Wind turbine generator load and power cooperative control method based on laser radar feedforward

Abstract

The invention relates to the technical field of wind turbine generator control, in particular to a wind turbine generator load and power cooperative control method based on laser radar feedforward. The method comprises the following steps: acquiring and normalizing wind power plant monitoring data flow, and outputting standardized wind power plant data; physical and virtual laser radar data are fused through a sliding window averaging method; outputting whole-field wind regime distribution in a future prediction time domain by using space-time Kriging interpolation; outputting a predicted inflow condition, wake flow velocity loss and a key component load spectrum through a wake flow load coupling prediction model; constructing a multi-objective optimization function to generate an optimal control instruction set; and executing the instruction, and updating model parameters through a recursive least square algorithm. According to the method, wind regime accurate prediction is achieved through space-time Kriging interpolation, interference between units is quantified in combination with a wake flow load coupling model, and the problem that traditional control is poor in generating capacity and load balance effect is solved.

Description

Technical Field

[0001] This invention relates to the field of wind turbine control technology, and more specifically, to a method for coordinated control of load and power in wind turbines based on lidar feedforward. Background Technology

[0002] The operational efficiency and lifespan of wind turbines rely heavily on accurate wind condition sensing and dynamic control. The core objective is to maximize power generation while minimizing load losses on critical components such as blades and towers. Existing wind turbine control methods largely depend on real-time feedback data from the turbine's own sensors, which suffers from issues such as delayed wind condition sensing and inaccurate wake interference prediction. This makes it difficult to achieve a dynamic balance between power generation and load in rapidly changing wind farm environments. Either the pursuit of power generation efficiency sacrifices equipment load safety, or the reduction of load limits power generation potential, and there is a lack of precise adaptation to the spatiotemporal distribution characteristics of wind farms. Therefore, this invention provides a wind turbine load and power coordinated control method based on lidar feedforward. Summary of the Invention

[0003] The purpose of this invention is to provide a load and power coordinated control method for wind turbines based on lidar feedforward, in order to solve the problems of existing wind turbine control methods mentioned in the background art, which mostly rely on real-time feedback data from the turbine's own sensors, resulting in lag in wind condition perception and inaccurate prediction of wake interference.

[0004] To achieve the above objectives, the present invention aims to provide a method for coordinated control of load and power of wind turbine generators based on lidar feedforward, comprising the following steps:

[0005] S1. Acquire and normalize the wind farm monitoring data stream, and output standardized wind farm data; the monitoring data stream includes physical lidar data, unit operation data, and virtual lidar data.

[0006] S2. The physical lidar data and virtual lidar data in the standardized wind field data are fused by the sliding window averaging method to obtain three-dimensional wind field data;

[0007] S3. Based on three-dimensional wind field data, spatial interpolation and temporal prediction are performed using spatiotemporal kriging interpolation, and the distribution of the entire field wind conditions in the future prediction time domain is output.

[0008] S4. Input the wind condition distribution of the entire field into the pre-trained wake load coupled prediction model, and output the predicted inflow conditions, wake velocity deficit and load spectrum of key components of each unit in the future prediction time domain.

[0009] S5. Construct a multi-objective optimization function with the objectives of maximizing power generation and minimizing load. Combine the predicted inflow conditions, wake velocity deficit, and load spectrum, and solve the optimal control instruction set by a distributed feedforward method based on priority rules.

[0010] S6. Execute the optimal control instruction set, collect actual operating data, calculate the deviation between the actual operating data and the optimization target, and update the wake load coupling prediction model parameters in S4 online through the recursive least squares algorithm.

[0011] As a further improvement to this technical solution, the specific steps involved in S1 for acquiring and normalizing the wind farm monitoring data stream and outputting standardized wind farm data are as follows:

[0012] The monitoring data stream is obtained from the wind farm real-time monitoring system, including physical lidar data, unit operation data, and virtual lidar data;

[0013] The monitoring data streams are aligned using a unified timestamp, and data loss due to different sampling frequencies is handled by interpolation.

[0014] For various types of monitoring data, set reasonable threshold ranges to identify and remove outliers;

[0015] For short-term missing data, linear interpolation between adjacent time points is used to fill the gaps.

[0016] Max-min normalization is performed on various monitoring data, and the normalized physical lidar data, unit operation data, and virtual lidar data are integrated into standardized wind field data according to time series.

[0017] As a further improvement to this technical solution, the specific steps involved in S2, which use the sliding window averaging method to fuse physical lidar data and virtual lidar data in standardized wind field data, are as follows:

[0018] Extract physical lidar data and virtual lidar data from standardized wind field data;

[0019] Set the sliding window length to The window sliding step size is ;

[0020] For each spatial grid point within each sliding window, calculate the average value of the physical lidar wind speed and the average value of the virtual lidar wind speed at that point within the window time, and then perform weighted fusion to obtain the fused wind speed value.

[0021] Simultaneously, the average value of the wind direction from the physical lidar and the average value of the wind direction from the virtual lidar at this point within the window time are calculated and weighted to obtain the wind direction fusion value.

[0022] For each sliding window, the wind speed fusion values ​​of all spatial locations are arranged in a three-dimensional grid to form the three-dimensional wind speed distribution corresponding to that window. The wind direction fusion values ​​of all spatial locations are arranged in the same way to form the three-dimensional wind direction distribution.

