having the advantages of improving the pitch angle control accuracy and response speed, improving the evenness of blade force, effectively coping with rapidly changing wind conditions, reducing the phase delay of the dynamic response of aerodynamic loads, and effectively compensating for the tower shadow effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flowchart of a pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] The purpose of the present application is to provide a pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization, which has the advantages of improving the pitch angle control accuracy and response speed, achieving precise capture and compensation of complex wind conditions, improving the evenness of blade force, effectively coping with rapidly changing wind conditions, reducing the phase delay of the dynamic response of aerodynamic loads, and effectively compensating for the tower shadow effect.

[0016] The above technical solutions will be described in detail below in combination with the drawings of the specification and specific embodiments to better understand the above technical solutions. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments only used to explain the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention. In addition, it should be noted that for the sake of description, only parts related to the present invention are shown in the drawings rather than all.

[0017] Embodiment 1

[0018] As Figure 1 shown in a flowchart of a pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization:

[0019] The present invention belongs to the field of intelligent control of wind turbines, and particularly relates to a pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization. The core inventive concept breaks through the limitations of traditional single-point wind speed measurement. By constructing a three-dimensional spatio-temporal synchronization observation network with a multi-radar array, a dynamic coupling relationship between the blade rotation coordinate system and the flow field evolution model is established to solve the phase delay problem of the dynamic response of aerodynamic loads under complex wind conditions.

[0020] The problem of the dynamic response of the blade aerodynamic load of a wind turbine under complex wind conditions has always been a technical challenge. Traditional single-point wind speed measurement methods are difficult to comprehensively capture the wind speed changes around the wind turbine rotor, resulting in the inability to accurately reflect the force conditions of the blades, thereby affecting the accuracy and reliability of pitch control. To solve this problem, the present invention proposes a pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization.

[0021] This method constructs a three-dimensional spatio-temporal velocity field of the wind turbine plane through the synchronized measurement data of a horizontal scanning radar array and a vertical profile radar array. These velocity field data are input into a flow field-structure coupling model to calculate the dynamic aerodynamic load distribution of each blade segment in real time. Based on the deviation between the aerodynamic load distribution data and the target load, an initial pitch compensation instruction set is generated. Then, the initial instruction set is input into the dynamic constraint model of the actuator for anti-saturation optimization, and the final pitch angle control instruction is output and synchronized to the pitch actuator.

[0022] Constructing a three-dimensional spatio-temporal velocity field of the wind turbine plane through the synchronized measurement data of a horizontal scanning radar array and a vertical profile radar array helps to comprehensively capture the wind speed changes around the wind turbine rotor. Inputting these velocity field data into a flow field-structure coupling model to calculate the dynamic aerodynamic load distribution of each blade segment in real time can accurately reflect the force conditions of the blades under different wind conditions. Generating an initial pitch compensation instruction set based on the deviation between the aerodynamic load distribution data and the target load makes the pitch control more accurate. Inputting the initial instruction set into the dynamic constraint model of the actuator for anti-saturation optimization ensures that the pitch control instruction does not exceed the capabilities of the actuator during execution, and outputting the final pitch angle control instruction and synchronizing it to the pitch actuator guarantees the reliability and effectiveness of the pitch control.

[0023] In practical applications, it is first necessary to synchronously obtain the wind speed data around the wind turbine through a horizontal scanning radar array and a vertical profile radar array. After processing these data, a three-dimensional spatio-temporal velocity field of the wind turbine plane is constructed. Then, these velocity field data are input into a fluid-structure coupling model, which can calculate the dynamic aerodynamic load distribution of each blade segment in real time. Based on these calculation results, a comparison is made with the target load to generate an initial pitch compensation instruction set. To ensure that the pitch control instructions do not exceed the capabilities of the actuator during execution, the initial instruction set needs to be input into the actuator dynamic constraint model for anti-saturation optimization, and the finally output pitch angle control instructions will be synchronized to the pitch actuator.

