A cross-scale physical field data-driven wind farm negative shear identification method and system

Abstract

The application is a wind farm negative shear identification method and system driven by cross-scale physical field data, belonging to the field of wind power generation. The application first relies on small and medium scale coupling high resolution wind field simulation to build a wind farm multi-working condition wind field sample library, extracts virtual measuring point multi-height wind parameters and unit front multi-height wind speed data; constructs the wind field data as neural network input and output feature vectors, divides the data set after standardization processing, optimizes hyperparameters through grid search and cross validation, trains a multi-input multi-output neural network model, and establishes a nonlinear mapping relationship between incoming flow wind conditions and unit front multi-layer wind speed; uses multiple evaluation indexes to verify the model prediction performance, solves the wind shear coefficient based on the predicted wind speed fitting wind profile, determines the unit front negative shear working condition according to the coefficient, and outputs the negative shear identification result and spatial distribution combined with the unit position. The application can obtain the wind farm multi-unit negative shear spatial distribution under the premise of preserving high resolution wind field information.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation technology, specifically relating to a cross-scale physical field data-driven method and system for identifying negative shear in wind farms. Background Technology

[0002] The hub height of current large wind turbines has exceeded 150 meters, and the swept diameter of blades exceeds 200 meters. Within such a vast vertical swept plane, especially in complex terrain with dramatic undulations and varying surface roughness, wind conditions exhibit extreme spatial heterogeneity and three-dimensional nonlinearity. Under these conditions, the variation of incoming wind speed with altitude in front of the turbine often deviates from the classic "logarithmic law" or "power law" profile in meteorology, frequently exhibiting a reverse distribution where high-level wind speeds are significantly lower than low-level wind speeds—a phenomenon known as "negative wind shear." Negative wind shear causes uneven wind loads on the upper and lower parts of the turbine rotor, affecting power output stability and potentially increasing fatigue loads and control risks. To achieve proactive safety warnings for wind farms in complex terrain, it is necessary to rapidly predict wind speeds at different heights in front of multiple turbines and identify negative wind shear accordingly.

[0003] High-fidelity small-scale coupled simulations can generate a large number of wind speed samples across multiple altitudes and turbine units. These methods can accurately analyze the details of microscopic wake dissipation and topographic forcing based on the first principles of fluid mechanics. However, they require extremely large amounts of computing power and computation time, making them unsuitable for frequent online assessments or rapid early warnings. CN104699936A proposes to circumvent the timeliness problem by establishing an offline CFD wind field speedup plugin database through pre-calculation. However, this static lookup method lacks memory of historical evolution and dynamic coupling of large-scale meteorological forcing, resulting in severe degradation of prediction accuracy when faced with highly unsteady turbulent disturbances and wake oscillations.

[0004] Data-driven methods based on traditional statistics or conventional shallow machine learning have been widely used in wind power prediction, but the following problems still exist: First, most traditional machine learning prediction models are designed as "single-input-single-output" or local "multi-input-single-output" structures. They cannot extract and internalize the deep physical correlations and wake memory across spatial nodes from a global perspective. Second, there is a lack of high-quality multidimensional training data. Wind farms generally have only a very small number of anemometer towers, lacking detailed measured samples that can characterize the three-dimensional profile in front of the wind turbines across the entire site.

[0005] Traditional numerical simulation methods and neural network prediction models have the following shortcomings in wind shear prediction for wind farms: Existing technologies still struggle to balance high-resolution wind field information with the timeliness of negative shear identification: high-resolution numerical simulations involve large computational loads, offline fixed samples have limited adaptability to complex and variable wind conditions, and existing neural network models are mostly designed for single turbine locations or local areas. When directly predicting wind shear coefficients, they are prone to amplifying wind speed errors at different height levels, making it difficult to quickly and stably obtain the negative shear distribution of multiple wind turbines in a wind farm. Summary of the Invention

[0006] The core objective of this invention is to integrate the spatial analytical advantages of high-precision numerical simulation with the prediction time advantages of deep learning algorithms, proposing a cross-scale physics-driven wind farm negative shear identification method and system. This method rapidly and stably obtains the negative shear distribution across multiple wind turbine units while preserving high-resolution wind field information. First, this invention constructs a training dataset using small-to-medium scale coupled high-resolution simulation results. Virtual measuring points are deployed around or at the center of the wind farm, and wind speeds and directions at different heights are extracted and used as inputs to a neural network. Simultaneously, wind speeds at multiple height levels in front of multiple turbine units within the farm are extracted as the outputs of the neural network. The nonlinear mapping relationship between the input free flow and the output wind profile in front of the turbine units is learned through an artificial neural network.

