Wind turbine generator system wind farm monitoring method and system

Abstract

The application discloses a kind of wind farm monitoring method and system of wind turbine unit, it is related to wind power generation technical field, including: obtaining wind speed information, blade load information and operating parameter information of wind turbine unit;Equivalent wind speed prediction information and corresponding equivalent compensation parameter prediction information are determined based on wind speed information, blade load information and operating parameter information prediction;Determine the corresponding control reference parameter information based on equivalent wind speed prediction information;Control reference parameter information is feedforward corrected based on equivalent compensation parameter prediction information, and control correction parameter information is determined;Variable pitch angle correction parameter information and torque correction parameter information in control correction parameter information are used to control wind turbine unit monitoring wind farm.The application predicts impeller plane equivalent wind speed and feedforward compensation parameter by fusing multi-source data, so as to actively correct variable pitch and torque instruction, realize accurate wind farm monitoring, to ensure structural safety.

Description

Technical Field

[0001] This application relates to the field of wind power generation technology, and in particular to wind farm monitoring methods and systems for wind turbine generator sets. Background Technology

[0002] When wind turbine generators operate in complex wind field environments, their power control and structural safety are highly dependent on the accurate perception of the actual inflow wind speed at the rotor plane. The equivalent wind speed at the rotor plane directly determines the aerodynamic load and energy capture of each section of the blade. Therefore, it is necessary to accurately track the dynamic changes of the equivalent wind speed at the rotor plane in order to effectively cope with flow field distortion and accurately represent the equivalent wind conditions of the wind field.

[0003] Currently, the existing practice involves directly measuring wind speed and direction at a single point using a mechanical or ultrasonic anemometer installed at the rear of the nacelle. This single-point measurement is then directly provided to the pitch and torque controllers as the effective wind speed at the rotor plane, allowing for unified adjustment based on a standardized pitch strategy. However, the measured data under complex wind conditions is highly irregular and susceptible to interference, failing to accurately represent wind field velocity or detect dynamic changes in effective wind speed. This leads to a mismatch between control actions and actual loads, causing the power curve to deviate from the design optimum, resulting in low power output accuracy. Furthermore, it exacerbates fatigue and extreme loads on components such as blades, main shaft, and tower, jeopardizing unit lifespan and operational safety. Therefore, how to more accurately monitor the wind field under complex wind conditions is a pressing issue that needs to be addressed. Summary of the Invention

[0004] The main purpose of this application is to provide a wind farm monitoring method and system for wind turbine generators, aiming to solve the technical problem of how to more accurately monitor the wind farm under complex wind conditions.

[0005] To achieve the above objectives, this application proposes a method for monitoring wind farms using wind turbine generators, the method comprising: To acquire wind speed information, blade load information, and operating parameter information of wind turbine generators; Based on the wind speed information, the blade load information, and the operating parameter information, predictions are made to determine the equivalent wind speed prediction information and the corresponding equivalent compensation parameter prediction information. Based on the equivalent wind speed prediction information, the corresponding control reference parameter information is determined; Based on the equivalent compensation parameter prediction information, the control reference parameter information is feedforward corrected to determine the control correction parameter information; The wind turbine generator is controlled to monitor the wind farm based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information.

[0006] In one embodiment, the step of predicting and determining equivalent wind speed prediction information and corresponding equivalent compensation parameter prediction information based on the wind speed information, the blade load information, and the operating parameter information includes: Based on the wind speed information, the blade load information and the operating parameter information, data preprocessing is performed to determine preprocessing information. The data preprocessing includes filtering and noise reduction, normalization and time-series fusion. Based on the preprocessed information, a predefined flow field reconstruction compensation model is input to predict the equivalent wind speed in the impeller plane, and the equivalent wind speed prediction information is determined. Based on the equivalent wind speed prediction information, the corresponding equivalent compensation parameter prediction information is determined.

[0007] In one embodiment, the step of predicting the equivalent wind speed in the impeller plane based on the pre-defined flow field reconstruction compensation model input with the preprocessed information, and determining the equivalent wind speed prediction information, includes: At least one of the single-point wind speed and radial wind speed in the preprocessed information is analyzed to obtain a first analysis result; At least one of the pendulum moment load and the oscillation moment load in the preprocessed information is analyzed to obtain a second analysis result; At least one of the generator speed parameters, pitch angle parameters, and output power parameters in the preprocessed information is analyzed to obtain a third analysis result; Based on the first analysis result, the second analysis result, and the third analysis result, a predefined flow field reconstruction compensation model is input to predict the equivalent wind speed in the impeller plane, thereby obtaining the equivalent wind speed prediction information.

[0008] In one embodiment, the step of predicting the equivalent wind speed in the impeller plane based on the first analysis result, the second analysis result, and the third analysis result by inputting a predefined flow field reconstruction compensation model to obtain equivalent wind speed prediction information includes: Obtain historical wind field monitoring information; The feature tensor information is determined based on the first analysis result, the second analysis result, and the third analysis result; Based on the historical wind field monitoring information and the feature tensor information, a predefined flow field reconstruction compensation model is input to predict the equivalent wind speed in the impeller plane, thereby obtaining equivalent wind speed prediction information. The equivalent wind speed prediction information includes equivalent prediction information of turbulence intensity, equivalent prediction information of wind shear index, and equivalent prediction information of inflow angle.

[0009] In one embodiment, the step of determining the corresponding equivalent compensation parameter prediction information based on the equivalent wind speed prediction information includes: Based on the equivalent wind speed prediction information, the implicit temporal features within a preset time period are extracted to determine the implicit temporal feature sequence. A pre-defined multi-head attention strategy is used to weight and fuse the focused feature vectors corresponding to the implicit temporal feature sequence to determine the focused feature vector fusion information. Based on the focused feature vector fusion information, the feedforward compensation coefficient of the preset control cycle is calculated to obtain the equivalent compensation parameter prediction information.

[0010] In one embodiment, the step of determining the corresponding control reference parameter information based on the equivalent wind speed prediction information includes: Obtain power limitation requirement information; Based on the power limitation demand information and the equivalent wind speed prediction information, the control reference parameters are calculated to obtain control reference parameter information, which includes pitch angle reference parameter information and torque reference parameter information.

[0011] In one embodiment, the step of feedforward correction of the control reference parameter information based on the equivalent compensation parameter prediction information to determine the control correction parameter information includes: Based on the pitch angle compensation value in the equivalent compensation parameter prediction information, the pitch angle reference parameter information in the control reference parameter information is feedforward corrected to obtain the pitch angle correction parameter information. Based on the torque compensation coefficient in the equivalent compensation parameter prediction information, the torque reference parameter information in the control reference parameter information is feedforward corrected to obtain the torque correction parameter information. Based on the pitch angle correction parameter information and the torque correction parameter information, control correction parameter information is obtained.

[0012] In one embodiment, the step of controlling the wind turbine generator to monitor the wind farm based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information includes: The target compensation angle information is determined based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information. The corresponding control command is determined based on the target compensation angle information; The wind turbine generator is controlled to monitor the wind farm based on the control commands.

[0013] In one embodiment, the step of determining the corresponding control command based on the target compensation angle information includes: Obtain theoretical power information; Based on the target compensation angle information, the corresponding actual output power is monitored to determine the actual output power monitoring information; Based on the comparison between the theoretical power information and the actual output power monitoring information, the power tracking deviation information is determined. The corresponding control command is determined based on the power tracking deviation information.

[0014] Furthermore, to achieve the above objectives, this application also proposes a wind farm monitoring system for wind turbine generators, the wind farm monitoring system comprising: The acquisition module is used to acquire wind speed information, blade load information, and operating parameter information of the wind turbine generator set; The processing module is used to make predictions based on the wind speed information, the blade load information and the operating parameter information, and to determine the equivalent wind speed prediction information and the corresponding equivalent compensation parameter prediction information. The processing module is also used to determine the corresponding control reference parameter information based on the equivalent wind speed prediction information; The execution module is used to perform feedforward correction on the control reference parameter information based on the equivalent compensation parameter prediction information, and determine the control correction parameter information; The execution module is also used to control the wind turbine generator set to monitor the wind field based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information.