[0024] Compared with the prior art, the present invention constructs a three-dimensional spatio-temporal synchronous observation network through multiple radar arrays, which can comprehensively capture the wind speed changes around the wind turbine and solve the limitations of traditional single-point wind speed measurement. Through the fluid-structure coupling model, the dynamic aerodynamic load distribution of each blade segment can be calculated in real time to ensure the accuracy of pitch control. Through the actuator dynamic constraint model for anti-saturation optimization, it is ensured that the pitch control instructions do not exceed the capabilities of the actuator during execution, thus guaranteeing the reliability and effectiveness of pitch control.

[0025] By synchronously measuring data through a horizontal scanning radar array and a vertical profile radar array to construct a three-dimensional spatio-temporal velocity field of the wind turbine plane, it helps to comprehensively capture the wind speed changes around the wind turbine. Inputting these velocity field data into the fluid-structure coupling model to calculate the dynamic aerodynamic load distribution of each blade segment in real time can accurately reflect the force conditions of the blades under different wind conditions. Based on the deviation between the aerodynamic load distribution data and the target load, an initial pitch compensation instruction set is generated, making the pitch control more accurate. Inputting the initial instruction set into the actuator dynamic constraint model for anti-saturation optimization, ensuring that the pitch control instructions do not exceed the capabilities of the actuator during execution, and outputting the final pitch angle control instructions and synchronizing them to the pitch actuator, guarantee the reliability and effectiveness of pitch control.

[0026] Furthermore, this application also proposes to receive the original point cloud data stream of multiple radar nodes, eliminate the coordinate offset caused by tower sway through a spatio-temporal registration algorithm; map the registered point cloud data to a rotating coordinate system to generate a three-dimensional grid velocity field with the impeller center as the origin; calculate the spatial distribution of the flow field acceleration field based on the difference in velocity field data between adjacent time steps.

[0027] By receiving the original point cloud data streams from multiple radar nodes and using a spatio-temporal registration algorithm to eliminate the coordinate offset caused by tower sway, the accuracy of the data can be ensured. Mapping the registered point cloud data to a rotating coordinate system to generate a three-dimensional meshed velocity field with the impeller center as the origin helps to better reflect the actual aerodynamic environment of the wind turbine plane. Based on the difference in velocity field data between adjacent time steps, calculating the spatial distribution of the flow field acceleration field can further improve the accurate prediction of the aerodynamic load changes in the wind turbine plane.

[0028] Receive the original point cloud data streams from multiple radar nodes and eliminate the coordinate offset caused by tower sway through a spatio-temporal registration algorithm. The spatio-temporal registration algorithm can achieve the purpose of eliminating the coordinate offset through various implementation methods, such as using a registration algorithm based on the least squares method or a registration method based on the iterative closest point (ICP) algorithm. Mapping the registered point cloud data to a rotating coordinate system can be achieved through a rotation matrix transformation. Generating a three-dimensional meshed velocity field with the impeller center as the origin can be achieved through a mesh generation algorithm, and common meshing methods include uniform mesh generation and adaptive mesh generation. Based on the difference in velocity field data between adjacent time steps, calculating the spatial distribution of the flow field acceleration field can be achieved through numerical differentiation methods, such as using backward differentiation, forward differentiation, or central differentiation methods.

[0029] Specifically, synchronize the measurement data of the horizontal scanning radar array and the vertical profile radar array to construct a three-dimensional spatio-temporal velocity field of the wind turbine plane to solve the problem that traditional single-point wind speed measurement can only obtain local wind speed information and cannot capture the spatio-temporal evolution characteristics of the three-dimensional wind field:

[0030] Basis for heterogeneous radar selection: X-band radar (horizontal scanning): Utilize its high range resolution (0.5 m) to capture the horizontal wind shear effect;

[0031] Millimeter-wave radar (vertical profile scanning): Obtain the vertical turbulence component through wide beam coverage (-30° to +45° pitch angle).

[0032] Dual-mode timing module (Beidou / GPS): Ensure that the time synchronization error of multi-radar data is <1 μs and eliminate the impact of time-domain misalignment on the three-dimensional flow field reconstruction.