[0007] During the application phase, the model quickly predicts the wind speed at multiple heights in front of each unit over a future period based on the measured incoming wind speed and direction. Then, the predicted wind profile is fitted or segmented to obtain the wind shear coefficient or vertical wind speed gradient. When the wind shear coefficient is negative or the vertical gradient is negative, it is determined that there is a negative shear risk for the corresponding unit and height segment.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: A cross-scale physics data-driven method for identifying negative shear in wind farms includes: 1) Based on the results of small-scale coupled high-resolution wind field simulation, a multi-condition wind field sample library of the target wind farm under different incoming wind speeds, wind directions and boundary conditions is constructed, and data is obtained from the multi-condition wind field sample library, including multi-height wind speed and wind direction data at virtual measuring points and multi-height wind speed data in front of each wind turbine. 2) Based on the acquired data, the wind speed and wind direction information at multiple heights at virtual measuring points are constructed into neural network input feature vectors, and the wind speeds at multiple heights in front of each wind turbine are constructed into output target vectors. The input features and output targets are then standardized and divided into training, validation, and test sets. 3) Utilize the obtained training set to construct and train a multi-input multi-output neural network prediction model, establish a nonlinear mapping relationship between atmospheric free-flow wind speed and direction and wind speed at multiple heights in front of multiple wind turbines, and obtain the trained multi-input multi-output neural network prediction model. 4) Using the obtained training set, determine the hyperparameters of the multi-input multi-output neural network prediction model through grid search and cross-validation; input the test set into the multi-input multi-output neural network prediction model after determining the hyperparameters to obtain the predicted wind speed at different heights in front of each wind turbine, and use MSE, RMSE, MAE and R² indicators to evaluate the prediction accuracy and generalization ability. 5) Based on the predicted wind speeds at different heights in front of each wind turbine, wind profiles are fitted, wind shear gradient values ​​or wind shear coefficients are calculated, and the presence of negative shear wind conditions in front of each wind turbine is determined according to the positive or negative value of the wind shear gradient value or wind shear coefficient. The negative shear determination results are associated with the location of the corresponding wind turbine, and the negative shear identification results and their spatial distribution are output.

[0009] A further improvement of the present invention is that, in step 1), a multi-condition wind farm sample library is constructed, including: setting up virtual measuring points around and inside the wind farm, and setting up multiple height layers in front of each virtual measuring point and each wind turbine; selecting medium- and small-scale coupled high-resolution wind farm simulation conditions according to preset incoming wind speed, wind direction and boundary conditions. Multi-height wind speed and direction data at virtual measuring points and multi-height wind speed data in front of each wind turbine are extracted from each simulated operating condition and stored according to the simulated operating conditions to form the multi-operating-condition wind field sample library.

[0010] A further improvement of this invention is that, in step 2), constructing the input feature vector and the output target vector, and performing standardization and dataset partitioning, includes: The multi-height wind speed and wind direction data at virtual measuring points are constructed into an input feature vector according to a preset order, and the multi-height wind speed data in front of each wind turbine are constructed into an output target vector according to the wind turbine number and height layer order. The input feature vector and the output target vector are standardized respectively, and the corresponding mean and standard deviation are saved for the inverse standardization restoration of the model prediction results. Under the premise that the sample data are independent and identically distributed, all samples are randomly shuffled and divided into training set and test set according to a preset ratio; the K-fold cross-validation method is used to verify the model performance in the training set.

[0011] A further improvement of this invention is that, in step 3), constructing and training a multi-input multi-output neural network prediction model includes: A multi-input multi-output (MIMO) neural network prediction model is constructed based on the multi-input multi-output (MIMO) regression task in supervised learning. This model takes wind speed and direction at different virtual measurement points and altitudes as inputs, and wind speeds at multiple altitudes in front of multiple wind turbines as outputs. The model includes an input layer, hidden layers, and an output layer. The number of neurons in the input layer is consistent with the dimension of the input feature vector, and the number of neurons in the output layer is consistent with the dimension of the output target vector. The model is trained using the training set to obtain the final MIMO prediction model.

[0012] A further improvement of this invention is that, in step 4), the hyperparameters of the neural network model are determined through grid search and cross-validation, and the MSE, RMSE, MAE, and R² indices are used to evaluate the model's prediction accuracy and generalization ability for wind speeds at multiple altitudes under unknown wind conditions, including: Hyperparameter grid tuning: To improve the prediction accuracy and generalization ability of the multi-input multi-output neural network model, hyperparameters such as the number of hidden layers, the number of hidden layer neurons, the initial learning rate, the batch size, the L2 regularization coefficient Alpha, and the maximum number of iterations were tuned. Candidate value ranges were set for each hyperparameter, and different parameter combinations were traversed using a grid search method. The model performance of each parameter configuration was evaluated in conjunction with the K-fold cross-validation results. Finally, the parameter combination with lower validation error and better generalization ability was selected as the final model configuration. Training a multi-input multi-output neural network prediction model: The artificial neural network model is trained using training set samples, enabling the model to learn the nonlinear mapping relationship between atmospheric free-flow wind speed and direction and wind speeds at multiple heights in front of each turbine in the wind farm. During training, the model error is monitored through a loss function, and the model's coefficient of determination, mean square error, and other indicators on the validation samples are evaluated through cross-validation. If the training loss gradually decreases and stabilizes as the number of iterations increases, while the coefficient of determination on the validation set gradually increases and stabilizes, it indicates that the model has good learning ability and convergence performance. To test the prediction accuracy and generalization ability of the multi-input multi-output neural network prediction model: the neural network model was trained by inputting the free-flow wind speed and direction from the test set to obtain the predicted wind speed at different heights in front of each target wind turbine; the predicted wind speed was compared with the actual wind speed in the simulation sample, and the model performance was evaluated using indicators such as mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²).