[0015] One or more technical solutions proposed in this application have at least the following technical effects: This embodiment proposes a wind farm monitoring method for wind turbine generator sets, which acquires wind speed information, blade load information, and operating parameter information of the wind turbine generator set; predicts the equivalent wind speed and corresponding equivalent compensation parameter prediction information based on the wind speed, blade load, and operating parameter information; determines the corresponding control reference parameter information based on the equivalent wind speed prediction information; performs feedforward correction on the control reference parameter information based on the equivalent compensation parameter prediction information to determine control correction parameter information; and controls the wind turbine generator set to monitor the wind farm based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information. This application acquires wind speed, blade load, and operating parameter information of wind turbine generators to sense inflow distortion and generator status, eliminating reliance on single-point data. It dynamically outputs equivalent wind speed prediction information and corresponding equivalent compensation parameter prediction information to compensate for flow field distortion and accurately characterize the true inflow wind speed at the impeller plane. This allows for the calculation of control reference parameters including pitch angle and torque, ensuring that the control target matches the actual wind conditions. The equivalent compensation parameter prediction information is used to feedforward correct the control reference parameters, and compensation commands are proactively injected before the controller operates, overcoming measurement lag and load mismatch. Based on the corrected pitch angle and torque correction parameters, the generator is controlled to monitor the wind field, achieving high-precision power tracking and structural safety assurance. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an embodiment of the wind farm monitoring method for wind turbine generators in this application. Figure 2 This is a flowchart illustrating Embodiment 2 of the wind farm monitoring method for wind turbine generators in this application. Figure 3 This is a schematic diagram of the module structure of the wind farm monitoring system for wind turbine generator sets according to an embodiment of this application; The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0020] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0021] The main solution of this application embodiment is as follows: acquiring wind speed information, blade load information, and operating parameter information of the wind turbine generator set; making predictions based on the wind speed information, blade load information, and operating parameter information to determine equivalent wind speed prediction information and corresponding equivalent compensation parameter prediction information; determining corresponding control reference parameter information based on the equivalent wind speed prediction information; performing feedforward correction on the control reference parameter information based on the equivalent compensation parameter prediction information to determine control correction parameter information; and controlling the wind turbine generator set to monitor the wind field based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information.

[0022] In this embodiment, for ease of description, the following description will focus on the wind farm monitoring equipment for identifying wind turbine generator sets.

[0023] Because existing technology measures data under complex wind conditions that are highly irregular and easily interfered with, it cannot accurately represent wind field speed or perceive dynamic changes in effective wind speed. This results in a mismatch between control actions and actual loads, causing the power curve to deviate from the design optimum, leading to low power output accuracy. It also exacerbates fatigue and extreme loads on components such as blades, main shafts, and towers, endangering the lifespan and operational safety of the unit.

[0024] This application provides a solution that, by acquiring wind speed, blade load, and operating parameter information of a wind turbine generator set, can sense inflow distortion and unit status, eliminating reliance on single-point data. It dynamically outputs equivalent wind speed prediction information and corresponding equivalent compensation parameter prediction information to compensate for flow field distortion and accurately characterize the true inflow wind speed at the impeller plane. This allows for the calculation of control reference parameters including pitch angle and torque, ensuring the control target matches the actual wind conditions. The equivalent compensation parameter prediction information is used to feedforward correct the control reference parameters, proactively injecting compensation commands before the controller operates, overcoming measurement lag and load mismatch. Based on the corrected pitch angle and torque correction parameters, the unit monitors the wind field, achieving high-precision power tracking and structural safety assurance.

[0025] Based on this, embodiments of this application provide a method for monitoring wind farms of wind turbine generators, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the wind farm monitoring method for wind turbine generator sets according to this application.

[0026] In this embodiment, the wind farm monitoring method for wind turbine generator sets includes steps S10 to S40: Step S10: Obtain wind speed information, blade load information, and operating parameter information of the wind turbine generator set; It should be noted that the wind speed information is measurement data used to characterize the spatial distribution and dynamic changes of the inflow wind conditions in front of the wind turbine rotor, including single-point wind speed, original wind direction measurement value, and average radial wind speed, used to provide inflow wind field characteristics at different scales. The blade load information is real-time acquired structural response data, including the bending moment measurement value of the blade in the flapping direction and the oscillation direction, used to indirectly characterize the non-uniform aerodynamic load distribution borne by each blade section on the rotor rotation plane. The operating parameter information is control and response variables acquired in real time to characterize the current operating state of the unit, including parameters such as generator speed, pitch angle, nacelle wind vane measurement value, and output power, used to assist in determining the operating conditions of the unit.

[0027] It is understandable that the wind turbine generator set is an electromechanical system that converts wind energy into electrical energy. It consists of components such as tower, nacelle, hub, and multiple blades. In complex wind field environments, such as high turbulence, strong wind shear, and complex terrain wakes, traditional power control systems rely solely on single-point measurements from the nacelle anemometer. This cannot accurately represent the inflow wind speed distribution that will act on the entire rotating rotor plane. Because single-point measurements lose spatial distribution information, they differ fundamentally from the actual aerodynamic loads borne by each blade section in key characteristics such as velocity, turbulence intensity, and angle of attack. Therefore, obtaining wind speed information, blade load information, and operating parameter information of the wind turbine generator set can provide a multi-dimensional and comprehensive understanding of inflow distortion and unit status, eliminating the reliance on single-point wind speed data.

[0028] In a specific embodiment, the wind speed information may include single-point wind speed, original wind direction measurement value, and average radial wind speed. The single-point wind speed and original wind direction measurement value can be obtained by a mechanical or ultrasonic anemometer installed at the tail of the nacelle, while the average radial wind speed can be obtained by the radial wind speed at multiple distance gates (such as 0.5 times the rotor radius, 1.0 times the rotor radius, and 1.5 times the rotor radius) in front of the impeller from a lidar wind measurement device deployed on the top of the nacelle. The blade load information can be obtained in real time by fiber Bragg grating (FBG) sensors or strain gauge sensors installed at the root or spanwise critical section of each blade. This can include the swing moment (the bending moment component that causes the blade to bend in the downwind direction) caused by aerodynamic thrust and the oscillation moment (the bending moment component that causes the blade to oscillate within the rotor's rotation plane) caused by gravity and aerodynamic torque. For example, when a blade bends under aerodynamic load, the composite material structure at the blade root region experiences strain, and the grating pitch of the attached fiber grating changes accordingly, causing a shift in the center wavelength of its reflection spectrum. This wavelength shift can be measured in real time using a high-speed spectrometer demodulator and converted into corresponding swing and oscillation moments. The operating parameter information can be obtained by real-time acquisition of basic state data of the wind turbine generator set, such as generator speed, pitch angle, nacelle wind vane measurement, and output power, characterizing the actual response state of the unit under current wind conditions.

[0029] Step S20: Based on the wind speed information, the blade load information and the operating parameter information, make a prediction to determine the equivalent wind speed prediction information and the corresponding equivalent compensation parameter prediction information; It should be noted that the equivalent wind speed prediction information is a real-time predicted value used to characterize the average effective inflow wind speed on the entire impeller rotation plane, including equivalent prediction information of turbulence intensity, equivalent prediction information of wind shear index, and equivalent prediction information of inflow angle. It is used to accurately characterize the aerodynamic energy input level that will be applied to the unit. The equivalent compensation parameter prediction information is a set of feedforward compensation coefficients or compensation values ​​calculated to actively correct the traditional controller commands, including pitch angle compensation value and torque compensation coefficient. It is used to directly inject compensation amount before the controller issues the final execution command to overcome control errors caused by flow field distortion, measurement lag, and load mismatch.

[0030] In a specific embodiment, data preprocessing is performed based on the wind speed information, blade load information, and operating parameter information to determine preprocessed information. The data preprocessing includes filtering and denoising, normalization, and time-series fusion. Specifically, the original timestamps of wind speed, blade load, and operating parameter information can be acquired synchronously and uniformly resampled to a preset high-frequency reference frequency. Filtering and denoising are then performed, using a combination of median filtering and low-pass filtering to remove abnormal transition points, effectively suppressing random noise and interference in the original sensor signal, improving the signal-to-noise ratio and accuracy of the input data. The data from each dimension are then normalized and mapped to a unified numerical range to eliminate numerical imbalances between different scales. This significantly improves the convergence speed and numerical stability of the deep learning model during training and inference. Subsequently, time-series fusion processing is performed, and the normalized multi-dimensional sequences of single-point wind speed, radial wind speed, flapping and swaying moments, generator speed, and pitch angle from each independent sensor within a preset time window are strictly aligned and spliced ​​into a multi-dimensional time-series feature matrix according to time steps. This yields preprocessed information, enabling the model to perceive the dynamic coupling relationship between wind condition changes and unit response.

