Wind power plant flow field characteristic prediction method based on characteristic sampling and related device
By constructing a feature parameter space and training a machine learning proxy model, the contradiction between computational cost and accuracy in wind farm flow field characteristic prediction is resolved, achieving efficient and accurate flow field characteristic prediction, reducing computational resource consumption and improving prediction efficiency.
Patent Information
- Application Number
- CN202511779794.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-03-06
AI Technical Summary
Existing technologies struggle to balance computational cost and accuracy in predicting wind farm flow field characteristics. High-precision CFD methods consume enormous computational resources and are inefficient, while low-cost empirical models lack sufficient accuracy to meet the accuracy requirements of engineering applications.
A feature sampling-based approach is adopted. By constructing an n-dimensional feature parameter space, representative sample points are selected using methods such as Latin hypercube sampling. High-fidelity CFD simulation is performed to build a flow field characteristic database and train a machine learning surrogate model to achieve flow field characteristic prediction.
It significantly reduces computational costs, improves prediction efficiency, and can achieve accurate flow field characteristic prediction within milliseconds or seconds. It also has good generalization ability, ensuring prediction accuracy and continuity.
Smart Images
Figure CN121615482A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of wind farm technology and relates to a method and related device for predicting wind farm flow field characteristics based on feature sampling. Background Technology
[0002] In the fields of wind farm micro-site selection, power generation prediction, and site optimization control, accurately reproducing the flow field (such as the distribution of wind speed and turbulence intensity) inside and around the wind farm under different inflow wind conditions is crucial. Currently, high-fidelity numerical simulation, such as computational fluid dynamics (CFD), is the mainstream technical means to obtain this kind of information, mainly in the following two aspects.
[0003] First, there's the CFD simulation method based on refined discretization. To accurately reproduce the wind farm and surrounding wind resource distribution under all inflow wind conditions, this method typically requires exhaustive refined discretization of boundary wind speeds, wind directions, and other conditions. For example, wind speeds are discretized at 0.5 m / s intervals from the cut-in wind speed to the cut-out wind speed, and wind directions are discretized across the entire 360° range at 1° intervals. Then, for each discretized combination of wind conditions (wind speed, wind direction), a complete CFD numerical simulation is performed. Finally, the simulation results for all conditions are summarized to form a de facto flow field database. This technique has two main drawbacks: First, it consumes enormous computational resources and is inefficient. The wind condition parameter space is vast, and exhaustive discretization leads to an exponential increase in the number of conditions to be simulated. Each CFD simulation itself requires significant computation time and expensive computational resources, making the total cost and time of building a complete database prohibitively high. Second, it lacks exhaustiveness and has poor generalization ability. Due to limitations in computing resources, true "exhaustive search" is impossible in practical applications. The setting of the discrete interval is subjective; too large an interval leads to the loss of crucial information, while too small an interval results in unbearable computational burden. Therefore, the database established by this method is inherently sparse and discontinuous. For intermediate operating conditions not directly simulated, the flow field characteristics usually need to be estimated through interpolation, resulting in unreliable accuracy and weak generalization ability. Secondly, there are traditional empirical or simplified model methods. To circumvent the high computational cost of CFD, some existing technologies employ rapid prediction methods based on empirical formulas or simplified physical models. These techniques mainly suffer from low prediction accuracy and poor universality. These methods often oversimplify physical processes such as complex terrain and the interaction of closely spaced wind turbine wakes, leading to significant deviations between the predicted results and the actual flow field. This makes it difficult to meet the accuracy requirements of engineering applications, especially unsuitable for wind farms in complex terrain.
[0004] In summary, the core contradiction in existing technologies is that high-precision CFD methods are computationally expensive and inefficient, while low-cost empirical models lack sufficient accuracy. Therefore, there is an urgent need in this field for a scheme to obtain wind farm flow field characteristics that can balance computational efficiency and prediction accuracy. Summary of the Invention
[0005] The purpose of this invention is to provide a method and related apparatus for predicting wind farm flow field characteristics based on feature sampling, so as to solve the technical problem that existing wind farm flow field characteristic prediction methods are difficult to balance computational cost and computational accuracy.
