A prediction method and system based on three-dimensional current flow of wind turbine

Abstract

The present disclosure proposes a method and system for predicting three-dimensional current flow in a wind turbine. The method includes acquiring historical operating data of the wind turbine to construct a training dataset. The historical operating data includes geometric data, rotational speed, and three-dimensional current flow data of the wind turbine blades. A three-dimensional flow prediction model is constructed. The three-dimensional flow prediction model includes a random deactivation module, a local feature calculation module, an intermediate convolution module, and a global feature calculation module. The three-dimensional flow prediction model is trained using the training dataset to obtain a trained three-dimensional flow prediction model. The geometric data and rotational speed of the wind turbine to be tested are acquired and input into the trained three-dimensional flow prediction model to obtain the three-dimensional current flow data of the wind turbine to be tested. The method of the present disclosure solves the problem of long flow prediction time and low efficiency when changing the geometry of a wind turbine in the prior art.

Description

Technical Field

[0001] The present disclosure relates to the technical field of wind turbine flow prediction, and in particular to a prediction method and system based on three-dimensional current flow of a wind turbine. Background Art

[0002] Computational Fluid Dynamics (CFD) is a branch of fluid mechanics that uses numerical analysis and data structures to analyze and solve problems involving fluid flow. CFD is applied to a wide range of research and engineering problems in many research fields and industries, including aerodynamics and aerospace analysis, weather simulation, natural sciences and environmental engineering, industrial system design and analysis, bioengineering, fluid flow and heat transfer, and engine and combustion analysis. CFD is used in engine analysis, for example, CFD flow simulation can be used to design wind turbines.

[0003] In the prior art, when designing wind turbines, within an optimization framework, the flow simulation utilized for different wind turbines varies. That is, different wind turbine designs require different flow simulation models, and a large amount of computing time is required when performing flow predictions for different wind turbine geometries. This leads to the problem of long flow prediction time and low efficiency when changing the wind turbine geometry. Summary of the Invention

[0004] The present disclosure aims to address, at least to some extent, one of the technical problems in the related art. To this end, the present disclosure provides a method and system for predicting three-dimensional current flow in a wind turbine. The primary purpose is to address the existing issues of long flow prediction times and low efficiency when changing wind turbine geometry.

[0005] According to a first embodiment of the present disclosure, a prediction method based on three-dimensional current flow of a wind turbine is provided, comprising:

[0006] Acquiring historical operating data of a wind turbine, and constructing the training dataset based on the historical operating data, wherein the historical operating data includes geometric shape data, rotational speed, and three-dimensional current flow data of wind turbine blades, wherein model input data is obtained based on the geometric shape data and rotational speed, and a label is obtained based on the three-dimensional current flow data;

[0007] Constructing a three-dimensional flow prediction model, the three-dimensional flow prediction model includes a random deactivation module, a local feature calculation module, an intermediate convolution module, and a global feature calculation module, wherein the random deactivation module is used to control the deactivation of nodes of the three-dimensional flow prediction model according to a set ratio, the local feature calculation module is used to perform local feature calculations on model input data, the intermediate convolution module is used to control weight reduction, and the global feature calculation module is used to perform global feature calculations to generate three-dimensional current flow data;

[0008] Using the training data set to train the three-dimensional flow prediction model to obtain a trained three-dimensional flow prediction model;

[0009] The geometric shape data and the rotation speed of the wind turbine to be tested are obtained, and the geometric shape data and the rotation speed of the wind turbine to be tested are input into the trained three-dimensional flow prediction model to obtain the three-dimensional current flow data of the wind turbine to be tested.

[0010] In one embodiment of the present disclosure, the geometric shape data of the wind turbine blade includes a geometric volume and a geometric image, and the model input data is obtained based on the geometric shape data and the rotation speed, including: obtaining the boundary condition inflow parameters and the boundary condition outflow parameters of the blade based on the geometric image; and using the geometric volume, the rotation speed, the boundary condition inflow parameters and the boundary condition outflow parameters as the model input data.

[0011] In one embodiment of the present disclosure, the local feature calculation module includes a plurality of first network structures, and the first network structure includes a plurality of network branches with different filtering masks.

[0012] In one embodiment of the present disclosure, the global feature calculation module includes multiple second network structures, and the second network structure includes multiple network branches with the same filtering mask.

[0013] In one embodiment of the present disclosure, the local feature calculation module, the intermediate convolution module, and the global feature calculation module respectively use a hyperbolic tangent function as an activation function.

[0014] In one embodiment of the present disclosure, it is further included that: before using the training data set to train the three-dimensional flow prediction model, the training data set needs to be standardized.

[0015] According to a second aspect of the present disclosure, there is also provided a prediction system based on three-dimensional current flow of a wind turbine, comprising:

[0016] a training data construction module, configured to obtain historical operating data of the wind turbine and construct the training data set based on the historical operating data, wherein the historical operating data includes geometric shape data, rotational speed, and three-dimensional current flow data of the wind turbine blades, wherein model input data is obtained based on the geometric shape data and rotational speed, and a label is obtained based on the three-dimensional current flow data;

[0017] a modeling module for constructing a three-dimensional flow prediction model, the three-dimensional flow prediction model including a random deactivation module, a local feature calculation module, an intermediate convolution module, and a global feature calculation module; the random deactivation module is used to control the deactivation of nodes of the three-dimensional flow prediction model according to a set ratio; the local feature calculation module is used to perform local feature calculations on model input data; the intermediate convolution module is used to control weight reduction; and the global feature calculation module is used to perform global feature calculations to generate three-dimensional current flow data;

[0018] A training module, configured to train the three-dimensional flow prediction model using the training data set to obtain a trained three-dimensional flow prediction model;

[0019] The prediction module is used to obtain geometric shape data and rotation speed of the wind turbine to be tested, and input the geometric shape data and rotation speed of the wind turbine to be tested into the trained three-dimensional flow prediction model to obtain three-dimensional current flow data of the wind turbine to be tested.