[0031] Based on the preprocessed information, a predefined flow field reconstruction compensation model is input to predict the equivalent wind speed in the impeller plane, thus determining the equivalent wind speed prediction information. This preprocessed information can be used as an input feature tensor to input into the trained flow field reconstruction compensation model. The flow field reconstruction compensation model includes an input layer, a multi-source feature encoding layer, a temporal dependency modeling layer, an attention fusion layer, and an output layer connected sequentially. The input layer receives the preprocessed information, which is a multi-dimensional temporal feature matrix after filtering, normalization, and time-series fusion. Its dimension is (time step T, number of feature channels C), where T is the preset time window length (e.g., 5 seconds, corresponding to 50 time steps), and C is the number of feature channels (including 7 channels: single-point wind speed, radial wind speed, flapping moment, oscillation moment, generator speed, blade pitch angle, and output power). The multi-source feature encoding layer is used for... The raw measurements of each channel are mapped to a unified feature vector, specifically through three parallel sub-coding units: The first sub-coding unit performs a one-dimensional convolution operation on the wind speed channel (single-point wind speed and radial wind speed) with a kernel size of 3, outputting a 64-dimensional feature vector to extract the spatial gradient and fluctuation characteristics of the inflow wind field; the second sub-coding unit performs a similar one-dimensional convolution operation on the load channel (flapping moment and swaying moment) with a kernel size of 3, outputting a 64-dimensional feature vector to characterize the non-uniform aerodynamic load distribution on the impeller plane; the third sub-coding unit performs a fully connected linear mapping on the operating parameter channel (generator speed, pitch angle, output power), outputting a 32-dimensional feature vector to encode the response characteristics of the unit at the current operating point. The outputs of the above three sub-coding units are concatenated to form a 160-dimensional feature vector, which serves as the joint representation input to the time-dependent modeling layer. The temporal dependency modeling layer employs a bidirectional gated recurrent unit (Bi-GRU) structure. A gated recurrent unit is a variant of a recurrent neural network that effectively captures long-short-term dependencies in a time series, featuring update and reset gates to reduce the number of parameters while avoiding gradient vanishing. The Bi-GRU layer encodes the time series from front to back and from back to front using forward and backward GRUs respectively, concatenating the hidden states in both directions at each time step to comprehensively capture the contextual information during wind condition evolution. The number of hidden units in the Bi-GRU layer can be set to 128 (64 units per direction), with a time step T=50, outputting a 256-dimensional hidden state vector sequence for each time step. The attention fusion layer employs a multi-head self-attention mechanism, which allows the model to dynamically allocate weights in the input sequence. By calculating the similarity score between the query vector Q, key vector K, and value vector V, weighted focusing on information from different time steps is achieved. Therefore, the multi-head self-attention layer of the flow field reconstruction compensation model can contain 4 attention heads, each with a feature dimension of 64 (obtained by linear transformation of 256-dimensional hidden states). The outputs of the 4 heads are concatenated and then linearly transformed to obtain a 256-dimensional fused feature vector.This layer automatically assigns weights to different time steps and different characteristic channels, focusing on analyzing the nonlinear mapping relationship between single-point measurements in the nacelle wake region (severely affected by wake interference) and the actual inflow at the impeller plane (indirectly characterized by load and lidar data). It also assesses complex flow field distortion effects such as vertical wind shear and horizontal crosswinds. The output layer includes two parallel branches: the first branch outputs equivalent wind speed prediction information, containing three nodes corresponding to the equivalent values ​​of turbulence intensity, wind shear index, and inflow angle, respectively, using a linear activation function to output continuous values; the second branch outputs equivalent compensation parameter prediction information, containing two nodes corresponding to the strain angle compensation value and torque compensation coefficient, also using a linear activation function. At this point, the bidirectional GRU layer can capture long-term dynamic response characteristics in the wind condition sequence, such as the cumulative change of turbulence intensity over time or the slow drift of wind shear. The multi-head self-attention layer can accurately identify the historical moments and sensor dimensions that have the greatest impact on the current impeller inflow state; for example, when gusts occur, the attention weights will focus on the wind speed and bending moment data at the moment of load abrupt change.

[0032] It should be understood that the training process of the flow field reconstruction compensation model may include an offline pre-training stage and an online adaptive fine-tuning stage. In the offline pre-training stage, a training dataset is constructed using large-scale wind field measured data or high-fidelity simulation data. Specifically, the wind speed distribution at multiple cross-sections in front of the impeller plane obtained by lidar measurement (converted to turbulence intensity, wind shear index, and inflow angle through energy equivalence) is used as the true label. Simultaneously, operating parameters from the nacelle anemometer, blade load sensor, and unit PLC are collected as input samples. During training, Mean Squared Error (MSE) is used as the loss function for the equivalent wind speed prediction branch, and Mean Absolute Error (MAE) is used as the loss function for the compensation parameter prediction branch. The overall loss function is a weighted sum of two terms (with weight coefficients of 0.7 and 0.3, respectively). The optimizer uses the Adam optimizer with an initial learning rate of 0.001, a batch size of 64, and 200 iterations. An early stopping strategy is employed (training stops when the validation set loss does not decrease for 10 consecutive iterations). After training, the optimal model parameters on the validation set are saved as the final model. In the online adaptive fine-tuning stage, after the model is deployed to the wind turbine main control unit, real-time operating data is continuously collected. Using the preset power tracking deviation information (i.e., the difference between theoretical power and actual output power) as a weak supervision signal, the model is fine-tuned online every hour: using the operating data of the most recent hour (about 3600 samples), with cross-entropy loss (which measures the severity of actual power tracking error) as an auxiliary loss, small-batch gradient descent updates are performed on all or part of the model parameters (only updating the parameters of the attention fusion layer and output layer, fixing the coding layer and GRU layer), and the learning rate is reduced to one-tenth of the initial learning rate (0.0001). This allows for rapid adaptation to seasonal changes in wind field conditions or dynamic characteristic drift caused by unit aging, achieving long-term adaptive and accurate perception.

[0033] Based on the equivalent wind speed prediction information, the corresponding equivalent compensation parameter prediction information is determined. That is, the attention mechanism layer can be used to extract the focused feature vector and perform weighted fusion to generate a wind distortion feature representation. Then, a specific feedforward compensation solution layer calculates the pitch angle compensation value and torque compensation coefficient for the current control cycle, quantifies the feedforward intervention amount required due to flow field distortion, and directly maps the abstract equivalent wind condition deviation into an executable feedforward compensation command. This enables feedforward active correction of control commands. Compared with the traditional architecture that only corrects the wind speed value and then the controller responds, this method directly injects the compensation amount from the mechanism, resulting in a faster response speed. The equivalent compensation parameter prediction information determined in this way can accurately offset the imbalance between quasi-static load and dynamic load caused by complex wind shear, inflow angle deviation, etc., so that the flow field distortion is compensated before the control command is applied to the unit, significantly improving the adaptive capability of the control system to the transient wind condition.

[0034] Step S30: Determine the corresponding control reference parameter information based on the equivalent wind speed prediction information; It should be noted that the control reference parameter information is the standard control output calculated to enable the wind turbine to track the optimal power curve under ideal steady-state wind conditions. It includes pitch angle reference parameter information and torque reference parameter information. The pitch angle reference parameter information is the uniform pitch angle command value of the blade calculated based on the deviation between the equivalent wind speed and the rated wind speed, which is used to adjust the wind energy capture efficiency. The torque reference parameter information is the electromagnetic torque command value of the generator calculated based on the relationship between the equivalent wind speed and the optimal tip speed ratio, which is used to control the speed and achieve maximum power tracking. Therefore, the control reference parameter information essentially represents the ideal output of the controller under the assumption of no flow field distortion and is the basic reference for feedforward compensation correction.

[0035] In a specific embodiment, power limitation requirement information is obtained. This information can come from the active power setting upper limit command issued by the system, the unit output upper limit value dynamically calculated within the wind farm to meet grid security or frequency regulation requirements, or the self-protective power reduction command triggered by the unit itself due to fault warning states such as converter overheating, generator overspeed, or excessive vibration of components. It is accessed in the form of a power upper limit value or a power limitation percentage as a prerequisite constraint for the control benchmark calculation. By actively acquiring and introducing power limitation requirement information, the calculation of control benchmark parameters is no longer solely aimed at maximizing power generation, but also has the ability to respond to external grid dispatch and meet the unit's safety boundary. This effectively avoids the risk of unit overload, over-generation, or equipment damage that may result from blindly pursuing the theoretically optimal power curve.

[0036] Based on the power limitation requirement information and the equivalent wind speed prediction information, the control reference parameters are calculated to obtain control reference parameter information, which includes pitch angle reference parameters and torque reference parameters. Specifically, based on the equivalent wind speed prediction information, the theoretically optimal tip speed ratio and wind energy capture coefficient at the current wind speed can be determined by looking up a table. According to the PID control algorithm, the pitch angle reference parameters and torque reference parameters are calculated respectively. For example, when the wind speed is below the rated value, the optimal pitch angle is kept constant, and the torque reference parameter follows the equivalent wind speed to achieve maximum power point tracking. When the wind speed approaches or exceeds the rated value... When setting values, the power limitation requirement information is used as the upper limit of the constraint to calculate the pitch angle reference parameters for stabilizing the rated power. The pitch angle reference parameters are reference values ​​for the uniform pitch angle of the blades. At the same time, the torque reference parameters are adjusted accordingly. The torque reference parameters are the given values ​​of the generator electromagnetic torque. Therefore, the calculated control reference parameter information constitutes a set of ideal steady-state control commands under the assumption of no flow field distortion. It can more accurately characterize the real aerodynamic energy input level of the impeller plane. At the same time, the power limitation requirement is incorporated into the constraint calculation to ensure that the control reference achieves optimal energy capture under the premise of ensuring safety and response scheduling.

[0037] In one feasible implementation, step S30 may include steps A11-A12: Step A11: Obtain power limitation requirement information; It should be noted that the power limitation requirement information is an instruction or parameter obtained in real time to constrain the current maximum allowable output power of the unit. For example, it may be the upper limit of active power set by the system, the single unit output limit dynamically generated by the wind farm according to the frequency regulation requirements or the power allocation strategy at the farm level, or the self-protective power reduction target value triggered by early warning signals such as the unit's own converter temperature exceeding the limit, generator winding overheating, or vibration acceleration of key components exceeding the standard. It is used as a hard constraint condition for the control benchmark calculation in the form of absolute power value or percentage relative to rated power, to ensure that the unit prioritizes meeting the multiple operating boundaries of grid safety, farm-level coordination, and equipment protection while pursuing maximum power tracking.