[0013] A further improvement of the present invention is that, in step 5), the predicted wind speed at different heights in front of each wind turbine is fitted with a wind profile, and the wind shear gradient value or wind shear coefficient is calculated based on the fitting results. After obtaining the predicted wind speed at multiple heights in front of each wind turbine, the vertical wind speed distribution of each wind turbine is fitted with a wind profile, and the wind shear gradient value or wind shear coefficient is calculated based on the wind speed variation relationship at different heights. When the wind speed decreases with increasing height, or when the wind shear gradient value or wind shear coefficient is negative, it is determined that there is a negative wind shear condition in front of the corresponding wind turbine. According to the wind turbine number, the negative shear determination result, wind shear gradient value or wind shear coefficient are associated with the location and height layer of the corresponding wind turbine, and the location of the negative shear and the range of the affected height are output.

[0014] A cross-scale physics data-driven wind farm negative shear identification system includes: Sample library construction module: Based on the results of small-scale coupled high-resolution wind field simulation, it constructs a multi-condition wind field sample library of the target wind farm under different incoming wind speeds, wind directions and boundary conditions, and obtains data from the multi-condition wind field sample library, including multi-height wind speed and wind direction data at virtual measuring points and multi-height wind speed data in front of each wind turbine. Feature construction and preprocessing module: Based on the acquired data, it constructs multiple height wind speed and wind direction information at virtual measuring points into neural network input feature vectors, constructs multiple height wind speeds in front of each wind turbine into output target vectors, and performs standardization processing on the input features and output targets, as well as dividing the training set, validation set, and test set. Multi-height wind speed prediction module: Used to construct and train a multi-input multi-output neural network prediction model using the obtained training set, establish a nonlinear mapping relationship between atmospheric free-flow wind speed and direction and multi-height wind speed in front of multiple wind turbines, and obtain the trained multi-input multi-output neural network prediction model. Model optimization and evaluation module: This module is used to determine the hyperparameters of the multi-input multi-output neural network prediction model using the obtained training set through grid search and cross-validation; the test set is input into the multi-input multi-output neural network prediction model after the hyperparameters are determined to obtain the predicted wind speed at different heights in front of each wind turbine, and the prediction accuracy and generalization ability are evaluated using MSE, RMSE, MAE and R² indices. Negative shear recognition output module: It is used to fit the wind profile based on the predicted wind speed at different heights in front of each wind turbine, calculate the wind shear gradient value or wind shear coefficient, and determine whether there is negative shear wind condition in front of each wind turbine according to the positive or negative of the wind shear gradient value or wind shear coefficient. It associates the negative shear judgment result with the position of the corresponding wind turbine and outputs the negative shear recognition result and its spatial distribution.

[0015] A further improvement of this invention is that, in the sample library construction module, a multi-condition wind farm sample library is constructed, including: setting up virtual measuring points around and inside the wind farm, and setting up multiple height layers in front of each virtual measuring point and each wind turbine; selecting medium- and small-scale coupled high-resolution wind farm simulation conditions according to preset incoming wind speed, wind direction and boundary conditions. Multi-height wind speed and direction data at virtual measuring points and multi-height wind speed data in front of each wind turbine are extracted from each simulated operating condition and stored according to the simulated operating conditions to form the multi-operating-condition wind field sample library.

[0016] A further improvement of this invention lies in that, in the feature construction and preprocessing module, the construction of the input feature vector and the output target vector, as well as the standardization processing and dataset partitioning, include: The multi-height wind speed and wind direction data at virtual measuring points are constructed into an input feature vector according to a preset order, and the multi-height wind speed data in front of each wind turbine are constructed into an output target vector according to the wind turbine number and height layer order. The input feature vector and the output target vector are standardized respectively, and the corresponding mean and standard deviation are saved for the inverse standardization restoration of the model prediction results. Under the premise that the sample data are independent and identically distributed, all samples are randomly shuffled and divided into training set and test set according to a preset ratio; the K-fold cross-validation method is used to verify the model performance in the training set.

[0017] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a cross-scale physical field data-driven wind farm negative shear identification method.

[0018] Compared with the prior art, the present invention has at least the following beneficial technical effects: This invention discloses a cross-scale physical field data-driven method and system for identifying negative shear in wind farms, which effectively solves the technical problems of high computational cost, poor identification stability, difficulty in characterizing the coupling characteristics of the entire field, and limited identification coverage in existing technologies, and has good engineering application value.

[0019] This invention relies on a pre-trained neural network model, eliminating the need for repeated high-resolution numerical simulations. It only requires inputting free-flow wind speed and direction parameters to quickly output wind speed results for multiple units at multiple heights, significantly reducing computing power and computation time, and enabling efficient and rapid prediction of negative shear conditions in wind farms.

[0020] Meanwhile, the present invention adopts a two-step identification strategy of first predicting the wind profile and then calculating the wind shear, avoiding direct fitting of the highly sensitive wind shear coefficient, effectively suppressing the identification deviation caused by wind speed fluctuations and boundary disturbances, and greatly improving the stability and accuracy of negative shear identification.

[0021] Compared to traditional empirical formulas and linear statistical models, this invention utilizes the excellent nonlinear fitting capability of neural networks to accurately characterize the complex coupling relationship between free flow, wind direction, terrain acceleration, and wake interference, realistically reproducing the three-dimensional wind field distribution of wind farms and adapting to the identification needs of complex wind field conditions.

[0022] Furthermore, this invention can simultaneously predict wind speed and determine negative shear across multiple wind turbines and at multiple height levels, outputting the spatial distribution results of negative shear, thus overcoming the limitations of traditional single-point identification. By combining data standardization, grid search tuning, and cross-validation mechanisms, it effectively suppresses model overfitting, significantly improves model generalization ability, and can adapt to different practical engineering application scenarios in wind farms. Attached Figure Description

[0023] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a flowchart of the overall process for neural network negative shear recognition method.

