Wind power plant unit load reduction optimization method and system based on cabin type laser radar real-time wind measurement

Abstract

The invention discloses a wind power plant unit load reduction optimization method and system based on cabin type laser radar real-time wind measurement, and relates to the technical field of wind power generation. Constructing a load estimation model; performing real-time detection on a blade root bending moment predicted value sequence and a tower bottom vibration acceleration predicted value sequence, constructing a multi-objective optimization function including blade root bending moment, tower bottom vibration acceleration and power generation power, and setting a dynamic weight coefficient of the multi-objective optimization function according to three-dimensional wind field prediction sequence data; obtaining an optimal control variable through a multi-objective genetic algorithm based on the multi-objective optimization function; and according to the optimal control variable and vertical wind shearing in the three-dimensional area, differential adjustment is conducted on the pitch angle of each blade, and a personalized variable pitch angle sequence of each blade is obtained. According to the method, by establishing the optimization objective function fusing the load and the power, multi-variable collaborative optimization is achieved, and the safety and the economical efficiency of wind power plant unit operation are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a method and system for optimizing load reduction of wind farm units based on real-time wind measurement using a nacelle-type laser radar. Background Art

[0002] Chinese patent publication number CN106368898A discloses a large-scale wind turbine generator adjustment and control method and device, including: real-time acquisition of the wind rotor acceleration value, when the wind rotor acceleration value is greater than the wind rotor acceleration set limit, multiplying the wind rotor acceleration value by a specific gain as the target pitch rate; otherwise, acquiring the real-time speed of the generator, when the generator speed is greater than the generator speed set limit, correcting the gain of the pitch PID controller according to the pitch angle of the generator set and the ratio of the generator speed exceeding the generator speed set limit, and using the corrected pitch PID controller to control the pitch rate of the blades.

[0003] A Chinese patent with publication number CN114876730B discloses a method for controlling gust-load reduction operation of a wind turbine, comprising: obtaining a measured wind speed at a wind turbine by using a laser wind measuring radar; obtaining a wind direction by using a wind vane on the top of a nacelle; if the deviation between the wind direction measured by the laser wind measuring radar and the wind direction measured by the wind vane on the top of the nacelle is within a set degree, performing a yaw action based on the wind direction measured by the laser radar; otherwise, performing a yaw action based on the wind direction measured by the wind vane on the top of the nacelle to ensure that the wind turbine is accurately facing the wind; obtaining an estimated wind speed value by using a Newton iteration algorithm based on the relationship between generator power and torque; measuring the wind speed at the wind turbine nacelle by using an anemometer to obtain a measured wind speed value at the nacelle, and obtaining a wind speed value at the nacelle based on the measured wind speed value at the nacelle and the estimated wind speed value; obtaining a pitch compensation coefficient based on the wind speed difference between the measured wind speed and the wind speed at the nacelle, obtaining a pitch compensation value based on the pitch coefficient, and controlling the pitch of the wind turbine based on the pitch compensation value.

[0004] With the advancement of wind power technology, the capacity of individual wind turbines has continued to increase, with rotor diameters exceeding 200 meters. The aerodynamic and mechanical loads on key turbine components (such as blades, towers, and gearboxes) have increased exponentially. Traditional wind measurement equipment (such as the anemometer at the rear of the nacelle) suffers from measurement lag and susceptibility to blade obstruction. This results in the inability of turbine control strategies to respond promptly to complex wind conditions (such as turbulence, wind shear, and gusts), exacerbating component fatigue damage and increasing operation and maintenance costs.

[0005] Existing feedback control strategies based on single-point wind measurement (such as PID pitch control) have limited effectiveness in suppressing dynamic loads, particularly in scenarios with three-dimensional wind field inhomogeneity, where blade load imbalance is a prominent issue. While some research has attempted to incorporate lidar feedforward control, these strategies have not fully utilized its three-dimensional wind field scanning capabilities and lack the deep synergy between multi-source data fusion and intelligent algorithms, leaving much room for improvement in load reduction effectiveness. Summary of the Invention

[0006] In order to solve the above technical problems, the present invention aims to provide a method for optimizing load reduction of wind farm units based on real-time wind measurement by a nacelle-type laser radar, comprising the following steps:

[0007] Step s1: Perform coordinate system conversion preprocessing on the scanning point data collected by the cabin-mounted lidar, delineate the three-dimensional area, obtain the comprehensive wind speed of each grid point, perform turbulence intensity and wind shear analysis on the comprehensive wind speed of each grid point, and construct three-dimensional wind field sequence data;

[0008] Step s2: Perform heterogeneous fusion of the wind turbine's multi-source sensor data and 3D wind field sequence data to construct a load estimation model, and output a blade root bending moment prediction value sequence, a tower bottom vibration acceleration prediction value sequence, and a 3D wind field prediction sequence data for the next control cycle;

[0009] Step s3: Perform real-time detection on the blade root bending moment prediction value sequence and the tower bottom vibration acceleration prediction value sequence, and determine whether to execute step s4 based on the detection results;

[0010] Step s4: Construct a multi-objective optimization function that includes blade root bending moment, tower bottom vibration acceleration, and power generation. Set the dynamic weight coefficient of the multi-objective optimization function according to the three-dimensional wind field prediction sequence data. Based on the multi-objective optimization function, use a multi-objective genetic algorithm to obtain the optimal control variables.

[0011] Step s5: According to the optimal control variables and the vertical wind shear in the three-dimensional area, the pitch angle of each blade is adjusted differentially to obtain a personalized variable pitch angle sequence for each blade.

[0012] Furthermore, the radar coordinate system ( r -x r y r z r ), the geometric center of the cabin-type laser radar is used as the origin of the radar coordinate system. The geometric center of the radar antenna is the physical reference point for the emission and reception of the laser beam. The spherical coordinates of all scanning points are calculated with this point as the origin. The laser emission direction of the cabin-type laser radar (the direction of the incoming flow, pointing to the front of the impeller) is used as the x-axis of the radar coordinate system. r Axis, y of the radar coordinate system r The axis is horizontal and perpendicular to the right r Axis, z of the radar coordinate system r The axis is vertically upward, opposite to the direction of gravity;

[0013] Construct wind turbine coordinate system (O w -XYZ), with the hub center of the fan as the origin of the fan coordinate system, the X axis of the fan coordinate system is rThe axes are parallel and point to the front of the impeller. The Y axis of the fan coordinate system is horizontal and perpendicular to the X axis. The Z axis of the fan coordinate system is vertically upward, opposite to the direction of gravity.

[0014] Furthermore, the scanning point data collected by the nacelle-type lidar is pre-processed by coordinate system conversion, a three-dimensional area is delineated, and the integrated wind speed of each grid point is obtained. The process includes:

[0015] Acquire a number of scanning point data in a radar coordinate system collected by a nacelle-type laser radar in a current control period, and convert the number of scanning point data in the radar coordinate system into a number of scanning point data in a wind turbine coordinate system;

[0016] The spherical coordinates of each scanning point in the radar coordinate system are Where: R is the scanning distance (such as 100-500 meters, interval 10 meters), θ is the horizontal azimuth (such as -25° to +25°, interval 0.5°), is the vertical elevation angle (0°, 15°, 30°);

[0017] The spherical coordinates of the radar coordinate system Transform to Cartesian coordinate system (x r 、y r 、z r ):

[0018] Define the installation position of the nacelle-mounted lidar in the wind turbine coordinate system as (X0, Y0, Z0). Considering the offset of the radar installation position (X0, Y0, Z0), the coordinates of the scanning point in the wind turbine coordinate system are:

[0019]

[0020] Taking the hub center of the fan as the origin, a three-dimensional area is defined in front of the fan impeller, and the three-dimensional area is divided into several grid points;