[0038] Step A12: Based on the power limitation demand information and the equivalent wind speed prediction information, the control reference parameters are calculated to obtain control reference parameter information, which includes pitch angle reference parameter information and torque reference parameter information.

[0039] It is understood that the pitch angle reference parameter information is the uniform pitch angle command value of the blade calculated according to the preset control PID algorithm under ideal steady-state wind conditions. The PID algorithm is a closed-loop automatic control algorithm based on deviation feedback. That is, it generates the control output by linearly combining the proportional, integral and derivative operations of the deviation between the target setpoint and the actual feedback value. The proportional link immediately generates the corrective force according to the current deviation. The larger the deviation, the stronger the adjustment force. The integral link responds to the cumulative amount of deviation over time to eliminate the steady-state residual error that the proportional control cannot eliminate. The derivative link applies damping suppression in advance according to the rate of change of deviation to prevent the system from overshooting and oscillation. In wind turbine generator sets, rated power or optimal tip speed ratio is used as the control target. The pitch angle reference command and torque reference command are calculated in real time based on the measured wind speed or power deviation. When the equivalent wind speed is lower than the rated wind speed, the pitch angle reference parameters maintain the optimal pitch angle to maximize wind energy capture. When the equivalent wind speed approaches or exceeds the rated wind speed, the pitch angle reference parameters dynamically increase to limit aerodynamic torque, ensuring the generator output power remains stable below the rated value or power limit. The torque reference parameters are the generator electromagnetic torque setpoint calculated under the same assumptions. Below the rated wind speed, the torque reference parameters follow the equivalent wind speed according to the optimal tip speed ratio to achieve maximum power tracking. Above the rated wind speed or under power limit constraints, the torque reference parameters are adjusted accordingly to coordinate with the pitch control action to maintain power balance.

[0040] Step S40: Based on the equivalent compensation parameter prediction information, the control reference parameter information is feedforward corrected to determine the control correction parameter information; It should be noted that the control correction parameter information is the actual control command set used to directly drive the actuator of the generator set, including pitch angle correction parameter information and torque correction parameter information. The pitch angle correction parameter information is the final blade unified pitch angle command formed by superimposing the pitch angle compensation value obtained by the compensation coefficient mapping on the pitch angle reference command. The torque correction parameter information is the final generator electromagnetic torque command formed by multiplying the torque reference command by the torque compensation coefficient.

[0041] It is understandable that feedforward correction is an active compensation mechanism that differs from feedback control. Feedback control waits for flow field distortion to cause deviations in power or speed before the controller passively responds and adjusts, resulting in inherent measurement lag and response delay. Feedforward correction, on the other hand, calculates the amount of compensation to be injected in advance based on the real-time perception of the current flow field distortion by a deep learning model before the deviation actually occurs, and directly corrects the control reference command.

[0042] In a specific embodiment, the pitch angle reference parameter information in the control reference parameter information is feedforward corrected based on the pitch angle compensation value in the equivalent compensation parameter prediction information to obtain the pitch angle correction parameter information. That is, the pitch angle compensation value can be extracted from the equivalent compensation parameter prediction information. This pitch angle compensation value is calculated based on the flow field distortion characteristics such as the current wind shear index, inflow angle deviation, and turbulence intensity, through attention-weighted fusion to obtain the unified blade pitch angle increment. This pitch angle compensation value is then superimposed on the pitch angle reference parameter information generated by the PID algorithm. When the flow field... When distortion causes the local wind speed on the impeller plane to be higher than the single-point measurement value, the compensation value is positive to increase the pitch angle in advance to reduce aerodynamic capture, and conversely, the compensation value is negative to release the pitch angle and improve energy absorption. The result after superposition is the pitch angle correction parameter information. By actively superimposing the feedforward compensation value on the reference pitch angle output by the PID, the pitch command includes the correction amount at the moment when the impeller faces wind shear or sudden change in inflow angle. There is no need to wait for power or speed deviation to occur before passive adjustment. It can effectively suppress the unbalanced load of the impeller caused by non-uniform inflow and significantly reduce the fatigue bending moment amplitude in the blade wagging direction.

[0043] Based on the torque compensation coefficient in the equivalent compensation parameter prediction information, the torque reference parameter information in the control reference parameter information is feedforward corrected to obtain the torque correction parameter information. That is, the torque compensation coefficient can be obtained from the equivalent compensation parameter prediction information. For example, when the equivalent wind speed is higher than the single-point measurement value, the coefficient is greater than 1 to increase the torque setpoint and capture more energy. When the equivalent wind speed is lower than the single-point measurement value, the coefficient is less than 1 to reduce the torque setpoint and avoid overload. Thus, the torque compensation coefficient is multiplied by the torque reference parameter information, that is, the torque correction parameter information is equal to the torque reference parameter information multiplied by the torque compensation coefficient. This completes the feedforward correction of the generator electromagnetic torque setpoint value, so that the generator torque command can follow the real equivalent wind speed change of the impeller plane in real time, rather than the single-point measurement value in the severely distorted wake region. This effectively avoids the low torque setpoint and power generation loss caused by the underestimation of wind speed, and prevents torque overshoot and transmission chain impact caused by the overestimation of wind speed. While improving energy capture efficiency, it significantly reduces the alternating torque amplitude of transmission components such as gearbox and main shaft.

[0044] Based on the pitch angle correction parameter information and the torque correction parameter information, control correction parameter information is obtained. That is, control correction parameter information can be obtained from the pitch angle correction parameter information and the torque correction parameter information. The control correction parameter information may include fields such as timestamp, corrected pitch angle target angle, corrected torque setpoint, and check code. It is directly written into the corresponding register of the main control PLC through the high-speed real-time bus, overwriting the original reference control instruction, and serving as the final execution instruction for the pitch actuator and converter in the current control cycle. This ensures strict synchronization and coordination of the two control loops in terms of timing, avoids the accumulation of control delay caused by multi-level calculations, and ensures that the feedforward compensated instruction can be issued to the physical actuator within a single control cycle, achieving dynamic optimal matching between wind turbine capture efficiency and generator output power.

[0045] In one feasible implementation, step S40 may include steps B11-B13: Step B11: Based on the pitch angle compensation value in the equivalent compensation parameter prediction information, the pitch angle reference parameter information in the control reference parameter information is feedforward corrected to obtain the pitch angle correction parameter information. It should be noted that the pitch angle correction parameter information is used to directly drive the final blade of the pitch actuator to uniformly control the pitch angle. For example, when the model senses that the impeller plane is abnormally subjected to local aerodynamic loads due to wind shear or inflow angle deviation, the pitch angle correction parameter information has been used to incrementally correct the reference pitch angle in advance, so that the blades can complete the angle adjustment before the flow field distortion effect actually acts on the unit, thus changing from passive response to active compensation.

[0046] Step B12: Based on the torque compensation coefficient in the equivalent compensation parameter prediction information, the torque reference parameter information in the control reference parameter information is feedforward corrected to obtain torque correction parameter information; It should be noted that the torque correction parameter information is the final set value used to directly control the electromagnetic torque of the generator. For example, when the equivalent wind speed is higher than the distortion measurement value in the wake region of the nacelle, the torque correction parameter information is correspondingly greater than the original reference torque, enabling the generator to capture power that matches the actual incoming energy. When the equivalent wind speed is lower than the distortion measurement value, the torque correction parameter information is correspondingly less than the original reference torque, preventing the drive train from being subjected to excessive mechanical torque impact.

[0047] Step B13: Based on the pitch angle correction parameter information and the torque correction parameter information, obtain the control correction parameter information.

[0048] It is understood that the control correction parameter information can be directly written into the control bus and sent to the pitch actuator and converter, which is the final control quantity actually executed by the unit.

[0049] Step S50: Control the wind turbine generator set to monitor the wind field based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information.

[0050] It is understood that the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information can directly control the operation of the unit, so that the actual operating state of the unit can closely approximate the theoretical optimal power curve, realize the accurate response to the real equivalent wind conditions of the impeller plane in complex wind field environment, and effectively reduce the accumulation of fatigue load on key components while maximizing power generation benefits.

[0051] In a specific embodiment, the target compensation angle information is determined based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information. That is, based on the difference between the pitch angle correction parameter information and the current actual pitch angle, the target compensation angle information required for the rotation of each blade pitch motor is calculated by the PID control algorithm. In other words, the precise rotation increment required for each blade to reach the target angle position required by the correction command from the current angle position is directly sent to the driver of each pitch motor in the form of pulse signal or analog signal, realizing the precise conversion from digital command to physical action, ensuring that the corrected unified pitch command can be executed quickly and accurately, and eliminating the accumulation of errors in the command transmission process.

[0052] Theoretical power information is obtained; based on the target compensation angle information, the corresponding actual output power is monitored to determine the actual output power monitoring information. That is, the theoretical power information represents the optimal electrical power that the unit should theoretically output under the current inflow conditions. At the same time, the active power output by the generator is collected in real time as the actual output power monitoring information. It is continuously recorded in the form of a time sequence and timestamped for comparison with the theoretical power. The theoretical power provides an ideal benchmark for evaluating the feedforward compensation effect, while the actual output power truly reflects the unit response after the compensation is executed.