[0023] The three-dimensional wind speed distribution and the three-dimensional wind direction distribution are smoothed in the time dimension. The smoothed three-dimensional wind speed distribution and the three-dimensional wind direction distribution are arranged in time order to form three-dimensional wind field data.

[0024] As a further improvement to this technical solution, in step S3, the specific steps involved in using spatiotemporal kriging interpolation for spatial interpolation and temporal prediction to output the overall wind distribution in the future prediction time domain are as follows:

[0025] Wind speed and direction data at the current moment and at several historical moments are extracted from the three-dimensional wind field data to form a spatiotemporal sample set;

[0026] Based on the spatiotemporal sample set, an empirical covariance function describing the spatial and temporal correlation of wind speed is established.

[0027] Based on the empirical covariance function, the spatiotemporal sample set... For each point, calculate the covariance between each pair and construct the covariance matrix. ;

[0028] For each spatiotemporal point to be predicted, calculate the covariance vector between that point and all sample points. Combined with the covariance matrix Solve for the Kriging weight vector ;

[0029] The wind speed and wind direction of the sample are weighted and averaged using the Kriging weight vector to obtain the predicted wind speed and predicted wind direction values ​​for that point.

[0030] By traversing all future prediction times and all spatial grid points, the wind speed prediction field and wind direction prediction field are obtained, and output as the overall wind condition distribution in the future prediction time domain.

[0031] As a further improvement to this technical solution, the specific steps involved in constructing the wake load coupling prediction model in step S4 are as follows:

[0032] Receive data on wind conditions across the entire site and unit operation;

[0033] For each turbine in the wind farm, the predicted wind speed and predicted wind direction values ​​are extracted from the wind condition distribution data of the entire farm and used as the predicted inflow conditions for that turbine.

[0034] For each pair of upstream and downstream units, the wake calculation module, combined with the distance between them and the thrust coefficient of the upstream unit, calculates the upstream unit's thrust. Wake in downstream units The proportion of wind speed reduction caused by location ;

[0035] For downstream units, the predicted inflow wind speed is superimposed with the wind speed attenuation effect caused by all upstream units to obtain the effective wind speed after considering wake interference.

[0036] Based on the effective wind speed of each unit, as well as the pitch angle and rotational speed in the unit's operating data, the load response module calculates the load values ​​that the key components of the unit will bear at each future moment; among which, the key components include the blade root and the bottom of the tower.

[0037] The load values ​​are arranged in chronological order to form the load spectrum of key components;

[0038] The wake calculation module is coupled with the load response module to form a wake load coupled prediction model.

[0039] As a further improvement to this technical solution, in step S4, the specific steps involved in outputting the predicted inflow conditions, wake velocity deficit, and load spectrum of key components for each unit in the future prediction time domain through the wake load coupling prediction model are as follows:

[0040] The predicted inflow conditions are input into the wake calculation module of the wake load coupled prediction model to calculate the effective wind speed of each unit after considering the wake effect.

[0041] The difference sequence between the predicted wind speed and the effective wind speed is the wake velocity deficit.

[0042] The effective wind speed and unit operating data are input into the load response module of the wake load coupling prediction model to calculate the load spectrum of key components of each unit.

[0043] Output the predicted inflow conditions, wake velocity deficit, and load spectrum of key components for each unit.

[0044] As a further improvement to this technical solution, the specific steps involved in constructing the multi-objective optimization function with the objectives of maximizing power generation and minimizing load in step S5 are as follows:

[0045] Based on the effective wind speed sequence, the average predicted power of the unit during the forecast period is calculated by querying the power curve of the unit model.

[0046] Extract the maximum blade load of each unit during the prediction period from the load spectrum of key components;

[0047] The total cost function is constructed by weighted summation based on the average predicted power and the maximum blade load. ;

[0048] Set a pitch angle change rate constraint to limit the pitch angle change of a single unit from exceeding its maximum allowable change rate.

[0049] Set a power limit constraint for generators to restrict the actual output power of a single unit from exceeding its rated capacity;

[0050] A multi-objective optimization function is constructed, which, under the conditions of satisfying the pitch angle change rate constraint and the generator power upper limit constraint, finds a series of optimal pitch angle setpoints for each unit in the future time domain, such that the total cost function... minimize.

[0051] As a further improvement to this technical solution, the specific steps involved in generating the optimal control instruction set in step S5 using a distributed feedforward method based on priority rules are as follows:

[0052] Based on the maximum blade load value and average predicted power of each unit, the control priority score of each unit is obtained by normalization and weighted summation. ;

[0053] The units are scored according to their control priority. Sort the units from highest to lowest quality and perform local optimizations on each unit in turn;

[0054] For the first For this generator unit, based on the predicted inflow conditions, the effective wind speed after the wake effect, and the load spectrum of key components, under the premise of satisfying the pitch angle change rate constraint and the generator power upper limit constraint, the gradient descent method is used to find the optimal pitch angle sequence that minimizes its local cost function; where the local cost function is obtained by decomposing the total cost function.

[0055] After each unit is optimized, the thrust coefficient of the unit is recalculated based on its optimal pitch angle sequence, and then the effective wind speed of all affected downstream units is updated through the wake calculation module.

