Laser radar assisted wind power generation wind shear modeling method

Abstract

The invention discloses a laser radar-assisted wind power generation wind shear modeling method, and relates to the technical field of wind shear modeling, and the method comprises the steps: collecting the wind speed measurement data of a laser radar at a plurality of heights in a target fan region, and constructing a measurement volume integral operator matrix; constructing a unit induction field correction model based on large eddy simulation in combination with a gradient boosting tree; coupling the measurement volume integral operator with the unit induction field correction model, and obtaining a high-precision wind speed vertical section and a corresponding wind shear parameter through inversion; on the basis of the high-precision wind speed vertical section and the corresponding wind shear parameters, physical constraints are fused, a wind shear model is generated, and uncertainty evaluation is carried out; according to the wind shear modeling method, the problems that the measurement error of the laser radar is difficult to correct, the wind shear modeling lacks the physical constraint and the result credibility cannot be quantified are solved by constructing the volume measurement integral operator and the unit induction field correction model for coupling inversion and fusing the physical constraint and the Bayesian uncertainty evaluation method.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of wind shear modeling, in particular to a laser radar assisted wind shear modeling method for wind power generation. BACKGROUND

[0002] As an important part of renewable energy, the power generation efficiency and operation safety of wind power generation depend to a large extent on the accurate acquisition and modeling of wind field characteristic parameters. Among them, wind shear, as an important parameter describing the variation of wind speed with height, not only affects the aerodynamic load and power output characteristics of the wind turbine, but also directly relates to the fatigue life of the wind turbine and the development of the overall power optimization strategy of the wind farm. Therefore, how to realize high-precision modeling and dynamic evaluation of wind shear has become a core problem of wind energy resource assessment and intelligent wind turbine control.

[0003] For example, the invention patent with the announcement number CN119778196B announces a wind turbine wind speed and direction acquisition method, which comprises the following steps: acquiring the wind speed and direction data of the wind turbine operation area in real time through a sensor; eliminating high-frequency disturbances based on a turbulence model; correcting the wind speed and direction data based on a wind shear model and compensating the wind speed gradient; constructing a wind field optimization model and dynamically optimizing the wind speed and direction data using a global optimization method; based on multi-scale decomposition, macro trends and local details are fused to generate the final wind speed and direction data; and outputting the high-precision corrected wind speed and direction data. By introducing turbulence correction and wind shear model, the main measurement errors caused by turbulence disturbance and height difference are eliminated for the complex wind field characteristics in the wind turbine operation area. This layered correction method effectively reduces the randomness and deviation of the wind speed and direction data, so that the output data has high reliability and accuracy in various wind field environments.

[0004] In the prior art, when the laser radar measures the wind speed profile, due to the mutual superposition of the measurement volume effect and the wind turbine induced disturbance, it is difficult to effectively separate them, resulting in systematic deviation of the measurement data and affecting the inversion accuracy. At the same time, the traditional wind speed vertical profile inversion algorithm is prone to instability in the case of strong noise or improper selection of regularization parameters, and it is difficult to obtain high-precision results. In addition, the existing wind shear modeling methods mostly rely on empirical formulas or pure data-driven models, lack of constraints on the physical laws of the atmospheric boundary layer, resulting in deviation of the extrapolation results from the actual wind field characteristics. Moreover, the current uncertainty evaluation mechanism is not perfect, and it cannot systematically quantify the confidence interval and risk level of the model extrapolation results, thereby limiting the reliable application of wind shear modeling results in wind turbine control and power prediction. In view of the above problems, the present application proposes a solution. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present application provide a laser radar assisted wind power wind shear modeling method, which solves the problems of laser radar measurement error difficult to correct, wind shear modeling lacking physical constraints and result reliability being unquantifiable by coupling inversion of a measurement volume integral operator and a unit induced field correction model and fusing physical constraints and Bayesian uncertainty evaluation.

[0006] To achieve the above object, the present application provides the following technical solutions: In a first aspect, the present application provides a laser radar assisted wind power wind shear modeling method, which comprises: collecting wind speed measurement data of a laser radar at several heights in a target wind turbine area, and constructing a measurement volume integral operator matrix; constructing a unit induced field correction model based on large eddy simulation combined with gradient boosting tree; coupling the measurement volume integral operator and the unit induced field correction model, and obtaining high-precision wind speed vertical profile and corresponding wind shear parameters through inversion; based on the high-precision wind speed vertical profile and the corresponding wind shear parameters, fusing physical constraints to generate a wind shear model and perform uncertainty evaluation.

[0007] In one of the embodiments, the wind speed measurement data of the laser radar at several heights in the target wind turbine area is collected, and the measurement volume integral operator matrix is constructed, specifically: obtaining the spatial positioning information of the target wind turbine area, and aligning the coordinates of the target wind turbine with the coordinates of the laser radar; constructing a radar scanning strategy optimization environment, optimizing based on a reinforcement learning algorithm to obtain an optimal radar scanning path; controlling the laser radar to collect Doppler shift data at each set elevation angle according to the radar scanning path; performing moving average filtering processing on the Doppler shift data, and calculating the line-of-sight wind speed component of each beam according to the filtered Doppler shift data; combining the line-of-sight velocity components at different elevations with the beam azimuth and elevation for geometric projection transformation, and reconstructing the three-dimensional wind speed vector components at each measurement point; extracting the horizontal component and the vertical component of the three-dimensional wind speed vector components as the preliminary true wind speed distribution at each vertical height layer of the target wind turbine area; and establishing the measurement volume integral operator according to the preliminary true wind speed distribution.