[0025] Figure 2 This is a schematic diagram of the arrangement of virtual measuring points, wind turbine units, and height layers in an example.

[0026] Figure 3 This is a schematic diagram illustrating the ANN architecture.

[0027] Figure 4 This represents the learning curve of the ANN model during the training process.

[0028] Figure 5 This is a schematic diagram showing the predicted and true values ​​of the incoming wind speed in front of the wind turbine. Detailed Implementation

[0029] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0030] In the description of this invention, it should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0031] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0032] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0033] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0034] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0035] Example 1 like Figure 1 As shown, this invention provides a cross-scale physics field data-driven method for identifying negative wind shear in wind farms. It constructs a sample library based on high-resolution wind farm simulation data, builds neural network feature vectors, and uses a two-step method to predict wind shear. Specifically, it includes: 1) Based on the results of small-to-medium scale coupled high-resolution wind field simulations, a multi-condition wind field sample library is constructed for the target wind farm under different incoming wind speeds, wind directions, and boundary conditions. Data is obtained from the multi-condition wind field sample library, including multi-height wind speed and wind direction data at virtual measuring points and multi-height wind speed data in front of each wind turbine. Figure 2 The left side shows the constructed virtual measuring points and the distribution of wind turbine units, while the right side shows the height layer of the extracted wind speed. 2) Based on the acquired data, the wind speed and direction information at multiple heights at virtual measuring points are used to construct the neural network input feature vector, and the wind speeds at multiple heights in front of each wind turbine are used to construct the output target vector, thus obtaining... Figure 3The ANN architecture is shown, and the input features and output targets are standardized and the training set, validation set, and test set are divided. 3) Using the obtained training set, construct and train a multi-input multi-output neural network prediction model to establish a nonlinear mapping relationship between atmospheric free-flow wind speed and direction and wind speeds at multiple heights in front of multiple wind turbines, thus obtaining the trained multi-input multi-output neural network prediction model. Figure 4 The learning curve of the ANN model during the training process; 4) Using the obtained training set, determine the hyperparameters of the multi-input multi-output neural network prediction model through grid search and cross-validation; input the test set into the multi-input multi-output neural network prediction model after determining the hyperparameters, and obtain... Figure 5 The predicted wind speeds at different heights in front of each wind turbine are shown, and the prediction accuracy and generalization ability are evaluated using MSE, RMSE, MAE and R² indices. 5) Based on the predicted wind speeds at different heights in front of each wind turbine, wind profiles are fitted, wind shear gradient values ​​or wind shear coefficients are calculated, and the presence of negative shear wind conditions in front of each wind turbine is determined according to the positive or negative value of the wind shear gradient value or wind shear coefficient. The negative shear determination results are associated with the location of the corresponding wind turbine, and the negative shear identification results and their spatial distribution are output.

[0036] In step 1) of this embodiment, a multi-condition wind farm sample library is constructed, including: setting up virtual measuring points around and inside the wind farm, and setting up multiple height layers in front of each virtual measuring point and each wind turbine; selecting medium- and small-scale coupled high-resolution wind farm simulation conditions according to preset incoming wind speed, wind direction and boundary conditions. Multi-height wind speed and direction data at virtual measuring points and multi-height wind speed data in front of each wind turbine are extracted from each simulated operating condition and stored according to the simulated operating conditions to form the multi-operating-condition wind field sample library.

[0037] In step 2) of this embodiment, the input feature vector and output target vector are constructed and standardized and the dataset is partitioned, including: The multi-height wind speed and wind direction data at virtual measuring points are constructed into an input feature vector according to a preset order, and the multi-height wind speed data in front of each wind turbine are constructed into an output target vector according to the wind turbine number and height layer order. The input feature vector and the output target vector are standardized respectively, and the corresponding mean and standard deviation are saved for the inverse standardization restoration of the model prediction results. Under the premise that the sample data are independent and identically distributed, all samples are randomly shuffled and divided into training set and test set according to a preset ratio; the K-fold cross-validation method is used to verify the model performance in the training set.

[0038] In step 3) of this embodiment, constructing and training a multi-input multi-output neural network prediction model includes: A multi-input multi-output (MIMO) neural network prediction model is constructed based on the multi-input multi-output (MIMO) regression task in supervised learning. This model takes wind speed and direction at different virtual measurement points and altitudes as inputs, and wind speeds at multiple altitudes in front of multiple wind turbines as outputs. The model includes an input layer, hidden layers, and an output layer. The number of neurons in the input layer is consistent with the dimension of the input feature vector, and the number of neurons in the output layer is consistent with the dimension of the output target vector. The model is trained using the training set to obtain the final MIMO prediction model. In step 4) of this embodiment, the hyperparameters of the neural network model are determined through grid search and cross-validation, and the MSE, RMSE, MAE, and R² indices are used to evaluate the model's prediction accuracy and generalization ability for wind speeds at multiple altitudes under unknown wind conditions, including: Hyperparameter grid tuning: To improve the prediction accuracy and generalization ability of the multi-input multi-output neural network model, hyperparameters such as the number of hidden layers, the number of hidden neurons, the initial learning rate, the batch size, the L2 regularization coefficient Alpha, and the maximum number of iterations were tuned. Candidate value ranges were set for each hyperparameter, and different parameter combinations were traversed using a grid search method. The model performance of each parameter configuration was evaluated based on the results of K-fold cross-validation. Finally, the parameter combination with lower validation error and better generalization ability was selected as the final model configuration. Training a multi-input multi-output neural network prediction model: The artificial neural network model is trained using training set samples, enabling the model to learn the nonlinear mapping relationship between atmospheric free-flow wind speed and direction and wind speeds at multiple heights in front of each turbine in the wind farm. During training, the model error is monitored through a loss function, and the model's coefficient of determination, mean square error, and other indicators on the validation samples are evaluated through cross-validation. If the training loss gradually decreases and stabilizes as the number of iterations increases, while the coefficient of determination on the validation set gradually increases and stabilizes, it indicates that the model has good learning ability and convergence performance. The prediction accuracy and generalization ability of the multi-input multi-output neural network prediction model were tested: the neural network model was trained using the free-flow wind speed and direction from the test set to obtain the predicted wind speed at different heights in front of each target wind turbine. The predicted wind speed was compared with the actual wind speed in the simulation samples, and the model performance was evaluated using metrics such as mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²).