[0056] Based on the updated effective wind speed, the load response module is invoked again to update the load spectrum of key components of the affected units;

[0057] After traversing all units and completing the optimization, the optimal pitch angle sequence of each unit is integrated in chronological order to form the optimal control command set.

[0058] As a further improvement to this technical solution, the specific steps involved in calculating the deviation between the actual operating data and the optimization target in step S6 are as follows:

[0059] After executing the optimal control command set, monitoring data of the wind turbine units during actual operation is collected, including the actual output power of each unit. Actual leaf root bending moment Actual pitch angle ;

[0060] Based on the wake load coupling prediction model, the actual inflow wind speed is used as input, and combined with the actual blade pitch angle, the predicted power is obtained. With predicted load ;

[0061] Based on actual output power Predicted power Calculate the power deviation sequence;

[0062] Based on actual leaf root bending moment Predicted load Calculate the load deviation sequence;

[0063] The average power deviation within each time period is calculated based on the power deviation sequence and the load deviation sequence. Deviation from average load ;

[0064] If the average power deviation is satisfied Exceeding the power deviation threshold, average load deviation If any one of the load deviation thresholds is exceeded, the model parameter update process is triggered.

[0065] As a further improvement to this technical solution, in step S6, the specific steps involved in updating the wake load coupling prediction model parameters in step S4 online using the recursive least squares algorithm are as follows:

[0066] Identify the key parameters that need to be updated online in the wake load coupling prediction model, including the wake attenuation coefficient and the bending moment coefficient;

[0067] Data pairs corresponding to the model's predictions are extracted from actual operational data and used as training samples for the recursive least squares algorithm. Each sample includes an input feature vector. and output observations The input feature vector includes the actual inflow wind speed. Actual pitch angle Actual tip speed ratio Output observations These are the actual effective wind speeds. and actual leaf root bending moment ;

[0068] The wake calculation module and the load response module are approximated by a first-order linear approximation around the current parameter values ​​to form a linear parameter estimation form.

[0069] The recursive least squares method with a forgetting factor is used to update the parameter estimation vector time step by time;

[0070] The updated parameter estimation vector is synchronized to the wake load coupled prediction model. The deviation between the predicted value and the actual value output by the updated model is compared. If the deviation still exceeds the set threshold, the update step is repeated; otherwise, the update process is stopped and the current parameters are retained for use in subsequent control cycles.

[0071] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0072] 1. In this method of load and power coordinated control of wind turbine based on lidar feedforward, physical lidar and virtual lidar data are integrated, and spatiotemporal kriging interpolation is used to achieve accurate prediction of future wind conditions. Combined with the wake load coupling model, the wake interference between units is quantified. Compared with the traditional control method that relies on real-time feedback, the method can predict the spatiotemporal distribution of the wind field and load changes in advance, and solves the problems of poor power generation and load balance caused by the lag in response to wind condition changes and insufficient consideration of wake influence in traditional control.

[0073] 2. In this method of load and power coordinated control of wind turbine based on lidar feedforward, the multi-objective optimization function takes into account both maximizing power generation and minimizing load. It adopts a distributed feedforward solution strategy based on priority rules and updates the model parameters online through recursive least squares algorithm. Compared with traditional fixed parameter control or single objective optimization methods, it can adapt to different wind farm conditions and operating requirements, and can dynamically correct model deviations to achieve continuous optimization of control accuracy, effectively improving the operating efficiency and lifespan of wind turbine. Attached Figure Description

[0074] Figure 1 This is a flowchart of the overall method of the present invention. Detailed Implementation

[0075] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0076] Example:

[0077] Please see Figure 1 As shown, this embodiment provides a method for coordinated control of wind turbine load and power based on lidar feedforward, including the following steps:

[0078] S1. Acquire and normalize the wind farm monitoring data stream, and output standardized wind farm data; the monitoring data stream includes physical lidar data, unit operation data, and virtual lidar data.

[0079] In this embodiment, monitoring data streams are obtained from the wind farm real-time monitoring system, including physical lidar data, turbine operation data, and virtual lidar data;

[0080] Among them, the physical lidar data is the wind field scanning data obtained by actual measurement, including wind speed and wind direction information at different azimuth and elevation angles in front of the wind turbine;

[0081] Unit operating data includes turbine speed, blade pitch angle, power output, load sensor signals, etc.

[0082] Virtual lidar data is predicted wind field data generated through wake simulation models (such as Jensen model, FLORIS, etc.), covering areas not scanned by physical lidar or wake-affected areas between units;

[0083] The monitoring data streams are aligned using a unified timestamp, and data loss due to different sampling frequencies is handled by interpolation.

[0084] For various types of monitoring data, set reasonable threshold ranges (such as wind speed not higher than the cut-out wind speed, not lower than the cut-in wind speed, power not exceeding the rated capacity, etc.) to identify and remove outliers.

[0085] For short-term missing data, linear interpolation of adjacent time points or historical data from the same period are used to fill the gaps.

[0086] Max-min normalization is performed on various monitoring data, and the normalized physical lidar data, unit operation data, and virtual lidar data are integrated into standardized wind field data according to time series.