[0006] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides a method for predicting the flow field characteristics of a wind farm based on feature sampling, comprising the following steps: Determine the key input parameters that affect the flow field characteristics of the wind farm and their value ranges to form an n-dimensional characteristic parameter space; Within the n-dimensional feature parameter space, a representative set of sample points is selected using a space-filling design method; For each feature sampling point in the sample point set, a high-fidelity CFD numerical simulation is performed to obtain flow field data, thereby constructing a flow field characteristic database. A machine learning proxy model is trained using the aforementioned flow field characteristic database to obtain a wind farm flow field characteristic prediction model. By inputting wind condition parameters into the wind farm flow field characteristic prediction model, the overall flow field distribution and key performance indicators are predicted.
[0007] Furthermore, the key input parameters affecting the flow field characteristics of the wind farm include at least the inflow wind speed, inflow wind direction, and inflow turbulence intensity.
[0008] Furthermore, the space-filling design method is the Latin hypercube sampling method, the Sobol sequence sampling method, or the Halton sequence sampling method.
[0009] Furthermore, the step of performing high-fidelity CFD numerical simulation to obtain flow field data for each feature sampling point in the sample point set, representing the operating condition, specifically includes: Construct a three-dimensional computational domain mesh that includes the actual terrain of the wind farm and the specific locations of the wind turbines; Based on the operating parameters corresponding to the feature sampling points, set the entry boundary conditions of the computational domain; Based on the pre-defined computational domain, a turbulence model and a fan model are used for simulation calculations to obtain flow field data.
[0010] Furthermore, each record in the flow field characteristic database includes an input vector and output data; the input vector is the feature sampling parameter corresponding to the working condition, and the output data includes full-field flow field data and / or key characteristic parameters.
[0011] Furthermore, the step of training a machine learning surrogate model using the flow field characteristic database to obtain a wind farm flow field characteristic prediction model specifically includes: The high-dimensional full-field flow field data in the flow field characteristic database are subjected to intrinsic orthogonal decomposition to reduce the dimensionality, and the flow field is represented as a linear combination of several POD basis modes. The learning objective of the model is then transformed into predicting low-dimensional POD coefficients. Using the feature sampling parameters as input features and the low-dimensional POD coefficients as output labels, a machine learning surrogate model is trained to establish a mapping relationship from wind condition parameters to the output flow field, thereby obtaining a wind farm flow field characteristic prediction model.
[0012] Furthermore, the machine learning proxy model is a Gaussian process regression, a neural network, or a support vector machine.
[0013] Secondly, the present invention provides a wind farm flow field characteristic prediction system based on feature sampling, comprising: The feature parameter space definition module is used to determine the key input parameters that affect the flow field characteristics of the wind farm and their value ranges, thus forming an n-dimensional feature parameter space. The feature sampling module is used to select a representative set of sample points in the n-dimensional feature parameter space using a space filling design method. The database construction module is used to perform high-fidelity CFD numerical simulations on the working conditions represented by each feature sampling point in the sample point set to obtain flow field data, thereby constructing a flow field characteristic database. The model training module is used to train a machine learning proxy model using the flow field characteristic database to obtain a wind farm flow field characteristic prediction model. The prediction module is used to input wind condition parameters into the wind farm flow field characteristic prediction model to predict the overall flow field distribution and key performance indicators.
[0014] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the wind farm flow field characteristic prediction method based on feature sampling as described above.
[0015] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the wind farm flow field characteristic prediction method based on feature sampling.
