Methods, equipment and media for aerodynamic correction of wind turbine airfoils
By extracting features from the aerodynamic information of wind turbine airfoils and training neural networks, the problem that low-precision data cannot meet the requirements of high precision is solved, and the effect of quickly acquiring high-precision aerodynamic data is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NORTH CHINA ELECTRIC POWER UNIV
- Filing Date
- 2025-02-18
- Publication Date
- 2026-05-26
AI Technical Summary
Existing methods are insufficient to obtain high-precision aerodynamic data from low-precision aerodynamic data, thus failing to meet practical needs.
A wind turbine airfoil aerodynamic correction method based on data fusion strategy is adopted. By acquiring the aerodynamic information of the first and second blades, feature extraction is performed and a neural network fusion model is trained. The trained model is then used to obtain high-precision aerodynamic information.
It enables the rapid acquisition of high-precision aerodynamic data for airfoils of different thicknesses, reducing reliance on traditional wind tunnel testing and improving data accuracy.
Smart Images
Figure CN120046493B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wind turbine technology, specifically providing a method, equipment, and medium for aerodynamic correction of wind turbine airfoils. Background Technology
[0002] High-precision aerodynamic data is the basis for the overall design of wind turbine airfoils and a fundamental prerequisite and important guarantee for the successful development of airfoils. Wind tunnel testing and physical models are the main research methods for analyzing the aerodynamic characteristics of wind turbine airfoils. Wind tunnel testing can conduct highly accurate aerodynamic characteristic analysis of airfoils and provide direct and realistic aerodynamic characteristic data, but it suffers from high cost and long testing cycles. Physical models are based on assumptions and theoretical derivations. Although they can provide a preliminary understanding and prediction of aerodynamic characteristics, their results often depend on the simplification of the model and the rationality of the assumptions, resulting in lower accuracy. They are suitable for the preliminary design stage and rapid evaluation. High-precision data (wind tunnel testing) has higher realism and reliability, but it is costly and scarce; low-precision data (physical models) is inexpensive and abundant, but may contain biases and errors.
[0003] Existing methods struggle to obtain higher-precision aerodynamic data from low-precision aerodynamic data, which fails to meet practical needs. Summary of the Invention
[0004] To overcome the aforementioned shortcomings, this application is proposed to provide a solution, or at least a partial solution, to the technical problem that existing methods cannot obtain higher-precision aerodynamic data from low-precision aerodynamic data. This application provides a method, apparatus, and medium for aerodynamic correction of wind turbine airfoils.
[0005] In a first aspect, this application provides a method for aerodynamic correction of wind-powered airfoils based on a data fusion strategy, the method comprising:
[0006] Acquire aerodynamic information of the first blade and aerodynamic information of the second blade, wherein the first blade aerodynamic information and the second blade aerodynamic information are aerodynamic information corresponding to the first thickness type and the first incoming flow information;
[0007] Feature extraction is performed on the aerodynamic information of the first blade and the first incoming flow information to obtain the aerodynamic features of the first blade;
[0008] The neural network fusion model is trained using the aerodynamic characteristics of the first blade and the aerodynamic information of the second blade to obtain a trained neural network fusion model.
[0009] Obtain the aerodynamic information of the third blade corresponding to the second thickness type and the second incoming flow information;
[0010] Based on the aerodynamic information of the third blade and the second incoming flow information, the aerodynamic correction information of the blade corresponding to the second thickness type is obtained using the trained neural network fusion model.
[0011] In one embodiment of this application, the neural network fusion model includes L layers, each layer containing multiple neurons. The first layer is an input layer containing n1 neurons, and the Lth layer is an output layer containing n neurons. L The neural network fusion model is composed of [number] neurons, and the model is derived from [number] neurons. arrive The mapping is given by , where L is a positive integer greater than or equal to 3 and R is the set of real numbers.
[0012] In one embodiment of this application, the step of training the neural network fusion model using the aerodynamic features of the first blade and the aerodynamic information of the second blade includes:
[0013] The aerodynamic characteristics of the first blade are input into the neural network fusion model to obtain the prediction result;
[0014] The aerodynamic information of the second blade is used as a label, and the model loss is calculated based on the prediction result and the aerodynamic information of the second blade.