[0021] For example: X direction: 100-500 meters (incoming flow direction, step length 10 meters), a total of 41 grids;

[0022] Y direction: -120m to +120m (impeller width ±1.2 times radius, step length 10m), a total of 25 grids;

[0023] Z direction: 0-200 meters (hub height ±100 meters, step length 10 meters), a total of 21 grids, forming 41×25×21=21525 grid points;

[0024] The wind speed of each grid point is obtained based on the data of several scanning points in the wind turbine coordinate system, and the wind speed of each grid point is processed by inverse distance weighted average to obtain the comprehensive wind speed of each grid point;

[0025] Specifically, the Doppler frequency shift of scattered light measured by the cabin-type laser radar is used to obtain the value of each grid point (X g 、Y g , Z g ) radial velocity Vr g ;

[0026]

[0027] Where δ represents the laser wavelength, Δf g represents the Doppler shift of the g-th grid point;

[0028] Based on the radial velocity Vr g Get each grid point (X g 、Y g , Z g )’s three-dimensional wind speed components:

[0029]

[0030] Among them, V g,X is the wind speed component along the incoming flow direction at the g-th grid point, V g,Y is the horizontal transverse wind speed component at the g-th grid point, V g,Z is the vertical wind speed component;

[0031] Each grid point (X g 、Y g , Z g ) wind speed V g for:

[0032]

[0033] Wind speed V g The corresponding wind direction angle is:

[0034] θ g =arctan2(V g,Y ,V g,X )(horizontal wind direction angle, with due north being 0°);

[0035] (vertical inclination);

[0036] For each grid point (X g 、Y g , Z g ), its comprehensive wind speed V1 g (By the adjacent scanning points (X i 、Y i , Z i ) wind speed V i The weighted average is:

[0037]

[0038] Among them, d i is the grid point (X g 、Y g , Z g ) and scanning point (X i 、Y i , Z i ) is the Euclidean distance, nq is the number of adjacent scanning points;

[0039] Comprehensive wind speed V1 g The corresponding wind direction angle is:

[0040] Horizontal wind direction angle (relative to true north): θ g =arctan2(V1 g,Y ,V1 g,X );

[0041] Vertical inclination angle (angle between wind direction and horizontal plane):

[0042] Furthermore, the turbulence intensity and wind shear analysis of the integrated wind speed at each grid point is performed to construct the three-dimensional wind field sequence data. The process includes:

[0043] Perform turbulence intensity and wind shear analysis on the comprehensive wind speed at each grid point at each moment in the current control cycle to obtain the turbulence intensity, vertical wind shear and horizontal wind shear at each grid point at each moment;

[0044] Turbulence intensity: the ratio of the standard deviation of wind speed to the average wind speed in a three-dimensional area: Among them, I u is the turbulence intensity, σ V is the standard deviation of wind speed, is the average wind speed;

[0045] Vertical wind shear (ΔV Z ):Wind speed difference at different Z height grid points in the same XY plane: ΔV Z =V Z＝100m -V Z＝0m ;

[0046] Horizontal wind shear (ΔV X ):The wind speed difference between grid points at different Y positions in the same XZ plane: ΔV X =V Y＝+100m -V Y＝-100m ;

[0047] Three-dimensional wind field sequence data are constructed based on the comprehensive wind speed, turbulence intensity, vertical wind shear and horizontal wind shear at each grid point at each moment.

[0048] Furthermore, the process of heterogeneously fusing the wind turbine's multi-source sensor data (including blade strain gauge load sequence data, tower vibration sequence data, and transmission chain torque sequence data) with the three-dimensional wind field sequence data to construct a load estimation model and output the blade root bending moment prediction value sequence, tower bottom vibration acceleration prediction value sequence, and three-dimensional wind field prediction sequence data for the next control cycle includes:

[0049] Obtain blade strain gauge load sequence data, tower vibration sequence data, transmission chain torque sequence data, and three-dimensional wind field sequence data of the wind turbine over several historical control cycles;

[0050] The blade strain gauge load data series, tower vibration data series, and transmission chain torque data series of the wind turbine in each historical control period are time-aligned with the three-dimensional wind field sequence data to generate a multi-source data set in a unified time coordinate system.

[0051] A load estimation model is constructed based on deep learning. Multi-source data sets within each historical control cycle are used as training sets and test sets. The training sets are input into the load estimation model for training until the loss function training is stable. The model parameters are saved and the load estimation model is tested with the test set until it meets the preset requirements. The load estimation model is then output.

[0052] The blade strain gauge load data sequence, tower vibration data sequence and transmission chain torque data sequence of the current control cycle are time-aligned with the three-dimensional wind field sequence data to generate a multi-source data set in a unified time coordinate system and input it into the load estimation model. According to the load estimation model, the blade root bending moment prediction value sequence M for the next control cycle is output. b (t), tower bottom vibration acceleration prediction value sequence A s (t) and three-dimensional wind field prediction sequence data W(t)=[V1(t),I u (t),ΔV Z (t),ΔV X (t)].

[0053] Furthermore, the blade root bending moment prediction value sequence and the tower bottom vibration acceleration prediction value sequence are detected in real time, and the process of determining whether to execute step s4 according to the detection results includes:

[0054] Preset safety thresholds corresponding to the blade root bending moment and the tower bottom vibration acceleration, compare the blade root bending moment prediction value sequence and the tower bottom vibration acceleration prediction value sequence with the corresponding safety thresholds, and obtain the cumulative time when the blade root bending moment prediction value is greater than or equal to the corresponding safety threshold and the cumulative time when the tower bottom vibration acceleration prediction value is greater than or equal to the corresponding safety threshold;

[0055] A cumulative time threshold is preset. If the cumulative time during which the predicted value of the blade root bending moment is greater than or equal to the corresponding safety threshold is greater than the cumulative time threshold, or the cumulative time during which the predicted value of the tower bottom vibration acceleration is greater than or equal to the corresponding safety threshold is greater than the cumulative time threshold, step s4 is executed.

[0056] Furthermore, a multi-objective optimization function is constructed that includes the blade root bending moment, tower bottom vibration acceleration, and power generation. The dynamic weight coefficient of the multi-objective optimization function is set according to the three-dimensional wind field prediction sequence data. The process of obtaining the optimal control variables through a multi-objective genetic algorithm based on the multi-objective optimization function includes the following steps:

[0057] A multi-objective optimization function is constructed that includes blade root bending moment, tower bottom vibration acceleration, and power generation. Control variables and constraints are preset. Dynamic weight coefficients of blade root bending moment, tower bottom vibration acceleration, and power generation in the multi-objective optimization function are set based on three-dimensional wind field prediction sequence data. The control variables include a collective pitch angle sequence, a yaw angle sequence, and a generator torque sequence. Chromosome encoding is performed on the control variables, and a population is initialized to generate an initial population. Adaptive mutation that changes with the number of iterations is introduced to improve the multi-objective genetic algorithm.

[0058] Based on the initial population, constraints, three-dimensional wind field prediction sequence data and multi-objective optimization function, the optimal control variables are obtained through the improved multi-objective genetic algorithm.