[0039] In step 5) of this embodiment, wind profiles are fitted to the predicted wind speeds at different heights in front of each wind turbine, and wind shear gradient values ​​or wind shear coefficients are calculated based on the fitting results. After obtaining the predicted wind speed at multiple heights in front of each wind turbine, the vertical wind speed distribution of each wind turbine is fitted with a wind profile, and the wind shear gradient value or wind shear coefficient is calculated based on the wind speed variation relationship at different heights. When the wind speed decreases with increasing height, or when the wind shear gradient value or wind shear coefficient is negative, it is determined that there is a negative wind shear condition in front of the corresponding wind turbine. According to the wind turbine number, the negative shear determination result, wind shear gradient value or wind shear coefficient are associated with the location and height layer of the corresponding wind turbine, and the location of the negative shear and the range of the affected height are output.

[0040] Example 2 This invention provides a cross-scale physics data-driven method for identifying negative shear in wind farms, comprising: Step 1, construct a multi-condition wind farm sample library: Based on the results of small-scale coupled high-resolution wind farm simulation, virtual measuring points are set up around and inside the wind farm to extract wind speed and direction data at multiple measuring points and multiple heights, as well as wind speed data at multiple heights in front of multiple turbines in the farm.

[0041] Step 2, construct the input feature vector and output target vector: use the free-flow wind speed and direction information of the atmosphere at virtual measuring points around and within the wind farm as the model input features, including wind speed data at different heights of multiple measuring points, and wind direction data at the corresponding locations or hub heights.

[0042] Using the wind speeds at different heights in front of each wind turbine in the wind farm as the output target, a multi-input multi-output regression task is constructed. For n wind turbines in the target wind farm, the output target vector can be represented as the set of wind speeds in front of the n turbines at multiple height levels, used to characterize the inflow wind profile of each turbine.

[0043] Step 3: Determine the two-step wind shear prediction strategy: In view of the problem that the wind shear coefficient is sensitive to wind speed disturbances at different height levels and that direct prediction is prone to error amplification and training instability, a two-step prediction scheme is adopted.

[0044] The first step is to use a neural network model to establish a nonlinear mapping relationship between the free-flow wind speed and direction and the wind speed at multiple altitudes in front of the generator set, and to predict the wind speed at different altitudes in front of each target generator set.

[0045] The second step involves fitting wind profiles based on multiple predicted height wind speed data, calculating the gradient value or wind shear coefficient that characterizes vertical wind shear, and determining whether negative shear exists based on its positive or negative sign.

[0046] Step 4, Sample Data Standardization: Standardization makes the distribution of features more uniform, preventing features with large values ​​from dominating the training process, and improving the convergence speed and training stability of gradient-based neural network models. The mean and standard deviation of the input and output data are saved during standardization for destandardization of the model's prediction results.

[0047] Step 5: Divide the data into training, validation, and test sets. Assuming the sample data are independent and identically distributed, all samples are randomly shuffled and divided into training and test sets according to a preset ratio. During training, K-fold cross-validation is further used to validate the model performance, reducing the impact of random partitioning and improving the reliability of the model evaluation results.

[0048] Step 6: Construct a multi-input multi-output neural network prediction model: Based on the multi-input multi-output regression task in supervised learning, an artificial neural network model is constructed. This model uses the wind speed and direction of the free atmospheric flow at different measuring points and altitudes as input, and the wind speed at multiple altitudes in front of multiple wind turbines as output. The neural network model includes an input layer, hidden layers, and an output layer. The number of neurons in the input layer is consistent with the dimension of the input feature vector, and the number of neurons in the output layer is consistent with the dimension of the output target vector.

[0049] Step 7, Hyperparameter Grid Tuning: To improve the prediction accuracy and generalization ability of the neural network model, hyperparameters such as the number of hidden layers, the number of hidden neurons, the initial learning rate, the batch size, the L2 regularization coefficient Alpha, and the maximum number of iterations are tuned. Candidate value ranges are set for each hyperparameter, and different parameter combinations are traversed using a grid search method. The model performance of each parameter configuration is evaluated based on the K-fold cross-validation results. Finally, the parameter combination with lower validation error and better generalization ability is selected as the final model configuration.