[0015] Determine whether the model loss is less than a preset loss threshold, or whether the number of iterations for training has reached a preset number threshold;
[0016] If so, a trained neural network fusion model is obtained.
[0017] In one embodiment of this application, the mean squared error function is used as the model loss function.
[0018] In one embodiment of this application, the method further includes: when the model loss is not less than a preset loss threshold, or the number of iterations for training has not reached a preset number threshold, using stochastic gradient descent to optimize the model loss.
[0019] In one embodiment of this application, obtaining the aerodynamic information of the first blade and the aerodynamic information of the second blade includes:
[0020] Based on the first thickness type and the first incoming flow information, aerodynamic data is calculated using a physical model based on potential flow theory to obtain the aerodynamic information of the first blade.
[0021] The aerodynamic information of the second blade was obtained by wind tunnel testing.
[0022] In one embodiment of this application, the feature extraction of the aerodynamic information of the first blade and the first incoming flow information includes:
[0023] Remove invalid information from the first blade aerodynamic information and the first incoming flow information;
[0024] The aerodynamic information of the first blade and the first incoming flow information after removing invalid information are converted into feature vectors to obtain the aerodynamic features of the first blade.
[0025] In one embodiment of this application, the second thickness type is obtained by interpolating the thickness of the first thickness type;
[0026] The step of obtaining the aerodynamic information of the third blade corresponding to the second thickness type and the second incoming flow information includes: calculating the aerodynamic data based on the second thickness type and the second incoming flow information using a physical model of potential flow theory to obtain the aerodynamic information of the third blade.
[0027] In a second aspect, an electronic device is provided, comprising:
[0028] At least one processor;
[0029] And, a memory communicatively connected to the at least one processor;
[0030] The memory stores a computer program, which, when executed by the at least one processor, is the aforementioned aerodynamic correction method for wind turbine airfoils.
[0031] In a third aspect, a computer-readable storage medium is provided, wherein a plurality of program codes are stored therein, the program codes being adapted to be loaded and run by a processor to perform the wind-powered airfoil aerodynamic correction method described in any of the preceding claims.
[0032] The above-described technical solutions of this application have at least one or more of the following features.
[0033] Beneficial effects:
[0034] The aerodynamic correction method for wind turbine airfoils in this application specifically includes: acquiring aerodynamic information of a first blade and a second blade, wherein the first blade aerodynamic information and the second blade aerodynamic information correspond to a first thickness type and a first incoming flow information; extracting features from the first blade aerodynamic information and the first incoming flow information to obtain aerodynamic features of the first blade; training a neural network fusion model using the first blade aerodynamic features and the second blade aerodynamic information to obtain a trained neural network fusion model; acquiring aerodynamic information of a third blade corresponding to a second thickness type and the second incoming flow information; and acquiring aerodynamic correction information for the blade corresponding to the second thickness type based on the third blade aerodynamic information and the second incoming flow information, using the trained neural network fusion model. In this way, aerodynamic data for airfoils of different thicknesses can be quickly acquired, reducing reliance on traditional wind tunnel tests and obtaining high-precision aerodynamic data. Attached Figure Description
[0035] The disclosure of this application will become more readily understood with reference to the accompanying drawings. It will be readily understood by those skilled in the art that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of this application. Furthermore, similar numbers in the drawings are used to denote similar components, wherein:
[0036] Figure 1 This is a schematic diagram of the main process of a wind turbine airfoil aerodynamic correction method in one embodiment of this application;
[0037] Figure 2 This is a schematic diagram of the structure of a neural network fusion model in one embodiment of this application;
[0038] Figure 3 This is a schematic diagram of the structure of a wind-powered airfoil aerodynamic correction device in one embodiment of this application;
[0039] Figure 4 This is a schematic diagram of the structure of an electronic device in one embodiment of this application. Detailed Implementation
[0040] Some embodiments of this application are described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of this application and are not intended to limit the scope of protection of this application.