[0020] In one embodiment of the present disclosure, the geometric shape data of the wind turbine blade includes a geometric volume and a geometric image, and the training data construction module is used to: obtain the boundary condition inflow parameters and the boundary condition outflow parameters of the blade based on the geometric image; and use the geometric volume, the rotation speed, the boundary condition inflow parameters and the boundary condition outflow parameters as model input data.

[0021] According to an embodiment of the third aspect of the present disclosure, a prediction device based on the three-dimensional current flow of a wind turbine is also provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the prediction method based on the three-dimensional current flow of a wind turbine proposed in the embodiment of the first aspect of the present disclosure.

[0022] According to a fourth aspect embodiment of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is also provided, wherein the computer instructions are used to enable the computer to execute the prediction method based on the three-dimensional current flow of the wind turbine proposed in the first aspect embodiment of the present disclosure.

[0023] In one or more embodiments of the present disclosure, historical operating data of a wind turbine is obtained, and a training data set is constructed based on the historical operating data, wherein the historical operating data includes geometric shape data, rotation speed and three-dimensional current flow data of the wind turbine blades, wherein model input data is obtained based on the geometric shape data and the rotation speed, and a label is obtained based on the three-dimensional current flow data; a three-dimensional flow prediction model is constructed, and the three-dimensional flow prediction model includes a random deactivation module, a local feature calculation module, an intermediate convolution module and a global feature calculation module, wherein the random deactivation module is used to control the deactivation of nodes of the three-dimensional flow prediction model according to a set ratio, the local feature calculation module is used to perform local feature calculation on the model input data, the intermediate convolution module is used to control the reduction of weights, and the global feature calculation module is used to perform global feature calculation to generate three-dimensional current flow data; the three-dimensional flow prediction model is trained using the training data set to obtain a trained three-dimensional flow prediction model; the geometric shape data and rotation speed of the wind turbine to be tested are obtained, and the geometric shape data and rotation speed of the wind turbine to be tested are input into the trained three-dimensional flow prediction model to obtain the three-dimensional current flow data of the wind turbine to be tested. In this case, a training data set is obtained by combining geometric shape data, rotation speed and three-dimensional current flow data, and the training data set is used to train a three-dimensional flow prediction model to obtain a trained three-dimensional flow prediction model. The three-dimensional flow prediction model is suitable for the prediction of various wind turbines with blades of different geometric shapes. In particular, when the geometric shape of the wind turbine needs to be replaced, the existing three-dimensional flow prediction model can be directly used for training, and the model required for the replaced geometric shape can be matched more quickly, thereby improving the prediction efficiency and solving the problem of long flow prediction time and low efficiency when replacing the geometric shape of the wind turbine in the prior art.

[0024] Additional aspects and advantages of the present disclosure will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The above and / or additional aspects and advantages of the present disclosure will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0026] Figure 1 A schematic flow chart of a method for predicting three-dimensional current flow in a wind turbine according to an embodiment of the present disclosure is shown;

[0027] Figure 2 A schematic diagram showing the structure of a three-dimensional flow prediction model provided by an embodiment of the present disclosure is shown;

[0028] Figure 3 A schematic diagram showing a first network structure provided by an embodiment of the present disclosure is shown;

[0029] Figure 4A schematic diagram showing a second network structure provided by an embodiment of the present disclosure is shown;

[0030] Figure 5 A block diagram of a prediction system based on three-dimensional current flow of a wind turbine provided by an embodiment of the present disclosure is shown;

[0031] Figure 6 The block diagram is a device for predicting three-dimensional current flow of a wind turbine, which is used to implement the method for predicting three-dimensional current flow of a wind turbine according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0032] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible implementations consistent with the present disclosure. Rather, they are merely examples of apparatuses and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

[0033] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and features of different embodiments or examples, unless they are mutually inconsistent.

[0034] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present disclosure, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. It should also be understood that the term "and / or" used in the present disclosure refers to and includes any or all possible combinations of one or more associated listed items.

[0035] The following describes in detail embodiments of the present disclosure, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present disclosure, and should not be construed as limiting the present disclosure.

[0036] The present disclosure provides a prediction method and system based on three-dimensional current flow of a wind turbine, the main purpose of which is to solve the problem of long flow prediction time and low efficiency when changing the geometric shape of a wind turbine in the prior art.

[0037] In a first embodiment, Figure 1 FIG. 1 is a flow chart showing a method for predicting three-dimensional current flow in a wind turbine according to an embodiment of the present disclosure. Figure 1 As shown, the prediction method based on the three-dimensional current flow of a wind turbine includes:

[0038] Step S11, obtaining historical operating data of the wind turbine, and constructing a training data set based on the historical operating data, the historical operating data including geometric shape data, rotation speed and three-dimensional current flow data of the wind turbine blades, wherein model input data is obtained based on the geometric shape data and rotation speed, and a label is obtained based on the three-dimensional current flow data.