[0059] Furthermore, the multi-objective optimization function is specifically:

[0060]

[0061] in:

[0062]

[0063] P(t)=T e (t)·ω e ;

[0064]

[0065] Among them, λ1, λ2, λ3 are the initial weight coefficients, N p is the control cycle duration (100ms), P represents the generated power, ρ is the air density, A is the blade swept area, is the average wind speed at time t, δ p (t) collective pitch angle at time t, φ(t) yaw angle at time t, T e (t) is the generator torque at the tth moment, C T (δ p (t)) represents the collective pitch angle δp The thrust coefficient at (t) is the collective pitch angle δ p (t) is a nonlinear function obtained through the fan aerodynamic model (such as blade element momentum theory), L is the blade arm, ΔC T (φ(t)) is the thrust coefficient deviation caused by the yaw angle φ(t), C P (δ p (t)) represents the collective pitch angle δ p (t) preset power coefficient, m x is the equivalent mass of the tower (kg), for example, the mass of a 10MW wind turbine tower is about 800-1000 tons, C y (φ(t)) is the preset lateral force coefficient corresponding to the yaw angle φ(t), is the rate of change of the generator torque at the tth moment, k drive is the transmission chain stiffness, ω e is the impeller speed. The above formulas are all calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained by simulating a large amount of data.

[0066] The constraints are:

[0067] Pitch rate ≤ 5° / s: Prevents pitch mechanism overload and mechanical damage (such as hydraulic system pressure exceeding the limit);

[0068] Yaw angular velocity ≤ 0.1° / s: reduces yaw bearing impact and yaw motor energy consumption;

[0069] Generator torque ≤ T_max: protects the gearbox and generator and prevents overload tripping (T_max is usually 1.2 times the rated torque).

[0070] The calculation formula of adaptive mutation that changes with the number of iterations is:

[0071]

[0072] Among them, D is the population diversity index, p m is the mutation rate, p m0 , p m1 are preset parameters.

[0073] The specific process of obtaining the optimal control variables through the improved multi-objective genetic algorithm includes:

[0074] The control variables are encoded as chromosomes. For example, a string of numbers is used to represent the collective pitch angle sequence, yaw angle sequence and generator torque sequence. A certain number of chromosomes are randomly generated to form an initial population. Each chromosome represents a possible control variable. The multi-objective optimization function J value corresponding to different chromosomes is obtained. The tournament selection method is used to select chromosomes with lower multi-objective optimization function J values ​​from the current population as parents. The parent chromosomes are crossed and some genes are exchanged to generate new daughter chromosomes. This simulates the exchange of biological genetic genes and generates new control variables. The daughter chromosomes are mutated and some genes are randomly changed to increase population diversity and avoid falling into local optimality. The above steps are repeated and iterated continuously until the termination conditions are met, such as reaching the maximum number of iterations or the fitness value no longer increases significantly, and the optimal control variables are output.

[0075] Furthermore, the process of setting the dynamic weight coefficient of the multi-objective optimization function according to the three-dimensional wind field prediction sequence data includes:

[0076] Obtain blade root bending moment sequences, tower bottom vibration acceleration sequences, and three-dimensional wind field sequence data within several historical control cycles, perform statistical analysis on the blade root bending moment sequences and tower bottom vibration acceleration sequences within several historical control cycles, and obtain aerodynamic load fluctuation coefficients and mechanical load fluctuation coefficients corresponding to different three-dimensional wind field sequence data;

[0077]

[0078] Among them, k a Indicates the aerodynamic load fluctuation coefficient, k m Indicates the mechanical load fluctuation coefficient;

[0079] Constructing a weight mapping comparison table based on the aerodynamic load fluctuation coefficient and the mechanical load fluctuation coefficient corresponding to different three-dimensional wind field sequence data, wherein the weight mapping comparison table includes weight coefficients of blade root bending moment, tower bottom vibration acceleration, and power generation corresponding to different three-dimensional wind field sequence data;

[0080] Furthermore, the process of obtaining weight coefficients of blade root bending moment, tower bottom vibration acceleration, and power generation power according to the aerodynamic load fluctuation coefficient and the mechanical load fluctuation coefficient includes:

[0081]

[0082] Among them, k1 is the slope coefficient, which controls the rate of change of weight (default k1 = 8), a1 is the threshold center, corresponding to λ1 = 0.5λ 1max k a value, λ 1max is the maximum upper limit of λ1, λ1 is the weight coefficient corresponding to the bending moment at the blade root;

[0083]

[0084] Among them, k2 is the slope coefficient, which controls the rate of change of weight (default k2 = 6), a2 is the threshold center, corresponding to λ2 = 0.5λ 2max k m value, λ 2max is the maximum upper limit of λ2, and λ2 is the weight coefficient corresponding to the vibration acceleration of the tower bottom;

[0085] λ3=λ2-λ1;

[0086] Among them, λ3 is the weight coefficient corresponding to the generated power;

[0087] According to the three-dimensional wind field prediction sequence data and the weight mapping comparison table, the dynamic weight coefficients of the blade root bending moment, tower bottom vibration acceleration and power generation in the multi-objective optimization function are obtained.

[0088] Step s4 seeks the global optimal solution for multiple variables, including pitch, yaw, and power, focusing on multi-objective balance (load minimization and power maximization). However, this approach is limited in its ability to handle local details of the wind field's spatial distribution characteristics (such as wind speed differences at individual blade positions). Step s4 requires simultaneous optimization of the pitch sequence for all three blades. Directly compensating the wind speed at each blade's spatial position (e.g., 150 scan points) would require exponential computational effort, making it difficult to meet the real-time requirements of a 100ms control cycle.

[0089] If only the global optimization in step s4 is relied upon, multiple iterative adjustments are required to converge. However, independent pitch compensation directly applies differentiated adjustments to each blade (e.g., +3° pitch angle for blade 1, -1.5° pitch angle for blade 2) through an explicit formula based on wind shear, achieving instant "what you see is what you get" load balancing. The response time is shortened from 500ms when relying solely on step s4 to 100ms (synchronized with the control cycle).

[0090] Furthermore, the pitch angles of the blades are adjusted differentially based on the optimal control variables and the vertical wind shear in the three-dimensional region. The process of obtaining a personalized pitch angle sequence for each blade includes:

[0091] Divide the blade into n equal segments (e.g., n=5), obtain the grid points corresponding to the midpoints of each segment in the three-dimensional region, obtain the vertical wind shear at the grid points corresponding to the midpoints of each segment, perform blade-length integration on the vertical wind shear at the grid points corresponding to the midpoints of each segment, and obtain the dynamic vertical wind shear corresponding to the blade;

[0092] The calculation formula for the blade length integral of the vertical wind shear at the grid point corresponding to the midpoint of each segment is:

[0093]

[0094] Where, ΔV Z,i (t) represents the dynamic vertical wind shear of the i-th blade at the t-th moment, R is the blade length, ΔV Z,i,k (t) is the vertical wind shear at the grid point corresponding to the midpoint of the kth segment of the i-th blade at time t;

[0095] The wind shear compensation coefficient is preset, and the collective pitch angle sequence in the optimal control variable is extracted. The collective pitch angle sequence is differentially adjusted according to the dynamic vertical wind shear and the wind shear compensation coefficient to obtain the personalized pitch angle sequence of the blade.

[0096] Furthermore, for the i-th blade (i=1, 2, 3), the calculation formula for differentially adjusting the collective pitch angle sequence according to the dynamic vertical wind shear and wind shear compensation coefficient is:

[0097] δ p,i (t) = δ p,avg (t)+k z ΔV Z,i (t);

[0098] Among them, δ p,i (t) is the personalized pitch angle of the i-th blade at the t-th moment, δ p,avg (t) is the collective pitch angle at the tth moment, k z is the preset wind shear compensation coefficient.