[0041] In the description of this application, "module" and "processor" can include hardware, software, or a combination of both. A module can include hardware circuitry, various suitable sensors, communication ports, memory, and can also include software components, such as program code, or a combination of software and hardware. A processor can be a central processing unit, microprocessor, image processor, digital signal processor, or any other suitable processor. The processor has data and / or signal processing capabilities. The processor can be implemented in software, in hardware, or a combination of both. Non-transitory computer-readable storage media includes any suitable medium capable of storing program code, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, etc. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The terms "at least one A or B" or "at least one of A and B" have a similar meaning to "A and / or B" and can include only A, only B, or A and B. The singular terms "a" or "this" can also include plural forms.
[0042] Currently, traditional methods struggle to obtain higher-precision aerodynamic data from low-precision aerodynamic data, which fails to meet practical needs.
[0043] Therefore, this application provides a method, device and medium for aerodynamic correction of wind turbine airfoils.
[0044] See appendix Figure 1 , Figure 1 This is a schematic flowchart of the main steps of a wind turbine airfoil aerodynamic correction method according to an embodiment of this application.
[0045] like Figure 1 As shown, the aerodynamic correction method for wind turbine airfoils in this embodiment mainly includes the following steps S10-S50.
[0046] Step S10: Obtain the aerodynamic information of the first blade and the aerodynamic information of the second blade. The aerodynamic information of the first blade and the aerodynamic information of the second blade are the aerodynamic information corresponding to the first thickness type and the first incoming flow information.
[0047] Step S20: Extract features from the aerodynamic information of the first blade and the first incoming flow information to obtain the aerodynamic features of the first blade.
[0048] Step S30: Use the aerodynamic characteristics of the first blade and the aerodynamic information of the second blade to train the neural network fusion model and obtain the trained neural network fusion model.
[0049] Step S40: Obtain the aerodynamic information of the third blade corresponding to the second thickness type and the second incoming flow information.
[0050] Step S50: Based on the aerodynamic information of the third blade and the second incoming flow information, obtain the blade aerodynamic correction information corresponding to the second thickness type using the trained neural network fusion model.
[0051] Based on steps S10-S50 above, firstly, aerodynamic information of the first blade and second blade are acquired. The first and second blade aerodynamic information correspond to the first thickness type and the first incoming flow information, respectively. Feature extraction is performed on the first blade aerodynamic information and the first incoming flow information to obtain the first blade aerodynamic features. The first blade aerodynamic features and the second blade aerodynamic information are then used to train a neural network fusion model, resulting in a trained neural network fusion model. Next, aerodynamic information of the third blade corresponding to the second thickness type and the second incoming flow information is acquired. Finally, based on the third blade aerodynamic information and the second incoming flow information, and using the trained neural network fusion model, aerodynamic correction information for the blade corresponding to the second thickness type is obtained. In this way, aerodynamic data for airfoils of different thicknesses can be quickly acquired, reducing reliance on traditional wind tunnel testing and obtaining high-precision aerodynamic data.
[0052] Next, we will first describe in detail the physical model of this application based on potential flow theory.
[0053] This application is applicable to any airfoil. In order to describe the aerodynamic correction method of the wind turbine airfoil in more detail, the NREL S series airfoil will be used as an example. First, three airfoils with different thicknesses are selected, namely S809, S816 and S823. Then, the low-precision aerodynamic data of the airfoil is quickly obtained using Xfoil software.
[0054] Xfoil is a numerical simulation tool based on the combination of potential flow theory and boundary layer theory, used for rapidly calculating the aerodynamic performance of two-dimensional airfoils. Potential flow theory assumes that the fluid flow is incompressible, irrotational, and inviscid, and its core principle is to describe the flow field using a velocity potential function φ.
[0055] The fluid is incompressible, and its potential function satisfies the Laplace equation:
[0056]
[0057] The fluid is irrotational, and the velocity vector can be expressed as the gradient of the velocity potential:
[0058]
[0059] Apply an impermeable condition to the airfoil surface, meaning the normal component of the fluid velocity is zero:
[0060]
[0061] The free-flow condition is satisfied in the far field:
[0062] (as r→∞)
[0063] Where n is the normal vector of the airfoil surface, V ∞ is the free-flow velocity, and r is the distance in the airfoil's external field.