[0008] In one of the embodiments, the radar scanning strategy optimization environment is constructed, and the optimal radar scanning path is obtained based on the reinforcement learning algorithm, specifically: defining a state vector of the scanning path optimization problem; defining an action space, which includes adjustable scanning parameters; designing a reward function, which includes a vertical wind speed profile error minimization term, a spatial coverage maximization term and a measurement signal-to-noise ratio constraint term; training the scanning strategy in the simulation environment using the reinforcement learning algorithm, iteratively optimizing the scanning path strategy until a preset training number is reached, and obtaining the optimal radar scanning path that maximizes the cumulative reward.

[0009] In one of the embodiments, a unit-induced field correction model is constructed based on large eddy simulation combined with gradient boosting tree, specifically: obtaining working environment and real-time aerodynamic characteristic data of a target wind turbine; establishing a large eddy simulation model and initializing according to the working environment and real-time aerodynamic characteristic data; running a wind speed field simulation based on the large eddy simulation model to obtain an induced flow field around the wind turbine, the induced flow field including local turbulence, vortex and wind speed distortion distribution around the nacelle caused by blade rotation; and constructing a unit-induced field correction model according to the induced flow field.

[0010] In one of the embodiments, the unit-induced field correction model is constructed according to the induced flow field, specifically: extracting wind speed disturbance information according to the induced flow field; quantitatively analyzing the distribution of the wind speed disturbance information in space and vertical height to generate an induced effect index; constructing an induced effect feature vector based on the induced effect index; corresponding matching the induced effect feature vector with the measured wind speed data around the wind turbine, mapping each measured point with the corresponding feature vector through spatial coordinates and height information to form a training data set, the training data set including the wind turbine induced effect feature vector as the input feature and the corresponding measured wind speed data as the output label; initializing a gradient boosting tree model, taking the average value of the training data output labels as the benchmark output value of each regression tree; iteratively training based on the training data set, and in each iteration, weighting and accumulating the output of the residual tree by multiplying the learning rate; and taking the gradient boosting tree model after iterative training as the unit-induced field correction model.

[0011] In one of the embodiments, the measurement volume integral operator is coupled with the unit-induced field correction model, specifically: mapping the wind speed component measured by the laser radar to the vertical height grid according to the measurement volume integral operator to form a preliminary wind speed measurement matrix; outputting the induced effect correction factor of the corresponding measurement point by using the unit-induced field correction model based on the preliminary wind speed measurement matrix; correcting the wind speed component of the corresponding height layer in the preliminary wind speed measurement matrix according to the induced effect correction factor to obtain a corrected wind speed measurement matrix; and obtaining a high-precision wind speed vertical profile and corresponding wind shear parameters through adaptive regularization inversion according to the corrected wind speed measurement matrix.

[0012] In one of the embodiments, according to the modified wind speed measurement matrix, a high-precision wind speed vertical profile and corresponding wind shear parameters are obtained through adaptive regularization inversion, specifically: based on the modified wind speed measurement matrix, a wind speed inversion equation set is established; based on the wind speed inversion equation set, an adaptive regularization term is introduced to construct an objective function, the objective function also includes adaptive regularization parameters corresponding to each height layer determined based on the measurement noise level, signal-to-noise ratio distribution and local gradient strength; an iterative solution algorithm is used to minimize the objective function, and the real wind speed profile vector to be inverted and the adaptive regularization parameter are updated in each iteration until the residual difference of adjacent two iterations is less than a set threshold, and the inversion converges; after the inversion converges, a high-resolution real wind speed vertical profile is output and wind shear parameters are calculated.

[0013] In one of the embodiments, based on the high-precision wind speed vertical profile and corresponding wind shear parameters, a wind shear model is generated by fusing physical constraints, specifically: a wind shear feature sample set is constructed according to the high-precision wind speed vertical profile and corresponding wind shear parameters; a constraint function is established based on the atmospheric boundary layer theory; a long short-term memory network model is constructed, which includes a data-driven prediction layer using a multi-layer LSTM network structure for capturing the nonlinear dynamic law of the wind speed vertical profile over time; the wind shear feature sample set is input into the long short-term memory network model for training, and the output is a wind shear data set corresponding to the height; during the training of the long short-term memory network model, the constraint function is introduced as a regularization term into the joint optimization objective function for fusion training, so that the model output meets the data accuracy and physical rationality at the same time, and the wind shear model is obtained.

[0014] In one of the embodiments, uncertainty evaluation is performed, specifically: according to the output of the wind shear model, a wind shear data set and its corresponding input feature set are extracted; based on the input feature set and the model structure parameters, is defined as a random variable, and each random variable is assigned a corresponding probability distribution; based on the probability distribution, the input variable is randomly sampled, and the uncertainty is propagated through the Bayesian sampling method to obtain the posterior distribution; according to the posterior distribution, the confidence interval and the uncertainty quantification index of each height layer are calculated; based on the confidence interval and the uncertainty index of each height layer, a wind shear uncertainty profile is constructed, and the wind shear prediction value and its confidence range at different height layers are visually expressed in the form of a continuous curve; the credibility and risk assessment of the wind shear change trend with height are judged through the uncertainty profile.