[0053] Based on the comparison between the theoretical power information and the actual output power monitoring information, the power tracking deviation information is determined. Based on the power tracking deviation information, the corresponding control command is determined, that is, the theoretical power information and the actual output power monitoring information at the same time are subtracted point by point to generate the power tracking deviation information. When the power tracking deviation information shows that the deviation continues to exceed the preset threshold, the feedforward compensation coefficient of the next control cycle is fine-tuned according to the direction and magnitude of the deviation, and the updated control correction parameters are generated as the new control command. If the deviation exceeds the effective compensation range of the model, the online fine-tuning module is triggered, and the parameters of the deep learning model are updated by rapid gradient descent using the recently collected running data, so that the model can readjust to the current wind conditions.

[0054] Based on the control commands, the wind turbine generator is controlled to monitor the wind farm. That is, the control commands are written into the actuator through the bus to control the operation of the generator and continuously monitor the power tracking effect. This enables the generator to autonomously optimize the control strategy when the wind farm conditions change over a long period of time, maintain high-precision power tracking in the short term, and achieve long-term adaptive capability to slowly changing factors such as seasonal changes and generator aging. This ensures the dual goals of maximizing power generation revenue and ensuring structural safety in complex wind farm environments.

[0055] In one feasible implementation, step S50 may include steps C11-C13: Step C11: Determine the target compensation angle information based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information; It should be noted that the target compensation angle information is the precise angular increment value required for each blade pitch motor to rotate in order to achieve the corrected pitch angle target. It is expressed in the form of a physical quantity that can be recognized by each blade pitch motor driver, and directly corresponds to the target position difference required by the motor encoder, ensuring that the corrected unified pitch command can be executed independently and accurately by each blade.

[0056] Step C12: Determine the corresponding control command based on the target compensation angle information; It should be noted that the control commands are a set of execution commands that can be directly sent to the pitch motor driver and converter control unit via the real-time bus. They include the target position setting value of each blade pitch angle and the given value of the generator electromagnetic torque. They are encapsulated into standard control data frames according to the industrial real-time bus protocol, so that complete fields such as target value, execution rate limit, and enable status can be obtained.

[0057] In one feasible implementation, step C12 may include steps D11 to D14: Step D11: Obtain theoretical power information; It should be noted that the theoretical power information represents the maximum active power that the unit can theoretically output under the current inflow conditions. It is generated in real time by a pre-calibrated power curve lookup table or analytical function and serves as an ideal benchmark for measuring the effect of compensation control. It does not consider the dynamic response error and actual loss of the actuator and represents the upper limit of the unit's optimal power generation capacity under perfect control conditions.

[0058] Step D12: Based on the target compensation angle information, monitor the corresponding actual output power to determine the actual output power monitoring information; It should be noted that the actual output power monitoring information is the sequence value of the active power actually output by the generator, which includes the actual power fluctuations caused by factors such as sensor accuracy limitations, converter efficiency losses and unit dynamic response characteristics. It is continuously recorded in a time-series format with timestamps to objectively reflect the actual power generation effect of the unit after the execution of compensation control commands.

[0059] Step D13: Based on the comparison between the theoretical power information and the actual output power monitoring information, determine the power tracking deviation information; It should be noted that the power tracking deviation information is a set of error characteristics obtained to measure the effect of feedforward compensation control, including the absolute power difference at each point, the average deviation within a specific time window, the standard deviation, and the Pearson correlation coefficient between the theoretical power curve and the actual power curve.

[0060] Step D14: Determine the corresponding control command based on the power tracking deviation information.

[0061] Understandably, when the power tracking deviation information shows that the actual power continuously deviates from the theoretical power and exceeds the allowable threshold, the system automatically determines the control commands to be adjusted in the next control cycle based on the magnitude, direction, and duration of the deviation. For example, when the deviation is small, the feedforward compensation coefficient is fine-tuned directly according to the deviation direction to generate updated pitch angle and torque correction commands. When the deviation continues to expand beyond the effective compensation range of the model, the online fine-tuning mechanism of the deep learning model is triggered. The model parameters are updated quickly by gradient descent using the recently collected operating data, so that the model relearns the optimal mapping relationship under the current wind conditions, thereby outputting control commands that adapt to the new operating conditions.

[0062] Step C13: Control the wind turbine generator set to monitor the wind field based on the control command.

[0063] It is understood that the control commands can be continuously sent to the pitch actuator and converter via a high-speed real-time bus to drive the blades and generator to operate according to the corrected commands, thereby achieving a precise response to the actual equivalent wind conditions of the impeller plane in complex wind field environments, and effectively reducing the accumulation of fatigue loads on key components while maximizing power generation benefits.

[0064] This embodiment proposes a wind farm monitoring method for wind turbine generator sets, which acquires wind speed information, blade load information, and operating parameter information of the wind turbine generator set; predicts the equivalent wind speed and corresponding equivalent compensation parameter prediction information based on the wind speed, blade load, and operating parameter information; determines the corresponding control reference parameter information based on the equivalent wind speed prediction information; performs feedforward correction on the control reference parameter information based on the equivalent compensation parameter prediction information to determine control correction parameter information; and controls the wind turbine generator set to monitor the wind farm based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information. This invention addresses the technical challenge of more accurately monitoring wind fields under complex wind conditions. Compared to existing technologies, this application acquires wind speed, blade load, and operating parameter information from wind turbine generators. This allows for the perception of inflow distortion and generator status, eliminating reliance on single-point data. It dynamically outputs equivalent wind speed prediction information and corresponding equivalent compensation parameter prediction information to compensate for flow field distortion and accurately characterize the true inflow wind speed at the impeller plane. This enables the calculation of control reference parameters, including pitch angle and torque, matching the control target with the actual wind conditions. The equivalent compensation parameter prediction information is used to feedforward correct the control reference parameters, proactively injecting compensation commands before the controller operates. This overcomes measurement lag and load mismatch. Based on the corrected pitch angle and torque correction parameters, the generator is controlled to monitor the wind field, achieving high-precision power tracking and structural safety assurance.

[0065] Based on the first embodiment of this application, in the second embodiment of this application, the same or similar content as the first embodiment described above can be referred to the above description, and will not be repeated hereafter.

[0066] In this embodiment, refer to Figure 2 , Figure 2 This is a flowchart illustrating Embodiment 2 of the wind farm monitoring method for wind turbine generators in this application. Step S20 specifically includes steps S21 to S23: Step S21: Based on the wind speed information, the blade load information and the operating parameter information, perform data preprocessing to determine preprocessing information. The data preprocessing includes filtering and noise reduction, normalization and time-series fusion. It should be noted that the preprocessed information is a structured multidimensional temporal feature matrix that can be directly input into the deep flow field reconstruction and compensation model. It is strictly aligned in the time dimension, and in the spatial dimension, it covers data sequences from multiple channels such as single-point wind speed in the nacelle, radial wind speed of lidar, blade root flapping and oscillation bending moment, generator speed, blade pitch angle and output power. In the numerical dimension, it is mapped to a unified numerical range to eliminate the influence of dimensional differences on model calculation.

[0067] It is understood that the data preprocessing refers to a series of normalization operations performed on the original multi-source heterogeneous data to ensure the quality, consistency, and usability of the input data for the deep learning model. This preprocessing may include filtering and denoising, normalization, and time-series fusion. Specifically, the filtering and denoising process employs digital signal processing techniques such as median filtering and low-pass filtering to remove and smooth high-frequency electromagnetic interference spikes superimposed on the original wind speed signal and pulse outliers caused by instantaneous sensor disconnections in the blade load signal. Median filtering replaces abnormal jump points by taking the median within a sliding window, while low-pass filtering filters out random noise above a preset cutoff frequency by retaining low-frequency effective components, thereby improving the signal-to-noise ratio. The normalization process... The processing refers to the process of eliminating the bias effect caused by the significant differences in the dimensions and numerical ranges of different physical quantities such as wind speed, load, speed, and power on the training of deep learning models. The filtered data of each dimension are linearly mapped according to their statistical characteristics. The minimum-maximum normalization method is usually used to uniformly scale all features to a similar numerical range. The time series fusion processing is the process of strictly aligning the data sequences of each independent sensor within a preset time window according to a unified time index and splicing them into a multi-dimensional time series feature tensor at a fixed time step after filtering, denoising and normalization. This ensures the correct correspondence between the multi-source data in terms of time causality, so that the model can learn the time series dependency between wind condition changes and the dynamic response of the unit.

[0068] In a specific embodiment, the original timestamps of the wind speed information, blade load information, and operating parameter information can be acquired synchronously and uniformly resampled to a preset reference frequency. For example, the wind speed information, blade load information, and operating parameter information can be filtered using a combination of median filtering and low-pass filtering. Then, the filtered data for each dimension can be processed by maximum and minimum normalization to eliminate differences in physical dimensions. The multi-dimensional sequences of single-point wind speed, radial wind speed, flapping and swaying bending moment, generator speed, and blade pitch angle from each independent sensor within the past preset time window can be timed. The time steps are strictly aligned and spliced ​​into a multi-dimensional temporal feature matrix, which serves as preprocessing information that can be directly input into the deep flow field reconstruction and compensation model. This effectively suppresses random noise and anomalous jumps in the original sensor signals, significantly improves the signal-to-noise ratio and authenticity of the input data, and eliminates numerical imbalances between different dimensions such as wind speed, load, and rotational speed. This allows the deep learning model to converge faster and be more numerically stable during training and inference, thereby ensuring the causal alignment of multi-source data in the time dimension and the integrity of the spatial structure. This enables the model to accurately perceive the temporal dependency between wind condition changes and the dynamic response of the unit.