[0087] S2. The physical lidar data and virtual lidar data in the standardized wind field data are fused by the sliding window averaging method to obtain three-dimensional wind field data;

[0088] In this embodiment, physical lidar data and virtual lidar data are extracted from standardized wind field data;

[0089] Both physical lidar data and virtual lidar data are stored in a unified time-series format with a time resolution of [missing information]. (For example =1 second), spatial resolution is , , (For example, a horizontal resolution of 10 meters and a vertical resolution of 5 meters).

[0090] Set the sliding window length to (For example =30 seconds), window sliding step size is (For example =10 seconds), each window contains Data at each point in time;

[0091] For each spatial grid point within each sliding window, calculate the average value of the physical lidar wind speed and the average value of the virtual lidar wind speed at that point within the window time, and then perform weighted fusion to obtain the fused wind speed value.

[0092] Simultaneously, the average value of the wind direction from the physical lidar and the average value of the wind direction from the virtual lidar at this point within the window time are calculated and weighted to obtain the wind direction fusion value.

[0093]

[0094] in, To integrate data; This represents the average value of the physical lidar data (wind speed or wind direction) within the window. This represents the average value of the virtual lidar data (wind speed or wind direction) within the window. For physical weighting coefficients, For virtual weight coefficients, satisfying ;

[0095] Weight allocation principle: If the spatial grid point is within the effective detection range of the physical lidar and the data quality indicators (such as signal-to-noise ratio and confidence level) are higher than the threshold, then set... , If the point is located in a radar blind zone or has low data quality, then set... , If the point has no physical data at all (such as an occluded area), then it relies entirely on virtual data, i.e. , ;

[0096] For each sliding window (window increments by step) (Slide) Arrange the combined wind speed values ​​of all spatial locations on a three-dimensional grid to form the three-dimensional wind speed distribution corresponding to the window. The wind direction values ​​of all spatial locations are merged and arranged in the same way to form a three-dimensional wind direction distribution. ;

[0097] The three-dimensional wind speed distribution and the three-dimensional wind direction distribution are smoothed in the time dimension, and the smoothed three-dimensional wind speed distribution and the three-dimensional wind direction distribution are arranged in time order to form three-dimensional wind field data.

[0098] Smoothing refers to applying a low-pass filter in the time dimension to the three-dimensional wind speed and direction distributions to eliminate data fluctuations and maintain spatiotemporal continuity.

[0099] S3. Based on three-dimensional wind field data, spatial interpolation and temporal prediction are performed using spatiotemporal kriging interpolation, and the distribution of the entire field wind conditions in the future prediction time domain is output.

[0100] In this embodiment, wind speed and wind direction data at the current moment and at several historical moments are extracted from the three-dimensional wind field data to form a spatiotemporal sample set;

[0101] Each sample is represented as:

[0102]

[0103] in, Spatial coordinates (unit: meters); Timestamp (unit: seconds); The wind speed (in m / s) or wind direction (in degrees) at that location.

[0104] Based on the spatiotemporal sample set, an empirical covariance function describing the spatial and temporal correlation of wind speed is established.

[0105] Assuming that wind speed is stationary in space and time, we establish an empirical covariance function:

[0106]

[0107] in, Spatial distance; For time intervals; For wind speed variance; For spatially relevant scales (e.g., 200 meters); For time-related scales (e.g., 20 seconds); The empirical covariance function value; It is an exponential function;

[0108] Based on the empirical covariance function, the spatiotemporal sample set... For each pair of points, calculate the covariance and construct... covariance matrix ;

[0109] For each spatiotemporal point to be predicted Calculate the covariance vector between this point and all sample points. Combined with the covariance matrix Solve for the Kriging weight vector ;

[0110]

[0111] in, For spacetime points The covariance vector between all sample points (length is...) ); This is the Kriging weight vector;

[0112] The wind speed and wind direction of the sample are weighted and averaged using the Kriging weight vector to obtain the predicted wind speed and predicted wind direction values ​​for that point.

[0113]

[0114] in, For the first Kriging weight vectors for each sample; Forecast wind speed or forecast wind direction;

[0115] By iterating through all future prediction times (e.g., 30 seconds in the future, one time per second) and all spatial grid points, the wind speed prediction field is obtained. Wind direction prediction field And output as the overall wind condition distribution in the future prediction time domain;

[0116] Among them, wind speed prediction field Indicates spatial location and future moments Predicted wind speed value, wind direction prediction field This represents the predicted wind direction value at the corresponding location and time. Together, these two values ​​constitute the complete wind vector information for each spatial point in the future time domain.

[0117] S4. Input the wind condition distribution of the entire field into the pre-trained wake load coupled prediction model, and output the predicted inflow conditions, wake velocity deficit and load spectrum of key components of each unit in the future prediction time domain.

[0118] In this embodiment, the wind conditions distribution across the entire field and the unit's operating data (including pitch angle, speed, generator torque, etc.) are received.

[0119] For each turbine in the wind farm, the predicted wind speed and predicted wind direction at a specific location in front of the rotor at each future moment are extracted from the wind condition distribution data of the entire farm and used as the predicted inflow conditions for that turbine.