[0064] The calculation process for airfoil aerodynamic data in Xfoil is as follows: The surface element method is used to solve for the airfoil surface pressure distribution, lift, drag, and moment based on potential flow theory. First, according to the airfoil shape, the airfoil surface is divided into a sufficient number of finite element units, i.e., surface elements. Then, surface sources or surface vortices are arranged on each surface element, and a system of linear equations is established that satisfies the non-penetrating boundary conditions at the control points on each surface element and the Kuka condition at the airfoil trailing edge. Finally, the system of equations is solved to determine the surface element strength (dipole strength or vortex strength), and the disturbance velocity potential is obtained to further calculate the airfoil surface pressure distribution, lift, drag, and moment.
[0065] Xfoil was used to calculate the aerodynamic data of the S809, S816, and S823 airfoils at different Reynolds numbers and angles of attack. The calculation results obtained by Xfoil were poor and were regarded as low-precision aerodynamic data. At the same time, wind tunnel experiments were carried out to obtain the corresponding high-precision aerodynamic data.
[0066] The following provides further explanation of steps S10 to S50.
[0067] In a specific embodiment of this application, the step S10 described above involves obtaining the aerodynamic information of the first blade and the second blade, which includes: calculating the aerodynamic data based on the first thickness type and the first incoming flow information using a physical model of potential flow theory to obtain the aerodynamic information of the first blade; and obtaining the aerodynamic information of the second blade using a wind tunnel test.
[0068] The first thickness type can be any thickness of airfoil. For example, for the NREL S-series airfoils, the first thickness type can be three different thicknesses: S809, S816, and S823. The first incoming flow information can include the airfoil's angle of attack and Reynolds number. The first blade aerodynamic information and the second blade aerodynamic information can include lift coefficient and drag coefficient, etc.
[0069] Specifically, the aerodynamic information of the first blade under the first thickness type and the first incoming flow information can be obtained through the aforementioned physical model based on potential flow theory.
[0070] The aerodynamic information of the second blade was obtained through wind tunnel testing based on the first thickness type and the first incoming flow information.
[0071] The above is a further explanation of step S10. Step S20 will be further explained below.
[0072] In one specific embodiment of this application, the feature extraction of the first blade aerodynamic information and the first incoming flow information includes: removing invalid information from the first blade aerodynamic information and the first incoming flow information; converting the first blade aerodynamic information and the first incoming flow information after removing invalid information into feature vectors to obtain the first blade aerodynamic features.
[0073] Specifically, for the aerodynamic information of the first blade and the first incoming flow information, preprocessing can be performed first, such as removing outliers and filling in missing values. Then, the preprocessed aerodynamic information of the first blade and the first incoming flow information can be converted into feature vectors that can be recognized by the neural network fusion model. For example, word embedding can be performed on the aerodynamic information of the first blade and the first incoming flow information to obtain feature vectors that can be recognized by the neural network fusion model.
[0074] The above is a further explanation of step S20. Before explaining step S30 in detail below, we will first introduce the neural network fusion model of this application in detail.
[0075] Since both the high-precision (wind tunnel experiment) and low-precision (physical model) data for wind turbine airfoils originate from the same physical system, a mapping relationship exists between the two. Assume the datasets obtained from the physical model and wind tunnel experiment for the airfoil are (X, Y) respectively. low ) and (X, Y high If a mapping f(*) is constructed between the high-precision aerodynamic data obtained from wind tunnel experiments and the low-precision aerodynamic data obtained from physical models, then:
[0076] Y high =f(X,Y) low )
[0077] In the formula, Y low and Y high These are low-precision and high-precision aerodynamic datasets, respectively. The aerodynamic data mainly includes the lift coefficient (C). L ), drag coefficient (C) D Aerodynamic data such as X; X represents the angle of attack and Reynolds number (Re) of the airfoil.
[0078] The neural network fusion model in this application consists of a neural network. Deep learning can automatically extract features from the raw data, thus using airfoil flow information X and low-precision aerodynamic data Y. low As input, supervised machine learning techniques are employed, in which high-precision aerodynamic data Y... high The labels are used to train the neural network. By training the neural network, it learns to "map" high-precision data from low-precision data, thus modeling the relationship between low-precision and high-precision aerodynamic data.