[0099] A wind farm unit load reduction optimization system based on nacelle-mounted laser radar real-time wind measurement includes a control center, which is communicatively connected to a data processing module, a load estimation module, a real-time detection module, a load optimization module, and an adaptive adjustment module;

[0100] The data processing module is used to perform coordinate system conversion preprocessing on the scanning point data collected by the cabin-mounted lidar, delineate the three-dimensional area, obtain the integrated wind speed of each grid point, perform turbulence intensity and wind shear analysis on the integrated wind speed of each grid point, and construct three-dimensional wind field sequence data;

[0101] The load estimation module is used to perform heterogeneous fusion of the wind turbine's multi-source sensor data and three-dimensional wind field sequence data, build a load estimation model, and output the blade root bending moment prediction value sequence, tower bottom vibration acceleration prediction value sequence, and three-dimensional wind field prediction sequence data for the next control cycle;

[0102] The real-time detection module is used to perform real-time detection on the blade root bending moment prediction value sequence and the tower bottom vibration acceleration prediction value sequence, and determine whether to execute the load optimization module based on the detection results;

[0103] The load optimization module is used to construct a multi-objective optimization function that includes blade root bending moment, tower bottom vibration acceleration, and power generation. The dynamic weight coefficient of the multi-objective optimization function is set according to the three-dimensional wind field prediction sequence data. Based on the multi-objective optimization function, the optimal control variables are obtained through a multi-objective genetic algorithm.

[0104] The adaptive adjustment module is used to differentially adjust the pitch angle of each blade according to the optimal control variable and the vertical wind shear in the three-dimensional area, and obtain a personalized variable pitch angle sequence for each blade.

[0105] Compared with the prior art, the present invention has the following beneficial effects:

[0106] 1. Through precise conversion between the radar coordinate system and the wind turbine coordinate system and inverse distance weighted interpolation, a three-dimensional wind field grid is constructed, with a wind measurement accuracy of ≤0.5m / s, more than three times higher than traditional single-point wind measurement (error ±1.5m / s). This system can capture spatial distribution characteristics such as vertical wind shear, horizontal wind shear, and turbulence intensity, providing more accurate input for load prediction. For example, in complex terrain wind fields, traditional wind towers cannot perceive local wind shear in real time. However, the present invention uses a three-dimensional grid to predict wind speed differences at different blade positions three seconds in advance, reducing load prediction errors to ≤8%, a significant reduction compared to existing technologies (error ≥15%).

[0107] 2. By integrating heterogeneous data from multiple sources, including lidar wind field data, blade strain gauges, tower vibration, and drive train torque, a load prediction model is constructed based on deep learning (e.g., LSTM). This model can predict blade root bending moment and tower vibration acceleration for the next control cycle in advance, providing a decision-making basis for active load reduction. Existing technologies often rely on single wind field data or simplified physical models. This invention combines data-driven and physical models to reduce load prediction errors by over 40% compared to traditional methods.

[0108] 3. Construct a multi-objective function encompassing blade bending moment, tower vibration acceleration, and power generation. A weight mapping table is constructed using historical data statistics. The weights of each objective are dynamically adjusted based on the real-time three-dimensional wind field (e.g., increasing load weighting at high wind speeds and emphasizing power weighting at low wind speeds), achieving a Pareto optimal balance between load reduction and power generation efficiency. For example, under extreme wind speeds of 25 m / s, dynamic weighting reduces peak blade bending moment by 22%, while keeping power generation losses to less than 5%, a significant improvement over traditional fixed-weight control (power loss ≥ 10%).

[0109] 4. A multi-objective genetic algorithm with adaptive mutation is introduced, which takes the collective pitch angle, yaw angle and generator torque as control variables, and solves the global optimal solution under the constraints (pitch rate ≤ 5° / s, yaw angular velocity ≤ 0.1° / s). Compared with the traditional MPC algorithm, the optimization efficiency is improved by 30%, and nonlinear constraint problems can be handled.

[0110] 5. Divide the blade into n equal segments, calculate dynamic vertical wind shear by integrating the blade length, and make differentiated adjustments to the collective pitch reference value (e.g., increasing the pitch angle when wind speed is higher in the upper section of the blade) to achieve precise load balance on each blade. For example, under the condition of vertical wind shear ΔVz = 3m / s, independent pitch control reduces blade load imbalance by 85% and tower vibration acceleration by 18%, significantly improving over traditional collective pitch control (imbalance reduction ≤ 50%).

[0111] 6. By real-time monitoring of blade bending moment and tower vibration acceleration and triggering load reduction strategies, the fatigue load of key components can be reduced by 25%-40%. It is expected that the unit operation and maintenance cycle will be extended from 3 months to 6 months, the service life will be extended by more than 20%, and the operation and maintenance costs will be reduced by 15%-20%.

[0112] 7. While reducing the load, the power generation capacity is maintained through dynamic weighting, and the annual equivalent full-load hours are increased by 3.2%-4.8%, with significant economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0113] Figure 1 This is a schematic diagram of a method for optimizing load reduction of wind farm units based on real-time wind measurement by a nacelle-mounted lidar according to an embodiment of the present application.

[0114] Figure 2 This is a schematic diagram of a wind farm unit load reduction optimization system based on real-time wind measurement by a nacelle-mounted lidar according to an embodiment of the present application. DETAILED DESCRIPTION

[0115] The following is a clear and complete description of the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0116] like Figure 1 As shown, the wind farm unit load reduction optimization method based on nacelle-type laser radar real-time wind measurement includes the following steps:

[0117] Step s1: Perform coordinate system conversion preprocessing on the scanning point data collected by the cabin-mounted lidar, delineate the three-dimensional area, obtain the comprehensive wind speed of each grid point, perform turbulence intensity and wind shear analysis on the comprehensive wind speed of each grid point, and construct three-dimensional wind field sequence data;

[0118] Step s2: Perform heterogeneous fusion of the wind turbine's multi-source sensor data and 3D wind field sequence data to construct a load estimation model, and output a blade root bending moment prediction value sequence, a tower bottom vibration acceleration prediction value sequence, and a 3D wind field prediction sequence data for the next control cycle;

[0119] Step s3: Perform real-time detection on the blade root bending moment prediction value sequence and the tower bottom vibration acceleration prediction value sequence, and determine whether to execute step s4 based on the detection results;

[0120] Step s4: Construct a multi-objective optimization function that includes blade root bending moment, tower bottom vibration acceleration, and power generation. Set the dynamic weight coefficient of the multi-objective optimization function according to the three-dimensional wind field prediction sequence data. Based on the multi-objective optimization function, use a multi-objective genetic algorithm to obtain the optimal control variables.

[0121] Step s5: According to the optimal control variables and the vertical wind shear in the three-dimensional area, the pitch angle of each blade is adjusted differentially to obtain a personalized variable pitch angle sequence for each blade.

[0122] It should be further explained that, in the specific implementation process, the radar coordinate system ( r -x r y r z r ), the geometric center of the cabin-type laser radar is used as the origin of the radar coordinate system. The geometric center of the radar antenna is the physical reference point for the emission and reception of the laser beam. The spherical coordinates of all scanning points are calculated with this point as the origin. The laser emission direction of the cabin-type laser radar (the direction of the incoming flow, pointing to the front of the impeller) is used as the x-axis of the radar coordinate system. r Axis, y of the radar coordinate system r The axis is horizontal and perpendicular to the right r Axis, z of the radar coordinate system r The axis is vertically upward, opposite to the direction of gravity;

[0123] Construct wind turbine coordinate system (O w -XYZ), with the hub center of the fan as the origin of the fan coordinate system, the X axis of the fan coordinate system is r The axes are parallel and point to the front of the impeller. The Y axis of the fan coordinate system is horizontal and perpendicular to the X axis. The Z axis of the fan coordinate system is vertically upward, opposite to the direction of gravity.

[0124] It should be further explained that, in the specific implementation process, the scanning point data collected by the cabin-mounted lidar is pre-processed by coordinate system conversion, the three-dimensional area is delineated, and the comprehensive wind speed of each grid point is obtained. The process includes:

[0125] The nacelle-mounted lidar adopts a "1 horizontal main beam + 2 vertical auxiliary beams" design. The main beam scans horizontally (±α°) to cover the impeller width, while the auxiliary beams scan the vertical wind profile at elevation angles of β1° and β2°, forming a "sector + layered" scanning coverage.