[0015] From the above technical solutions, it can be seen that the embodiments of the present application have the following advantages: By organically combining the laser radar measurement volume integral operator, the large eddy simulation, the gradient boosting tree correction model, the adaptive regularization inversion and the long short-term memory network model with physical constraints, a complete high-precision analysis system from wind speed data acquisition to wind shear modeling and uncertainty evaluation is constructed. The advantages of the scheme are as follows: the three-dimensional wind speed field around the wind turbine can be reconstructed based on the real physical process, the measurement error caused by the cabin and blade induction effect is significantly reduced; the balance between the smoothness and the detail preservation of the wind speed vertical profile is realized through the adaptive regularization inversion, and the stability and precision of the inversion result are ensured; the deep learning structure with physical constraints is introduced in the wind shear modeling, so that the model output not only conforms to the data law, but also conforms to the physical characteristics of the atmospheric boundary layer; finally, the quantification and risk visualization evaluation of the prediction credibility are realized through the Bayesian uncertainty analysis. Overall, the method realizes the whole-chain optimization of wind field measurement, correction, inversion, modeling and credibility evaluation, and has the significant advantages of high precision, physical consistency, interpretability and engineering reliability. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A laser radar assisted wind shear modeling method for wind power generation is provided. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0018] Referring to Figure 1 The laser radar assisted wind shear modeling method for wind power generation provided by the present application includes the following steps: S1, collecting wind speed measurement data of the laser radar at several heights in the target wind turbine area, and constructing a measurement volume integral operator matrix, the measurement volume integral operator matrix being used for mathematical mapping of the measured speed value of the laser radar and the actual vertical height profile wind speed.

[0019] In the embodiment, the wind speed measurement data of the laser radar at several heights in the target wind turbine area is collected, and a measurement volume integral operator matrix is constructed, specifically as follows: The spatial positioning information of the target wind turbine area is obtained, the wind turbine tower position, the cabin center and the blade swept range are measured by using a laser range finder, and the coordinates of the target wind turbine are aligned with the laser radar coordinates, thereby providing basic geometric information for subsequent measurement data; An environment for optimizing a radar scanning strategy is constructed, and an optimal radar scanning path is obtained based on a reinforcement learning algorithm; The laser radar is controlled to emit a plurality of laser pulses at each set elevation angle according to the radar scanning path, and Doppler shift data of each laser pulse on a return path is collected; The Doppler shift data is subjected to a moving average filtering process, and a line-of-sight wind speed component of each light beam is calculated based on the filtered Doppler shift data; The line-of-sight wind speed component is specifically calculated according to the following formula: In the formula, is the line-of-sight wind speed component, is the Doppler shift data, is the wavelength of the laser.

[0020] The line-of-sight wind speed component at different elevations is combined with the azimuth angle and the elevation angle to perform geometric projection transformation, and three-dimensional wind speed vector components at each measurement point are reconstructed by the least square method, specifically as follows: The relationship between the line-of-sight wind speed of each light beam and the three-dimensional wind speed can be expressed as: In the formula, V is a three-dimensional wind speed vector of a target point, is a unit direction vector of the radar beam indicating that the beam points in the three-dimensional space, wherein, is an X-axis direction vector is a y-axis direction vector is a z-axis direction vector , is the beam elevation angle, is the beam azimuth angle, wherein, u is a horizontal east-west wind speed component, is a horizontal north-south wind speed component, is a vertical wind speed component.

[0021] A linear equation set is constructed , wherein, is a set of line-of-sight wind speed components measured by N beams passing through the same measurement point, is a beam direction matrix, each row corresponding to a direction vector of a beam ; The reconstructed three-dimensional wind speed vector components are obtained by the least square method based on the linear equation set; The three-dimensional wind speed vector components are specifically calculated according to the following formula: In the formula, is the three-dimensional wind speed vector component, is the transpose of the light beam direction matrix.

[0022] extracting the horizontal component and the vertical component of the three-dimensional wind speed vector component as a preliminary real wind speed distribution at each vertical height layer in the target wind turbine area; establishing a measurement volume integral operator according to the preliminary real wind speed distribution.

[0023] The measurement volume integral operator has the following specific calculation formula: wherein, is the wind speed component measured by the laser radar, is a weight, indicating the contribution degree of each measurement point in the integration, is the preliminary real wind speed distribution at each spatial point (x, y, z) in the target area, including and respectively represent the components of the wind speed in the x-axis, the y-axis and the z-axis, is the light beam volume.