[0039] In step S11, the historical operation data includes geometric shape data, rotation speed and three-dimensional current flow data of the wind turbine blades. The historical operation data is stored in a memory in the form of a file.

[0040] Specifically, files containing historical operating data such as geometry data, rotational speed, and three-dimensional current flow data can be stored in the CFD General Notation System (CGNS) file format, thereby enabling cross-platform and program-independent work with geometry data, rotational speed, and three-dimensional current flow data. As is easy to understand, CGNS is based on a database format (Hierachical Data Format Version 5, HDF5), which stores data hierarchically in the form of nodes and sub-nodes. In addition to possible sub-nodes, metadata names, labels, and the data type of any stored file are also stored on each node. The specifications for the structure and content of CGNS files are unified descriptors / labels for various parameters (such as labels for coordinate / flow nodes), structural descriptions (node structures for structured networks, unstructured networks, etc.), and further specifications for storing important aerodynamic characteristics.

[0041] In step S11 , the geometric shape data of the wind turbine blade includes a geometric volume and a geometric image.

[0042] In step S11, model input data is obtained based on the geometric shape data and the rotational speed, including: obtaining the boundary condition inflow parameters and boundary condition outflow parameters of the blade based on the geometric image; and using the geometric volume, rotational speed, boundary condition inflow parameters, and boundary condition outflow parameters as model input data. The geometric image of the wind turbine blade is input into an existing image boundary algorithm to obtain boundary condition inflow parameters and boundary condition outflow parameters that can reflect the blade geometry. The boundary condition inflow parameters and boundary condition outflow parameters are based on the geometric mesh. The boundary condition inflow parameters and boundary condition outflow parameters can be referred to as a flow solution. A flow solution is divided into different subnetworks or regions, each of which is assigned a separate node in the CGNS file.

[0043] In step S11 , the three-dimensional current flow data may include three types of data: current velocity in the x-, y-, and z-axis directions, current density, and current pressure.

[0044] In step S11, a training data set can be constructed based on the model input data and the corresponding labels.

[0045] In step S11, considering that only layers with convolution operations are used to process bounding boxes, for a single model, it is assumed that the blocks of all bounding boxes have the same size in the original space. In order to standardize the different numerical ranges of different physical quantities, the training dataset needs to be standardized after it is constructed, and then the standardized training dataset is used to train the three-dimensional flow prediction model.

[0046] Specifically, for the model input data in the training data set, the reference quantity X set in advance can be used. ref , and normalized by formula (1). Formula (1) satisfies:

[0047]

[0048] Where X represents the model input data before standardization, and X' represents the model input data after standardization.

[0049] For the labels in the training data set (i.e., the three-dimensional current flow data in the training data set), the average value and standard deviation of various data in the three-dimensional current flow data are used to perform normalization processing using formula (2). Formula (2) satisfies:

[0050]

[0051] Taking the velocity of current as an example, Y represents the velocity of current before normalization, Y' represents the velocity of current after normalization, μY represents the average value of the velocity of current, σY represents the standard deviation of the velocity of current, and ∈ is a numerical stability constant, typically within the range of 1e-8 (i.e., 1*10 to the power of -8). The current density and current pressure can also be standardized using the same standardization method as the current velocity. In addition, the average values and standard deviations of the various variables included in the three-dimensional current flow data can be formed on randomly selected training data to cover as wide a range of values as possible.

[0052] Step S12, constructing a three-dimensional flow prediction model, which includes a random deactivation module, a local feature calculation module, an intermediate convolution module and a global feature calculation module. The random deactivation module is used to control the deactivation of nodes of the three-dimensional flow prediction model according to a set ratio, the local feature calculation module is used to perform local feature calculations on model input data, the intermediate convolution module is used to control the reduction of weights, and the global feature calculation module is used to perform global feature calculations to generate three-dimensional current flow data.

[0053] In step S12, the input data for the 3D flow prediction model is the model input data obtained in step S11. Specifically, the inputs to the 3D flow prediction model include geometric volume, rotational velocity, boundary condition inflow parameters, and boundary condition outflow parameters. The output data of the 3D flow prediction model is 3D current flow data. Specifically, the output of the 3D flow prediction model includes current velocity, current density, and current pressure in the x, y, and z axes.

[0054] Figure 2 A schematic structural diagram of a three-dimensional flow prediction model provided by an embodiment of the present disclosure is shown.

[0055] In step S12, as Figure 2As shown, the 3D flow prediction model includes an input layer. This layer receives model input data, including geometric volume, rotational velocity, boundary condition inflow parameters, and boundary condition outflow parameters. The rotational velocity, boundary condition inflow parameters, and boundary condition outflow parameters are included as partial model inputs because the boundary conditions for the blade rows physically exist only at the inlet and outlet interfaces. To transfer these variables to the neural network input space, these variables are also applied by default only to the regions within the bounding boxes marked with the inlet and outlet interfaces. For the remaining regions, these channels are set to zero. However, applying boundary conditions to all regions within a bounding box can be advantageous. Boundary conditions have a decisive influence on the entire flow solution. Therefore, in some cases, it may be easier for the neural network if these parameters apply everywhere, not just to the front and rear boundary regions. The same applies to the rotor rotational velocity. Since, in principle, only the geometry is moving, not the entire flow field, it makes sense from a physical perspective to apply them only to the regions described as solid walls. However, even here, setting the rotational speed over the entire rotor surface can be advantageous. Therefore, taking the rotation speed, boundary condition inflow parameter and boundary condition outflow parameter as part of the model input is conducive to better prediction of the model.