[0126] Take a 10MW wind turbine as an example:

[0127] Horizontal scanning range: ±25° (α=25°), angular resolution 0.5°, 101 scanning points per circle;

[0128] Vertical layer angle: 0° (main beam), 15° (lower layer), 30° (upper layer) (β1=15, β2=30), each layer is scanned horizontally independently;

[0129] Scanning distance: 100-500 meters, distance resolution 10 meters, single control cycle 100ms;

[0130] Acquire a number of scanning point data in a radar coordinate system collected by a nacelle-type laser radar in a current control period, and convert the number of scanning point data in the radar coordinate system into a number of scanning point data in a wind turbine coordinate system;

[0131] The spherical coordinates of each scanning point in the radar coordinate system are Where: R is the scanning distance (such as 100-500 meters, interval 10 meters), θ is the horizontal azimuth (such as -25° to +25°, interval 0.5°), is the vertical elevation angle (0°, 15°, 30°);

[0132] The spherical coordinates of the radar coordinate system Transform to Cartesian coordinate system (x r 、y r 、z r ):

[0133] Define the installation position of the nacelle-mounted lidar in the wind turbine coordinate system as (X0, Y0, Z0). Considering the offset of the radar installation position (X0, Y0, Z0), the coordinates of the scanning point in the wind turbine coordinate system are:

[0134]

[0135] Taking the hub center of the fan as the origin, a three-dimensional area is defined in front of the fan impeller, and the three-dimensional area is divided into several grid points;

[0136] For example: X direction: 100-500 meters (incoming flow direction, step length 10 meters), a total of 41 grids;

[0137] Y direction: -120m to +120m (impeller width ±1.2 times radius, step length 10m), a total of 25 grids;

[0138] Z direction: 0-200 meters (hub height ±100 meters, step length 10 meters), a total of 21 grids, forming 41×25×21=21525 grid points;

[0139] The wind speed of each grid point is obtained based on the data of several scanning points in the wind turbine coordinate system, and the wind speed of each grid point is processed by inverse distance weighted average to obtain the comprehensive wind speed of each grid point;

[0140] Specifically, the Doppler frequency shift of scattered light measured by the cabin-type laser radar is used to obtain the value of each grid point (X g 、Y g , Z g ) radial velocity Vr g ;

[0141]

[0142] Where δ represents the laser wavelength, Δf g represents the Doppler shift of the g-th grid point;

[0143] Based on the radial velocity Vr g Get each grid point (X g 、Y g , Z g )’s three-dimensional wind speed components:

[0144]

[0145] Among them, V g,X is the wind speed component along the incoming flow direction at the g-th grid point, V g,Y is the horizontal transverse wind speed component at the g-th grid point, V g,Z is the vertical wind speed component;

[0146] Each grid point (X g 、Y g , Z g ) wind speed V g for:

[0147]

[0148] Wind speed V g The corresponding wind direction angle is:

[0149] θ g =arctan2(V g,Y ,V g,X )(horizontal wind direction angle, with due north being 0°);

[0150] (vertical inclination);

[0151] For each grid point (X g 、Y g , Z g ), its comprehensive wind speed V1 g (By the adjacent scanning points (X i 、Yi , Z i ) wind speed V i The weighted average is:

[0152]

[0153] Among them, d i is the grid point (X g 、Y g , Z g ) and scanning point (X i 、Y i , Z i ) is the Euclidean distance, nq is the number of adjacent scanning points;

[0154] Comprehensive wind speed V1 g The corresponding wind direction angle is:

[0155] Horizontal wind direction angle (relative to true north): θ g =arctan2(V1 g,Y ,V1 g,X );

[0156] Vertical inclination angle (angle between wind direction and horizontal plane):

[0157] It should be further explained that, in the specific implementation process, the turbulence intensity and wind shear analysis of the integrated wind speed at each grid point and the construction of three-dimensional wind field sequence data include:

[0158] Perform turbulence intensity and wind shear analysis on the comprehensive wind speed at each grid point at each moment in the current control cycle to obtain the turbulence intensity, vertical wind shear and horizontal wind shear at each grid point at each moment;

[0159] Turbulence intensity: the ratio of the standard deviation of wind speed to the average wind speed in a three-dimensional area: Among them, I u is the turbulence intensity, σ V is the standard deviation of wind speed, is the average wind speed;

[0160] Vertical wind shear (ΔV Z ):Wind speed difference at different Z height grid points in the same XY plane: ΔV Z =V Z＝100m -V Z＝0m ;

[0161] Horizontal wind shear (ΔV X ):The wind speed difference between grid points at different Y positions in the same XZ plane: ΔV X =V Y＝+100m -V Y＝-100m ;

[0162] Three-dimensional wind field sequence data are constructed based on the comprehensive wind speed, turbulence intensity, vertical wind shear and horizontal wind shear at each grid point at each moment.

[0163] It should be further explained that, in the specific implementation process, the process of heterogeneously fusing the wind turbine's multi-source sensor data (including blade strain gauge load sequence data, tower vibration sequence data, and drive train torque sequence data) and 3D wind field sequence data to construct a load estimation model and output the blade root bending moment prediction value sequence, tower bottom vibration acceleration prediction value sequence, and 3D wind field prediction sequence data for the next control cycle includes the following:

[0164] Obtain blade strain gauge load sequence data, tower vibration sequence data, transmission chain torque sequence data, and three-dimensional wind field sequence data of the wind turbine over several historical control cycles;

[0165] The blade strain gauge load data series, tower vibration data series, and transmission chain torque data series of the wind turbine in each historical control period are time-aligned with the three-dimensional wind field sequence data to generate a multi-source data set in a unified time coordinate system.

[0166] A load estimation model is constructed based on deep learning. Multi-source data sets within each historical control cycle are used as training sets and test sets. The training sets are input into the load estimation model for training until the loss function training is stable. The model parameters are saved and the load estimation model is tested with the test set until it meets the preset requirements. The load estimation model is then output.

[0167] The blade strain gauge load data sequence, tower vibration data sequence and transmission chain torque data sequence of the current control cycle are time-aligned with the three-dimensional wind field sequence data to generate a multi-source data set in a unified time coordinate system and input it into the load estimation model. According to the load estimation model, the blade root bending moment prediction value sequence M for the next control cycle is output. b (t), tower bottom vibration acceleration prediction value sequence A s (t) and three-dimensional wind field prediction sequence data W(t)=[V1(t),I u (t),ΔV Z (t),ΔV X (t)].

[0168] Building a deep learning-based load estimation model is a complex process involving multiple steps, including model selection, training, validation, and testing. The following is a detailed supplementary explanation of this process:

[0169] The Long Short-Term Memory (LSTM) network, suitable for time series analysis, was chosen as the deep learning architecture. After finalizing the model architecture, the mean squared error loss function was selected as the optimization objective. The prepared training set was then fed into the selected deep learning model to begin training. During training, the weights were continuously updated using the backpropagation algorithm, gradually reducing the loss function until a steady state was reached. During this process, techniques such as early stopping were utilized to prevent overfitting. In addition to the basic training process, various model parameters, including the learning rate, batch size, and regularization coefficient, were tuned using grid search.

[0170] After model training is complete and parameter adjustments are complete, a final evaluation is performed on the test set to obtain the model's evaluation results. These results include classification metrics such as accuracy, recall, and F1 score. Based on the evaluation results on the test set, it is determined whether the model meets the expected standards. If the requirements are met, the model parameters are saved and prepared for deployment. If not, it is necessary to return to a previous stage to re-examine issues such as data quality, model structure, or training strategy.