[0033] Dynamic coordinate registration algorithm: 1. Tower shadow effect compensation: Based on the tower sway model, real-time correct the radar coordinate offset;

[0034] 2. Extended Kalman filter (EKF): Fuse the IMU attitude data and laser ranging data to establish a rotating coordinate system conversion model to ensure the accurate mapping of the point cloud data to the impeller center coordinate system.

[0035] 3D grid wind field construction: Map the registered point cloud to the impeller rotation coordinate system to generate a 0.5m×0.5m×0.5m grid velocity field; Calculate the acceleration distribution of the flow field based on the velocity field difference between adjacent time steps.

[0036] This application eliminates the coordinate offset caused by tower sway through a spatio-temporal registration algorithm to ensure data accuracy; Map the registered point cloud data to the rotation coordinate system to generate a 3D grid velocity field with the impeller center as the origin, which can better reflect the actual aerodynamic environment of the wind turbine plane; Calculate the spatial distribution of the flow field acceleration field based on the velocity field data difference between adjacent time steps to further improve the accurate prediction of the aerodynamic load change in the wind turbine plane. These technical features have significant advantages compared with the prior art, can more accurately solve the coordinate offset problem of multi-radar node data, and improve the prediction accuracy of the aerodynamic load change in the wind turbine plane.

[0037] Furthermore, this application also proposes to input the dynamic load deviation vector into a pre-trained neural network model to map and generate the reference pitch angle increment of each blade; Superimpose the tower shadow interference feature vector constructed based on historical operation data to generate an initial pitch command with pre-compensation; Fuse the initial command set with the current pitch angle state data to generate a time-continuous pitch command sequence.

[0038] This application includes all the features of the preamble part and the characterizing part. The preamble part relates to the acquisition of the dynamic load deviation vector, and the characterizing part includes inputting the vector into a pre-trained neural network model to generate the reference pitch angle increment, superimposing the tower shadow interference feature vector to generate the initial pitch command, and finally fusing the current pitch angle state data to generate a time-continuous pitch command sequence. These technical features cooperate with each other to solve the dynamic load compensation problem of the blade under complex wind conditions. By inputting the dynamic load deviation vector into a pre-trained neural network model, the reference pitch angle increment of each blade can be accurately mapped and generated. Superimposing the tower shadow interference feature vector further optimizes the initial pitch command, making it more targeted and effective. Finally, through the fusion with the current pitch angle state data, a time-continuous pitch command sequence is generated, thus realizing the dynamic load compensation of the blade under complex wind conditions.

[0039] The acquisition of the dynamic load deviation vector can be achieved in various ways. For example, it can be calculated by combining the three-dimensional spatio-temporal velocity field data constructed by a multi-radar array with a flow field-structure coupling model. The pre-trained neural network model can be trained using deep learning techniques. After inputting the dynamic load deviation vector, it outputs the reference pitch angle increment. The generation of the tower shadow interference feature vector can be based on historical operation data and constructed through data mining and pattern recognition techniques. The generation of the initial pitch instruction set requires superimposing the reference pitch angle increment and the tower shadow interference feature vector. Finally, the fusion of the initial instruction set and the current pitch angle state data can be achieved through state estimation and control algorithms to generate a time-continuous pitch instruction sequence.

[0040] This application optimizes the pitch angle control method by introducing a neural network model and historical data analysis, and can more accurately respond to dynamic load changes under complex wind conditions. Compared with traditional control methods, this application has higher accuracy and response speed when dealing with complex wind conditions, and can significantly improve the operating stability and efficiency of wind turbines.

[0041] Furthermore, this application also proposes to obtain the historical wind direction data stream from the SCADA system, construct a spatio-temporal distribution map of the tower wake interference pattern; extract the near-tower area velocity profile in the radar data in real time, and perform convolutional similarity matching with the distribution map; extract the wake vortex intensity coefficient and phase delay parameter according to the matching result, and construct a time-varying interference feature vector.