[0056] In step S12, the random deactivation module is connected to the input layer. The random deactivation module is used to control the nodes of the three-dimensional flow prediction model to be deactivated according to a set ratio. The random deactivation module can be Figure 2 The Dropout layer is shown. The Dropout layer is connected to the input layer. It should be noted that during the actual prediction process, the Dropout layer does not control node deactivation; rather, it passes all input layer data to the subsequent network. During model training, the Dropout layer controls node deactivation in the 3D flow prediction model according to a set ratio. For example, a Dropout value of 0.8 deactivates 20% of the nodes in the 3D flow prediction model, retaining 80% of the nodes for training.

[0057] In step S12, the Dropout layer is added to the three-dimensional flow prediction model to take into account that the calculation of the bounding box may result in inactivated blocks, which is particularly evident in high-resolution and edge areas. In order to make the neural network achieve a certain invariance with respect to these holes, the dropout method is adopted, that is, the Dropout layer is added. That is, during the training process, the Dropout layer randomly closes a predetermined proportion of neurons (i.e., neural network nodes), which forces the remaining neurons to learn better predictions without adjacent neurons and offsets the well-known problem of overfitting the network to the training data set. The dropout method is used in a way that a certain number of input nodes are randomly closed during training, which forces the algorithm to learn correct current flow data even without a complete input space.

[0058] In step S12, the three-dimensional flow prediction model further includes an initial convolution module. The initial convolution module is connected to the random inactivation module. The initial convolution module is used to perform preliminary convolution processing on the model input data. Figure 2 As shown in Figure 3, the initial convolution module can be composed of two convolution layers. The first convolution layer is connected to the Dropout layer, and the second convolution layer is connected to the local feature calculation module.

[0059] In step S12, the local feature calculation module is connected to the initial convolution module. The local feature calculation module is used to perform local feature calculation on the model input data processed by the initial convolution module to detect low-level features.

[0060] In step S12, the local feature calculation module includes a plurality of first network structures, and the first network structure includes a plurality of network branches with different filter masks. The first network structure may be a network structure A. Figure 2 As shown in FIG, the local feature calculation module can be composed of 12 network structures A. The 12 network structures A are connected in series, with the first network structure A connected to the initial convolution module and the last network structure A connected to the intermediate convolution module.

[0061] Figure 3 A structural diagram of a first network structure provided by an embodiment of the present disclosure is shown. Figure 3 The first network structure (ie, network structure A) shown includes four network branches with different filter masks, such as Figure 3 As shown, the four network branches are:

[0062] First network branch: This network branch includes filters with a convolution kernel of 1x1x1. Using a filter mask with a convolution kernel of 1x1x1, new information can be obtained with very few weights and in a very local way by combining different feature maps.

[0063] Second network branch: This branch includes a filter with a convolution kernel of 1x1x1 and a filter with a convolution kernel of 3x3x3 (symmetrically padded). The 1x1x1 filter is connected to the initial convolution module as the initial filter of the second network branch. This initial filter can reduce the number of feature maps of the larger filter core to save weights. The 3x3x3 filter connected to the 1x1x1 filter can be used to detect local features. The boundary area is symmetrically expanded, thereby being able to accommodate erroneous information that may be generated by the multiple weighting of the data in the boundary area.

[0064] Third network branch: This branch includes a 1x1x1 convolutional kernel and a 3x3x3 convolutional kernel (zero-padded, dilation = 2). The 1x1x1 filter is connected to the initial convolutional module as the initial filter of the third network branch. This initial filter can reduce the number of feature maps. Due to the expansion of the convolution, a wider range of effects should be detected. Edge regions are expanded here with zero padding to minimize any impact of the multi-weighting of edge regions.

[0065] Fourth network branch: This branch includes a convolution kernel with a 3x3x3 filter (symmetric padding) and a convolution kernel with a 3x3x3 filter (zero padding). Similar to the previous filtering, this branch is more computationally intensive and expensive. By using two 3x3x3 filters, the feature map can learn more features between the two filters. On the one hand, the symmetry, on the other hand, the zero padding, as an edge treatment, balances the advantages and disadvantages of the two methods.

[0066] like Figure 3 As shown, the first network structure also includes a fully connected layer, and the four network branches are all connected to the fully connected layer (i.e., Filter connection).

[0067] like Figure 3 As shown, the first network structure also includes an activation layer, which is connected to the fully connected layer. Considering that the tanh function is continuous, it is more consistent with the process of fluid solution (i.e., fluid mechanics solution). Its point symmetry leads to better activation and represents a larger range of values, especially for inputs with negative signs. Therefore, the activation layer can use the hyperbolic tangent function (i.e., tangent hyperbolic tanh function) as the activation function.

[0068] In step S12, the intermediate convolution module is used to control the weight reduction. The intermediate convolution module is connected to the local feature calculation module and the global feature calculation module. Figure 2As shown, the intermediate convolutional module can consist of a single convolutional layer with a 1x1x1 kernel. In this case, considering traditional image recognition methods in two-dimensional space, filter kernels use 4D tensors, where the last two dimensions specify the size of the input feature map fin and the output feature map fout. Due to the additional dimension of three-dimensional space, the filter kernel requires a fifth dimension. Therefore, with a filter size of x×y×z, the resulting convolution tensor is (x×y×z×fin×fout). Therefore, a single 3x3x3 filter with 20 input and 20 output feature maps already has 10,800 trainable weights. If filters are also combined into an Inception module, the number of trainable weights becomes very large, even with a small number of layers. Therefore, the weights can be reduced by using intermediate convolutional modules (i.e., 1x1x1 convolutional layers). In addition, dilated convolutions are used to expand the receptive field of filters with the same number of trainable weights. Dilated convolutions play an important role in the later layers of neural networks to detect high-level features so that all current influences and effects can be mapped, even with a smaller number of weights, and connections can be established between all input and output nodes without the need for more layers.