[0120] Based on the wind farm layout, the upstream turbines of each unit are identified. For each upstream and downstream turbine group, the upstream turbine's thrust coefficient is calculated using the wake calculation module, taking into account the distance between them. Wake in downstream units The proportion of wind speed reduction caused by location ;

[0121]

[0122] in, For the first The thrust coefficient of the turbine unit is obtained by looking up the current pitch angle and the predicted wind speed from a table; This is the wake attenuation coefficient (typically 0.04). For the unit and The horizontal distance between them; The diameter of the wind turbine; For upstream units Wake in downstream units The proportion of wind speed reduction caused by location;

[0123] For downstream units, the predicted inflow wind speed is superimposed with the wind speed attenuation effect caused by all upstream units to obtain the effective wind speed after considering wake interference.

[0124]

[0125] in, To take into account the wake effect of the upstream unit, the first The effective wind speed of the unit; When the wake is not considered Inflow velocity of the unit (obtained from predicted inflow conditions); For all those located in the first The collection of units upstream of the Taiwanese unit;

[0126] Based on the effective wind speed of each unit, as well as the pitch angle and rotational speed in the unit's operating data, the load response module calculates the load values ​​that the key components of the unit will bear at each future moment; among which, the key components include the blade root and the bottom of the tower.

[0127] The load response module converts wind speed changes into loads on critical components, such as the bending moment at the blade root. For example:

[0128]

[0129] in, for The bending moment value at the root of the blade at any given time. The density of air is 1.225 kg / m³. Where is the blade radius; The bending moment coefficient is the pitch angle. Speed ​​ratio of leaf tip The function is obtained through a predefined lookup table; To account for the effective wind speed after considering the wake effect, i.e. A sequence in time;

[0130] The load values ​​are arranged in chronological order to form the load spectrum of key components, which is a predicted sequence of load changes over time.

[0131] The wake calculation module is coupled with the load response module to form a wake load coupled prediction model.

[0132] Key parameters in the wake load coupling prediction model (such as the wake attenuation coefficient) Bending moment coefficient The query table (etc.) is set according to the actual turbine model and layout of the wind farm during initialization and is continuously updated in S6 through online data.

[0133] In this embodiment, the predicted inflow conditions are input into the wake calculation module of the wake load coupling prediction model to calculate the effective wind speed of each unit after considering the wake effect.

[0134] The difference sequence between the predicted wind speed and the effective wind speed is the wake velocity deficit.

[0135] The effective wind speed and unit operating data are input into the load response module of the wake load coupling prediction model to calculate the load spectrum of key components of each unit.

[0136] Output the predicted inflow conditions, wake velocity deficit, and load spectrum of key components for each unit.

[0137] S5. Construct a multi-objective optimization function with the objectives of maximizing power generation and minimizing load. Combine the predicted inflow conditions, wake velocity deficit, and load spectrum, and solve the optimal control instruction set by a distributed feedforward method based on priority rules.

[0138] In this embodiment, based on the effective wind speed sequence, the average predicted power of the unit model during the prediction period is calculated by querying the power curve of the unit model.

[0139] Extract the maximum blade load of each unit during the prediction period from the load spectrum of key components;

[0140] The total cost function is constructed by weighted summation based on the average predicted power and the maximum blade load. ;

[0141]

[0142] in, For the total cost function, the smaller the value, the better the control effect (high power generation, low load). This represents the total number of wind turbine units. For the first The average predicted power (megawatts) of the turbine unit in the prediction time domain is derived from the effective wind speed. The sequence is calculated from the power curve of the unit, and the power curve lookup table is pre-calibrated based on the unit model. For the first The maximum blade load of the turbine unit in the prediction time domain is derived from the load spectrum. The sequence, taking the maximum value after taking the absolute value, that is... ; For power weighting coefficients, For load weighting coefficients, satisfying ;

[0143] Scenario 1 (Normal Power Generation Mode): If grid demand is high and wind conditions are stable, the following can be set: Prioritize power generation;

[0144] Scenario 2 (Lifespan Protection Mode): If encountering gusts of wind or an imminent unit maintenance schedule, the following settings can be configured: Prioritize load reduction;

[0145] Set a pitch angle change rate constraint to limit the pitch angle change of a single unit from exceeding its maximum allowable change rate:

[0146]

[0147] in, For the first Taiwanese crew The pitch angle at that moment, For the first Taiwanese crew The propeller pitch angle at that moment; The maximum allowable rate of change of pitch angle is typically found in mainstream megawatt-class wind turbines. ;

[0148] Set a power limit constraint on generators to restrict the actual output power of a single unit from exceeding its rated capacity:

[0149]

[0150] in, For the first Taiwanese crew Output power at any given moment; The rated power of the wind turbine is the nameplate parameter of the turbine, for example, 3MW;

[0151] Construct a multi-objective optimization function, that is, under the conditions of satisfying the pitch angle change rate constraint and the generator power upper limit constraint, optimize the function by setting a target for each unit ( Find a series of optimal pitch angle settings in the future time domain. This makes the total cost function minimize.

