Wind turbine generator set incoming wind condition prediction method and system based on meteorological background and wind power plant characteristics

Abstract

The invention relates to the technical field of wind power generation, in particular to a wind turbine generator incoming wind condition prediction method and system based on a meteorological background and wind power plant characteristics, and the method comprises the steps: obtaining historical incoming wind condition data through an anemometer tower, and calculating a historical incoming wind condition sequence affected by an upstream terrain wake flow effect in front of a wind wheel of a wind turbine generator; obtaining weather forecast data and interpolating the weather forecast data of different air pressure layers to a unit target height; constructing a spatial downscaling model, and establishing a spatial mapping relationship between the historical weather forecast data of the unit target height and the incoming wind condition; constructing a time feature extraction model, and extracting time features based on historical weather forecast data and unit operation data; and combining weather forecast data, historical incoming flow wind condition data of the unit and operation data of the unit, and outputting an incoming flow wind condition prediction result by utilizing the spatial-temporal feature extraction model. The objective of the invention is to realize simultaneous prediction of incoming flow wind speed and wind direction and improve the accuracy of incoming flow wind condition prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a method and system for predicting incoming wind conditions of a wind turbine generator set based on meteorological background and wind farm characteristics. Background Art

[0002] In the field of wind power generation, accurately predicting the incoming wind conditions in front of wind turbines is a key strategy to improve power generation efficiency. Currently, most existing studies focus on the prediction of a single wind speed or wind direction, lacking a comprehensive analysis of the complex wind farm environment and meteorological background. In addition, most existing models rely on wind speed data obtained by the SCADA system for prediction. Due to the influence of the rotation of the wind turbine hub, the SCADA wind speed cannot accurately reflect the incoming wind conditions, which limits its effectiveness in practical applications.

[0003] However, the wind direction prediction in the incoming wind condition is more complex than the wind speed prediction. Since the wind direction is directional, it is difficult for traditional statistical methods to capture its inherent changing patterns. Based on this, some studies have introduced machine learning methods, such as neural networks and Bayesian analysis, in order to improve the accuracy of wind direction prediction. Nevertheless, the existing methods still focus on the prediction of a single parameter of wind speed or wind direction, and lack a comprehensive method for simultaneously predicting wind speed and wind direction, which affects the final prediction accuracy. Summary of the invention

[0004] In order to achieve the simultaneous prediction of incoming wind speed and wind direction and improve the accuracy of incoming wind condition prediction, the present invention provides a method and system for predicting incoming wind conditions of wind turbines based on meteorological background and wind farm characteristics. The technical solutions adopted are as follows:

[0005] The technical solution of the first aspect of the present invention provides a method for predicting wind conditions of wind turbines based on meteorological background and wind farm characteristics, the method comprising:

[0006] The historical incoming wind condition data is obtained by using the wind tower, and the historical incoming wind condition sequence in front of the computer group wind rotor affected by the upstream terrain wake effect is calculated;

[0007] Obtain weather forecast data and interpolate weather forecast data of different pressure layers to the crew's target altitude;

[0008] Construct a spatial downscaling model to establish the spatial mapping relationship between the historical meteorological forecast data of the unit target altitude and the incoming wind conditions;

[0009] Construct a time feature extraction model to extract time features based on historical weather forecast data and unit operation data;

[0010] Combining meteorological forecast data, historical incoming wind condition data of the unit and unit operation data, the spatiotemporal feature extraction model is used to output the incoming wind condition prediction results.

[0011] Furthermore, the historical wind condition sequence in front of the computer group wind rotor affected by the upstream terrain wake effect includes:

[0012] Divide wind direction quadrants according to wind farm layout parameters;

[0013] Calculate the horizontal distance between the target unit and the upstream unit under a given wind direction;

[0014] Calculate the initial wake radius of the upstream unit based on the unit rotor radius and the preset axial flow induction factor;

[0015] The position information of the target unit within the initial wake radius is extracted, and the incoming wind speed of the target unit under the influence of the terrain wake effect of the upstream unit is calculated using the three-dimensional wake model.