[0069] In step S12, the global feature calculation module is connected to the intermediate convolution module. The global feature calculation module is used to perform global feature calculation to generate three-dimensional current flow data.

[0070] In step S12, the global feature calculation module includes multiple second network structures, and the second network structure includes multiple network branches with the same filter mask. The second network structure can be network structure B. Figure 2 As shown in Figure 1, the global feature calculation module can be composed of 6 network structures B. The 6 network structures B are connected in series, with the first network structure B connected to the middle convolution module and the last network structure B connected to the output layer.

[0071] Figure 4 A structural diagram of a second network structure provided by an embodiment of the present disclosure is shown. Figure 4 The second network structure shown (i.e., network structure B) includes three network branches with the same filter mask. Each network branch includes a filter with a convolution kernel of 3x3x3.

[0072] like Figure 4 As shown, the second network structure also includes a fully connected layer, and the three network branches are all connected to the fully connected layer (i.e., Filter connection).

[0073] like Figure 4As shown, the second network structure also includes an activation layer, which is connected to the fully connected layer. The activation layer can use the hyperbolic tangent function (i.e., the tangent hyperbolic tanh function) as the activation function. The input of the activation layer includes the output of the fully connected layer and the output of the intermediate convolution module.

[0074] In step S12, the second network structure is compared with the first network structure. The combination of the first network structure network branches is more targeted at the detection of local effects and features. For the calculation of global features, a different starting layer structure is used in the back layer of the neural network, such as Figure 2 The second network structure is shown. The difference between the second network structure and the first network structure lies in the different dilation factors in the 3x3x3 filter and the omission of the feature map reduction method using the 1x1x1 filter. To avoid false positives caused by symmetrical edge processing, only zero padding is used when detecting global features. Because the filter kernel here is very wide, it often includes edge regions, which can reduce the impact of deformation during model training.

[0075] In step S12, as Figure 2 As shown, the 3D flow prediction model includes an output layer. The output layer is connected to the global feature calculation module. The output layer is used to output 3D current flow data, namely the current velocity, current density, and current pressure in the x, y, and z axes.

[0076] In this embodiment, in the three-dimensional flow prediction model, a combination of residual layers and inception layers is used for the individual layers themselves. Therefore, the advantages of these two types of layers can be combined. On the one hand, this makes it possible to train different filter masks for different filter widths of the same layer. On the other hand, the skip connections of the residual layers ensure that each layer has error information, even if the individual layers no longer pass on error information. Only the first two layers of the neural network consist of simple convolutional layers to achieve an ideal scale for the number of feature maps used. The initial layers at the lower levels of the network (e.g. the local feature calculation module) are designed to detect local, limited properties of the input data through their filters with small receptive fields, while the initial layers at the higher levels (e.g. the global feature calculation module) cover a much wider area and can therefore detect global structures.

[0077] In this embodiment, except for the output layer, the ends of the other layers in the three-dimensional flow prediction model all use the hyperbolic tangent function as the activation function, so as to be able to produce outputs that exceed the positive and negative standard normal distribution.

[0078] Step S13: Using the training data set to train the three-dimensional flow prediction model to obtain a trained three-dimensional flow prediction model.

[0079] In step S13, during the training process, the training data set is used, and the geometric volume, rotation speed, boundary condition inflow parameters and boundary condition outflow parameters in the training data set are used as model inputs. The speed of the current in the x-, y-, and z-axis directions, the current density, and the current pressure in the training data set are used as labels to train the three-dimensional flow prediction model, thereby obtaining a trained three-dimensional flow prediction model.

[0080] In step S13, during the training process, the Adadelta algorithm can be used as an optimization strategy to accelerate the training process. During the training process, for example, a total of 446,527 trainable weights are used in the model.

[0081] Step S14 , obtaining geometric shape data and rotation speed of the wind turbine to be tested, and inputting the geometric shape data and rotation speed of the wind turbine to be tested into the trained three-dimensional flow prediction model to obtain three-dimensional current flow data of the wind turbine to be tested.

[0082] In step S14 , the geometric shape of the blades of the wind turbine to be tested is the same as the geometric shape of the blades of the wind turbine used to obtain the training dataset.

[0083] In step S14 , before inputting the geometric shape data and the rotation speed into the trained three-dimensional flow prediction model, the geometric shape data and the rotation speed may be standardized first and then fed into the trained three-dimensional flow prediction model.

[0084] In step S14, the geometric shape data includes the geometric volume of the blades of the wind turbine to be tested and the geometric image of the blades of the wind turbine to be tested. The geometric shape data and the rotational speed of the wind turbine to be tested are input into the trained three-dimensional flow prediction model to obtain three-dimensional current flow data of the wind turbine to be tested, including:

[0085] Based on the geometric shape data, the geometric volume, boundary condition inflow parameters and boundary condition outflow parameters of the blades of the wind turbine to be tested are obtained; the geometric volume, rotation speed, boundary condition inflow parameters and boundary condition outflow parameters are input into the trained three-dimensional flow prediction model for prediction, and the current velocity, current density and current pressure in the x-, y- and z-axis directions of the wind turbine to be tested are output.