[0152] In this embodiment, the control priority score for each unit is obtained by normalizing and weighting the maximum blade load value and average predicted power of each unit. ;

[0153]

[0154] in, This represents the maximum blade load value for the unit during the predicted period; This represents the maximum load value among all the units in the wind farm. This represents the average predicted power of the unit during the forecast period. This represents the maximum predicted power of all units in the wind farm. For power weighting coefficients, This refers to the load weighting coefficient; To control the priority score, the higher the value, the greater the impact of the unit on the overall objective in the current time period, and the more likely it should be prioritized for control optimization;

[0155] The units are scored according to their control priority. Sort the units from highest to lowest quality and perform local optimizations on each unit in turn;

[0156] For the first Based on the predicted inflow conditions, effective wind speed after wake effect, and load spectrum of key components of the generator unit, and under the premise of satisfying the pitch angle change rate constraint and generator power upper limit constraint, the gradient descent method is used to find the local cost function. Minimum optimal pitch angle sequence The local cost function is obtained by decomposing the total cost function.

[0157] Local cost function:

[0158]

[0159] in, For the first Local cost function of the unit;

[0160] After each unit is optimized, the thrust coefficient of the unit is recalculated based on its optimal pitch angle sequence, and then the effective wind speed of all affected downstream units is updated through the wake calculation module.

[0161]

[0162] in, For upstream units The change in wind speed caused by the adjustment of the pitch angle can be estimated by the change in its thrust coefficient; For upstream units For downstream units The wind speed attenuation ratio; The updated effective wind speed for downstream units; The effective wind speed before the update;

[0163] Based on the updated effective wind speed, the load response module is invoked again to update the load spectrum of key components of the affected units;

[0164] After traversing all units and completing the optimization, the optimal pitch angle sequence for each unit is... Integrate them in chronological order to form the optimal control instruction set;

[0165] The optimal control instruction set is indexed by time and contains the pitch angle setting value for each unit at each control moment in the future.

[0166] S6. Execute the optimal control instruction set, collect actual operating data, calculate the deviation between the actual operating data and the optimization target, and update the parameters of the wake load coupling prediction model in S4 online through the recursive least squares algorithm.

[0167] In this embodiment, after executing the optimal control instruction set, monitoring data of the wind turbine units during actual operation is collected, including the actual output power of each unit. Actual leaf root bending moment (Acquired via load sensor) Actual pitch angle ;

[0168] Based on the wake load coupling prediction model, the predicted power is obtained by using the actual inflow wind speed (which can be measured by physical lidar) as input and combining it with the actual blade pitch angle. With predicted load ;

[0169] Based on actual output power Predicted power Calculate the power deviation sequence;

[0170]

[0171] in, Indicates the first Taiwanese crew The deviation between the actual power and the predicted power at any given time; a positive value indicates that the actual power generation is higher than the prediction.

[0172] Based on actual leaf root bending moment Predicted load Calculate the load deviation sequence;

[0173]

[0174] in, This indicates the deviation between the actual load and the predicted load; a positive value indicates that the actual load is higher than the predicted load.

[0175] The average power deviation within each time period is calculated based on the power deviation sequence and the load deviation sequence. Deviation from average load ;

[0176]

[0177]

[0178] in, To assess the length of the time period (e.g., 30 seconds). This represents the total number of generating units. This represents the average power deviation. This represents the average load deviation.

[0179] If the average power deviation is satisfied Exceeding the power deviation threshold (e.g., 10kW), average load deviation If any of the load deviation thresholds (such as 5 kN·m) are exceeded, the model parameter update process is triggered.

[0180] In this embodiment, key parameters that need to be updated online in the wake load coupling prediction model are identified, including the wake attenuation coefficient. and bending moment coefficient ;

[0181] Among them, the wake attenuation coefficient The bending moment coefficient reflects the wind speed attenuation characteristics during wake propagation. Indicates a specific pitch angle Speed ​​ratio of leaf tip The proportional relationship between the bending moment on the lower blade and the wind speed;

[0182] Data pairs corresponding to the model's predictions are extracted from actual operational data and used as training samples for the recursive least squares algorithm. Each sample includes an input feature vector. and output observations The input feature vector includes the actual inflow wind speed. Actual pitch angle Actual tip speed ratio Output observations These are the actual effective wind speeds. (Used to update the wake attenuation coefficient) ) and actual blade root bending moment (used to update bending moment coefficient) );

[0183] The wake calculation module and the load response module are approximated by a first-order linear approximation around the current parameter values ​​to form a linear parameter estimation form:

[0184]

[0185] in, For example, the parameter vector to be estimated. ; To observe noise; For the corresponding input feature vector, construct according to the model structure, for example, for the wake flow model, Can contain and Combination items; To output the observed values;

[0186] The recursive least squares method with a forgetting factor is used to update the parameter estimation vector time step by time;

[0187] The updated formula is as follows:

[0188]

[0189]

[0190]

[0191] in, for The parameter estimation vector at time step [time]. for The parameter estimation vector at time step; for The parameter estimation error covariance matrix at time t. for The parameter estimation error covariance matrix at time t; This is the gain matrix; Forgetting factor, usually taken as This is used to mitigate the impact of old data and enhance the model's adaptability to current operating conditions; The identity matrix has dimensions and parameter vectors. The dimensions are consistent;

[0192] Update the parameter estimation vector The updated model is synchronized to the wake load coupling prediction model. The deviation between the predicted value and the actual value is compared. If the deviation still exceeds the set threshold, the update step is repeated. Otherwise, the update process is stopped, and the current parameters are retained for use in subsequent control cycles.