[0042] Obtain the historical wind direction data stream from the SCADA system, construct a spatio-temporal distribution map of the tower wake interference pattern for identifying and predicting tower wake interference. Extract the near-tower area velocity profile in the radar data in real time, and perform convolutional similarity matching with the distribution map to ensure the matching accuracy. Extract the wake vortex intensity coefficient and phase delay parameter according to the matching result, and construct a time-varying interference feature vector. Through these technical features, the influence of tower wake interference on pitch angle control can be effectively identified and compensated, and the accuracy and reliability of pitch angle dynamic compensation can be improved.

[0043] Furthermore, the historical wind direction data stream of the SCADA system can be processed through data mining and analysis techniques to construct a more accurate tower wake interference pattern. The process of convolutional similarity matching can be realized through an efficient algorithm to ensure real-time performance and accuracy. The extraction of the wake vortex intensity coefficient and phase delay parameter can be carried out through comparative analysis and optimization algorithms to improve the accuracy and reliability of the time-varying interference feature vector. For example, algorithms based on machine learning can be used to optimize the matching process and parameter extraction process, thereby improving the performance and response speed of the overall system.

[0044] Furthermore, wake database construction:

[0045] Extract 10-year historical wind direction data from the SCADA system to construct a tower wake interference pattern atlas;

[0046] The convolutional neural network CNN matches the real-time radar data and outputs the wake vorticity intensity coefficient and phase delay.

[0047] Superimpose a tower shadow interference pre-compensation term on the pitch command:

[0048] This application constructs a spatio-temporal distribution atlas of the tower wake interference pattern, combines the convolutional similarity matching of real-time radar data, extracts the wake vorticity intensity coefficient and phase delay parameters, and can effectively solve the influence of tower wake interference on the dynamic compensation of the pitch angle. Compared with the prior art, the method of this application can provide higher matching accuracy and real-time performance, improve the accuracy and reliability of the dynamic compensation of the pitch angle, and thus better adapt to complex wind conditions.

[0049] Furthermore, this application also proposes to receive the real-time status data stream of the pitch actuator, including hydraulic oil pressure, actuator displacement, and temperature monitoring values; input the status data into the actuator dynamic model to predict the pitch rate limit curve within the future time window; use a constrained optimization algorithm to re-plan the initial instruction set so that the pitch trajectory falls within the rate limit curve range.

[0050] By receiving the real-time status data stream of the pitch actuator, including hydraulic oil pressure, actuator displacement, and temperature monitoring values, the working state of the actuator can be understood in real time; inputting these status data into the actuator dynamic model can predict the pitch rate limit curve within the future time window, thereby predicting the working ability of the actuator in advance; using a constrained optimization algorithm to re-plan the initial instruction set ensures that the pitch trajectory falls within the rate limit curve range and avoids the actuator saturation problem.

[0051] In practical applications, receiving the real-time status data stream of the pitch actuator can be achieved through sensors installed on the actuator, and these sensors can monitor key parameters such as hydraulic oil pressure, actuator displacement, and temperature in real time. These data will be transmitted to the control system in real time. Inputting these status data into the dynamic model of the actuator can be realized using physical model-based or data-driven modeling methods. Physical models usually involve detailed modeling of the mechanical and hydraulic characteristics of the actuator, while data-driven methods can use machine learning techniques to make predictions based on historical data. Using a constrained optimization algorithm to re-plan the initial instruction set, specifically, linear or nonlinear programming methods can be used to adjust the initial instruction set in combination with the pitch rate limit curve to ensure that the pitch trajectory does not exceed the actuator's ability range within the future time window.

[0052] By monitoring the status data of the pitch actuator in real time and combining a dynamic model and an optimization algorithm, this application can effectively avoid the actuator saturation problem. Compared with the prior art, this method can more accurately predict and control the pitch rate, thereby improving the operating stability and efficiency of the wind turbine generator set.

[0053] Furthermore, this application also proposes injecting a swept-frequency test signal during the idle period of the pitch system to collect the frequency response data of the actuator. Based on the response data, the damping coefficient and stiffness parameters in the dynamic model are identified. When the change amount of the parameters exceeds the adaptive threshold, the weight matrix of the dynamic model is updated.