[0050] Step 8: Train the multi-height wind speed prediction model: The artificial neural network model is trained using the training set samples, enabling the model to learn the nonlinear mapping relationship between the free-flowing wind speed and direction and the wind speed at multiple heights in front of each turbine in the wind farm. During training, the model error is monitored using a loss function, and cross-validation is used to evaluate the model's coefficient of determination, mean squared error, and other indicators on the validation samples. If, with increasing iterations, the training loss gradually decreases and stabilizes, while the validation set coefficient of determination gradually increases and stabilizes, it indicates that the model possesses good learning ability and convergence performance.

[0051] Step 9: Test the model's prediction accuracy and generalization ability: Input the free-flow wind speed and direction from the test set into the trained neural network model to obtain the predicted wind speed at different heights in front of each target wind turbine. Compare the predicted wind speed with the actual wind speed in the simulation samples, and evaluate the model performance using metrics such as mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²).

[0052] Step 10: Fitting wind profiles based on predicted wind speeds and calculating wind shear parameters: After obtaining the predicted wind speeds at multiple heights in front of each wind turbine, a wind profile is fitted to the vertical wind speed distribution of each turbine, and the wind shear coefficient is calculated based on the wind speed variation relationship at different heights. Since this wind shear parameter is not directly predicted by the neural network, but is obtained by further fitting and calculating the predicted multi-height wind speeds, it can reduce the training instability problem caused by noise amplification when directly predicting the wind shear coefficient.

[0053] Step 11, Determine if there is negative wind shear: Determine if there is negative wind shear in front of the wind turbine based on the fitting results. When the wind speed decreases with increasing height, or the calculated wind shear coefficient is negative, it is determined that there is negative wind shear in front of the corresponding turbine; when the wind speed increases with increasing height, or the wind shear coefficient is positive, it is determined to be normal positive wind shear.

[0054] Step 12, Output the wind farm negative shear identification results: Output the wind speeds at various heights in front of each turbine predicted by the model, the fitted wind shear coefficients, and the negative shear determination results. The output results include the wind profile in front of each turbine, the wind shear gradient value, the location of the negative shear, and the corresponding wind condition number.

[0055] This enables a complete technical process, from inputting atmospheric free-flow wind speed and direction, to predicting the wind profile in front of multiple wind turbines within the wind farm, and then to identifying negative shear wind conditions.

[0056] Example 3 This invention provides a cross-scale physics field data-driven wind farm negative shear identification system, comprising: Sample library construction module: Based on the results of small-scale coupled high-resolution wind field simulation, it constructs a multi-condition wind field sample library of the target wind farm under different incoming wind speeds, wind directions and boundary conditions, and obtains data from the multi-condition wind field sample library, including multi-height wind speed and wind direction data at virtual measuring points and multi-height wind speed data in front of each wind turbine. Feature construction and preprocessing module: Based on the acquired data, it constructs multiple height wind speed and wind direction information at virtual measuring points into neural network input feature vectors, constructs multiple height wind speeds in front of each wind turbine into output target vectors, and performs standardization processing on the input features and output targets, as well as dividing the training set, validation set, and test set. Multi-height wind speed prediction module: Used to construct and train a multi-input multi-output neural network prediction model using the obtained training set, establish a nonlinear mapping relationship between atmospheric free-flow wind speed and direction and multi-height wind speed in front of multiple wind turbines, and obtain the trained multi-input multi-output neural network prediction model. Model optimization and evaluation module: This module is used to determine the hyperparameters of the multi-input multi-output neural network prediction model using the obtained training set through grid search and cross-validation; the test set is input into the multi-input multi-output neural network prediction model after the hyperparameters are determined to obtain the predicted wind speed at different heights in front of each wind turbine, and the prediction accuracy and generalization ability are evaluated using MSE, RMSE, MAE and R² indices. Negative shear recognition output module: It is used to fit the wind profile based on the predicted wind speed at different heights in front of each wind turbine, calculate the wind shear gradient value or wind shear coefficient, and determine whether there is negative shear wind condition in front of each wind turbine according to the positive or negative of the wind shear gradient value or wind shear coefficient. It associates the negative shear judgment result with the position of the corresponding wind turbine and outputs the negative shear recognition result and its spatial distribution.

[0057] Example 4 The following example uses a wind farm with 40 turbines in a complex terrain, but the invention is not limited to this scale. First, high-resolution three-dimensional wind field simulation results are used to set up multiple virtual measuring points around or within the wind farm. The wind speed and direction at different heights at the measuring points are extracted, and the wind speed at different heights in front of each turbine is also extracted.

[0058] Secondly, the samples were processed. In this embodiment, 5769 independent operating conditions were extracted from the simulation results. The free-flow wind speed and direction at various altitudes in each operating condition were concatenated into an input feature vector, and the wind speeds in front of the 40 units at 8 altitude levels were concatenated into an output target vector.

[0059] Next, an MLPRegressor model is built in a Python environment based on scikit-learn or an equivalent machine learning platform. The input and output are standardized, and the samples are randomly shuffled and divided into training and test sets at 80% and 20% respectively. Five-fold cross-validation is performed on the training set to monitor the training loss and verify the trend of R² changes, avoiding overfitting.

[0060] Then, the number of hidden layers, the number of hidden layer neurons, the learning rate, the batch size, and the L2 regularization coefficient are tuned using a grid. Optimal parameters can be set as follows: L2 regularization Alpha = 0.005, batch size = 150, initial learning rate = 0.005, maximum number of iterations = 200, number of hidden layer neurons = 45, activation function = ReLU, solver = Adam, and learning rate is adaptive.