[0016] Furthermore, the expression for the horizontal distance between the target unit and the upstream unit under a given wind direction is:

[0017]

[0018] Where, d mn Represents the horizontal distance between the upstream unit m and the target unit n; Indicates the angle of a given wind direction; (x m ,y m ) represents the location of the upstream unit; (x n ,y n ) indicates the location of the target unit.

[0019] Further, obtaining weather forecast data and interpolating weather forecast data of different pressure layers to the target altitude of the crew includes:

[0020] Select multiple grid points closest to the coordinate position of the target unit;

[0021] The meteorological forecast data of multiple grid points in different pressure layers are interpolated to the target unit location using the Kriskin vector interpolation method;

[0022] Calculate the forecast wind speed and direction data based on the interpolated weather forecast data.

[0023] Furthermore, using the Kriskin vector interpolation method to interpolate the meteorological forecast data of multiple grid points at different pressure layers to the target unit location includes:

[0024] Calculate the distance from each grid point to the target unit coordinate position;

[0025] With the goal of minimizing the variance between the estimated value and the true value and unbiased estimation, the weight coefficients on different wind speed components are solved;

[0026] The wind speed is interpolated using the weight coefficient of the wind speed vector, and the wind speed component interpolation at the target unit location in different wind directions is extracted.

[0027] Furthermore, the wind speed component interpolation of the target unit coordinate position in different wind directions is extracted, including:

[0028] The wind speed component interpolation of the target unit coordinate position in the east-west direction is extracted, and its expression is:

[0029]

[0030] Where u represents the east-west direction of wind speed on the horizontal plane; n (u) represents the weight coefficient of the u component; n represents the number of grid points under the u component;

[0031] The wind speed component interpolation of the target unit coordinate position in the north-south direction is extracted, and its expression is:

[0032]

[0033] Where, v component represents the east-west direction of wind speed on the horizontal plane; m (v) represents the weight coefficient of the v component; m represents the number of grid points under the v component.

[0034] Furthermore, a spatial downscaling model is constructed to establish the spatial mapping relationship between the historical meteorological forecast data of the unit target height and the incoming wind conditions, including:

[0035] The LightGBM model is used to establish the mapping relationship between the unit target altitude meteorological forecast data and the historical incoming wind conditions and extract spatial features.

[0036] Furthermore, a time feature extraction model is constructed to extract time features based on historical meteorological forecast data and unit operation data, including:

[0037] The encoder-decoder structure is used in combination with a multi-head cross-attention neural network to extract temporal features from historical weather forecast data and unit operation data.

[0038] Furthermore, by combining the meteorological forecast data, the historical wind condition data of the unit and the unit operation data, the spatiotemporal feature extraction model is used to output the wind condition forecast results, including:

[0039] The temporal and spatial feature extraction models are fused, and residual connections and data normalization are performed to finally output the forecast results of incoming wind speed and direction.

[0040] The technical solution of the second aspect of the present invention provides a system for predicting wind conditions of wind turbines based on meteorological background and wind farm characteristics, the system comprising:

[0041] A data acquisition module is configured to obtain historical incoming wind condition data using a wind tower, and to calculate a historical incoming wind condition sequence in front of a wind rotor affected by an upstream terrain wake effect;

[0042] A data interpolation module configured to obtain weather forecast data and interpolate the weather forecast data of different pressure layers to the target altitude of the crew;

[0043] A spatial downscaling module is configured to construct a spatial mapping relationship between the historical meteorological forecast data of the unit target height and the incoming wind conditions;

[0044] A time feature extraction module configured to extract time features based on historical weather forecast data and unit operation data;

[0045] The wind condition prediction module is configured to combine the meteorological forecast data, the historical incoming wind condition data of the unit and the unit operation data, and use the spatiotemporal feature extraction model to output the incoming wind condition prediction result.