[0054] By injecting a swept-frequency test signal during the idle period of the pitch system, collecting the frequency response data of the actuator, and identifying the damping coefficient and stiffness parameters in the dynamic model based on these data, the dynamic model of the actuator can be monitored and updated in real time. When the change amount of the parameters exceeds the adaptive threshold, the weight matrix of the dynamic model is updated. This process ensures that the pitch system can maintain high response accuracy and stability under different environmental and operating conditions, thus effectively solving the problems of response accuracy and stability of the pitch system in a dynamic environment.

[0055] The swept-frequency test signal can be generated by a conventional signal generator, and the signal frequency range can be adjusted according to the operating frequency of the pitch system. The acquisition of the frequency response data can be achieved by sensors installed on the actuator, and these sensors need to have high sensitivity and fast response capabilities. The identification of the dynamic model can adopt existing system identification algorithms, such as the least squares method or the recursive least squares method. Through these algorithms, the damping coefficient and stiffness parameters can be accurately extracted. When it is detected that the change amount of the parameters exceeds the preset adaptive threshold, the system will automatically trigger the update process of the weight matrix. The update of the weight matrix can be achieved through an optimization algorithm to ensure that the dynamic model can accurately reflect the current system state.

[0056] This technical solution ensures high response accuracy and stability of the system under different environmental and operating conditions by monitoring and updating the dynamic model of the pitch system in real time. Compared with the prior art, this solution introduces a swept-frequency test signal and an adaptive threshold mechanism, which can timely detect and correct the changes in system parameters and avoid the degradation of system performance caused by parameter drift. Thus, the reliability and control accuracy of the pitch system have been significantly improved.

[0057] Furthermore, this application also proposes performing proper orthogonal decomposition on the radar time-series velocity field data to extract the spatio-temporal evolution coefficients of the dominant flow modes; inputting the evolution coefficients into a pre-trained LSTM network to predict the flow field evolution trend in multiple future rotation periods; and injecting the predicted flow field data into the flow field-structure coupling model in advance to generate an advanced compensation instruction component.

[0058] This application includes the technical feature of forward-looking flow field compensation. By performing proper orthogonal decomposition on the radar time-series velocity field data, the spatio-temporal evolution coefficients of the dominant flow modes are extracted. These evolution coefficients are input into a pre-trained LSTM network to predict the evolution trend of the flow field in multiple future rotation cycles. Then, the predicted flow field data is injected into the flow field-structure coupling model in advance to generate an advanced compensation command component. Through the cooperation of these technical features, the influence of future wind field changes on pitch angle control can be predicted and compensated in advance, thereby improving the dynamic response ability and control accuracy of wind turbines under complex wind conditions. Through this solution, the problem of the influence of future wind field changes on pitch angle control can be effectively solved, ensuring that the wind turbine can maintain a stable operating state under different wind conditions and improving the power generation efficiency.

[0059] Furthermore, perform proper orthogonal decomposition on the radar time-series velocity field data to extract the spatio-temporal evolution coefficients of the dominant flow modes. Through the proper orthogonal decomposition method, the main flow modes can be extracted from the complex wind field data, and these modes represent the main change trends of the wind field. These evolution coefficients are input into a pre-trained LSTM network, and using the time series prediction ability of the LSTM network, the evolution trend of the flow field in multiple future rotation cycles is predicted. The predicted flow field data is injected into the flow field-structure coupling model in advance, and the future aerodynamic load changes are calculated through the coupling model to generate an advanced compensation command component. As a preferred implementation, various different proper orthogonal decomposition algorithms can be used, such as the standard POD method or the improved POD method, to improve the accuracy of flow mode extraction. The training of the LSTM network can be based on a large amount of historical wind field data to ensure the accuracy and reliability of the prediction.