[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention discloses a method and related apparatus for predicting wind farm flow field characteristics based on feature sampling. The invention first constructs an n-dimensional feature parameter space, then replaces thousands of exhaustive simulations with tens to hundreds of representative CFD simulations through intelligent feature sampling, reducing the computational cost of database construction by one to two orders of magnitude. Finally, the surrogate model trained on the database can achieve flow field predictions at the millisecond or second level, with extremely high efficiency. Furthermore, the prediction basis of this invention originates from high-fidelity CFD simulations. By learning the inherent physical laws of these high-precision data, the surrogate model can provide continuous, smooth, and accurate predictions across the entire parameter space, overcoming the shortcomings of traditional interpolation methods in terms of poor accuracy at unsimulated points. Therefore, this invention can significantly reduce computational costs, greatly improve efficiency, and ensure prediction accuracy, enabling rapid and accurate prediction of the entire field flow field characteristics under any wind conditions.
[0017] Furthermore, this invention employs space-filling sampling methods such as Latin hypercube sampling, Sobol sequence sampling, or Halton sequence sampling, enabling the database to effectively capture the global characteristics of the flow field response. This allows the surrogate model to possess good predictive ability (generalization ability) even for untrained wind conditions. In addition, new CFD data can be easily added to the database for updating and optimizing the surrogate model, continuously improving system performance. Attached Figure Description
[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the system of the present invention; Figure 3 This is a schematic diagram illustrating the implementation process of a wind farm flow field characteristic prediction method based on feature sampling according to an embodiment of the present invention. Figure 4 This is a schematic diagram of the computer device structure of the present invention. Detailed Implementation
[0020] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.
[0021] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.
[0022] See Figure 1 and Figure 3 This invention discloses a method for predicting the flow field characteristics of a wind farm based on feature sampling, comprising the following steps: S1, determine the key input parameters that affect the flow field characteristics of the wind farm and their value range, and construct an n-dimensional characteristic parameter space; Identify the key input parameters affecting the flow field characteristics of a wind farm, define their value ranges, and construct a... n The feature parameter space is defined as follows. The key input parameters include at least: inflow wind speed (U), inflow wind direction (θ), and inflow turbulence intensity (I). Optionally, they may also include wind shear index, fan layout parameters (such as row and column spacing), etc.
[0023] S2, within the n-dimensional feature parameter space, a representative set of sample points is selected using a space-filling design method; Defined in step S1 n Within the dimensional feature parameter space, a space-filling design method is employed to select a highly representative set of sample points. Preferably, the Latin hypercube sampling method is used. This method ensures that each probability stratification interval of each input parameter has only one sample point, thus achieving uniform coverage of the entire parameter space with the fewest possible sample points and avoiding the "curse of dimensionality" problem of grid discretization.
[0024] Preferably, in addition to Latin hypercube sampling, other low-bias sequences such as Sobol sequences and Halton sequences, or optimal Latin hypercube sampling, can also be used as feature sampling methods.
[0025] S3, For each feature sampling point in the sample point set, perform a high-fidelity CFD numerical simulation to obtain flow field data, thereby constructing a flow field characteristic database; For each feature sampling point obtained in step S2, a high-fidelity CFD numerical simulation is performed. Specifically, this includes: Establish a computational domain grid that includes the wind farm topography and wind turbine locations.
[0026] Set boundary conditions (such as inlet wind speed profile, wind direction, and turbulence intensity) corresponding to the sampling point parameters.
[0027] Select appropriate turbulence models (such as the k-ω SST model) and fan models (such as the actuation disk model or actuation line model).
[0028] The solution is calculated to obtain convergent, high-precision flow field data.
[0029] Extract and organize all CFD simulation results to construct a structured flow field characteristic database. Each record in the database includes: 1) Input vector: Feature sampling parameters (such as U, θ, I) corresponding to this working condition.
[0030] 2) Output data: including full-field flow field data (such as three-dimensional velocity field, pressure field) and / or key characteristic parameters (such as wind speed at each turbine hub height, wake velocity deficit, total power of the entire field, etc.).