[0046] The present invention has the following beneficial effects:

[0047] The invention provides a method for predicting the incoming wind conditions of wind turbines based on meteorological background and wind farm characteristics, and proposes a method for predicting the incoming wind conditions of wind turbines in two steps. First, a historical incoming wind condition data is obtained by using a wind measuring tower to extract a historical incoming wind condition sequence affected by the upstream wake effect in front of the wind rotor of the turbine, and the coupling effect of the wind farm terrain wake effect on the wind condition is considered, so as to improve the accuracy of the incoming wind condition prediction; then, by interpolating the meteorological forecast data of different pressure layers to the target height of the turbine, and interpolating the medium-sized meteorological forecast data to the target height of the turbine, the data dimension is effectively reduced, and the calculation complexity of the model is reduced; further, a spatial mapping relationship between the meteorological forecast data of the target height of the turbine and the historical incoming wind conditions is established, and the spatiotemporal characteristics of wind speed are extracted by the model fusion technology of the encoder-decoder structure, and the influence of multiple factors on the wind conditions is considered, so as to achieve a more precise and comprehensive prediction of the incoming wind conditions of the wind turbines, and the wind speed and wind direction can be predicted at the same time, and the wake effect, spatial downscaling, time characteristics and spatiotemporal characteristics fusion are comprehensively considered, which can significantly improve the accuracy of the incoming wind condition prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1A method flow chart of a method for predicting wind conditions of wind turbines based on meteorological background and wind farm characteristics provided by one embodiment of the present invention;

[0050] Figure 2 A schematic diagram of the structure of a system for predicting incoming wind conditions for wind turbines based on meteorological background and wind farm characteristics provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the method and system for predicting the incoming wind conditions of a wind turbine set based on the meteorological background and the characteristics of a wind farm according to the present invention, in combination with the accompanying drawings and preferred embodiments, as well as its specific implementation method, structure, features and effects. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0052] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0053] The following is a detailed description of a method and system for predicting incoming wind conditions for wind turbines based on meteorological background and wind farm characteristics provided by the present invention in conjunction with the accompanying drawings.

[0054] See also Figure 1 , which shows a method flow chart of a method for predicting wind conditions of wind turbines based on meteorological background and wind farm characteristics provided by the first aspect of the technical solution of the present invention, the method comprising:

[0055] Step S100: Use a wind tower to obtain historical incoming wind condition data, and extract the historical incoming wind condition sequence in front of the unit's wind rotor affected by the upstream terrain wake effect; wherein, the wind tower, as a wind speed measurement device for the wind farm, can accurately capture the wind speed and wind direction distribution at its location. However, inside the wind farm, the incoming wind conditions in front of the wind rotors of different units are affected by the coupling of the wake between units and terrain factors, and will present different wind speed distributions; the terrain wake coupling effect of the upstream unit will cause a decrease in the free flow wind speed and an increase in turbulence, affecting the incoming wind speed of the downstream unit; therefore, when you want to use the historical incoming wind condition data obtained by the wind tower to calculate the incoming wind speed in front of the unit's wind rotor, you need to consider the superposition interference of wakes between units and the influence of terrain factors, and then determine the incoming wind conditions in front of the wind rotor of each unit in the wind farm;

[0056] When the installation distance between the units in the wind farm is relatively close, under the influence of the incoming wind direction, the downstream units in the wind farm may be affected by the wake generated by the upstream units; therefore, when calculating the incoming wind speed in front of the wind rotor of the target unit, it is first necessary to determine whether the target unit is affected by the wake effect of the upstream unit under different wind directions; step S100 specifically includes:

[0057] Step S110: Divide the wind direction quadrants according to the wind farm layout parameters; for example, the total number of wind farm units is N, the upstream unit is numbered m, and the position in the wind farm is (x m ,y m ); the downstream unit number is n, and the position is (x n ,y n ); the hub altitudes of the upstream unit m and the downstream unit n are z m 、z n , the distance between the hub centers is D, and the horizontal distance between the two units is d mn , that is, the projection of D on the horizontal plane; for the convenience of analysis, a three-dimensional coordinate system (x, y, z) is established within the scope of the wind farm, and the relative positions of each unit are marked, where x and y represent the coordinates in the horizontal direction, and z represents the coordinates in the vertical direction; in the two-dimensional analysis plane, the north direction is taken as the zero-degree direction of the incoming wind, and the wind direction quadrant is divided every π / 2 angle, which is divided into four quadrants in total, as follows: the first quadrant: 0°≤wind direction<90°; the second quadrant: 90°≤wind direction<180°; the third quadrant: 180°≤wind direction<270°; the fourth quadrant: 270°≤wind direction<360;

[0058] Step S120: Calculate the horizontal distance between the target unit and the upstream unit under a given wind direction; specifically, the horizontal distance between the two units can be expressed as:

[0059]

[0060] In formula (1), d mn Represents the horizontal distance between the upstream unit m and the target unit n; Angle indicating a given wind direction;

[0061] By transforming and simplifying formula (1), we can get:

[0062]

[0063] In formula (2), Indicates the tangent value of a given wind direction angle; Represents the absolute value of the cosine of a given wind direction angle;

[0064] According to the trigonometric function transformation to correctly handle the angle values ​​in different quadrants, it can be expressed as:

[0065]

[0066] The horizontal distance between the two units can be expressed as:

[0067]

[0068] Step S120 provides a method for calculating the horizontal distance between the target unit and the upstream unit under a given wind direction, which improves the calculation accuracy while simplifying the calculation process and enhancing the robustness of the model; by accurately calculating the horizontal distance, the influence of the wake effect can be better evaluated;

[0069] Step S130: Calculate the initial wake radius of the upstream unit according to the unit rotor radius and the preset axial flow induction factor; the initial wake radius can be expressed as:

[0070] R w =R+a*d mn (6)

[0071] In formula (6), R w represents the initial wake radius of the upstream unit; R represents the rotor radius of the upstream unit; a represents the preset axial flow induction factor, which is used to reflect the speed of wake diffusion; d mn Represents the horizontal distance between the upstream unit m and the downstream unit n;

[0072] The rotor radius of the upstream unit can be calculated as follows:

[0073]

[0074] In equations (7) and (8), r represents the radius of the upstream wind turbine rotor, C T represents the axial thrust coefficient, that is, the ratio of the axial force generated by the wind turbine during operation to the incoming flow pressure; λ represents the axial induction factor, which is a dimensionless parameter that describes the influence of the wind turbine on the incoming wind speed; by calculating the initial wake radius of the upstream unit, the influence of the wake effect can be more accurately evaluated. Different wind directions and unit spacings will result in different wake radii, which will affect the incoming wind speed of the downstream unit.

[0075] Step S140: extract the position information of the target unit within the initial wake radius, and use the three-dimensional wake model to calculate the incoming wind speed of the target unit under the coupling influence of the terrain wake effect of the upstream unit; specifically, calculate the position of the downstream wind turbine within the wake radius of the upstream wind turbine to determine whether the downstream wind turbine is affected by the wake of the upstream wind turbine; for example, in the second and third quadrants, the maximum or minimum horizontal distance affected by the wake of the upstream unit can be expressed as:

[0076]

[0077] In the first and fourth quadrants, the maximum or minimum horizontal distance affected by the upstream unit wake can be expressed as:

[0078]

[0079] If the target unit is within the wake influence radius and height range of the upstream unit, it means that the target unit is affected by the wake effect of the upstream unit. The degree to which the target unit is affected by the wake of the upstream unit can be determined using a three-dimensional wake model, which can be expressed as:

[0080]

[0081] Where U m n It represents the incoming wind speed of the upstream unit m after being affected by the wake of the downstream unit n; U0 represents the free incoming wind speed; z represents the height difference between the wind rotors of the upstream unit m and the downstream unit n; η represents the wind shear index; ε y It can be described as r y =2.58*ε y =k y *x+R,k y represents the wake expansion coefficient on the horizontal plane, which is determined by the turbulence intensity on the horizontal plane; r z Indicates the wake radius in the vertical direction; ε z Represents the standard deviation of the vertical Gaussian distribution; z hub represents the hub height of the target unit; C represents the velocity shape characteristic parameter;

[0082] In an actual wind farm, the wind speed in front of the wind turbine rotor may be affected by the wake effect generated by multiple upstream turbines at the same time. At this time, the wind speed in front of the turbine considering the superimposed wake effect is:

[0083]

[0084] Where U n It represents the incoming wind speed after considering the influence of the superimposed wake; U0 represents the free incoming wind speed;

[0085] In summary, the incoming wind condition in front of the wind rotor of the target unit can be obtained, which serves as the basis for predicting the incoming wind condition of the unit.