[0079] In one specific embodiment of this application, the neural network fusion model includes L layers, each layer containing multiple neurons. The first layer is the input layer, containing n1 neurons, and the Lth layer is the output layer, containing n... L The neural network fusion model is composed of [number] neurons, and the model is derived from [number] neurons. arrive The mapping is given by , where L is a positive integer greater than or equal to 3 and R is the set of real numbers.
[0080] like Figure 2 As shown, the neural network fusion model of this application includes an input layer, a hidden layer, and an output layer, wherein the total number of layers (input, hidden, and output) is L. For example, the number of propagation layers in the neural network, L, can be set to 3 to 5 layers, where the 1st and Lth layers represent the input and output layers, respectively. Let the 1st layer have n... l If there are n neurons, then n1 is the number of neurons in the input layer, i.e., the dimension of the input data. A neural network is derived from R... n1 To R nL The mapping makes The vector b represents the weights of neuron j in layer l acting on the output values of neuron k in layer l-1. [l] ∈R nl This represents the bias of the l-th layer, therefore the bias of neuron j in the l-th layer is: The ReLU function is used as the activation function:
[0081] σ(x) = max(0, x)
[0082] The final output of the neural network is then obtained as follows:
[0083]
[0084]
[0085] in The parameters represent the network training parameters, where u is the network output and x is the network input. In the formula... This represents element-wise multiplication, also known as the Hadamard product. The Hadamard product is calculated element-wise from corresponding elements of two matrices (or vectors), rather than performing matrix multiplication. In neural networks, This usually represents operations between layers.
[0086] In one specific embodiment of this application, step S30 can be implemented by the following steps S301 to S303.
[0087] Step S301: Input the aerodynamic characteristics of the first blade into the neural network fusion model to obtain the prediction result.
[0088] Step S302: Use the aerodynamic information of the second blade as a label, and calculate the model loss based on the prediction result and the aerodynamic information of the second blade.
[0089] Step S303: Determine whether the model loss is less than a preset loss threshold, or whether the number of iterations for training has reached a preset number threshold. If so, obtain the trained neural network fusion model.
[0090] The preset loss threshold and preset number of attempts threshold can be values obtained in advance based on experiments. They can be adapted and adjusted according to the actual use scenario, and no specific limitations are imposed on them.
[0091] Specifically, the aerodynamic characteristics of the first blade can be input into the neural network fusion model to perform forward propagation calculations and obtain prediction results through the output layer. Error backpropagation uses an optimizer to propagate the model loss error backward using algorithms such as gradient descent and updates the network's trainable parameters. In a training iteration, a batch (i.e., a subset of the training set) is used for both forward and backpropagation. One iteration refers to the complete process of forward and backpropagation of the entire training set. After feedforward calculation and error backpropagation, if the expected convergence is not achieved, the aforementioned training steps are repeated. The end of training can be determined by setting a preset loss threshold or a maximum number of iterations. Furthermore, if the model loss is not less than the preset loss threshold, or if the number of iterations has not reached the preset threshold, stochastic gradient descent is used to further optimize the model loss.
[0092] Backpropagation in a neural network is an optimization process that learns unknown parameters by minimizing a loss function. Commonly used loss functions include mean squared error (MSE), cross-entropy, and classification hinge. Since the mean squared error function is more suitable for regression tasks, i.e., tasks that predict continuous values, this application can use the mean squared error function as the model loss function. The loss function for training samples is defined as follows:
[0093]
[0094] in, y represents the actual value of the label. H is the output value of the neural network, w is the network weight. N represents the number of high- and low-precision datasets, and Adam's stochastic gradient descent algorithm is used to optimize the loss function.
[0095] The above is a further explanation of step S30. Next, we will provide a further explanation of step S40.
[0096] Regarding step S40, in one specific embodiment of this application, the second thickness type is obtained by interpolating the thickness of the first thickness type; obtaining the aerodynamic information of the third blade corresponding to the second thickness type and the second incoming flow information includes: calculating the aerodynamic data based on the second thickness type and the second incoming flow information using a physical model of potential flow theory to obtain the aerodynamic information of the third blade.