[0193] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A wind turbine load and power collaborative control method based on lidar feedforward, characterized in that: The method comprises the following steps: S1, acquiring and normalizing wind farm monitoring data streams, and outputting standardized wind farm data; wherein the monitoring data streams comprise physical laser radar data, unit operation data, and virtual laser radar data; S2, fusing the physical laser radar data and the virtual laser radar data in the standardized wind farm data by a sliding window average method, and obtaining three-dimensional wind field data; S3, based on the three-dimensional wind field data, performing spatial interpolation and time prediction by using a spatiotemporal Kriging interpolation, and outputting full-field wind condition distribution in a future prediction time domain; S4, inputting the full-field wind condition distribution into a pre-trained wake load coupling prediction model, and outputting predicted inflow conditions, wake velocity loss, and key component load spectrum of each unit in the future prediction time domain; S5, constructing a multi-objective optimization function with the maximum power generation and the minimum load as the objectives, combining the predicted inflow conditions, the wake velocity loss, and the load spectrum, and solving by using a distributed feedforward method based on priority rules to generate an optimal control instruction set; S6, executing the optimal control instruction set, collecting actual operation data, calculating the deviation between the actual operation data and the optimization objectives, and updating the wake load coupling prediction model parameters in S4 by using a recursive least squares algorithm.

2. The wind turbine generator load and power collaborative control method based on lidar feedforward according to claim 1, characterized in that: In S1, the specific steps involved in acquiring and normalizing wind farm monitoring data streams and outputting standardized wind farm data are as follows: acquiring monitoring data streams, including physical laser radar data, unit operation data, and virtual laser radar data, from a wind farm real-time monitoring system; aligning the monitoring data streams by using a unified timestamp, and processing data missing caused by different sampling frequencies by using an interpolation method; setting reasonable threshold ranges for various types of monitoring data, identifying and removing outliers; for short-term missing data, filling in linear interpolation of adjacent time points; performing maximum-minimum normalization on various types of monitoring data, and integrating the normalized physical laser radar data, unit operation data, and virtual laser radar data into standardized wind farm data in time sequence.

3. The wind turbine generator load and power coordinated control method based on lidar feedforward according to claim 1, characterized in that: In S2, the specific steps involved in fusing the physical laser radar data and the virtual laser radar data in the standardized wind farm data by using a sliding window average method are as follows: extracting the physical laser radar data and the virtual laser radar data from the standardized wind farm data; The length of the sliding window is set as , and the window sliding step is ; for each spatial grid point in each sliding window, calculating the average value of the physical laser radar wind speed value and the average value of the virtual laser radar wind speed value in the window time, and performing weighted fusion to obtain a wind speed fusion value; at the same time, calculating the average value of the physical laser radar wind direction value and the average value of the virtual laser radar wind direction value in the window time, and performing weighted fusion to obtain a wind direction fusion value; for each sliding window, arranging the wind speed fusion values of all spatial position points in a three-dimensional grid to form a three-dimensional wind speed distribution corresponding to the window, and arranging the wind direction fusion values of all spatial position points in the same way to form a three-dimensional wind direction distribution; performing smoothing processing on the three-dimensional wind speed distribution and the three-dimensional wind direction distribution in the time dimension, respectively, arranging the smoothed three-dimensional wind speed distribution and the three-dimensional wind direction distribution in time sequence, and forming three-dimensional wind field data.

4. The wind turbine generator load and power coordinated control method based on lidar feedforward according to claim 1, characterized in that: In the S3, the spatial interpolation and time prediction are performed by using the space-time Kriging interpolation, and the specific steps of outputting the full-field wind distribution in the future prediction time domain include: extracting the wind speed and wind direction data at the current time and historical times from the three-dimensional wind field data to form a space-time sample set; establishing an empirical covariance function based on the space-time sample set to describe the correlation of the wind speed in space and time; Based on the empirical covariance function, the covariances between each pair of points in the spatio-temporal sample set are calculated to construct a covariance matrix ; and the covariance matrix is decomposed to obtain a plurality of principal components . For each spatio-temporal point to be predicted, compute the covariance vector between this point and all sample points , combine the covariance matrices , solve for the Kriging weight vector ; using the Kriging weight vector to perform weighted average on the wind speed and wind direction of the sample respectively to obtain the predicted wind speed value and the predicted wind direction value of the point; traversing all future prediction times and all spatial grid points to obtain the wind speed prediction field and the wind direction prediction field, and outputting the full-field wind distribution in the future prediction time domain.

5. The wind turbine generator load and power coordinated control method based on lidar feedforward according to claim 1, characterized in that: In the S4, the specific steps of constructing the wake load coupling prediction model include: receiving the full-field wind distribution and the unit operation data; for each unit in the wind farm, extracting the predicted wind speed value and the predicted wind direction value from the full-field wind distribution data as the predicted inflow condition of the unit; For each upstream and downstream group, the wake calculation module calculates the wind speed decay ratio caused by the wake of the upstream group wake in the downstream group position ; for the downstream unit, superimposing the predicted inflow wind speed on the wind speed attenuation effect caused by all upstream units to obtain the effective wind speed considering the wake interference; based on the effective wind speed of each unit and the pitch angle and the rotating speed in the unit operation data, calculating the load value of the key components of the unit at each time in the future through the load response module; wherein the key components include the blade root and the tower bottom; arranging the load values in time sequence to form the key component load spectrum; coupling the wake calculation module and the load response module to form the wake load coupling prediction model.