[0086] Step S200: Acquire weather forecast data and interpolate weather forecast data of different pressure layers to the target height of the unit; in step S100, the acquisition of the incoming wind speed takes into account the influence of the actual wind field terrain and the wake coupling effect between units. This implementation further considers the influence of weather forecast data on the unit output to achieve accurate forecast of the incoming wind speed; wherein the data source of the weather forecast data is the mesoscale weather forecast data provided by the ERA5 data set, and the mesoscale weather forecast data can provide fine spatial resolution and time resolution; the ERA5 data set includes weather forecast data of seven different pressure layers, and each pressure layer data contains four different meteorological data points; the wind speed data of each coordinate point provided by the ERA5 data set is divided into u component and v component, which represent the east-west and north-south directions of the wind speed on the horizontal plane respectively; according to meteorological regulations, 0° represents due north wind, at which time u is equal to 0 and v is less than 0; 90° represents due east wind, at which time u is less than 0 and v is equal to 0;

[0087] Step S200 specifically includes:

[0088] Step S210: Select multiple grid points closest to the target unit coordinate position; for example, find four grid points closest to the target unit coordinate position in the ERA5 data set, and the four grid points form a rectangular area surrounding the target position;

[0089] Step S220: interpolating the weather forecast data of multiple grid points at different pressure layers to the target unit position using the Kriskin vector interpolation method;

[0090] Step S220 specifically includes:

[0091] Step S221: Calculate the distance from each grid point to the target unit coordinate position, that is, calculate the Euclidean distance from each grid point to the target unit coordinate position;

[0092] Step S222: solving the weight coefficients on different wind speed components with the goal of minimizing the variance between the estimated value and the true value and unbiased estimation;

[0093] The method to minimize the difference between the estimated value and the true value can be expressed as:

[0094]

[0095] In formulas (14) and (15), λ m Represents the weight coefficient of the v component; λ n Represents the weight coefficient of the u component; Represents the estimated value of the wind speed v component at the target location obtained by interpolation; Z i (v) represents the true value of the wind speed v component actually observed at the target location; represents the estimated value of the wind speed u component at the target location obtained by interpolation; Z i (u) represents the true value of the wind speed u component actually observed at the target location;

[0096] The unbiased estimate can be expressed as:

[0097]

[0098] In formulas (16) and (17), E(.) represents the expected value, which is used to ensure that the expectation of the estimated value is equal to the true value, that is, the estimated value has no systematic deviation;

[0099] Step S223: interpolating the wind speed using the weight coefficient of the wind speed vector to extract the interpolated wind speed components at the target unit location in different wind directions;

[0100] Among them, the interpolation of the u component can be expressed as:

[0101]

[0102] In formula (18), λ m (v) represents the weight coefficient of the v component; m represents the grid point under the u component;

[0103] The interpolation of the u component can be expressed as:

[0104]

[0105] In formula (19), λ n (u) represents the weight coefficient of the u component, and n represents the grid point under the v component;

[0106] In this embodiment, using multiple nearest grid points for interpolation can better capture changes in local meteorological conditions and improve the spatial resolution of the interpolation results; by using the Kriskin interpolation method, combined with the weight coefficients on different wind direction components, the wind speed component of the target location is calculated. This method not only improves the accuracy of the incoming wind speed prediction, but also takes into account the changes in local meteorological conditions and enhances the robustness of the model; the scale of meteorological forecast data at different air pressures and different coordinate points is large, and direct use will increase the computational complexity. In this embodiment, the data of four coordinate points of each pressure layer are interpolated to the vertical height of the unit to reduce the data dimension, thereby reducing the amount of model calculation;

[0107] Step S230: Calculating wind speed and direction data at the height of the computer according to the interpolated weather forecast data;

[0108] The expression for wind speed is:

[0109]

[0110] In formula (20), u i represents the wind speed of the i-th unit;

[0111] The expression for wind direction is:

[0112]

[0113] In formulas (21) and (22), represents the wind direction angle; arctan2(v,u) represents the four-quadrant inverse tangent function; this formula can convert the wind direction into a range of 0° to 360°, and can reduce the four coordinate point data of each pressure layer to one target height point data, thereby simplifying the data set and improving data analysis efficiency; in this implementation case, the wind speed component obtained by interpolation is Converting to actual wind speed and direction provides high-precision weather forecast data input. This method not only simplifies the data set, but also takes into account the changes in local meteorological conditions, which can further improve the accuracy of incoming wind speed prediction and enhance the robustness of the model.