[0097] The second thickness type can be obtained by interpolating the thickness of the first thickness type. For example, when the first thickness type is an airfoil with three different thicknesses, S809, S816, and S823, the second thickness type can be a thickness type such as S812, S817, and S821 obtained by interpolating the thickness of the first thickness type.
[0098] Specifically, the aerodynamic information of the third blade corresponding to the second thickness type and the second incoming flow information can be obtained through the physical model of the aforementioned potential flow theory.
[0099] The above is a further explanation of step S40. Next, we will provide a further explanation of step S50.
[0100] In step S50, the aerodynamic information of the third blade and the incoming flow information can be further converted into feature vectors that can be recognized by the neural network fusion model. By inputting the feature vectors into the trained neural network fusion model, the blade aerodynamic correction information can be obtained. The blade aerodynamic correction information is high-precision blade aerodynamic information.
[0101] It should be noted that although the steps in the above embodiments are described in a specific order, those skilled in the art will understand that in order to achieve the effect of this application, different steps do not necessarily have to be executed in such an order. They can be executed simultaneously (in parallel) or in other orders, and these variations are all within the scope of protection of this application.
[0102] Furthermore, this application also provides a wind-powered airfoil-shaped aerodynamic correction device.
[0103] See appendix Figure 3 , Figure 3 This is a main structural block diagram of a wind turbine airfoil aerodynamic correction device according to an embodiment of this application.
[0104] like Figure 3 As shown, the aerodynamic correction device for wind-powered airfoils in this embodiment mainly includes a first acquisition module 11, a feature extraction module 12, a model training module 13, a second acquisition module 14, and a prediction module 15. In some embodiments, one or more of the first acquisition module 11, feature extraction module 12, model training module 13, second acquisition module 14, and prediction module 15 can be combined into a single module.
[0105] In some embodiments, the first acquisition module 11 can be configured to acquire first blade aerodynamic information and second blade aerodynamic information, wherein the first blade aerodynamic information and the second blade aerodynamic information are aerodynamic information corresponding to the first thickness type and the first incoming flow information.
[0106] The feature extraction module 12 can be configured to extract features from the aerodynamic information of the first blade and the first incoming flow information to obtain the aerodynamic features of the first blade.
[0107] The model training module 13 can be configured to train the neural network fusion model using the aerodynamic features of the first blade and the aerodynamic information of the second blade, so as to obtain a trained neural network fusion model.
[0108] The second acquisition module 14 can be configured to acquire the third blade aerodynamic information corresponding to the second thickness type and the second incoming flow information.
[0109] The prediction module 15 can be configured to obtain blade aerodynamic correction information corresponding to the second thickness type based on the third blade aerodynamic information and the second incoming flow information, using the trained neural network fusion model.
[0110] In one implementation, a description of the specific functions can be found in steps S10-S50.
[0111] The aforementioned wind turbine airfoil aerodynamic correction device is used for performing Figure 1 The wind turbine airfoil aerodynamic correction method embodiments shown are similar in technical principle, the technical problems solved, and the technical effects produced. Those skilled in the art can clearly understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the wind turbine airfoil aerodynamic correction device can be found in the embodiments of the wind turbine airfoil aerodynamic correction method, and will not be repeated here.
[0112] Furthermore, it should be understood that since the various modules are only provided to illustrate the functional units of the device described in this application, the physical devices corresponding to these modules may be the processor itself, or a part of the processor's software, hardware, or a combination of both. Therefore, the number of modules shown in the figures is merely illustrative.
[0113] Those skilled in the art will understand that the various modules in the device can be adaptively split or combined. Such splitting or combining of specific modules will not cause the technical solution to deviate from the principles of this application; therefore, the technical solutions after splitting or combining will fall within the protection scope of this application.
[0114] Those skilled in the art will understand that all or part of the processes in the method of the above-described embodiment can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable storage medium can include any entity or device capable of carrying the computer program code, a medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.
[0115] Furthermore, this application also provides an electronic device, which may include at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program, which, when executed by the at least one processor, implements the wind turbine airfoil aerodynamic correction method described in any of the above embodiments. See also Figure 4 As shown, Figure 4 The structure of an electronic device, including a processor 100 and a memory 200, is illustrated by way of example.