[0086] In a prediction method based on three-dimensional current flow of a wind turbine according to an embodiment of the present disclosure, historical operating data of the wind turbine is obtained, and a training data set is constructed based on the historical operating data, wherein the historical operating data includes geometric shape data, rotational speed and three-dimensional current flow data of the wind turbine blades, wherein model input data is obtained based on the geometric shape data and rotational speed, and a label is obtained based on the three-dimensional current flow data; a three-dimensional flow prediction model is constructed, and the three-dimensional flow prediction model includes a random deactivation module, a local feature calculation module, an intermediate convolution module and a global feature calculation module, wherein the random deactivation module is used to control the deactivation of nodes of the three-dimensional flow prediction model according to a set ratio, the local feature calculation module is used to perform local feature calculation on the model input data, the intermediate convolution module is used to control the reduction of weights, and the global feature calculation module is used to perform global feature calculation to generate three-dimensional current flow data; the three-dimensional flow prediction model is trained using the training data set to obtain a trained three-dimensional flow prediction model; the geometric shape data and rotational speed of the wind turbine to be tested are obtained, and the geometric shape data and rotational speed of the wind turbine to be tested are input into the trained three-dimensional flow prediction model to obtain the three-dimensional current flow data of the wind turbine to be tested. In this case, a training data set is obtained by combining geometric shape data, rotation speed, and three-dimensional current flow data. The training data set is used to train a three-dimensional flow prediction model to obtain a trained three-dimensional flow prediction model. The three-dimensional flow prediction model is suitable for prediction of various wind turbines with blades of different geometric shapes. In particular, when the wind turbine geometry needs to be replaced, the existing three-dimensional flow prediction model can be directly used for training, and the model required for the replaced geometry can be matched more quickly, thereby improving prediction efficiency and solving the problem of long flow prediction time and low efficiency when changing the wind turbine geometry in the prior art. In order to obtain prediction results as quickly as possible, the method disclosed in the present invention analyzes the geometry in a short time. First, a current database including geometric shape data, rotation speed, and three-dimensional current flow data is generated. Then, an artificial neural network model, namely the three-dimensional flow prediction model, is analyzed and established. The model utilizes modern image recognition technology in the field of deep learning. The method disclosed in the present invention is verified and evaluated using a test set based on statistical and aerodynamic indicators. The model of the method disclosed in the present invention can be applied to prediction of various wind turbines with blades of different geometric shapes. In particular, when the wind turbine geometry needs to be replaced, it can be trained on an existing basis, so that it can be transferred to applications with other geometries in a short time.

[0087] The following are system embodiments of the present disclosure, which can be used to implement the method embodiments of the present disclosure. For details not disclosed in the system embodiments of the present disclosure, please refer to the method embodiments of the present disclosure.

[0088] See Figure 5 , Figure 5A block diagram of a prediction system based on three-dimensional current flow in a wind turbine according to an embodiment of the present disclosure is shown. The prediction system based on three-dimensional current flow in a wind turbine can be implemented as all or part of a system through software, hardware, or a combination of both. The prediction system 10 based on three-dimensional current flow in a wind turbine includes a training data construction module 11, a modeling module 12, a training module 13, and a prediction module 14, wherein:

[0089] a training data construction module 11 for acquiring historical operating data of the wind turbine and constructing a training data set based on the historical operating data, wherein the historical operating data includes geometric shape data, rotational speed, and three-dimensional current flow data of the wind turbine blades, wherein model input data is obtained based on the geometric shape data and rotational speed, and labels are obtained based on the three-dimensional current flow data;

[0090] a modeling module 12 for constructing a three-dimensional flow prediction model, the three-dimensional flow prediction model including a random deactivation module, a local feature calculation module, an intermediate convolution module, and a global feature calculation module. The random deactivation module is used to control the deactivation of nodes in the three-dimensional flow prediction model according to a set ratio. The local feature calculation module is used to perform local feature calculations on model input data. The intermediate convolution module is used to control weight reduction. The global feature calculation module is used to perform global feature calculations to generate three-dimensional current flow data.

[0091] A training module 13 is used to train the three-dimensional flow prediction model using the training data set to obtain a trained three-dimensional flow prediction model;

[0092] The prediction module 14 is used to obtain geometric shape data and rotation speed of the wind turbine to be tested, and input the geometric shape data and rotation speed of the wind turbine to be tested into the trained three-dimensional flow prediction model to obtain three-dimensional current flow data of the wind turbine to be tested.

[0093] Optionally, the geometric shape data of the wind turbine blade includes a geometric volume and a geometric image, and the training data construction module 11 is used to: obtain the boundary condition inflow parameters and boundary condition outflow parameters of the blade based on the geometric image; and use the geometric volume, rotation speed, boundary condition inflow parameters and boundary condition outflow parameters as model input data.

[0094] Optionally, the local feature calculation module includes multiple first network structures, and the first network structure includes multiple network branches with different filtering masks.

[0095] Optionally, the global feature calculation module includes multiple second network structures, and the second network structure includes multiple network branches with the same filtering mask.

[0096] Optionally, the local feature calculation module, the intermediate convolution module and the global feature calculation module respectively use a hyperbolic tangent function as an activation function.