[0031] S4. The machine learning proxy model is trained using the flow field characteristic database to obtain the wind farm flow field characteristic prediction model. Using input vectors from the database as features and corresponding output data as labels, a machine learning surrogate model is trained. This surrogate model learns the complex nonlinear mapping relationship from input parameters to the output flow field. Specifically: For high-dimensional outputs such as full-field flow data, dimensionality reduction can be achieved first using the Orthogonal Eigenvalue Decomposition (POD) method, representing the flow field as a linear combination of several POD basis modes. The learning objective of the model then becomes predicting the low-dimensional POD coefficients.
[0032] Using the feature sampling parameters as input features and the low-dimensional POD coefficients as output labels, a machine learning surrogate model is trained to establish a mapping relationship from wind condition parameters to the output flow field, thereby obtaining a wind farm flow field characteristic prediction model.
[0033] Preferably, the surrogate model can be Gaussian process regression (GPR), neural network (NN), or support vector machine (SVR), etc. The advantage of Gaussian process regression is that it can provide an estimate of prediction uncertainty.
[0034] Preferably, the output of the surrogate model can bypass POD dimensionality reduction and directly use structures such as convolutional neural networks (CNN) or U-Net to process the flow field data on the regular grid, thereby achieving end-to-end input-output mapping.
[0035] S5. Input the wind condition parameters into the wind farm flow field characteristic prediction model to predict the flow field distribution and key performance indicators of the entire field.
[0036] When it is necessary to obtain the flow field characteristics under a new wind condition (U_new, θ_new, I_new) that has not been CFD simulated, simply input the wind condition parameters into the pre-trained surrogate model, and the model can output the predicted full-field flow field distribution and key performance indicators within seconds.
[0037] Preferably, the method of the present invention can be further combined with an active learning loop: the surrogate model can actively recommend the next most valuable CFD simulation scenario based on the uncertainty of its prediction, thereby achieving optimal model performance in an iterative manner with the fewest number of CFD simulations.
[0038] See Figure 2 This invention discloses a wind farm flow field characteristic prediction system based on feature sampling, comprising a feature parameter space definition module, a feature sampling module, a database construction module, a model training module, and a prediction module. The feature parameter space definition module is used to determine the key input parameters affecting the wind farm flow field characteristics and their value ranges, forming an n-dimensional feature parameter space. The feature sampling module is used to select a representative set of sample points within the n-dimensional feature parameter space using a space-filling design method. The database construction module is used to perform high-fidelity CFD numerical simulations on the operating conditions represented by each feature sampling point in the sample point set to obtain flow field data, thereby constructing a flow field characteristic database. The model training module is used to train a machine learning surrogate model using the flow field characteristic database to obtain a wind farm flow field characteristic prediction model. The prediction module is used to input wind condition parameters into the wind farm flow field characteristic prediction model to predict the overall flow field distribution and key performance indicators. This invention does not aim for exhaustive coverage of the wind condition parameter space. Instead, it captures the inherent laws and main characteristics of wind farm flow field changes through limited but scientifically designed high-fidelity CFD simulations (feature sampling). These "feature samples" are then used to train an efficient surrogate model, which infers and predicts flow field characteristics across the entire parameter space. By employing intelligent sampling and machine learning techniques, and utilizing a small amount of high-fidelity CFD simulation data, a surrogate model capable of rapidly and accurately predicting the full-field flow field characteristics under any wind condition is constructed, thus achieving a balance between high efficiency and high accuracy.
[0039] In one embodiment of the invention, see [link to embodiment]. Figure 4A computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions from the computer storage medium to achieve a corresponding method flow or function. The processor described in this embodiment can be used in the operation of a wind farm flow field characteristic prediction method based on feature sampling.
[0040] This invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the wind farm flow field characteristic prediction method based on feature sampling in the above embodiments.
[0041] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention 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.
[0042] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. 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 illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0043] 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 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0044] 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.