[0116] Furthermore, this application also provides a computer-readable storage medium. In one embodiment of the computer-readable storage medium according to this application, the computer-readable storage medium can be configured to store a program that performs the wind turbine airfoil aerodynamic correction method of the above-described method embodiments. This program can be loaded and run by a processor to implement the above-described wind turbine airfoil aerodynamic correction method. For ease of explanation, only the parts related to the embodiments of this application are shown; for specific technical details not disclosed, please refer to the method section of the embodiments of this application. The computer-readable storage medium can be a memory device formed by various electronic devices. Optionally, in the embodiments of this application, the computer-readable storage medium is a non-transitory computer-readable storage medium.
[0117] The technical solution of this application has been described in conjunction with the specific embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. Without departing from the principles of this application, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of this application.
Claims
1. A method of aerodynamic correction of a wind turbine airfoil, characterized in that, The method includes: Acquiring aerodynamic information of the first blade and the second blade, wherein the first blade aerodynamic information and the second blade aerodynamic information are aerodynamic information corresponding to the first thickness type and the first incoming flow information, includes: calculating aerodynamic data based on the first thickness type and the first incoming flow information using a physical model of potential flow theory to obtain the aerodynamic information of the first blade; and obtaining the aerodynamic information of the second blade using wind tunnel testing. Feature extraction is performed on the aerodynamic information of the first blade and the first incoming flow information to obtain the aerodynamic features of the first blade; The method involves training a neural network fusion model using the aerodynamic features of the first blade and the aerodynamic information of the second blade to obtain a trained neural network fusion model. This includes: inputting the aerodynamic features of the first blade into the neural network fusion model to obtain a prediction result; using the aerodynamic information of the second blade as a label, calculating the model loss based on the prediction result and the second blade aerodynamic information; determining whether the model loss is less than a preset loss threshold, or whether the number of iterations has reached a preset threshold; if so, obtaining the trained neural network fusion model. The aerodynamic information of the third blade corresponding to the second thickness type and the second incoming flow information is obtained, wherein the second thickness type is obtained by interpolating the thickness of the first thickness type; the aerodynamic information of the third blade corresponding to the second thickness type and the second incoming flow information includes: calculating the aerodynamic data based on the second thickness type and the second incoming flow information using a physical model of potential flow theory to obtain the aerodynamic information of the third blade; Based on the aerodynamic information of the third blade and the second incoming flow information, the aerodynamic correction information of the blade corresponding to the second thickness type is obtained using the trained neural network fusion model.
2. The aerodynamic correction method for wind turbine airfoils according to claim 1, characterized in that, The neural network fusion model comprises L layers, each layer comprising a plurality of neurons, the first layer being an input layer, the first layer comprising neurons, the Lth layer being an output layer, the Lth layer comprising neurons, the neural network fusion model being a mapping from to , wherein L is a positive integer greater than or equal to 3, is a real set.
3. The aerodynamic correction method for wind turbine airfoils according to claim 1, characterized in that, The mean squared error function is used as the model loss function.
4. The aerodynamic correction method for wind turbine airfoils according to claim 1, characterized in that, The method further includes: when the model loss is not less than a preset loss threshold, or the number of iterations for training has not reached a preset number threshold, using stochastic gradient descent to optimize the model loss.
5. The aerodynamic correction method for wind turbine airfoils according to claim 1, characterized in that, The feature extraction of the first blade aerodynamic information and the first incoming flow information includes: Remove invalid information from the first blade aerodynamic information and the first incoming flow information; The aerodynamic information of the first blade and the first incoming flow information after removing invalid information are converted into feature vectors to obtain the aerodynamic features of the first blade.
6. An electronic device, characterized in that, include: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores a computer program, which, when executed by the at least one processor, implements the aerodynamic correction method for wind turbine airfoils as described in any one of claims 1 to 5.
7. A computer-readable storage medium storing a plurality of program codes, characterized in that, The program code is adapted to be loaded and run by a processor to perform the aerodynamic correction method for wind-powered airfoils as described in any one of claims 1 to 5.