[0060] Specifically, adaptive compensation control

[0061] 1. Generate a three-dimensional velocity field in real time through heterogeneous radars. After EKF registration, it is input into the flow field-structure coupling model; the model outputs the dynamic aerodynamic load distribution of each blade segment; the dynamic aerodynamic load distribution is input into a pre-trained radial basis function network, and the reference pitch angle is output; its network structure: 7 nodes in the input layer, 32 nodes in the hidden layer, and 3 nodes in the output layer (independent compensation amounts for three blades);

[0062] Input features of the input layer: [average bending moment fluctuation, standard deviation of bending moment, wind speed at hub height, impeller speed, current pitch angle, hydraulic system oil temperature, time interval since the last pitch change];

[0063] 2. Anti-saturation optimization: Actuator state monitoring: Real-time collect the hydraulic oil pressure range, displacement, and temperature;

[0064] Dynamic constraint model: Predict the pitch rate limit curve in the next 5s, and re-plan the command using the quadratic programming algorithm;

[0065] 3. Establish a linear approximate dynamic model by extracting the dominant modes from the spatio-temporal data of the flow field.

[0066] 1. Data matrix construction: Arrange the sequence of velocity fields measured by the radar into a spatio-temporal matrix: , where n is the number of spatial points and m is the number of time steps;

[0067] 2. Singular value decomposition (SVD): Truncate to the first r principal modes;

[0068] 3. Construct an approximate Kooman matrix: , where is the time shift matrix;

[0069] 4. Future flow field prediction: where the prediction step k = 3 is about 1.2 seconds, and the mechanical delay of the actuator is compensated in advance.

[0070] 5. Weight fusion: When the prediction error > 15%, linearly reduce the weight of the lead compensation.

[0071] Thus, through the forward-looking flow field compensation technology, this application can predict and compensate in advance the impact of future wind field changes on the pitch angle control. Compared with the prior art, this application significantly improves the dynamic response ability and control accuracy of the wind turbine under complex wind conditions, ensures that the wind turbine can maintain a stable operating state under different wind conditions, and improves the power generation efficiency. This technical solution not only solves the problem of the impact of future wind field changes on the pitch angle control, but also improves the intelligent control level of the wind turbine through a forward-looking prediction and compensation mechanism.

[0072] Furthermore, this application also proposes to calculate the real-time error amount between the actual flow field data and the predicted data, construct a prediction credibility weight factor, and use a weighted fusion algorithm to mix the lead compensation instruction and the real-time compensation instruction. When the prediction error continues to increase, gradually reduce the weight ratio of the lead compensation instruction until it is turned off.

[0073] By calculating the real-time error amount between the actual flow field data and the predicted data and constructing a prediction credibility weight factor, the reliability of the predicted data can be evaluated. Using a weighted fusion algorithm to mix the lead compensation instruction and the real-time compensation instruction can flexibly adjust the weight of the compensation instruction in different situations. When the prediction error continues to increase, gradually reducing the weight ratio of the lead compensation instruction until it is turned off can prevent unreliable predicted data from having a negative impact on the compensation effect, thereby improving the accuracy and reliability of the compensation instruction.

[0074] Calculating the real-time error amount between the actual flow field data and the predicted data is achieved by comparing the currently measured actual flow field data with the flow field data predicted based on historical data. The construction of the prediction credibility weight factor is based on the magnitude of the error amount. The smaller the error amount, the higher the credibility of the predicted data, and the larger the weight factor. The weighted fusion algorithm generates the final compensation instruction by weighted averaging the lead compensation instruction and the real-time compensation instruction according to the weight factor. When the prediction error continuously increases, the system will automatically reduce the weight proportion of the lead compensation instruction until it completely relies on the real-time compensation instruction to ensure the reliability of the compensation instruction.

[0075] Through the above method, this application can effectively improve the accuracy and reliability of the compensation instruction. Especially in complex and variable wind conditions, it can flexibly adjust the compensation strategy to avoid the problem of compensation failure caused by prediction errors. Compared with the prior art, this application provides a more intelligent and adaptive compensation method, which can significantly improve the operation efficiency and stability of the wind turbine generator set.