[0097] Optionally, the wind turbine three-dimensional current flow prediction system 10 further includes a pre-processing module, which is used to perform standardization processing on the training data set before using the training data set to train the three-dimensional flow prediction model.

[0098] It should be noted that the above-described embodiments of the wind turbine three-dimensional current flow prediction system, when implementing the wind turbine three-dimensional current flow prediction method, only illustrate the division of the above-described functional modules. In actual applications, the above-described functions can be assigned to different functional modules as needed. That is, the internal structure of the wind turbine three-dimensional current flow prediction device can be divided into different functional modules to complete all or part of the functions described above. In addition, the wind turbine three-dimensional current flow prediction system and the wind turbine three-dimensional current flow prediction method embodiment provided in the above-described embodiments are based on the same concept. The implementation process is detailed in the method embodiment and will not be repeated here.

[0099] The serial numbers of the above-mentioned embodiments of the present disclosure are for description only and do not represent the advantages or disadvantages of the embodiments.

[0100] In a prediction system based on three-dimensional current flow of a wind turbine according to an embodiment of the present disclosure, a training data construction module is used to obtain historical operating data of the wind turbine and construct a training data set based on the historical operating data. The historical operating data includes geometric shape data, rotational speed and three-dimensional current flow data of the wind turbine blades, wherein model input data is obtained based on the geometric shape data and rotational speed, and a label is obtained based on the three-dimensional current flow data; a modeling module is used to construct a three-dimensional flow prediction model, and the three-dimensional flow prediction model includes a random deactivation module, a local feature calculation module, an intermediate convolution module and a global feature calculation module. The random deactivation module is used to control the deactivation of nodes of the three-dimensional flow prediction model according to a set ratio, the local feature calculation module is used to perform local feature calculation on the model input data, the intermediate convolution module is used to control the reduction of weights, and the global feature calculation module is used to perform global feature calculation to generate three-dimensional current flow data; a training module is used to train the three-dimensional flow prediction model using the training data set to obtain a trained three-dimensional flow prediction model; and a prediction module is used to obtain geometric shape data and rotational speed of the wind turbine to be tested, and input the geometric shape data and rotational speed of the wind turbine to be tested into the trained three-dimensional flow prediction model to obtain the three-dimensional current flow data of the wind turbine to be tested. In this case, a training data set is obtained by combining geometric data, rotational speed, and three-dimensional current flow data. The training data set is used to train a three-dimensional flow prediction model to obtain a trained three-dimensional flow prediction model. The three-dimensional flow prediction model is suitable for prediction of various wind turbines with blades of different geometric shapes. In particular, when the wind turbine geometry needs to be replaced, the existing three-dimensional flow prediction model can be directly used for training, and the model required for the replaced geometry can be matched more quickly, thereby improving prediction efficiency and solving the problem of long flow prediction time and low efficiency when changing the wind turbine geometry in the prior art. In order to obtain prediction results as quickly as possible, the system of the present disclosure analyzes the geometry in a short time. First, a current database including geometric data, rotational speed, and three-dimensional current flow data is generated. Then, an artificial neural network model, namely the three-dimensional flow prediction model, is analyzed and established. The model utilizes modern image recognition technology in the field of deep learning. The system of the present disclosure is verified and evaluated using a test set based on statistical and aerodynamic indicators. The model of the system of the present disclosure is suitable for prediction of various wind turbines with blades of different geometric shapes. In particular, when the wind turbine geometry needs to be replaced, it can be trained based on the existing foundation, so that it can be transferred to applications with other geometries in a short time.

[0101] According to an embodiment of the present disclosure, the present disclosure further provides a prediction device based on three-dimensional current flow of a wind turbine, a readable storage medium, and a computer program product.

[0102] Figure 61 is a block diagram of a device for predicting three-dimensional current flow in a wind turbine, used to implement a method for predicting three-dimensional current flow in a wind turbine according to an embodiment of the present disclosure. The device for predicting three-dimensional current flow in a wind turbine is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The device for predicting three-dimensional current flow in a wind turbine may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable electronic devices, and other similar computing devices. The components, connections and relationships of the components, and functions of the components shown in this disclosure are merely examples and are not intended to limit the implementation of the present disclosure as described and / or claimed in this disclosure.

[0103] like Figure 6 As shown, a prediction device 20 for predicting three-dimensional current flow in a wind turbine includes a computing unit 21, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 22 or loaded from a storage unit 28 into a random access memory (RAM) 23. RAM 23 may also store various programs and data required for the operation of the prediction device 20 for predicting three-dimensional current flow in a wind turbine. Computing unit 21, ROM 22, and RAM 23 are interconnected via a bus 24. An input / output (I / O) interface 25 is also connected to bus 24.

[0104] Multiple components in the wind turbine three-dimensional current flow prediction device 20 are connected to the I / O interface 25, including: an input unit 26, such as a keyboard, a mouse, etc.; an output unit 27, such as various types of displays, speakers, etc.; a storage unit 28, such as a magnetic disk, an optical disk, etc., which is communicatively connected to the computing unit 21; and a communication unit 29, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 29 allows the wind turbine three-dimensional current flow prediction device 20 to exchange information / data with other wind turbine three-dimensional current flow prediction devices via a computer network such as the Internet and / or various telecommunication networks.