[0069] Step S22: Based on the preprocessed information, input a predefined flow field reconstruction compensation model to predict the equivalent wind speed in the impeller plane and determine the equivalent wind speed prediction information; It is understood that the flow field reconstruction compensation model is used to learn the spatiotemporal dynamic mapping relationship between the distorted single-point measurement value at the rear of the nacelle and the true equivalent wind speed on the impeller plane. It captures the long-term dependence and dynamic response characteristics in the wind condition sequence by using stacked GRU layers, and automatically focuses on the most critical features for the prediction target at different times and data dimensions by introducing a multi-head attention mechanism.

[0070] In a specific embodiment, at least one of the single-point wind speed and radial wind speed in the preprocessed information is analyzed to obtain a first analysis result. That is, the single-point wind speed and radial wind speed can be extracted from the preprocessed information and feature encoded to extract the temporal fluctuation characteristics of the single-point wind speed and the spatial gradient characteristics of the radial wind speed along different distance gates. The two are then fused into the first analysis result. When only the single-point wind speed of the nacelle is input, the model only extracts the temporal features of that channel. Through the hierarchical analysis of the single-point wind speed and radial wind speed, the model can simultaneously perceive the disturbance characteristics of the wake region of the nacelle and the spatial distribution characteristics of the inflow wind field in front of the impeller.

[0071] At least one of the swaying moment load and the oscillation moment load in the preprocessed information is analyzed to obtain a second analysis result. That is, the swaying moment load and the oscillation moment load can be extracted from the preprocessed information, and dynamic feature encoding is performed through temporal convolution operation to extract the periodic fluctuation pattern of the load with the azimuth angle of the impeller rotation, the amplitude characteristics of the abrupt change, and the asymmetry of the load distribution between each blade, forming a second analysis result. This second analysis result characterizes the non-uniform aerodynamic load distribution of the inflow wind on the impeller sweep surface. Through in-depth analysis of the blade swaying moment and oscillation moment, the model can obtain the aerodynamic load perception capability of the impeller plane, and can realize accurate reconstruction of the equivalent wind speed under non-uniform inflow conditions.

[0072] At least one of the generator speed parameters, pitch angle parameters, and output power parameters in the preprocessed information is analyzed to obtain a third analysis result. That is, the generator speed parameters, pitch angle parameters, and output power parameters can be extracted from the preprocessed information, and feature mapping is performed to extract the dynamic change rate of generator speed, the correspondence between pitch angle and output power, and the position characteristics of the current operating point of the unit on the power curve. The third analysis result is formed to characterize the current operating state and control response characteristics of the unit. By analyzing the unit's operating state parameters, the model can distinguish the differences in the dynamic response characteristics of the unit under different operating conditions, avoid the equivalent wind speed prediction error caused by the switching of control strategies between the low wind speed maximum power tracking area and the high wind speed constant power area, and significantly improve the prediction robustness of the model across the entire wind speed range.

[0073] Historical wind field monitoring information is acquired; feature tensor information is determined based on the first, second, and third analysis results; based on the historical wind field monitoring information and the feature tensor information, a predefined flow field reconstruction compensation model is input to predict the equivalent wind speed in the impeller plane, obtaining equivalent wind speed prediction information. The equivalent wind speed prediction information includes equivalent prediction information for turbulence intensity, wind shear index, and inflow angle. This allows for the acquisition of historical wind field monitoring information recorded within a preset time period, which may include the sequence of equivalent wind speed prediction values ​​output by the model in the preceding control cycle and the corresponding confidence level markers. The first, second, and third analysis results are concatenated along the feature dimensions and stacked according to time steps to form a three-dimensional feature tensor information. The three dimensions of the feature tensor information represent the time step, data channel, and feature mapping, respectively. The feature tensor information and historical wind field monitoring information are input into a predefined flow field reconstruction compensation model. The flow field reconstruction compensation model is trained. For example, the stacked encoding layer of the flow field reconstruction compensation model encodes the temporal dependencies in the feature tensor, capturing the long-term and short-term dynamic features of wind condition evolution and unit response. The multi-head attention mechanism calculates attention weights at different time steps and different data channels, automatically focusing on the most critical historical moments and sensor dimensions for predicting the current impeller plane equivalent wind speed. Thus, multi-dimensional equivalent wind speed prediction information is generated through regression of the fully connected output layer. The equivalent wind speed prediction information includes turbulence intensity equivalent prediction information, wind shear index equivalent prediction information, and inflow angle equivalent prediction information. Among them, the turbulence intensity equivalent prediction information is obtained by statistical analysis of the high-frequency fluctuation components of the wind speed sequence, the wind shear index equivalent prediction information is obtained by gradient analysis of radial wind speed at different heights, and the inflow angle equivalent prediction information is calculated by the ratio of the lateral wind speed component to the axial wind speed component. Thus, it is possible to comprehensively perceive the type and degree of flow field distortion that will affect the unit.

[0074] In one feasible implementation, step S22 may include steps E11 to E14: Step E11: Analyze at least one of the single-point wind speed and radial wind speed in the preprocessed information to obtain the first analysis result; It should be noted that the first analytical result is a low-dimensional feature vector used to characterize the spatiotemporal distribution characteristics of the inflow wind field. It integrates the temporal fluctuation characteristics of the wind speed at a single point at the tail of the nacelle and the spatial gradient characteristics of the radial wind speed of the lidar along different distance gates, and encodes the pulsation intensity of the inflow wind in the time dimension and the velocity distribution trend in the spatial dimension in the form of a numerical vector.

[0075] Understandably, single-point wind speed is the wind speed and direction measurement value collected by a mechanical or ultrasonic anemometer installed at the tail of the nacelle in the downstream wake region of the impeller at a single spatial location, while radial wind speed refers to the average wind speed component along the laser line of sight at multiple preset distance gate positions in front of the impeller, obtained by a lidar deployed on the top of the nacelle in a fixed beam or simple scanning mode.

[0076] Step E12: Analyze at least one of the swing moment load and the oscillation moment load in the preprocessed information to obtain a second analysis result; It should be noted that the second analytical result is a feature vector used to indirectly characterize the non-uniform aerodynamic load distribution characteristics on the impeller rotation plane. It can encode the periodic fluctuation pattern of the swing moment with the impeller rotation azimuth angle, the abrupt change amplitude, and the asymmetry of the load distribution between blades.

[0077] Understandably, the flapping moment load is acquired in real time by fiber Bragg grating sensors or strain gauge sensors installed at the blade root or key cross sections in the spanwise direction. The measured value of the bending moment of the blade in the direction perpendicular to the plane of rotation characterizes the aerodynamic components acting on each cross section of the blade in the spanwise direction perpendicular to the rotor's plane of rotation. It is highly sensitive to the inflow velocity non-uniformity caused by vertical wind shear and tower shadow effect. The flapping moment load is the measured value of the bending moment of the blade in the direction in the plane of rotation acquired by the same sensor. It characterizes the aerodynamic components acting on each cross section of the blade in the spanwise direction parallel to the rotor's plane of rotation. It is sensitive to the changes in lateral aerodynamic load caused by inflow angle deviation and horizontal crosswind.

[0078] Step E13: Analyze at least one of the generator speed parameters, pitch angle parameters, and output power parameters in the preprocessed information to obtain a third analysis result; It should be noted that the third analytical result is a feature vector used to characterize the current operating conditions and control response characteristics of the unit. It can encode the dynamic rate of change of generator speed, the correspondence between pitch angle and output power, and the operating condition range of the unit on the power curve.

[0079] Understandably, the generator speed parameter characterizes the mechanical response state after the aerodynamic torque of the impeller and the electromagnetic torque of the generator are balanced. Its dynamic change rate is directly related to the fluctuation of the inflow wind speed. The blade pitch angle parameter is the commanded value or actual feedback value of the blade pitch angle currently acting on the blade unified pitch mechanism. It characterizes the degree of intervention of the control system in wind energy capture efficiency. In low wind speed areas, it usually keeps the optimal value unchanged, while in high wind speed areas, it dynamically increases to limit power. The output power parameter is the actual active power value generated by the unit in real time through the sensor at the output end of the converter. It characterizes the energy capture effect, that is, the actual execution effect of the control strategy.

[0080] Step E14: Based on the first analysis result, the second analysis result, and the third analysis result, a predefined flow field reconstruction compensation model is input to predict the equivalent wind speed in the impeller plane, thereby obtaining the equivalent wind speed prediction information.

[0081] It is understandable that the impeller plane equivalent wind speed is the non-uniform and unsteady inflow wind speed distribution at various spatial positions on the entire rotating impeller sweep surface of the wind turbine generator set, converted into a unified average effective wind speed value through the principle of energy equivalence. Unlike the single-point distorted wind speed measured by the single-point anemometer at the tail of the traditional nacelle under wake interference, the impeller plane equivalent wind speed represents the total effect of the true inflow velocity corresponding to the aerodynamic load borne by each section of the blade at different azimuth angle positions.