[0024] It should be noted that the measurement volume integral operator is used to describe the relationship between the laser radar measurement results and the real wind speed distribution in the target area. It forms a mathematical model by mapping the wind speed component measured by the radar with the wind speed of each measurement point in space. The operator is based on the line-of-sight wind speed data collected by the radar at different heights and different positions in the target area, integrates these data with the spatial volume, and calculates the three-dimensional distribution of the wind speed. Specifically, the measurement volume integral operator considers the direction, elevation angle and azimuth angle of the measurement beam, and integrates and maps the wind speed field. The advantage of constructing the measurement volume integral operator is that it can convert the linear measurement data of the laser radar into the real wind speed field distribution in space, overcoming the problem of insufficient data in a single measurement direction. Through the integral operator, a more accurate wind speed vertical profile and three-dimensional wind speed distribution can be constructed in the target area, especially in complex terrain and dynamic wind turbine environment, which can improve the reconstruction accuracy of the wind speed field. In addition, the operator can provide high-precision data support for subsequent wind shear modeling, wind energy resource assessment, etc., thereby optimizing the operation of wind turbines and the planning of wind farms.

[0025] Further, a radar scanning strategy optimization environment is constructed, and an optimal radar scanning path is obtained based on a reinforcement learning algorithm, specifically: The radar scanning strategy optimization environment includes a three-dimensional spatial coordinate system of the target wind turbine area, a wind turbine impeller height range, an installation position of the laser radar, a radar light beam divergence angle, a ranging resolution, and a signal-to-noise ratio characteristic, so as to form a state space and a measurement simulation environment required for reinforcement learning training, which can reflect the complex wind speed field of the wind turbine area and the detection capability of the laser radar; define a state vector of the scan path optimization problem, which includes the current elevation angle, azimuth angle, scan height layer number, historical measurement error, signal strength, and target area wind speed variation of the radar; define an action space, which includes adjustable scan parameters such as angle increment, azimuth angle increment, scan frequency adjustment, and beam focusing strategy; design a reward function to quantify the pros and cons of the scan path, which includes a vertical wind speed profile error minimization term, a spatial coverage maximization term, and a measurement signal-to-noise ratio constraint term; The vertical wind speed profile error minimization term minimizes the error between the wind speed vertical profile obtained by the radar inversion and the true profile by comparing with the reference profile or historical data. The spatial coverage maximization term is the proportion of the radar beam covering the key area, especially the high wind speed area in the key wind turbine area. The measurement signal-to-noise ratio constraint term is used to punish the scanning of weak echo or signal attenuation area, to ensure the quality of data acquisition and avoid errors caused by unclear signals.

[0026] Use reinforcement learning algorithm to train scan strategy in simulation environment, optimize scan path strategy through iteration of each action selection, state update and reward feedback until reaching preset training rounds, and get optimal radar scan path that maximizes cumulative reward.

[0027] It should be noted that the optimal radar scan path obtained based on reinforcement learning has significant advantages, mainly in dynamically adjusting the scan strategy according to the changes of the actual environment and measurement requirements. Through the reinforcement learning algorithm, the system can autonomously learn and optimize the scan path in multiple iterations, not only improving the spatial coverage of the wind turbine area, but also being able to real-time respond to the complexity of the wind speed field, radar signal attenuation and other factors, minimizing errors and signal loss. At the same time, the reward function in the optimization process considers multiple dimensions such as vertical wind speed profile error, spatial coverage and signal-to-noise ratio, ensuring that the final scan path can provide high-quality wind speed and direction data. This adaptive, flexible and efficient path optimization method provides reliable technical support for accurate measurement of wind speed field and efficient operation of wind farm.

[0028] S2, based on large eddy simulation combined with gradient boosting tree, a unit induced field correction model is constructed to correct the profile distortion deviation caused by the induction effect of the cabin and the blade disc in the measurement.

[0029] In this embodiment, based on large eddy simulation combined with gradient boosting tree, a unit induced field correction model is constructed, specifically: acquire the working environment of the target wind turbine and real-time aerodynamic characteristic data, the working environment including wind speed field, installation position of the wind turbine and nacelle parameters, and the real-time aerodynamic characteristic data including wind turbine blade geometry, length, bending angle, blade surface roughness, wind turbine working state parameters such as rotating speed, lift and drag coefficients, and flow field parameters around the wind turbine such as background wind speed, wind direction and turbulence intensity; According to the working environment and real-time aerodynamic characteristic data, a large eddy simulation model is established and initialized, the initialization including demarcating a calculation domain around the wind turbine and determining a mesh division scheme, wherein the mesh resolution satisfies the requirement of capturing small-scale vortices around the wind turbine blades and nacelle; setting boundary conditions including blade no-slip condition, tower surface condition and flow field inlet wind speed condition; Based on the large eddy simulation model, a wind speed field simulation is run to obtain an induced flow field around the wind turbine, the induced flow field including local turbulence, vortices caused by blade rotation and wind speed distortion distribution around the nacelle; The large eddy simulation can directly simulate large-scale vortices generated by the wind turbine blade rotation and nacelle structure in air flow and the interaction thereof with small-scale turbulence. By setting the wind turbine geometry, blade motion and boundary conditions in the calculation domain, LES can realistically reproduce the complex structure of air flow around the wind turbine, including tip vortices, nacelle disturbance and tower interference, thereby accurately reflecting the change of the wind turbine induced effect on the local wind speed field. In other words, LES analyzes the three-dimensional flow characteristics of the wind turbine and its surroundings, so that the simulation results can directly reflect the distortion and disturbance of the wind turbine blade rotation and nacelle on the surrounding wind speed field, thereby obtaining the induced flow field.