6. The wind turbine generator load and power coordinated control method based on lidar feedforward according to claim 5, characterized in that: In the S4, the specific steps of outputting the predicted inflow condition, the wake speed loss and the key component load spectrum of each unit in the future prediction time domain through the wake load coupling prediction model include: inputting the predicted inflow condition into the wake calculation module of the wake load coupling prediction model to calculate the effective wind speed of each unit considering the wake effect; calculating the difference sequence of the predicted wind speed value and the effective wind speed, which is the wake speed loss; inputting the effective wind speed and the unit operation data into the load response module of the wake load coupling prediction model to calculate the key component load spectrum of each unit; outputting the predicted inflow condition, the wake speed loss and the key component load spectrum of each unit.

7. The wind turbine generator load and power coordinated control method based on lidar feedforward according to claim 1, characterized in that: In the S5, the specific steps of constructing the multi-objective optimization function with the maximum power generation and the minimum load as the target include: based on the effective wind speed sequence, calculating the average predicted power of the unit in the prediction period by querying the power curve of the unit type; extracting the maximum blade load of each unit in the prediction period from the key component load spectrum; Based on the average predicted power and the maximum value of the blade load, a total cost function is constructed by weighted summation ; setting the pitch angle change rate constraint to limit the pitch angle change of a single unit to be less than the maximum allowed change rate; setting the generator power upper limit constraint to limit the actual output power of a single unit to be less than the rated capacity; A multi-objective optimization function is constructed, i.e. under the condition of meeting the pitch angle change rate constraint and the generator power upper limit constraint, by finding a series of optimal pitch angle set values for each unit in the future time domain, so that the total cost function is minimized.

8. The wind turbine generator load and power coordinated control method based on lidar feedforward according to claim 7, characterized in that: In the S5, the specific steps of generating the optimal control instruction set by using the distributed feedforward method based on the priority rule include: Based on the maximum load value of each unit blade and the average predicted power, normalization and weighted summation are performed to obtain the control priority score of each unit ; Ranking units by control priority score Ranking from high to low, and sequentially optimizing each unit locally; To the Based on the predicted inflow conditions, the effective wind speed after the wake influence and the key component load spectrum of the unit, the optimal pitch angle sequence that makes the local cost function minimum is searched by using the gradient descent method under the premise of meeting the pitch angle change rate constraint and the generator power upper limit constraint; wherein, the local cost function is obtained by decomposing the total cost function; after completing the optimization of each unit, recalculating the thrust coefficient of the unit based on the optimal pitch angle sequence, and then updating the effective wind speed of all affected units downstream of the unit through the wake calculation module; Based on the updated effective wind speed, the load response module is re-invoked to update the key component load spectrum of the affected unit; After traversing all units to complete optimization, the optimal pitch angle sequence of each unit is integrated in time sequence to form the optimal control instruction set.

9. The wind turbine generator load and power coordinated control method based on lidar feedforward according to claim 1, characterized in that: The specific steps involved in calculating the deviation of the actual operation data from the optimization target in S6 are as follows: After executing the optimal control instruction set, monitoring data of the wind turbine in actual operation is collected, including actual output power of each unit , actual blade root bending moment , actual pitch angle ; Based on the wake load coupling prediction model, using the actual inflow wind speed as input, combined with the actual pitch angle, the predicted power is obtained With the predicted load ; based on actual output power , predicted power calculating a power deviation sequence; Based on actual blade root bending moment , predicted load calculating load bias sequence; Based on the power deviation sequence and the load deviation sequence, average power deviations in the time period are calculated and average load deviations ; If any of the following conditions are met: average power deviation exceeds a power deviation threshold, average load deviation exceeds a load deviation threshold, then a model parameter update procedure is triggered.

10. The wind turbine generator load and power coordinated control method based on lidar feedforward according to claim 9, characterized in that: The specific steps involved in updating the wake load coupling prediction model parameters in S4 online through the recursive least squares algorithm in S6 are as follows: Identify the key parameters in the wake load coupling prediction model that need to be updated online, including the wake decay coefficient and the bending moment coefficient. extracting from the actual operation data a pair of data corresponding to the model prediction as a training sample of the recursive least square algorithm, each sample including an input feature vector and an output observation ; wherein the input feature vector comprises actual inflow wind speed , actual pitch angle , actual tip speed ratio ; the output observations are actual effective wind speed and actual blade root bending moment , respectively. First-order linear approximation of the wake calculation module and the load response module around the current parameter value is performed to form a linear parameter estimation form. Recurrence least squares method with forgetting factor is used to update the parameter estimation vector at each time step. The updated parameter estimation vector is synchronized to the wake load coupling prediction model, and the deviation between the predicted value and the actual value of the updated model output is compared. If the deviation still exceeds the set threshold, repeat the update step; otherwise, stop the current update process and retain the current parameters for use in the subsequent control period.

Images