[0114] Step S300: Construct a spatial downscaling model to establish a spatial mapping relationship between the historical weather forecast data of the unit target height and the incoming wind conditions; specifically, use the LightGBM model to establish a mapping relationship between the unit target height weather forecast data and the historical incoming wind conditions and extract spatial features; the spatial downscaling of weather forecast data can establish the relationship between the mesoscale weather forecast data and the microscale unit incoming wind speed, and use the weather forecast data after interpolation processing in step S200 and the historical incoming wind condition sequence extracted in step S100 as the training data set; divide the data set into multiple time windows, each window contains data of H time steps, and set Light The parameters of the tGBM model include at least the learning rate, the number of trees, and the maximum depth; then the LightGBM model is trained to improve the model performance through techniques such as gradient-based one-sided sampling, histogram tree node splitting, Leaf-wise tree growth strategy, mutually exclusive feature bundling method, sparse optimization, and parallel processing strategy; the mean square error can be used as the loss function during training, and the LightGBM model minimizes the loss function by iteratively adding base learners; the LightGBM model uses efficient feature processing and optimization strategies to more accurately capture the mapping relationship between mesoscale meteorological forecast data and microscale actual wind speed, thereby improving the accuracy of wind speed prediction.

[0115] Step S400: construct a time feature extraction model, extract time features based on historical weather forecast data and unit operation data; specifically, use a time convolutional neural network with an encoder-decoder structure, combined with a multi-head cross attention mechanism, to extract time features from historical weather forecast data and unit operation data, and the decoder outputs the time feature prediction result; since the natural wind in the meteorological data is the gas movement caused by the pressure difference in the atmosphere and the rotation of the earth, the wind speed has a time trend, and the reasonable extraction of time features can effectively enhance the robustness and accuracy of the prediction model;

[0116] The encoder-decoder structure is as follows:

[0117] The encoder includes: Dilated Causal Convolution Layer: Use Dilated Causal Convolution to capture long-term series dependencies; Dilated convolution increases the receptive field by introducing intervals in the convolution kernel while reducing the number of parameters; Normalization Layer: Use Batch Normalization or other normalization methods to speed up the training process and improve model stability; Non-linear Activation Layer: Use ReLU activation function to improve the network's expressiveness and learning efficiency; Regularization Layer: Use Dropout or other regularization methods to prevent overfitting; Residual Connection: Use Residual Connection Connections) to avoid the gradient vanishing problem, making deep networks easier to train; decoder: void causal convolution layer: also use void causal convolution layer to generate future prediction sequences; normalization layer: use batch normalization or other normalization methods; nonlinear activation layer: use ReLU activation function; regularization layer: use Dropout or other regularization methods; residual connection: use residual connection to avoid the gradient vanishing problem; encoder and decoder are used to capture the complex time patterns in historical weather forecast data, where the encoder is used to capture the time patterns and features in the input sequence, and the decoder is used to generate future prediction sequences, which can be expressed as:

[0118] P(y [t+1:t+H] |x [t-s:t] ,y [t-s:t] )(twenty four)

[0119] In formula (23), y [t+1:t+H] represents the wind speed forecast value for the next H time steps; x [t-s:t] represents the historical weather forecast data from time ts to time t; y [t-s:t] represents the historical actual wind speed observation value from time ts to time t; s represents the length of the historical sequence;

[0120] Then, after the time series features pass through the encoder-decoder structure, they will enter the multi-head cross-attention mechanism module; the multi-head cross-attention mechanism connects the encoder and decoder outputs, and each attention can focus on the relationship between different inputs and outputs, improving the quality of the output sequence and the generalization ability of the model; at the same time, the output of the decoder is also connected to the output of the multi-head cross-attention mechanism. This residual path allows the decoder to obtain longer-term sequence dependencies and global information from the output sequence; the encoder-decoder structure combined with the multi-head cross-attention mechanism can capture complex patterns in time series, including seasonal changes, trends, and periodic fluctuations, thereby improving the accuracy of wind speed prediction.