[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A wind farm flow field characteristic prediction method based on feature sampling, characterized in that, The method comprises the following steps: determining key input parameters affecting the flow field characteristics of a wind farm and their value ranges to form an n-dimensional characteristic parameter space; in the n-dimensional characteristic parameter space, a representative sample point set is selected by using a space-filling design method; for each characteristic sampling point in the sample point set, a high-fidelity CFD numerical simulation is performed on the working condition represented by the characteristic sampling point to obtain flow field data, thereby constructing a flow field characteristic database; a wind farm flow field characteristic prediction model is obtained by training a machine learning agent model using the flow field characteristic database; wind condition parameters are input into the wind farm flow field characteristic prediction model to predict the full-field flow field distribution and key performance indicators.
2. The wind farm flow field characteristic prediction method based on feature sampling according to claim 1, characterized in that, The key input parameters affecting the flow field characteristics of the wind farm at least include incoming wind speed, incoming wind direction, and incoming turbulence intensity.
3. The wind farm flow field characteristic prediction method based on feature sampling according to claim 1, characterized in that, The space-filling design method is a Latin hypercube sampling method, a Sobol sequence sampling method, or a Halton sequence sampling method.
4. The wind farm flow field characteristic prediction method based on feature sampling according to claim 1, characterized in that, The step of performing a high-fidelity CFD numerical simulation on each characteristic sampling point in the sample point set to obtain flow field data specifically comprises: constructing a three-dimensional calculation domain grid containing the actual terrain of the wind farm and the specific positions of the wind turbines; setting the inlet boundary conditions of the calculation domain according to the working condition parameters corresponding to the characteristic sampling points; based on the set calculation domain, selecting a turbulence model and a wind turbine model for simulation calculation to obtain flow field data.
5. The wind farm flow field characteristic prediction method based on feature sampling according to claim 1, characterized in that, Each record in the flow field characteristic database includes an input vector and output data; the input vector is the characteristic sampling parameter corresponding to the working condition, and the output data includes full-field flow field data and / or key characteristic parameters.
6. The wind farm flow field characteristic prediction method based on feature sampling according to claim 1, characterized in that, The step of training a machine learning agent model using the flow field characteristic database to obtain a wind farm flow field characteristic prediction model specifically comprises: performing intrinsic orthogonal decomposition dimension reduction processing on the high-dimensional full-field flow field data in the flow field characteristic database, representing the flow field as a linear combination of a plurality of POD basis modes, and converting the learning goal of the model to predicting low-dimensional POD coefficients; training a machine learning agent model by taking the characteristic sampling parameters as input features and the low-dimensional POD coefficients as output labels to establish a mapping relationship from wind condition parameters to output flow field, thereby obtaining a wind farm flow field characteristic prediction model.
7. The wind farm flow field characteristic prediction method based on feature sampling according to claim 1, characterized in that, The machine learning agent model is a Gaussian process regression, a neural network, or a support vector machine.
8. A wind farm flow field characteristic prediction system based on feature sampling, characterized by, The method comprises: a characteristic parameter space definition module for determining key input parameters affecting the flow field characteristics of a wind farm and their value ranges to form an n-dimensional characteristic parameter space; a characteristic sampling module for selecting a representative sample point set by using a space-filling design method in the n-dimensional characteristic parameter space; a database construction module for performing a high-fidelity CFD numerical simulation on the working condition represented by each characteristic sampling point in the sample point set to obtain flow field data, thereby constructing a flow field characteristic database; a model training module for training a machine learning agent model using the flow field characteristic database to obtain a wind farm flow field characteristic prediction model; and a prediction module for inputting wind condition parameters into the wind farm flow field characteristic prediction model to predict the full-field flow field distribution and key performance indicators. A prediction module is configured to input the wind condition parameters into the wind farm flow field characteristic prediction model to predict the full-field flow field distribution and key performance indicators.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The computer program is executed by the processor to implement the steps of the feature sampling-based wind farm flow field characteristic prediction method according to any one of claims 1-7.
10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the steps of the feature sampling-based wind farm flow field characteristic prediction method according to any one of claims 1-7.