[0171] It should be further explained that, in a specific implementation process, the blade root bending moment prediction value sequence and the tower bottom vibration acceleration prediction value sequence are detected in real time, and the process of determining whether to execute step s4 based on the detection results includes:

[0172] Preset safety thresholds corresponding to the blade root bending moment and the tower bottom vibration acceleration, compare the blade root bending moment prediction value sequence and the tower bottom vibration acceleration prediction value sequence with the corresponding safety thresholds, and obtain the cumulative time when the blade root bending moment prediction value is greater than or equal to the corresponding safety threshold and the cumulative time when the tower bottom vibration acceleration prediction value is greater than or equal to the corresponding safety threshold;

[0173] A cumulative time threshold is preset. If the cumulative time during which the predicted value of the blade root bending moment is greater than or equal to the corresponding safety threshold is greater than the cumulative time threshold, or the cumulative time during which the predicted value of the tower bottom vibration acceleration is greater than or equal to the corresponding safety threshold is greater than the cumulative time threshold, step s4 is executed.

[0174] It should be further explained that, in the specific implementation process, a multi-objective optimization function including blade root bending moment, tower bottom vibration acceleration, and power generation is constructed, and the dynamic weight coefficient of the multi-objective optimization function is set according to the three-dimensional wind field prediction sequence data. The process of obtaining the optimal control variables through the multi-objective genetic algorithm based on the multi-objective optimization function includes the following:

[0175] A multi-objective optimization function is constructed that includes blade root bending moment, tower bottom vibration acceleration, and power generation. Control variables and constraints are preset. Dynamic weight coefficients of blade root bending moment, tower bottom vibration acceleration, and power generation in the multi-objective optimization function are set based on three-dimensional wind field prediction sequence data. The control variables include a collective pitch angle sequence, a yaw angle sequence, and a generator torque sequence. Chromosome encoding is performed on the control variables, and a population is initialized to generate an initial population. Adaptive mutation that changes with the number of iterations is introduced to improve the multi-objective genetic algorithm.

[0176] Based on the initial population, constraints, three-dimensional wind field prediction sequence data and multi-objective optimization function, the optimal control variables are obtained through the improved multi-objective genetic algorithm.

[0177] It should be further explained that, in the specific implementation process, the multi-objective optimization function is specifically:

[0178]

[0179] in:

[0180]

[0181] P(t)=T e (t)·ω e ;

[0182]

[0183] Among them, λ1, λ2, λ3 are the initial weight coefficients, N p is the control cycle duration (100ms), P represents the generated power, ρ is the air density, A is the blade swept area, is the average wind speed at time t, δ p (t) collective pitch angle at time t, φ(t) yaw angle at time t, T e (t) is the generator torque at the tth moment, C T (δ p (t)) represents the collective pitch angle δ p The thrust coefficient at (t) is the collective pitch angle δ p (t) is a nonlinear function obtained through the fan aerodynamic model (such as blade element momentum theory), L is the blade arm, ΔC T (φ(t)) is the thrust coefficient deviation caused by the yaw angle φ(t), C P (δ p (t)) represents the collective pitch angle δ p (t) preset power coefficient, m xis the equivalent mass of the tower (kg), for example, the mass of a 10MW wind turbine tower is about 800-1000 tons, C y (φ(t)) is the preset lateral force coefficient corresponding to the yaw angle φ(t), is the rate of change of the generator torque at the tth moment, k drive is the transmission chain stiffness, ω e is the impeller speed. The above formulas are all calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained by simulating a large amount of data.

[0184] The constraints are:

[0185] Pitch rate ≤ 5° / s: Prevents pitch mechanism overload and mechanical damage (such as hydraulic system pressure exceeding the limit);

[0186] Yaw angular velocity ≤ 0.1° / s: reduces yaw bearing impact and yaw motor energy consumption;

[0187] Generator torque ≤ T_max: protects the gearbox and generator and prevents overload tripping (T_max is usually 1.2 times the rated torque).

[0188] The calculation formula of adaptive mutation that changes with the number of iterations is:

[0189]

[0190] Among them, D is the population diversity index, p m is the mutation rate, p m0 , p m1 are preset parameters.

[0191] The specific process of obtaining the optimal control variables through the improved multi-objective genetic algorithm includes:

[0192] The control variables are encoded as chromosomes. For example, a string of numbers is used to represent the collective pitch angle sequence, yaw angle sequence and generator torque sequence. A certain number of chromosomes are randomly generated to form an initial population. Each chromosome represents a possible control variable. The multi-objective optimization function J value corresponding to different chromosomes is obtained. The tournament selection method is used to select chromosomes with lower multi-objective optimization function J values ​​from the current population as parents. The parent chromosomes are crossed and some genes are exchanged to generate new daughter chromosomes. This simulates the exchange of biological genetic genes and generates new control variables. The daughter chromosomes are mutated and some genes are randomly changed to increase population diversity and avoid falling into local optimality. The above steps are repeated and iterated continuously until the termination conditions are met, such as reaching the maximum number of iterations or the fitness value no longer increases significantly, and the optimal control variables are output.

[0193] It should be further explained that, in the specific implementation process, the process of setting the dynamic weight coefficient of the multi-objective optimization function according to the three-dimensional wind field prediction sequence data includes:

[0194] Obtain blade root bending moment sequences, tower bottom vibration acceleration sequences, and three-dimensional wind field sequence data within several historical control cycles, perform statistical analysis on the blade root bending moment sequences and tower bottom vibration acceleration sequences within several historical control cycles, and obtain aerodynamic load fluctuation coefficients and mechanical load fluctuation coefficients corresponding to different three-dimensional wind field sequence data;

[0195]

[0196] Among them, k a Indicates the aerodynamic load fluctuation coefficient, k m Indicates the mechanical load fluctuation coefficient;

[0197] Constructing a weight mapping comparison table based on the aerodynamic load fluctuation coefficient and the mechanical load fluctuation coefficient corresponding to different three-dimensional wind field sequence data, wherein the weight mapping comparison table includes weight coefficients of blade root bending moment, tower bottom vibration acceleration, and power generation corresponding to different three-dimensional wind field sequence data;

[0198] It should be further explained that, in a specific implementation process, the process of obtaining the weight coefficients of the blade root bending moment, the tower bottom vibration acceleration, and the power generation power according to the aerodynamic load fluctuation coefficient and the mechanical load fluctuation coefficient includes:

[0199]

[0200] Among them, k1 is the slope coefficient, which controls the rate of change of weight (default k1 = 8), a1 is the threshold center, corresponding to λ1 = 0.5λ 1max k a value, λ 1max is the maximum upper limit of λ1, λ1 is the weight coefficient corresponding to the bending moment at the blade root;

[0201]

[0202] Among them, k2 is the slope coefficient, which controls the rate of change of weight (default k2 = 6), a2 is the threshold center, corresponding to λ2 = 0.5λ 2max k m value, λ 2max is the maximum upper limit of λ2, and λ2 is the weight coefficient corresponding to the vibration acceleration of the tower bottom;

[0203] λ3=λ2-λ1;

[0204] Among them, λ3 is the weight coefficient corresponding to the generated power;

[0205] According to the three-dimensional wind field prediction sequence data and the weight mapping comparison table, the dynamic weight coefficients of the blade root bending moment, tower bottom vibration acceleration and power generation in the multi-objective optimization function are obtained.

[0206] Step s4 seeks the global optimal solution for multiple variables, including pitch, yaw, and power, focusing on multi-objective balance (load minimization and power maximization). However, this approach is limited in its ability to handle local details of the wind field's spatial distribution characteristics (such as wind speed differences at individual blade positions). Step s4 requires simultaneous optimization of the pitch sequence for all three blades. Directly compensating the wind speed at each blade's spatial position (e.g., 150 scan points) would require exponential computational effort, making it difficult to meet the real-time requirements of a 100ms control cycle.