[0061] After the model training is completed, new free-flow wind speed and direction are input, and the predicted wind speed values ​​at multiple heights in front of each unit are output. After destandardizing the predicted values, the wind shear index or vertical gradient is calculated for each unit, negative shear is identified, and a list of units with negative shear, their risk level, and height range are output.

[0062] In a preferred verification embodiment, the average loss during model training gradually decreases and stabilizes at around 0.07, while the R² on the validation set gradually increases and stabilizes at around 0.89. On the test set, the average MSE is 0.7581, the average RMSE is 0.8707, the average MAE is 0.5991, and the average R² is 0.8902, which meets the requirements of engineering applications.

[0063] Example 5 The present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of a cross-scale physical field data-driven wind farm negative shear identification method.

[0064] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0065] This application is described with reference to flowchart illustrations and / or block diagrams of methods, systems, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0066] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0067] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0068] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0069] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for identifying negative shear in wind farms driven by cross-scale physical field data, characterized in that, include: 1) Based on the results of small-scale coupled high-resolution wind field simulation, a multi-condition wind field sample library of the target wind farm under different incoming wind speeds, wind directions and boundary conditions is constructed, and data is obtained from the multi-condition wind field sample library, including multi-height wind speed and wind direction data at virtual measuring points and multi-height wind speed data in front of each wind turbine. 2) Based on the acquired data, the wind speed and wind direction information at multiple heights at virtual measuring points are constructed into neural network input feature vectors, and the wind speeds at multiple heights in front of each wind turbine are constructed into output target vectors. The input features and output targets are then standardized and divided into training, validation, and test sets. 3) Utilize the obtained training set to construct and train a multi-input multi-output neural network prediction model, establish a nonlinear mapping relationship between atmospheric free-flow wind speed and direction and wind speed at multiple heights in front of multiple wind turbines, and obtain the trained multi-input multi-output neural network prediction model. 4) Using the obtained training set, determine the hyperparameters of the multi-input multi-output neural network prediction model through grid search and cross-validation; input the test set into the multi-input multi-output neural network prediction model after determining the hyperparameters to obtain the predicted wind speed at different heights in front of each wind turbine, and use MSE, RMSE, MAE and R² indicators to evaluate the prediction accuracy and generalization ability. 5) Based on the predicted wind speeds at different heights in front of each wind turbine, wind profiles are fitted, wind shear gradient values ​​or wind shear coefficients are calculated, and the presence of negative shear wind conditions in front of each wind turbine is determined according to the positive or negative value of the wind shear gradient value or wind shear coefficient. The negative shear determination results are associated with the location of the corresponding wind turbine, and the negative shear identification results and their spatial distribution are output.

2. The method for identifying negative shear in wind farms driven by cross-scale physical field data according to claim 1, characterized in that, In step 1), a multi-condition wind farm sample library is constructed, including: setting up virtual measuring points around and inside the wind farm, and setting up multiple height layers in front of each virtual measuring point and each wind turbine; selecting small-scale coupled high-resolution wind farm simulation conditions according to preset incoming wind speed, wind direction and boundary conditions. Multi-height wind speed and direction data at virtual measuring points and multi-height wind speed data in front of each wind turbine are extracted from each simulated operating condition and stored according to the simulated operating conditions to form the multi-operating-condition wind field sample library.

3. The method for identifying negative shear in wind farms driven by cross-scale physical field data according to claim 1, characterized in that, In step 2), the input feature vector and output target vector are constructed, and standardization and dataset partitioning are performed, including: The multi-height wind speed and wind direction data at virtual measuring points are constructed into an input feature vector according to a preset order, and the multi-height wind speed data in front of each wind turbine are constructed into an output target vector according to the wind turbine number and height layer order. The input feature vector and the output target vector are standardized respectively, and the corresponding mean and standard deviation are saved for the inverse standardization restoration of the model prediction results. Under the premise that the sample data are independent and identically distributed, all samples are randomly shuffled and divided into training set and test set according to a preset ratio; the K-fold cross-validation method is used to verify the model performance in the training set.

4. The method for identifying negative shear in wind farms driven by cross-scale physical field data according to claim 1, characterized in that, In step 3), a multi-input multi-output neural network prediction model is constructed and trained, including: A multi-input multi-output (MIMO) neural network prediction model is constructed based on the multi-input multi-output (MIMO) regression task in supervised learning. This model takes wind speed and direction at different virtual measurement points and altitudes as inputs, and wind speeds at multiple altitudes in front of multiple wind turbines as outputs. The model includes an input layer, hidden layers, and an output layer. The number of neurons in the input layer is consistent with the dimension of the input feature vector, and the number of neurons in the output layer is consistent with the dimension of the output target vector. The model is trained using the training set to obtain the final MIMO prediction model.