[0105] The computing unit 21 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the computing unit 21 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 21 performs the various methods and processes described above, such as executing a prediction method based on three-dimensional current flow in a wind turbine. For example, in some embodiments, the prediction method based on three-dimensional current flow in a wind turbine can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as a storage unit 28. In some embodiments, part or all of the computer program can be loaded and / or installed into the prediction device 20 based on three-dimensional current flow in a wind turbine via the ROM 22 and / or the communication unit 29. When the computer program is loaded into the RAM 23 and executed by the computing unit 21, one or more steps of the prediction method based on three-dimensional current flow in a wind turbine described above can be performed. Alternatively, in other embodiments, the computing unit 21 may be configured in any other appropriate manner (for example, by means of firmware) to execute the prediction method based on the three-dimensional current flow of the wind turbine.

[0106] Various embodiments of the systems and techniques described above in the present disclosure can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0107] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0108] In the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device for predicting three-dimensional current flow in a wind turbine. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or electronic device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage electronic device, a magnetic storage electronic device, or any suitable combination of the foregoing.

[0109] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0110] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.

[0111] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact via a communication network. This client-server relationship is established by computer programs running on the respective computers, establishing a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host, a host product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosts and VPS services ("Virtual Private Servers" or simply "VPS"). The server may also be a server in a distributed system or a server integrated with blockchain.

[0112] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This disclosure is not limited here.

[0113] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A prediction method based on three-dimensional current flow in a wind turbine, characterized in that: include: Acquiring historical operating data of the wind turbine and constructing a training dataset based on the historical operating data, wherein the historical operating data includes geometric shape data, rotational speed, and three-dimensional current flow data of the wind turbine blades, wherein model input data is obtained based on the geometric shape data and rotational speed, and a label is obtained based on the three-dimensional current flow data; Constructing a three-dimensional flow prediction model, the three-dimensional flow prediction model includes a random deactivation module, a local feature calculation module, an intermediate convolution module, and a global feature calculation module, wherein the random deactivation module is used to control the deactivation of nodes of the three-dimensional flow prediction model according to a set ratio, the local feature calculation module is used to perform local feature calculations on model input data, the intermediate convolution module is used to control weight reduction, and the global feature calculation module is used to perform global feature calculations to generate three-dimensional current flow data; Using the training data set to train the three-dimensional flow prediction model to obtain a trained three-dimensional flow prediction model; Acquiring geometric data and a rotational speed of the wind turbine to be tested, and inputting the geometric data and the rotational speed of the wind turbine to be tested into the trained three-dimensional flow prediction model to obtain three-dimensional current flow data of the wind turbine to be tested; The geometric shape data of the wind turbine blade includes a geometric volume and a geometric image. The step of obtaining the model input data based on the geometric shape data and the rotation speed includes: obtaining a boundary condition inflow parameter and a boundary condition outflow parameter of the blade based on the geometric image; using the geometric volume, the rotational speed, the boundary condition inflow parameter, and the boundary condition outflow parameter as model input data; The local feature calculation module includes a plurality of first network structures, each of which includes a plurality of network branches with different filtering masks; The global feature calculation module includes multiple second network structures, and the second network structure includes multiple network branches with the same filtering mask.

2. The prediction method based on three-dimensional current flow of a wind turbine according to claim 1, characterized in that: The local feature calculation module, the intermediate convolution module and the global feature calculation module respectively use the hyperbolic tangent function as the activation function.

3. The prediction method based on three-dimensional current flow of a wind turbine according to claim 1, characterized in that: Also includes: Before using the training data set to train the three-dimensional flow prediction model, the training data set needs to be standardized.

4. A prediction system based on three-dimensional current flow in wind turbines, characterized in that: include: a training data construction module, configured to obtain historical operating data of the wind turbine and construct a training data set based on the historical operating data, wherein the historical operating data includes geometric shape data, rotational speed, and three-dimensional current flow data of the wind turbine blades, wherein model input data is obtained based on the geometric shape data and rotational speed, and a label is obtained based on the three-dimensional current flow data; a modeling module for constructing a three-dimensional flow prediction model, the three-dimensional flow prediction model including a random deactivation module, a local feature calculation module, an intermediate convolution module, and a global feature calculation module; the random deactivation module is used to control the deactivation of nodes of the three-dimensional flow prediction model according to a set ratio; the local feature calculation module is used to perform local feature calculations on model input data; the intermediate convolution module is used to control weight reduction; and the global feature calculation module is used to perform global feature calculations to generate three-dimensional current flow data; A training module, configured to train the three-dimensional flow prediction model using the training data set to obtain a trained three-dimensional flow prediction model; a prediction module, configured to obtain geometric shape data and a rotational speed of the wind turbine to be tested, and input the geometric shape data and the rotational speed of the wind turbine to be tested into the trained three-dimensional flow prediction model to obtain three-dimensional current flow data of the wind turbine to be tested; The geometric shape data of the wind turbine blade includes a geometric volume and a geometric image, and the training data construction module is configured to: obtain boundary condition inflow parameters and boundary condition outflow parameters of the blade based on the geometric image; and use the geometric volume, the rotation speed, the boundary condition inflow parameters, and the boundary condition outflow parameters as model input data; The local feature calculation module includes a plurality of first network structures, each of which includes a plurality of network branches with different filtering masks; The global feature calculation module includes multiple second network structures, and the second network structure includes multiple network branches with the same filtering mask.

5. A prediction device based on three-dimensional current flow in a wind turbine, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the prediction method based on three-dimensional current flow of a wind turbine according to any one of claims 1 to 3.

6. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to enable the computer to execute the prediction method based on three-dimensional current flow of a wind turbine according to any one of claims 1-3.

Images