[0030] According to the induced flow field, a unit induced field correction model is constructed.

[0031] It should be noted that the advantage of constructing the unit induced field correction model by large eddy simulation (LES) is that LES can directly analyze the complex large-scale vortices generated by the wind turbine blade rotation and nacelle structure in air flow and the interaction thereof with small-scale turbulence, thereby realistically reproducing the three-dimensional flow structure around the wind turbine, including tip vortices, nacelle disturbance and tower interference. This enables the simulation results to accurately reflect the distortion and disturbance of the wind turbine induced effect on the local wind speed field, providing high-precision, physically interpretable three-dimensional wind speed disturbance information for subsequent establishment of the unit induced field correction model, significantly improving the accuracy and reliability of wind speed profile correction, while avoiding reliance on a large amount of expensive measured data.

[0032] Further, according to the induced flow field, a unit induced field correction model is constructed, specifically: According to the induced flow field, wind speed disturbance information is extracted, the wind speed disturbance information including local vortices generated by blade rotation, nacelle disturbance and wind speed deviation caused by the tower; Quantitative analysis is performed on the distribution of wind speed disturbance information in space and vertical height to generate an induction effect index for characterizing the induction effect, the induction effect index including a wind speed disturbance amplitude, a disturbance direction deviation, and a local velocity gradient; Based on the induction effect index, an induction effect feature vector is constructed, the induction effect feature vector including the wind speed disturbance amplitude, the disturbance direction deviation, the local velocity gradient, a blade length, a blade bending angle, a blade surface roughness, a nacelle geometric parameter, a tower position, and a wind turbine real-time working state parameter; The induction effect feature vector is correspondingly matched with measured wind speed data around the wind turbine, wherein the measured wind speed data includes three-dimensional wind speed components and measurement errors collected by a laser radar or other high-precision wind speed sensor at different height layers of a wind turbine blade disc, a nacelle, and a tower area; A mapping relationship is established between each measured point and the corresponding feature vector through spatial coordinates and height information to form a training data set, the training data set including the wind turbine induction effect feature vector as an input feature and the corresponding measured wind speed data as an output label; A gradient boosting tree model is initialized, the number of trees, the maximum depth, the learning rate, and the minimum leaf node sample number are set, and the benchmark output value of each regression tree is initialized as the average value of the training data output labels to provide an initial reference for subsequent iterative fitting residuals; Based on the training data set, iterative training is performed, residuals between a current model prediction value and the output label are calculated each time, and a new regression tree is trained to fit the residuals, the model output is gradually updated, and in each iteration, the output of the residual tree is weighted and accumulated by multiplying the learning rate, so that the gradient boosting tree can gradually approach the nonlinear mapping relationship between the input feature and the output correction factor; The gradient boosting tree model after iterative training is used as a unit induction field correction model, the model can receive the wind turbine induction effect feature vector input and output the corresponding dynamic induction effect correction factor, the induction effect correction factor including a wind speed distortion amplitude, a direction deviation, and a local velocity gradient correction value of each spatial point, so as to correct the wind speed disturbance under different heights and different working conditions and make the corrected wind speed profile more accurately reflect the real wind speed field.

[0033] S3, the measurement volume integral operator is coupled with the unit induction field correction model, and high-precision wind speed vertical profiles and corresponding wind shear parameters are obtained through inversion.

[0034] In this embodiment, the measurement volume integral operator is coupled with the unit induction field correction model, specifically: The wind speed components measured by the laser radar are mapped to the vertical height grid according to the measurement volume integral operator to form a preliminary wind speed measurement matrix; Based on the preliminary wind speed measurement matrix, a unit induced field correction model is input into the spatial coordinates, height information, fan blade geometric parameters, nacelle and tower structure parameters, blade rotating speed and fan real-time working state parameters of each measurement point, to obtain an induced effect correction factor corresponding to the measurement point; According to the induced effect correction factor, a wind speed component of a corresponding height layer in the preliminary wind speed measurement matrix is corrected, to obtain a corrected wind speed measurement matrix; The wind speed component of the corresponding height layer is corrected, and a specific calculation formula is as follows: In the formula, is a corrected wind speed component, is a corresponding induced effect correction factor, is a direction deviation.

[0035] According to the corrected wind speed measurement matrix, a high-precision wind speed vertical profile and corresponding wind shear parameters are obtained through adaptive regularization inversion.

[0036] Further, according to the corrected wind speed measurement matrix, a high-precision wind speed vertical profile and corresponding wind shear parameters are obtained through adaptive regularization inversion, and specifically: Based on the corrected wind speed measurement matrix, a wind speed inversion equation set is established; The wind speed inversion equation set has a specific calculation formula as follows: In the formula, is a measurement volume integral operator matrix, is a real wind speed profile vector to be inverted, is a measurement error vector.

[0037] Based on the wind speed inversion equation set, an adaptive regularization term is introduced to construct an objective function, to avoid the ill-posedness caused by direct solving. The objective function further includes adaptive regularization parameters corresponding to each height layer, which are determined based on a measurement noise level, a signal-to-noise ratio distribution and a local gradient intensity; The objective function has a specific calculation formula as follows: In the formula, is an objective function, is an adaptive regularization parameter varying with height, is a smoothing operator matrix.