[0121] Step S500: Combine weather forecast data, historical incoming wind condition data of the unit and unit operation data, and use the spatiotemporal feature extraction model to output the incoming wind condition prediction result; specifically, fuse the spatial downscaling model and the temporal feature extraction model, and use weather forecast data and unit operation data to train the incoming wind condition prediction model, and output the wind speed and wind direction prediction results; use the fusion model to extract the spatiotemporal information of wind speed; the data used include future weather forecast data: including the wind speed u component and v component of weather forecast data for the next H time steps, historical weather forecast data, historical incoming wind condition sequence and unit operation status data; divide the data into multiple time windows, each of which is a time window. A window contains S time steps of historical data and H time steps of future data; the LightGBM model is used to extract the mapping relationship between the target height weather forecast data and the historical incoming wind condition sequence to obtain spatial features; the encoder-decoder structured temporal convolutional neural network is used to capture the complex temporal patterns in the historical weather forecast data and the actual incoming wind condition observation data; the spatial features extracted by LightGBM and the temporal features extracted by the encoder-decoder are concatenated; then the wind condition prediction model is trained, and the root mean square loss function can be used; the trained model is applied to the real-time weather forecast data to predict the future wind speed and wind direction prediction results;

[0122] Step S500 constructs a comprehensive wind prediction model by fusing the spatial features extracted by LightGBM and the temporal features extracted by the encoder-decoder structure; normalizes the spatial features of wind speed extracted by LightGBM and the temporal features of wind speed fused at multiple scales to accelerate the convergence of the model, improve the generalization ability and stability of the model, and performs residual connection on the output of the decoder to solve the gradient disappearance problem; the final model extracts combined features through the Dense layer to fit the complex spatiotemporal feature relationship, and by fusing the spatial features extracted by LightGBM and the temporal features extracted by the encoder-decoder structure, it can capture the spatiotemporal characteristics of wind speed more comprehensively, improve the model's ability to capture complex data, and thus improve the prediction accuracy of the wind conditions of the wind turbine inflow.

[0123] See also Figure 2 , which shows a wind turbine set incoming wind condition prediction system based on meteorological background and wind farm characteristics provided by the technical solution of the second aspect of the present invention, the system comprising: a data acquisition module, configured to use a wind tower to obtain historical incoming wind condition data, and calculate the historical incoming wind condition sequence in front of the wind rotor of the computer set affected by the upstream terrain wake effect; a data interpolation module, configured to obtain meteorological forecast data and interpolate the meteorological forecast data of different pressure layers to the target height of the unit; a spatial downscaling module, configured to establish a spatial mapping relationship between the historical meteorological forecast data of the target height of the unit and the incoming wind condition; a time feature extraction module, configured to extract time features based on historical meteorological forecast data and unit operation data; a wind condition prediction module, configured to combine meteorological forecast data, historical incoming wind condition data of the unit and unit operation data, and use a spatiotemporal feature extraction model to output the incoming wind condition prediction result.

[0124] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0125] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for predicting wind conditions for wind turbines based on meteorological background and wind farm characteristics, characterized in that: The method comprises: The historical incoming wind condition data is obtained by using the wind tower, and the historical incoming wind condition sequence in front of the computer group wind rotor affected by the upstream terrain wake effect is calculated; Obtain weather forecast data and interpolate weather forecast data of different pressure layers to the crew's target altitude; Construct a spatial downscaling model to establish the spatial mapping relationship between the historical meteorological forecast data of the unit target altitude and the incoming wind conditions; Construct a time feature extraction model to extract time features based on historical weather forecast data and unit operation data; Combining meteorological forecast data, historical incoming wind condition data of the unit and unit operation data, the spatiotemporal feature extraction model is used to output the incoming wind condition prediction results.

2. The method for predicting wind conditions of wind turbines based on meteorological background and wind farm characteristics according to claim 1, characterized in that: The historical wind condition sequence in front of the computer group wind rotor affected by the upstream terrain wake effect includes: Divide wind direction quadrants according to wind farm layout parameters; Calculate the horizontal distance between the target unit and the upstream unit under a given wind direction; Calculate the initial wake radius of the upstream unit based on the unit rotor radius and the preset axial flow induction factor; The position information of the target unit within the initial wake radius is extracted, and the incoming wind speed of the target unit under the influence of the terrain wake effect of the upstream unit is calculated using the three-dimensional wake model.