[0207] If only relying on the global optimization of step s4, multiple iterative adjustments are required to converge. Independent pitch compensation directly applies differentiated adjustments to each blade (such as blade 1 pitch angle +3°, blade 2 pitch angle -1.5°) through an explicit formula based on wind shear, achieving instant load balancing. The response time is shortened from 500ms when relying only on step s4 to 100ms (synchronized with the control cycle).

[0208] It should be further explained that, in the specific implementation process, the pitch angle of each blade is adjusted differently according to the optimal control variable and the vertical wind shear in the three-dimensional area, and the process of obtaining the personalized pitch angle sequence for each blade includes:

[0209] Divide the blade into n equal segments (e.g., n=5), obtain the grid points corresponding to the midpoints of each segment in the three-dimensional region, obtain the vertical wind shear at the grid points corresponding to the midpoints of each segment, perform blade-length integration on the vertical wind shear at the grid points corresponding to the midpoints of each segment, and obtain the dynamic vertical wind shear corresponding to the blade;

[0210] The calculation formula for the blade length integral of the vertical wind shear at the grid point corresponding to the midpoint of each segment is:

[0211]

[0212] Where, ΔV Z,i (t) represents the dynamic vertical wind shear of the i-th blade at the t-th moment, R is the blade length, ΔV Z,i,k (t) is the vertical wind shear at the grid point corresponding to the midpoint of the kth segment of the i-th blade at time t;

[0213] The wind shear compensation coefficient is preset, and the collective pitch angle sequence in the optimal control variable is extracted. The collective pitch angle sequence is differentially adjusted according to the dynamic vertical wind shear and the wind shear compensation coefficient to obtain the personalized pitch angle sequence of the blade.

[0214] It should be further explained that, in the specific implementation process, for the i-th blade (i=1, 2, 3), the calculation formula for differentially adjusting the collective pitch angle sequence according to the dynamic vertical wind shear and wind shear compensation coefficient is:

[0215] δ p,i (t) = δ p,avg (t)+k z ΔV Z,i (t);

[0216] Among them, δ p,i (t) is the personalized pitch angle of the i-th blade at the t-th moment, δ p,avg (t) is the collective pitch angle at the tth moment, k z is the preset wind shear compensation coefficient.

[0217] like Figure 2 As shown, a wind farm unit load reduction optimization system based on nacelle-type laser radar real-time wind measurement includes a control center, which is communicatively connected to a data processing module, a load estimation module, a real-time detection module, a load optimization module, and an adaptive adjustment module;

[0218] The data processing module is used to perform coordinate system conversion preprocessing on the scanning point data collected by the cabin-mounted lidar, delineate the three-dimensional area, obtain the integrated wind speed of each grid point, perform turbulence intensity and wind shear analysis on the integrated wind speed of each grid point, and construct three-dimensional wind field sequence data;

[0219] The load estimation module is used to perform heterogeneous fusion of the wind turbine's multi-source sensor data and three-dimensional wind field sequence data, build a load estimation model, and output the blade root bending moment prediction value sequence, tower bottom vibration acceleration prediction value sequence, and three-dimensional wind field prediction sequence data for the next control cycle;

[0220] The real-time detection module is used to perform real-time detection on the blade root bending moment prediction value sequence and the tower bottom vibration acceleration prediction value sequence, and determine whether to execute the load optimization module based on the detection results;

[0221] The load optimization module is used to construct a multi-objective optimization function that includes blade root bending moment, tower bottom vibration acceleration, and power generation. The dynamic weight coefficient of the multi-objective optimization function is set according to the three-dimensional wind field prediction sequence data. Based on the multi-objective optimization function, the optimal control variables are obtained through a multi-objective genetic algorithm.

[0222] The adaptive adjustment module is used to differentially adjust the pitch angle of each blade according to the optimal control variable and the vertical wind shear in the three-dimensional area, and obtain a personalized variable pitch angle sequence for each blade.

[0223] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A wind farm unit load reduction optimization method based on real-time wind measurement using a nacelle-mounted laser radar, characterized in that: The following steps are involved: Step s1: Perform coordinate system conversion preprocessing on the scanning point data collected by the cabin-mounted lidar, delineate the three-dimensional area, obtain the comprehensive wind speed of each grid point, perform turbulence intensity and wind shear analysis on the comprehensive wind speed of each grid point, and construct three-dimensional wind field sequence data; Step s2: Perform heterogeneous fusion of the wind turbine's multi-source sensor data and 3D wind field sequence data to construct a load estimation model, and output a blade root bending moment prediction value sequence, a tower bottom vibration acceleration prediction value sequence, and a 3D wind field prediction sequence data for the next control cycle; Step s3: Perform real-time detection on the blade root bending moment prediction value sequence and the tower bottom vibration acceleration prediction value sequence, and determine whether to execute step s4 based on the detection results; Step s4: Construct a multi-objective optimization function that includes blade root bending moment, tower bottom vibration acceleration, and power generation. Set the dynamic weight coefficient of the multi-objective optimization function according to the three-dimensional wind field prediction sequence data. Based on the multi-objective optimization function, use a multi-objective genetic algorithm to obtain the optimal control variables. Step s5: According to the optimal control variables and the vertical wind shear in the three-dimensional area, the pitch angle of each blade is adjusted differentially to obtain a personalized variable pitch angle sequence for each blade.

2. The method for optimizing load reduction of wind farm units based on real-time wind measurement using a nacelle-type laser radar according to claim 1 is characterized in that: Construct radar coordinate system (O r -x r y r z r ), the geometric center of the cabin-type laser radar is used as the origin of the radar coordinate system, and the laser emission direction of the cabin-type laser radar is used as the x-axis of the radar coordinate system. r Axis, y of the radar coordinate system r The axis is horizontal and perpendicular to the right r Axis, z of the radar coordinate system r The axis is vertically upward; Construct wind turbine coordinate system (O w -XYZ), with the hub center of the fan as the origin of the fan coordinate system, the X axis of the fan coordinate system is r The axes are parallel, the Y axis of the fan coordinate system is horizontal to the right and perpendicular to the X axis, and the Z axis of the fan coordinate system is vertically upward.

3. The method for optimizing load reduction of wind farm units based on real-time wind measurement using a nacelle-type laser radar according to claim 2, characterized in that: The process of performing coordinate system conversion preprocessing on the scan point data collected by the cabin-mounted lidar, demarcating the three-dimensional area, and obtaining the integrated wind speed at each grid point includes: Acquire a number of scanning point data in a radar coordinate system collected by a nacelle-type laser radar in a current control period, and convert the number of scanning point data in the radar coordinate system into a number of scanning point data in a wind turbine coordinate system; Taking the hub center of the fan as the origin, a three-dimensional area is defined in front of the fan impeller, and the three-dimensional area is divided into several grid points; The wind speed of each grid point is obtained according to the data of several scanning points in the wind turbine coordinate system, and the wind speed of each grid point is processed by inverse distance weighted average to obtain the comprehensive wind speed of each grid point.

4. The method for optimizing load reduction of wind farm units based on real-time wind measurement using a nacelle-type laser radar according to claim 3 is characterized in that: The process of analyzing the turbulence intensity and wind shear of the integrated wind speed at each grid point and constructing three-dimensional wind field sequence data includes: Perform turbulence intensity and wind shear analysis on the comprehensive wind speed at each grid point at each moment in the current control cycle to obtain the turbulence intensity, vertical wind shear and horizontal wind shear at each grid point at each moment; Three-dimensional wind field sequence data are constructed based on the comprehensive wind speed, turbulence intensity, vertical wind shear and horizontal wind shear at each grid point at each moment.