[0038] The adaptive regularization parameter has a specific calculation formula as follows: In the formula, an adaptive regularization parameter for the i-th height layer, an initial regularization parameter, a measurement error variance for the current height layer, a signal-to-noise ratio, a local wind speed gradient amplitude. The adaptive regularization parameter adaptively adjusts the smoothing strength, so that the high signal-to-noise ratio area retains details, and the low signal-to-noise ratio area enhances stability.

[0039] An iterative solution algorithm is used to minimize the objective function, and the real wind speed profile vector to be inverted and the adaptive regularization parameter are updated in each iteration. Through the residual reduction criterion, convergence is judged until the residual difference of adjacent two iterations is less than the set threshold, and the inversion converges; After the inversion converges, the high-resolution real wind speed vertical profile is output and the wind shear parameter is calculated.

[0040] The wind shear parameter, specifically the calculation formula is as follows: In the formula, a wind shear parameter, and are the average wind speed values for different height layers, and are different reference heights.

[0041] It should be noted that the measurement volume integral operator is organically coupled with the unit-induced field correction model, and an adaptive regularization inversion mechanism is introduced, thereby realizing high-precision reconstruction of the wind speed profile and dynamic extraction of the wind shear parameter. Specifically, the measurement volume integral operator is responsible for establishing the mapping relationship between the lidar observation and the real wind speed field, so that the spatially discrete radar observation information can be projected into a continuous vertical height grid; the unit-induced field correction model further compensates and corrects the local flow field distortion caused by wind turbine blade rotation, nacelle interference, and tower shielding, significantly improving the physical authenticity of the wind speed data. On this basis, through adaptive regularization inversion, an objective function is established that includes an observation error term and a height-adaptive smoothing constraint term, so that the algorithm can automatically adjust the regularization strength according to the signal-to-noise ratio, measurement variance, and local wind speed gradient characteristics of different height layers: in high signal-to-noise ratio areas, the wind speed detail structure is maintained, and in low signal-to-noise ratio areas, the stability and robustness of the solution is enhanced. Finally, through iterative optimization, a high-resolution, continuous, and smooth wind speed vertical profile is obtained, and the wind shear parameter is calculated accordingly, realizing high-fidelity characterization of the wind energy distribution and wind shear characteristics in the wind turbine operating area. This method not only significantly reduces the ill-posed error in traditional inversion algorithms, but also enhances the ability to identify the real wind speed structure in complex flow fields, providing more reliable basic data support for wind energy resource assessment, wind turbine power prediction, and structural safety analysis.

[0042] S4, generating a wind shear model based on the high-precision wind speed vertical profile and the corresponding wind shear parameter, and fusing physical constraints, and performing uncertainty evaluation.

[0043] In this embodiment, based on the high-precision wind speed vertical profile and the corresponding wind shear parameter, a wind shear model is generated by fusing physical constraints and performing uncertainty evaluation, specifically; constructing a wind shear feature sample set according to the high-precision wind speed vertical profile and the corresponding wind shear parameter; Based on the atmospheric boundary layer theory, a constraint function including the friction velocity and the Obukhov length is established, and the constraint function is used to limit the physical consistency between the wind speed gradient output by the machine learning model and the stability parameter; The constraint function has the following specific calculation formula: In the formula, is a constraint function of a physical residual, is the wind speed, is the height, is the Obukhov length, which is a physical quantity representing the stability of the atmospheric boundary layer, and is obtained through the Monin-Obukhov similarity theory, is the friction velocity, which represents the friction intensity of the surface wind on the air, is used to represent the physical relationship between the wind speed gradient, the stability parameter, and the friction velocity, and the input physical quantity is mapped to a residual value The residual measures whether the wind speed profile is consistent with the atmospheric boundary layer theory. When =0, it indicates that the wind speed gradient completely satisfies the theoretical constraints.