[0082] In one feasible implementation, step E14 may include steps F11 to F13: Step F11: Obtain historical wind field monitoring information; It should be noted that the historical wind field monitoring information is the equivalent wind speed prediction sequence output by the flow field reconstruction compensation model in the preceding control cycle within a preset time period in the past, along with the corresponding model prediction confidence marker and the actual operating status record of the unit during that period, i.e., the time evolution trajectory of the wind field status, which enables the model to maintain the temporal consistency and physical continuity of the prediction results in continuous control.

[0083] Step F12: Determine the feature tensor information based on the first parsing result, the second parsing result, and the third parsing result; It should be noted that the feature tensor information is a three-dimensional structured numerical array formed by concatenating the first, second, and third analytical results along the feature dimension and stacking them at a unified time step. The three dimensions of the feature tensor information correspond to the time step sequence, multi-source data channels, and feature mapping vector, respectively. This integrates the originally dispersed and heterogeneous inflow wind field features, blade aerodynamic load response features, and unit operating status features into a compact joint representation, which serves as the standard input format for the flow field reconstruction compensation model.

[0084] Step F13: Based on the historical wind field monitoring information and the feature tensor information, input the predefined flow field reconstruction compensation model to predict the equivalent wind speed in the impeller plane, and obtain the equivalent wind speed prediction information. The equivalent wind speed prediction information includes the equivalent prediction information of turbulence intensity, the equivalent prediction information of wind shear index, and the equivalent prediction information of inflow angle.

[0085] It is understandable that the turbulence intensity equivalent prediction information is an equivalent parameter used to characterize the severity of the inflow wind speed fluctuations on the impeller swept surface, and is directly related to the amplitude of the alternating aerodynamic load borne by the blades. The wind shear index equivalent prediction information is an equivalent index value used to characterize the variation law of the inflow wind speed along the vertical height direction, and characterizes the severity of the wind speed difference between the upper and lower halves of the impeller plane. The inflow angle equivalent prediction information is an equivalent value used to characterize the deviation angle between the inflow wind direction and the impeller rotation axis. When the inflow angle deviates significantly from zero degrees, it indicates that the unit is being subjected to the continuous effect of lateral yaw wind conditions.

[0086] Step S23: Determine the corresponding equivalent compensation parameter prediction information based on the equivalent wind speed prediction information.

[0087] It is understandable that by using the equivalent compensation parameter prediction information, the amount of correction that the control command needs to be superimposed can be calculated and fed forward correction can be performed before the deviation actually occurs, so that the unit can complete the pre-compensation action before the flow field distortion causes the power deviation.

[0088] In a specific embodiment, implicit temporal features within a preset time period are extracted based on the equivalent wind speed prediction information to determine the implicit temporal feature sequence. That is, based on the equivalent wind speed prediction information, all implicit state vectors output at each time step within the past preset time window are retained and arranged sequentially to obtain the implicit temporal feature sequence. Each time step vector in the implicit temporal feature sequence encodes the accumulated wind condition evolution features, unit dynamic response status, and interactive coupling information between the two from the beginning of the sequence to the current moment. This maximizes the preservation of the evolution process information of wind field distortion features in the time dimension, enabling the model to trace back the different impacts of different historical moments on the current compensation requirements and avoid the loss of dynamic details caused by information compression.

[0089] By employing a pre-defined multi-head attention strategy, the focused feature vectors corresponding to the hidden temporal feature sequence are weighted and fused to determine the focused feature vector fusion information. Specifically, after extracting the hidden temporal feature sequence, a multi-head attention mechanism is used to independently calculate the attention weights of multiple parallel attention heads from different feature subspaces. Each head automatically assesses the importance of each historical moment in the sequence to the current feedforward compensation requirement by querying the similarity score between the query vector and the key vector. The weighted feature vectors output by each head are concatenated and fused through a linear transformation to obtain the focused feature vector fusion information. This effectively improves the model's analytical ability for complex wind field nonlinear temporal dependencies and significantly enhances the accuracy and robustness of feedforward compensation parameter prediction.

[0090] Based on the focused feature vector fusion information, the feedforward compensation coefficient of the preset control cycle is calculated to obtain the equivalent compensation parameter prediction information. That is, the feedforward compensation coefficient of the preset control cycle is calculated according to the focused feature vector fusion information. For example, the pitch angle compensation value required for the current control cycle is calculated, or the torque compensation coefficient is calculated. In this way, the equivalent compensation parameter prediction information can be obtained to offset the comprehensive impact of flow field distortion and achieve seamless connection from sensing to compensation.

[0091] In one feasible implementation, step S23 may include steps G11~G13: Step G11: Based on the equivalent wind speed prediction information, extract the implicit temporal features within a preset time period to determine the implicit temporal feature sequence. It should be noted that the implicit time-series feature sequence is a sequence of implicit state vectors output at each time step within a preset time window arranged in chronological order. Each time step vector is a high-dimensional abstract encoding of the accumulated wind condition evolution features, unit dynamic response status, and the interaction and coupling relationship between the two from the beginning of the sequence to the current moment.

[0092] Understandably, the preset time is the length of the historical time window that the system pre-sets for extracting hidden features, which is usually a time span of several seconds in the past.

[0093] Step G12: Use a preset multi-head attention strategy to perform weighted fusion of the focused feature vectors corresponding to the hidden temporal feature sequence to determine the focused feature vector fusion information; It should be noted that the focused feature vector fusion information is obtained by aggregating useful information that was originally scattered across multiple time steps into a compact fusion representation.

[0094] It is understandable that weighted fusion is an operation in which each attention head in the multi-head attention mechanism assigns different attention weights to the latent state vectors at each historical moment based on the similarity score between the query vector and the key vector, and performs a weighted summation of the vectors at each moment according to the weight. Among them, the key moments that have a great impact on the compensation needs are given higher weights, and their corresponding feature vectors dominate in the fusion result, while the weights of moments with little impact approach zero, and their contributions are effectively suppressed.

[0095] Step G13: Based on the focused feature vector fusion information, the feedforward compensation coefficient of the preset control cycle is calculated to obtain the equivalent compensation parameter prediction information.

[0096] It is understandable that the feedforward compensation coefficient of the preset control cycle is for the pitch angle compensation value and torque compensation coefficient required in the current single control cycle. The pitch angle compensation value is a pitch angle increment with a positive or negative sign, which is directly superimposed on the pitch angle reference command output by the PID controller. The torque compensation coefficient is a dimensionless multiplicative factor with a scaling center of 1.0, which is directly multiplied by the torque reference command to scale the generator electromagnetic torque setpoint. The feedforward compensation coefficient is recalculated in each control cycle to ensure that the compensation action follows the dynamic changes of the wind field in real time.

[0097] This embodiment proposes a wind farm monitoring method for wind turbine generators. Based on the wind speed information, blade load information, and operating parameter information, data preprocessing is performed to determine preprocessed information. The data preprocessing includes filtering and denoising, normalization, and time-series fusion. Based on the preprocessed information, a predefined flow field reconstruction compensation model is input to predict the equivalent wind speed in the impeller plane, determining the equivalent wind speed prediction information. Based on the equivalent wind speed prediction information, the corresponding equivalent compensation parameter prediction information is determined. This invention addresses the technical challenge of more accurately monitoring wind fields under complex wind conditions. Compared to existing technologies, this application preprocesses wind speed, blade load, and operating parameter information to detect inflow distortion and unit status, eliminating reliance on single-point distortion measurements at the nacelle tail. The preprocessed information is then input into a flow field reconstruction compensation model to predict the equivalent wind speed at the impeller plane, thus compensating for flow field distortion and accurately representing the true inflow wind speed at the impeller plane. This allows for the calculation of the equivalent compensation parameter prediction information, which in turn feeds forward to correct the control reference parameters. Compensation commands are proactively injected before the controller operates, achieving the dual goals of high-precision power tracking and structural safety assurance in complex wind field environments.

[0098] This application also provides a wind farm monitoring system for wind turbine generator sets; please refer to [reference needed]. Figure 3 The wind farm monitoring system for the wind turbine generator set includes: The acquisition module 10 is used to acquire wind speed information, blade load information and operating parameter information of the wind turbine generator set; Processing module 20 is used to make predictions based on the wind speed information, the blade load information and the operating parameter information, and determine the equivalent wind speed prediction information and the corresponding equivalent compensation parameter prediction information. The processing module 20 is also used to determine the corresponding control reference parameter information based on the equivalent wind speed prediction information; Execution module 30 is used to perform feedforward correction on the control reference parameter information based on the equivalent compensation parameter prediction information, and determine the control correction parameter information; The execution module 30 is also used to control the wind turbine generator set to monitor the wind field based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information.

[0099] The processing module 20 is also used to perform data preprocessing based on the wind speed information, the blade load information and the operating parameter information to determine preprocessing information. The data preprocessing includes filtering and noise reduction processing, normalization processing and time series fusion processing. Based on the preprocessed information, a predefined flow field reconstruction compensation model is input to predict the equivalent wind speed in the impeller plane, and the equivalent wind speed prediction information is determined. Based on the equivalent wind speed prediction information, the corresponding equivalent compensation parameter prediction information is determined.