5. The method for identifying negative shear in wind farms driven by cross-scale physical field data according to claim 1, characterized in that, In step 4), the hyperparameters of the neural network model are determined through grid search and cross-validation. The model's prediction accuracy and generalization ability for wind speeds at multiple altitudes under unknown wind conditions are evaluated using MSE, RMSE, MAE, and R² metrics, including: Hyperparameter grid tuning: To improve the prediction accuracy and generalization ability of the multi-input multi-output neural network model, hyperparameters such as the number of hidden layers, the number of hidden layer neurons, the initial learning rate, the batch size, the L2 regularization coefficient Alpha, and the maximum number of iterations were tuned. Candidate value ranges were set for each hyperparameter, and different parameter combinations were traversed using a grid search method. The model performance of each parameter configuration was evaluated in conjunction with the K-fold cross-validation results. Finally, the parameter combination with lower validation error and better generalization ability was selected as the final model configuration. Training a multi-input multi-output neural network prediction model: The artificial neural network model is trained using training set samples, enabling the model to learn the nonlinear mapping relationship between atmospheric free-flow wind speed and direction and wind speeds at multiple heights in front of each turbine in the wind farm. During training, the model error is monitored through a loss function, and the model's coefficient of determination, mean square error, and other indicators on the validation samples are evaluated through cross-validation. If the training loss gradually decreases and stabilizes as the number of iterations increases, while the coefficient of determination on the validation set gradually increases and stabilizes, it indicates that the model has good learning ability and convergence performance. To test the prediction accuracy and generalization ability of the multi-input multi-output neural network prediction model: the neural network model was trained by inputting the free-flow wind speed and direction from the test set to obtain the predicted wind speed at different heights in front of each target wind turbine; the predicted wind speed was compared with the actual wind speed in the simulation sample, and the model performance was evaluated using indicators such as mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²).

6. The method for identifying negative shear in wind farms driven by cross-scale physical field data according to claim 1, characterized in that, In step 5), wind profiles are fitted to the predicted wind speeds at different heights in front of each wind turbine, and wind shear gradient values ​​or wind shear coefficients are calculated based on the fitting results. After obtaining the predicted wind speed at multiple heights in front of each wind turbine, the vertical wind speed distribution of each wind turbine is fitted with a wind profile, and the wind shear gradient value or wind shear coefficient is calculated based on the wind speed variation relationship at different heights. When the wind speed decreases with increasing height, or when the wind shear gradient value or wind shear coefficient is negative, it is determined that there is a negative wind shear condition in front of the corresponding wind turbine. According to the wind turbine number, the negative shear determination result, wind shear gradient value or wind shear coefficient are associated with the location and height layer of the corresponding wind turbine, and the location of the negative shear and the range of the affected height are output.

7. A cross-scale physics field data-driven wind farm negative shear identification system, characterized in that, include: Sample library construction module: Based on the results of small-scale coupled high-resolution wind field simulation, it constructs a multi-condition wind field sample library of the target wind farm under different incoming wind speeds, wind directions and boundary conditions, and obtains data from the multi-condition wind field sample library, including multi-height wind speed and wind direction data at virtual measuring points and multi-height wind speed data in front of each wind turbine. Feature construction and preprocessing module: Based on the acquired data, it constructs multiple height wind speed and wind direction information at virtual measuring points into neural network input feature vectors, constructs multiple height wind speeds in front of each wind turbine into output target vectors, and performs standardization processing on the input features and output targets, as well as dividing the training set, validation set, and test set. Multi-height wind speed prediction module: Used to construct and train a multi-input multi-output neural network prediction model using the obtained training set, establish a nonlinear mapping relationship between atmospheric free-flow wind speed and direction and multi-height wind speed in front of multiple wind turbines, and obtain the trained multi-input multi-output neural network prediction model. Model optimization and evaluation module: This module is used to determine the hyperparameters of the multi-input multi-output neural network prediction model using the obtained training set through grid search and cross-validation; the test set is input into the multi-input multi-output neural network prediction model after the hyperparameters are determined to obtain the predicted wind speed at different heights in front of each wind turbine, and the prediction accuracy and generalization ability are evaluated using MSE, RMSE, MAE and R² indices. Negative shear recognition output module: It is used to fit the wind profile based on the predicted wind speed at different heights in front of each wind turbine, calculate the wind shear gradient value or wind shear coefficient, and determine whether there is negative shear wind condition in front of each wind turbine according to the positive or negative of the wind shear gradient value or wind shear coefficient. It associates the negative shear judgment result with the position of the corresponding wind turbine and outputs the negative shear recognition result and its spatial distribution.

8. The cross-scale physical field data-driven wind farm negative shear identification system according to claim 7, characterized in that, In the sample library construction module, a multi-condition wind farm sample library is constructed, including: setting up virtual measuring points around and inside the wind farm, and setting up multiple height layers in front of each virtual measuring point and each wind turbine; selecting medium- and small-scale coupled high-resolution wind farm simulation conditions according to preset incoming wind speed, wind direction and boundary conditions. Multi-height wind speed and direction data at virtual measuring points and multi-height wind speed data in front of each wind turbine are extracted from each simulated operating condition and stored according to the simulated operating conditions to form the multi-operating-condition wind field sample library.

9. A cross-scale physics field data-driven wind farm negative shear identification system according to claim 7, characterized in that, The feature construction and preprocessing module constructs the input feature vector and output target vector, performs standardization, and partitions the dataset, including: The multi-height wind speed and wind direction data at virtual measuring points are constructed into an input feature vector according to a preset order, and the multi-height wind speed data in front of each wind turbine are constructed into an output target vector according to the wind turbine number and height layer order. The input feature vector and the output target vector are standardized respectively, and the corresponding mean and standard deviation are saved for the inverse standardization restoration of the model prediction results. Under the premise that the sample data are independent and identically distributed, all samples are randomly shuffled and divided into training set and test set according to a preset ratio; the K-fold cross-validation method is used to verify the model performance in the training set.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the cross-scale physical field data-driven wind farm negative shear identification method according to any one of claims 1-6.

Images