[0044] A long short-term memory network model is constructed, which includes a data-driven prediction layer using a multi-layer LSTM network structure for capturing the nonlinear dynamic law of the wind speed vertical profile over time; The wind shear feature sample set is input into the long short-term memory network model for training, and the output is a wind shear data set corresponding to the height. In the training process, the Adam optimization algorithm or the gradient descent method is used, and the loss function is the prediction error plus the regularization term. The wind shear data set includes wind shear value, wind direction angle, and boundary layer stability change; In the training process of the long short-term memory network model, the constraint function is introduced as a regularization term into the joint optimization objective function for fusion training, so that the model output meets the data accuracy and physical rationality at the same time, and a wind shear model is obtained; The joint optimization objective function has the following specific calculation formula: wherein, is a joint optimization objective function, is a data prediction error term, is a weight coefficient. Further, uncertainty evaluation is performed, specifically: According to the output of the wind shear model, a wind shear dataset and its corresponding input feature set are extracted, the input feature set including wind speed vertical profile, wind shear gradient, turbulence intensity, boundary layer stability, and environmental temperature; Based on the input feature set and model structure parameters, input measurement error, model parameter fluctuation, and environmental disturbance are defined as random variables, and corresponding probability distributions are assigned to each random variable, which can be determined according to historical observation statistics or empirical prior, for example, measurement error follows Gaussian distribution, model weight disturbance follows Laplace distribution, and environmental disturbance follows lognormal distribution; Based on the probability distribution, the input variables are randomly sampled, and the uncertainty propagation is performed by the Bayesian sampling method to obtain the posterior distribution, wherein the uncertainty propagation calculation is performed by the Bayesian sampling method, specifically, a probabilistic graphical model of wind shear prediction is established, the model parameters are regarded as random variables, and the posterior distribution of wind shear prediction result is obtained by sampling through the maximum posterior estimation method; According to the posterior distribution, the confidence interval and the uncertainty quantification index of each height layer are calculated, the confidence interval is obtained by extracting the quantile point of the prediction distribution, for example, the 95% confidence interval is determined by the 2.5% and 97.5% quantile points, and the uncertainty index is realized by calculating the standard deviation, variance or entropy value of the prediction distribution, which is used to represent the fluctuation range and stability of the prediction result; Based on the confidence interval and the uncertainty index of each height layer, a wind shear uncertainty profile graph is constructed, and the wind shear prediction values and their confidence ranges of different height layers are visually expressed in the form of continuous curves; Through the uncertainty profile graph, the reliability of the wind shear change trend with height is judged and the risk is evaluated, when the confidence interval is narrowed and the standard deviation is less than the preset threshold, the region represents that the model prediction result is reliable, when the confidence interval is expanded or the standard deviation is greater than the preset threshold, the region represents that there is potential uncertainty risk.

[0045] It should be noted that by introducing the uncertainty propagation mechanism based on Bayesian sampling in the wind shear modeling process, the comprehensive quantification and dynamic evaluation of input errors, model fluctuations and environmental disturbances are realized, which can reflect the confidence and stability of the wind shear prediction results from the probability point of view. This scheme can not only output the prediction value of wind shear, but also provide the confidence interval and uncertainty index of each height layer at the same time, so that the prediction result has the ability of credibility quantization and risk identification. By constructing the wind shear uncertainty profile, the prediction reliability distribution of different height layers can be intuitively displayed, and the explainability enhancement and risk visualization expression of wind shear change trend are realized. Compared with the traditional deterministic model, the robustness, transparency and engineering application safety of wind shear prediction are significantly improved, which provides more reliable decision basis for the operation optimization, load evaluation and wind energy utilization strategy of wind turbines.

[0046] The above formulas are dimensionless numerical calculations, and the formulas are obtained by collecting a large amount of data to simulate the latest real situation, and the preset parameters in the formula are set by the person skilled in the art according to the actual situation.

[0047] The above embodiments can be realized wholly or partially by software, hardware, firmware or any combination thereof. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product.

[0048] Those skilled in the art can realize that the modules and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The person skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0049] In addition, the functional modules in each embodiment of the present application can be integrated in one processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0050] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0051] Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application, should be included in the scope of protection of the present application.

Claims

1. A lidar-assisted wind shear modeling method for wind power generation, characterized in that, include: Collect wind speed measurement data at several heights within the target wind turbine area using lidar, and construct a measurement volume integral operator matrix; A unit induced field correction model was constructed based on large eddy simulation combined with gradient boosting tree. The measurement volume integral operator is coupled with the unit induced field correction model, and a high-precision vertical wind speed profile and corresponding wind shear parameters are obtained through inversion. Based on high-precision vertical wind speed profiles and corresponding wind shear parameters, a wind shear model is generated by integrating physical constraints, and uncertainty assessment is performed.

2. The lidar-assisted wind shear modeling method for wind power generation according to claim 1, characterized in that, The acquisition of wind speed measurement data at several heights within the target wind turbine area using lidar, and the construction of a measurement volume integration operator matrix, are as follows: Acquire spatial positioning information of the target wind turbine area and align the coordinates of the target wind turbine with the coordinates of the lidar. A radar scanning strategy optimization environment is constructed, and optimization is performed based on reinforcement learning algorithms to obtain the optimal radar scanning path; The lidar is controlled to collect Doppler frequency shift data at each set elevation angle according to the radar scanning path; The Doppler frequency shift data is processed by moving average filtering, and the line-of-sight wind speed component of each beam is calculated based on the filtered Doppler frequency shift data. By combining the line-of-sight velocity components at different elevation angles with the beam azimuth and elevation angles, a geometric projection transformation is performed to reconstruct the three-dimensional wind speed vector components at each measuring point. At each vertical height layer in the target wind turbine area, the horizontal and vertical components of the three-dimensional wind speed vector components are extracted as the preliminary true wind speed distribution. Based on the preliminary actual wind speed distribution, a measurement volume integral operator is established.

3. The lidar-assisted wind shear modeling method for wind power generation according to claim 2, characterized in that, The construction of the radar scanning strategy optimization environment, based on reinforcement learning algorithms, yields the optimal radar scanning path, specifically as follows: Define the state vector for the scan path optimization problem; Define an action space, which includes adjustable scan parameters; Design a reward function, which includes a term for minimizing vertical wind speed profile error, a term for maximizing spatial coverage, and a term for constraining measurement signal-to-noise ratio. A reinforcement learning algorithm is used to train the scanning strategy in a simulation environment, and the scanning path strategy is iteratively optimized until a preset number of training rounds are reached to obtain the optimal radar scanning path that maximizes the cumulative reward.