[0100] The processing module 20 is further configured to analyze at least one of the single-point wind speed and radial wind speed in the preprocessed information to obtain a first analysis result; At least one of the pendulum moment load and the oscillation moment load in the preprocessed information is analyzed to obtain a second analysis result; At least one of the generator speed parameters, pitch angle parameters, and output power parameters in the preprocessed information is analyzed to obtain a third analysis result; Based on the first analysis result, the second analysis result, and the third analysis result, a predefined flow field reconstruction compensation model is input to predict the equivalent wind speed in the impeller plane, thereby obtaining the equivalent wind speed prediction information.

[0101] The processing module 20 is also used to acquire historical wind field monitoring information; The feature tensor information is determined based on the first analysis result, the second analysis result, and the third analysis result; Based on the historical wind field monitoring information and the feature tensor information, a predefined flow field reconstruction compensation model is input to predict the equivalent wind speed in the impeller plane, thereby obtaining equivalent wind speed prediction information. The equivalent wind speed prediction information includes equivalent prediction information of turbulence intensity, equivalent prediction information of wind shear index, and equivalent prediction information of inflow angle.

[0102] The processing module 20 is also used to extract implicit temporal features within a preset time period based on the equivalent wind speed prediction information, and determine the implicit temporal feature sequence. A pre-defined multi-head attention strategy is used to weight and fuse the focused feature vectors corresponding to the implicit temporal feature sequence to determine the focused feature vector fusion information. Based on the focused feature vector fusion information, the feedforward compensation coefficient of the preset control cycle is calculated to obtain the equivalent compensation parameter prediction information.

[0103] The processing module 20 is also used to obtain power limitation requirement information; Based on the power limitation demand information and the equivalent wind speed prediction information, the control reference parameters are calculated to obtain control reference parameter information, which includes pitch angle reference parameter information and torque reference parameter information.

[0104] The execution module 30 is further configured to perform feedforward correction on the pitch angle reference parameter information in the control reference parameter information based on the pitch angle compensation value in the equivalent compensation parameter prediction information, so as to obtain pitch angle correction parameter information. Based on the torque compensation coefficient in the equivalent compensation parameter prediction information, the torque reference parameter information in the control reference parameter information is feedforward corrected to obtain the torque correction parameter information. Based on the pitch angle correction parameter information and the torque correction parameter information, control correction parameter information is obtained.

[0105] The execution module 30 is also used to determine the target compensation angle information based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information; The corresponding control command is determined based on the target compensation angle information; The wind turbine generator is controlled to monitor the wind farm based on the control commands.

[0106] The execution module 30 is also used to acquire theoretical power information; Based on the target compensation angle information, the corresponding actual output power is monitored to determine the actual output power monitoring information; Based on the comparison between the theoretical power information and the actual output power monitoring information, the power tracking deviation information is determined. The corresponding control command is determined based on the power tracking deviation information.

[0107] The wind turbine wind farm monitoring system provided in this application, employing the wind turbine wind farm monitoring method described in the above embodiments, can solve the technical problem of how to more accurately monitor the wind farm under complex wind conditions. Compared with the prior art, the beneficial effects of the wind turbine wind farm monitoring system provided in this application are the same as those of the wind turbine wind farm monitoring method provided in the above embodiments, and other technical features of the wind turbine wind farm monitoring system are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0108] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for monitoring wind farms of wind turbine generator sets, characterized in that, The method includes: To acquire wind speed information, blade load information, and operating parameter information of wind turbine generators; Based on the wind speed information, the blade load information, and the operating parameter information, predictions are made to determine the equivalent wind speed prediction information and the corresponding equivalent compensation parameter prediction information. Based on the equivalent wind speed prediction information, the corresponding control reference parameter information is determined; Based on the equivalent compensation parameter prediction information, the control reference parameter information is feedforward corrected to determine the control correction parameter information; The wind turbine generator is controlled to monitor the wind farm based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information.

2. The method as described in claim 1, characterized in that, The step of predicting and determining the equivalent wind speed prediction information and the corresponding equivalent compensation parameter prediction information based on the wind speed information, the blade load information, and the operating parameter information includes: Based on the wind speed information, the blade load information and the operating parameter information, data preprocessing is performed to determine preprocessing information. The data preprocessing includes filtering and noise reduction, normalization and time-series fusion. Based on the preprocessed information, a predefined flow field reconstruction compensation model is input to predict the equivalent wind speed in the impeller plane, and the equivalent wind speed prediction information is determined. Based on the equivalent wind speed prediction information, the corresponding equivalent compensation parameter prediction information is determined.

3. The method as described in claim 2, characterized in that, The step of predicting the equivalent wind speed in the impeller plane based on the pre-defined flow field reconstruction compensation model inputted with the pre-processed information, and determining the equivalent wind speed prediction information, includes: At least one of the single-point wind speed and radial wind speed in the preprocessed information is analyzed to obtain a first analysis result; At least one of the pendulum moment load and the oscillation moment load in the preprocessed information is analyzed to obtain a second analysis result; At least one of the generator speed parameters, pitch angle parameters, and output power parameters in the preprocessed information is analyzed to obtain a third analysis result; Based on the first analysis result, the second analysis result, and the third analysis result, a predefined flow field reconstruction compensation model is input to predict the equivalent wind speed in the impeller plane, thereby obtaining the equivalent wind speed prediction information.

4. The method as described in claim 3, characterized in that, The step of predicting the equivalent wind speed in the impeller plane based on the first analysis result, the second analysis result, and the third analysis result, and obtaining the equivalent wind speed prediction information, includes: Obtain historical wind field monitoring information; The feature tensor information is determined based on the first analysis result, the second analysis result, and the third analysis result; Based on the historical wind field monitoring information and the feature tensor information, a predefined flow field reconstruction compensation model is input to predict the equivalent wind speed in the impeller plane, thereby obtaining equivalent wind speed prediction information. The equivalent wind speed prediction information includes equivalent prediction information of turbulence intensity, equivalent prediction information of wind shear index, and equivalent prediction information of inflow angle.

5. The method as described in claim 2, characterized in that, The step of determining the corresponding equivalent compensation parameter prediction information based on the equivalent wind speed prediction information includes: Based on the equivalent wind speed prediction information, the implicit temporal features within a preset time period are extracted to determine the implicit temporal feature sequence. A pre-defined multi-head attention strategy is used to weight and fuse the focused feature vectors corresponding to the implicit temporal feature sequence to determine the focused feature vector fusion information. Based on the focused feature vector fusion information, the feedforward compensation coefficient of the preset control cycle is calculated to obtain the equivalent compensation parameter prediction information.

6. The method as described in claim 1, characterized in that, The step of determining the corresponding control reference parameter information based on the equivalent wind speed prediction information includes: Obtain power limitation requirement information; Based on the power limitation demand information and the equivalent wind speed prediction information, the control reference parameters are calculated to obtain control reference parameter information, which includes pitch angle reference parameter information and torque reference parameter information.

7. The method as described in claim 1, characterized in that, The step of feedforward correction of the control reference parameter information based on the equivalent compensation parameter prediction information to determine the control correction parameter information includes: Based on the pitch angle compensation value in the equivalent compensation parameter prediction information, the pitch angle reference parameter information in the control reference parameter information is feedforward corrected to obtain the pitch angle correction parameter information. Based on the torque compensation coefficient in the equivalent compensation parameter prediction information, the torque reference parameter information in the control reference parameter information is feedforward corrected to obtain the torque correction parameter information. Based on the pitch angle correction parameter information and the torque correction parameter information, control correction parameter information is obtained.

8. The method as described in claim 1, characterized in that, The step of controlling the wind turbine generator to monitor the wind farm based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information includes: The target compensation angle information is determined based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information. The corresponding control command is determined based on the target compensation angle information; The wind turbine generator is controlled to monitor the wind farm based on the control commands.

9. The method as described in claim 8, characterized in that, The step of determining the corresponding control command based on the target compensation angle information includes: Obtain theoretical power information; Based on the target compensation angle information, the corresponding actual output power is monitored to determine the actual output power monitoring information; Based on the comparison between the theoretical power information and the actual output power monitoring information, the power tracking deviation information is determined. The corresponding control command is determined based on the power tracking deviation information.

10. A wind farm monitoring system for wind turbine generator sets, characterized in that, The system includes: The acquisition module is used to acquire wind speed information, blade load information, and operating parameter information of the wind turbine generator set; The processing module is used to make predictions based on the wind speed information, the blade load information and the operating parameter information, and to determine the equivalent wind speed prediction information and the corresponding equivalent compensation parameter prediction information. The processing module is also used to determine the corresponding control reference parameter information based on the equivalent wind speed prediction information; The execution module is used to perform feedforward correction on the control reference parameter information based on the equivalent compensation parameter prediction information, and determine the control correction parameter information; The execution module is also used to control the wind turbine generator set to monitor the wind field based on the pitch angle correction parameter information and torque correction parameter information in the control correction parameter information.

Images