5. The method for optimizing load reduction of wind farm units based on real-time wind measurement using a nacelle-type laser radar according to claim 4 is characterized in that: The process of heterogeneously fusing the wind turbine's multi-source sensor data and three-dimensional wind field sequence data, building a load estimation model, and outputting a blade root bending moment prediction value sequence, a tower bottom vibration acceleration prediction value sequence, and a three-dimensional wind field prediction sequence data for the next control cycle includes: Obtain blade strain gauge load sequence data, tower vibration sequence data, transmission chain torque sequence data, and three-dimensional wind field sequence data of the wind turbine over several historical control cycles; The blade strain gauge load data series, tower vibration data series, and transmission chain torque data series of the wind turbine in each historical control period are time-aligned with the three-dimensional wind field sequence data to generate a multi-source data set in a unified time coordinate system. A load estimation model is constructed based on deep learning. Multi-source data sets within each historical control cycle are used as training sets and test sets. The training sets are input into the load estimation model for training until the loss function training is stable. The model parameters are saved and the load estimation model is tested with the test set until it meets the preset requirements. The load estimation model is then output. The blade strain gauge load data sequence, tower vibration data sequence, and transmission chain torque data sequence of the current control cycle are time-aligned with the three-dimensional wind field sequence data to generate a multi-source data set in a unified time coordinate system and input it into the load estimation model. According to the load estimation model, the blade root bending moment prediction value sequence, tower bottom vibration acceleration prediction value sequence, and three-dimensional wind field prediction sequence data of the next control cycle are output.

6. The method for optimizing load reduction of wind farm units based on real-time wind measurement using a nacelle-type laser radar according to claim 5 is characterized in that: The process of performing real-time detection on the blade root bending moment prediction value sequence and the tower bottom vibration acceleration prediction value sequence and determining whether to execute step s4 based on the detection results includes: Preset safety thresholds corresponding to the blade root bending moment and the tower bottom vibration acceleration, compare the blade root bending moment prediction value sequence and the tower bottom vibration acceleration prediction value sequence with the corresponding safety thresholds, and obtain the cumulative time when the blade root bending moment prediction value is greater than or equal to the corresponding safety threshold and the cumulative time when the tower bottom vibration acceleration prediction value is greater than or equal to the corresponding safety threshold; A cumulative time threshold is preset. If the cumulative time during which the predicted value of the blade root bending moment is greater than or equal to the corresponding safety threshold is greater than the cumulative time threshold, or the cumulative time during which the predicted value of the tower bottom vibration acceleration is greater than or equal to the corresponding safety threshold is greater than the cumulative time threshold, step s4 is executed.

7. The method for optimizing wind farm unit load reduction based on real-time wind measurement using a nacelle-type laser radar according to claim 6, characterized in that: A multi-objective optimization function is constructed that includes blade root bending moment, tower bottom vibration acceleration, and power generation. The dynamic weight coefficients of the multi-objective optimization function are set according to the three-dimensional wind field prediction sequence data. The process of obtaining the optimal control variables based on the multi-objective optimization function through a multi-objective genetic algorithm includes the following: A multi-objective optimization function is constructed that includes blade root bending moment, tower bottom vibration acceleration, and power generation. Control variables and constraints are preset. Dynamic weight coefficients of blade root bending moment, tower bottom vibration acceleration, and power generation in the multi-objective optimization function are set based on three-dimensional wind field prediction sequence data. The control variables include a collective pitch angle sequence, a yaw angle sequence, and a generator torque sequence. Chromosome encoding is performed on the control variables, and a population is initialized to generate an initial population. Adaptive mutation that changes with the number of iterations is introduced to improve the multi-objective genetic algorithm. Based on the initial population, constraints, three-dimensional wind field prediction sequence data and multi-objective optimization function, the optimal control variables are obtained through the improved multi-objective genetic algorithm.

8. The method for optimizing load reduction of wind farm units based on real-time wind measurement using a nacelle-type laser radar according to claim 7 is characterized in that: The process of setting the dynamic weight coefficients of the multi-objective optimization function based on the three-dimensional wind field prediction sequence data includes: Obtain blade root bending moment sequences, tower bottom vibration acceleration sequences, and three-dimensional wind field sequence data within several historical control cycles, perform statistical analysis on the blade root bending moment sequences and tower bottom vibration acceleration sequences within several historical control cycles, and obtain aerodynamic load fluctuation coefficients and mechanical load fluctuation coefficients corresponding to different three-dimensional wind field sequence data; Constructing a weight mapping comparison table based on the aerodynamic load fluctuation coefficient and the mechanical load fluctuation coefficient corresponding to different three-dimensional wind field sequence data, wherein the weight mapping comparison table includes weight coefficients of blade root bending moment, tower bottom vibration acceleration, and power generation corresponding to different three-dimensional wind field sequence data; According to the three-dimensional wind field prediction sequence data and the weight mapping comparison table, the dynamic weight coefficients of the blade root bending moment, tower bottom vibration acceleration and power generation in the multi-objective optimization function are obtained.

9. The method for optimizing load reduction of wind farm units based on real-time wind measurement using a nacelle-type laser radar according to claim 8, characterized in that: The process of differentially adjusting the pitch angle of each blade based on the optimal control variable and the vertical wind shear in the three-dimensional region to obtain a personalized pitch angle sequence for each blade includes: Divide the blade into n equal segments, obtain the grid points corresponding to the midpoints of each segment in the three-dimensional area, obtain the vertical wind shear of the grid points corresponding to the midpoints of each segment, perform blade length integration on the vertical wind shear of the grid points corresponding to the midpoints of each segment, and obtain the dynamic vertical wind shear corresponding to the blade; The wind shear compensation coefficient is preset, and the collective pitch angle sequence in the optimal control variable is extracted. The collective pitch angle sequence is differentially adjusted according to the dynamic vertical wind shear and the wind shear compensation coefficient to obtain the personalized pitch angle sequence of the blade.

10. A wind farm unit load reduction optimization system based on nacelle-mounted laser radar real-time wind measurement, specifically applied to a wind farm unit load reduction optimization method based on nacelle-mounted laser radar real-time wind measurement according to any one of claims 1 to 9, characterized in that: The control center includes a data processing module, a load estimation module, a real-time detection module, a load optimization module, and an adaptive adjustment module. The data processing module is used to perform coordinate system conversion preprocessing on the scanning point data collected by the cabin-mounted lidar, delineate the three-dimensional area, obtain the integrated wind speed of each grid point, perform turbulence intensity and wind shear analysis on the integrated wind speed of each grid point, and construct three-dimensional wind field sequence data; The load estimation module is used to perform heterogeneous fusion of the wind turbine's multi-source sensor data and three-dimensional wind field sequence data, build a load estimation model, and output the blade root bending moment prediction value sequence, tower bottom vibration acceleration prediction value sequence, and three-dimensional wind field prediction sequence data for the next control cycle; The real-time detection module is used to perform real-time detection on the blade root bending moment prediction value sequence and the tower bottom vibration acceleration prediction value sequence, and determine whether to execute the load optimization module based on the detection results; The load optimization module is used to construct a multi-objective optimization function that includes blade root bending moment, tower bottom vibration acceleration, and power generation. The dynamic weight coefficient of the multi-objective optimization function is set according to the three-dimensional wind field prediction sequence data. Based on the multi-objective optimization function, the optimal control variables are obtained through a multi-objective genetic algorithm. The adaptive adjustment module is used to differentially adjust the pitch angle of each blade according to the optimal control variable and the vertical wind shear in the three-dimensional area, and obtain a personalized variable pitch angle sequence for each blade.

Images