4. The lidar-assisted wind shear modeling method for wind power generation according to claim 1, characterized in that, The unit induced field correction model constructed based on large eddy simulation combined with gradient boosting tree is as follows: Acquire the target fan's operating environment and real-time aerodynamic characteristics data; Based on the working environment and real-time aerodynamic characteristic data, a large eddy simulation model was established and initialized. Based on the large eddy simulation model, the wind speed field simulation is performed to obtain the induced flow field around the wind turbine. The induced flow field includes local turbulence and eddies caused by blade rotation and the wind speed distortion distribution around the nacelle. A correction model for the induced flow field of the unit is constructed based on the induced flow field.

5. The lidar-assisted wind shear modeling method for wind power generation according to claim 4, characterized in that, The construction of the unit induced field correction model based on the induced flow field is specifically as follows: Based on the induced flow field, wind speed disturbance information is extracted; Quantitative analysis of the spatial and vertical distribution of wind speed disturbance information is performed to generate an induced effect index. Based on the induction effect index, construct the induction effect feature vector; The induced effect feature vector is matched with the measured wind speed data around the wind turbine. The spatial coordinates and height information are used to establish a mapping relationship between each measured point and the corresponding feature vector to form a training dataset. The training dataset includes the wind turbine induced effect feature vector as input features and the corresponding measured wind speed data as output labels. Initialize the gradient boosting tree model, and set the baseline output value of each regression tree to the average value of the output labels of the training data; Iterative training is performed based on the training dataset, and in each iteration, the output of the residual tree is weighted and accumulated by multiplying by the learning rate; The gradient boosting tree model, which has been trained iteratively, is used as the unit induced field correction model.

6. The lidar-assisted wind shear modeling method for wind power generation according to claim 1, characterized in that, The coupling of the measurement volume integral operator with the unit induced field correction model is specifically as follows: The wind speed components measured by the lidar are mapped to the vertical height grid based on the measurement volume integral operator to form a preliminary wind speed measurement matrix; Based on the preliminary wind speed measurement matrix, the induced field correction model of the unit is used to output the induced effect correction factor for the corresponding measuring point; The wind speed components at the corresponding height levels in the preliminary wind speed measurement matrix are corrected according to the induction effect correction factor to obtain the corrected wind speed measurement matrix. Based on the corrected wind speed measurement matrix, a high-precision vertical wind speed profile and corresponding wind shear parameters are obtained through adaptive regularization inversion.

7. The lidar-assisted wind shear modeling method for wind power generation according to claim 6, characterized in that, Based on the corrected wind speed measurement matrix, a high-precision vertical wind speed profile and corresponding wind shear parameters are obtained through adaptive regularization inversion, specifically as follows: Based on the corrected wind speed measurement matrix, a set of wind speed inversion equations is established. Based on the wind speed inversion equations, an adaptive regularization term is introduced to construct an objective function. The objective function also includes determining the adaptive regularization parameters corresponding to each height layer based on the measurement noise level, signal-to-noise ratio distribution, and local gradient intensity. An iterative algorithm is used to minimize the objective function. In each iteration, the true wind speed profile vector to be inverted and the adaptive regularization parameter are updated until the difference between the residuals of two adjacent iterations is less than a set threshold, at which point the inversion converges. After the inversion converges, a high-resolution true wind speed vertical profile is output and the wind shear parameters are calculated.

8. The lidar-assisted wind shear modeling method for wind power generation according to claim 7, characterized in that, The wind shear model is generated based on high-precision vertical wind speed profiles and corresponding wind shear parameters, and by incorporating physical constraints. Specifically, the model is as follows: A wind shear feature sample set is constructed based on high-precision vertical wind speed profiles and corresponding wind shear parameters; Based on atmospheric boundary layer theory, constraint functions are established; A long short-term memory network model is constructed, which includes a data-driven prediction layer using a multi-layer LSTM network structure to capture the nonlinear dynamic law of wind speed vertical profile changing over time. The wind shear feature sample set is input into the long short-term memory network model for training, and the output is the wind shear dataset at the corresponding height; During the training of the long short-term memory network model, the constraint function is introduced as a regularization term into the joint optimization objective function for fusion training, so that the model output can simultaneously satisfy data accuracy and physical rationality, thus obtaining the wind shear model.

9. The lidar-assisted wind shear modeling method for wind power generation according to claim 8, characterized in that, The uncertainty assessment is specifically performed as follows: Based on the output of the wind shear model, extract the wind shear dataset and its corresponding input feature set; Based on the input feature set and model structure parameters, they are defined as random variables, and corresponding probability distributions are assigned to each random variable. Based on the probability distribution, the input variables are randomly sampled, and uncertainty propagation is performed using the Bayesian sampling method to obtain the posterior distribution. Based on the posterior distribution, calculate the confidence intervals and uncertainty quantification indices for each altitude layer; Based on the confidence intervals and uncertainty indices of each height level, a wind shear uncertainty profile is constructed, and the predicted wind shear values ​​and their confidence ranges at different height levels are visualized in the form of continuous curves. Reliability assessment and risk evaluation of wind shear variation with altitude using uncertainty profile diagrams.

Images