[0076] It should be understood that the disclosed embodiments of the present invention and the above descriptions enable those skilled in the art to implement the present invention using the present invention. At the same time, the present invention is not limited to the part of the embodiments mentioned above. It should be understood that those of ordinary skill in the art can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization, characterized in that, Comprising: Construct a three-dimensional spatio-temporal velocity field of the wind turbine plane through the synchronized measurement data of the horizontal scanning radar array and the vertical profile radar array to comprehensively capture the wind speed changes around the wind turbine; Input the velocity field data into the fluid-structure coupling model to calculate the dynamic aerodynamic load distribution of each blade segment in real time, reflecting the force conditions of the blades under different wind conditions; Generate an initial pitch compensation instruction set based on the deviation between the aerodynamic load distribution data and the target load; Input the initial pitch compensation instruction set into the actuator dynamic constraint model for anti-saturation optimization, output the final pitch angle control instruction and synchronize it to the pitch actuator, where the instruction is re-planned through the quadratic programming algorithm to limit the pitch rate within the dynamic constraint curve corresponding to the hydraulic system oil pressure threshold; The generation of the initial pitch compensation instruction set includes: Input the dynamic load deviation vector into the pre-trained neural network model to map and generate the reference pitch angle increment of each blade; Superimpose the tower shadow interference feature vector constructed based on historical operation data to generate an initial pitch instruction with pre-compensation; Fuse the initial pitch compensation instruction set with the current pitch angle state data to generate a time-continuous pitch instruction sequence; The generation of the tower shadow interference feature vector includes: Obtain the historical wind direction data stream from the SCADA system and construct a time-space distribution map of the tower wake interference pattern; Extract the near-tower area velocity profile in the radar data in real time and perform convolution similarity matching with the distribution map; Extract the wake vortex intensity coefficient and phase delay parameter according to the matching result to construct a time-varying interference feature vector.

2. The pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization according to claim 1, wherein The construction process of the three-dimensional spatio-temporal velocity field includes: Receive the original point cloud data stream of multiple radar nodes and eliminate the coordinate offset caused by tower swing through the spatio-temporal registration algorithm; Map the registered point cloud data to the rotating coordinate system to generate a three-dimensional grid velocity field with the impeller center as the origin; Calculate the spatial distribution of the flow field acceleration field based on the difference between the velocity field data of adjacent time steps.

3. The pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization according to claim 1, characterized in that The data processing of the anti-saturation optimization includes: Receive the real-time state data stream of the pitch actuator, including hydraulic oil pressure, actuator displacement, and temperature monitoring values; Input the state data into the actuator dynamic model to predict the pitch rate limit curve within the future time window; Use a constrained optimization algorithm to re-plan the initial pitch compensation instruction set to make the pitch trajectory fall within the range of the rate limit curve.

4. The pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization according to claim 3, characterized in that The update of the actuator dynamic model includes: Inject a swept-frequency test signal during the idle period of the pitch system and collect the frequency response data of the actuator; Identify the damping coefficient and stiffness parameter in the dynamic model based on the response data; When the parameter change amount exceeds the adaptive threshold, update the weight matrix of the dynamic model.

5. The pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization according to claim 1, characterized in that The method further includes prospective flow field compensation: Perform proper orthogonal decomposition on the radar time-series velocity field data and extract the spatio-temporal evolution coefficients of the dominant flow modes; Input the evolution coefficients into the pre-trained LSTM network to predict the flow field evolution trend in the future multiple rotation periods; Inject the predicted flow field data into the fluid-structure coupling model in advance to generate an advanced compensation instruction component.

6. The pitch angle dynamic compensation method based on multi-radar spatio-temporal synchronization according to claim 5, wherein The fusion of the advanced compensation instructions includes: Calculate the real-time error amount between the actual flow field data and the predicted data, and construct a prediction credibility weight factor; Use a weighted fusion algorithm to mix the lead compensation instruction and the real-time compensation instruction; When the prediction error continues to increase, gradually reduce the weight ratio of the lead compensation instruction until it is turned off.

Images