3. The method for predicting wind conditions of wind turbines based on meteorological background and wind farm characteristics according to claim 2, characterized in that: The expression for the horizontal distance between the target unit and the upstream unit under a given wind direction is: Where, d mn Represents the horizontal distance between the upstream unit m and the target unit n; Indicates the angle of a given wind direction; (x m ,y m ) represents the location of the upstream unit; (x n ,y n ) indicates the location of the target unit.

4. The method for predicting wind conditions of wind turbines based on meteorological background and wind farm characteristics according to claim 1, characterized in that: Obtaining weather forecast data and interpolating weather forecast data of different pressure layers to the crew's target altitude includes: Select multiple grid points closest to the target unit coordinate position; The meteorological forecast data of multiple grid points in different pressure layers are interpolated to the target unit location using the Kriskin vector interpolation method; Calculate the forecast wind speed and direction data based on the interpolated weather forecast data.

5. The method for predicting wind conditions of wind turbines based on meteorological background and wind farm characteristics according to claim 4, characterized in that: Using the Kriskin vector interpolation method to interpolate the meteorological forecast data of multiple grid points in different pressure layers to the target unit location includes: Calculate the distance from each grid point to the target unit coordinate position; With the goal of minimizing the variance between the estimated value and the true value and unbiased estimation, the weight coefficients on different wind speed components are solved; The wind speed is interpolated using the weight coefficient of the wind speed vector, and the wind speed component interpolation at the target unit location in different wind directions is extracted.

6. The method for predicting wind conditions of wind turbines based on meteorological background and wind farm characteristics according to claim 5, characterized in that: Extract the wind speed component interpolation of the target unit coordinate position in different wind directions, including: The wind speed component interpolation of the target unit coordinate position in the east-west direction is extracted, and its expression is: Where, u component represents the east-west direction of wind speed on the horizontal plane; n (u) represents the weight coefficient of the u component; n represents the number of grid points under the u component; The wind speed component interpolation of the target unit coordinate position in the north-south direction is extracted, and its expression is: Where, v component represents the east-west direction of wind speed on the horizontal plane; m (v) represents the weight coefficient of the v component; m represents the number of grid points under the v component.

7. The method for predicting wind conditions of wind turbines based on meteorological background and wind farm characteristics according to any one of claims 1 to 6, characterized in that: Constructing a spatial downscaling model and establishing the spatial mapping relationship between the historical meteorological forecast data of the unit target height and the incoming wind conditions include: The LightGBM model is used to establish the mapping relationship between the unit target altitude meteorological forecast data and the historical incoming wind conditions and extract spatial features.

8. The method for predicting wind conditions of wind turbines based on meteorological background and wind farm characteristics according to claim 7, characterized in that: Construct a time feature extraction model, and extract time features based on historical weather forecast data and unit operation data, including: The encoder-decoder structure is used in combination with a multi-head cross-attention neural network to extract temporal features from historical weather forecast data and unit operation data.

9. The method for predicting wind conditions of wind turbines based on meteorological background and wind farm characteristics according to claim 8, characterized in that: Combining meteorological forecast data, historical wind data of the unit and unit operation data, the spatiotemporal feature extraction model is used to output the wind forecast results, including: The temporal and spatial feature extraction models are fused, and residual connections and data normalization are performed to finally output the forecast results of incoming wind speed and direction.

10. A wind turbine inflow wind condition prediction system based on meteorological background and wind farm characteristics, characterized in that: The system comprises: A data acquisition module is configured to obtain historical incoming wind condition data using a wind tower, and to calculate a historical incoming wind condition sequence in front of a wind rotor affected by an upstream terrain wake effect; A data interpolation module configured to obtain weather forecast data and interpolate the weather forecast data of different pressure layers to the target altitude of the crew; A spatial downscaling module is configured to construct a spatial mapping relationship between the historical meteorological forecast data of the unit target height and the incoming wind conditions; A time feature extraction module configured to extract time features based on historical weather forecast data and unit operation data; The wind condition prediction module is configured to combine the meteorological forecast data, the historical incoming wind condition data of the unit and the unit operation data, and use the spatiotemporal feature extraction model to output the incoming wind condition prediction result.

Images