A wind turbine blade inflow prediction method and system based on multi-source operation data

Abstract

The application discloses a wind turbine blade inflow prediction method and system based on multi-source operation data, and belongs to the technical field of wind power generation, and comprises the following steps: obtaining multi-source operation data; data preprocessing and basic time sequence feature construction; input feature construction; CNN-BiGRU-Attention prediction model construction and training; and outputting a prediction result; real-time nacelle wind speed and real wind direction of all units in a wind farm are taken as real-time input features, the obtained data standardizer is used to normalize the real-time input features, and then the normalized real-time input features are input into the trained CNN-BiGRU-Attention prediction model, and a future 15-minute target wind turbine blade front inflow wind speed prediction result is output. The application realizes wind turbine blade inflow wind speed prediction in a future period of time.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation technology, specifically relating to a method and system for predicting the inflow of wind turbine blades based on multi-source operating data. Background Technology

[0002] As wind power installed capacity continues to increase, wind turbine operation control is gradually evolving from single-unit control to wind farm-level collaborative optimization control. The inflow wind speed in front of the wind turbine blades is a crucial input condition for yaw control and load optimization. Based on the trend of changes in the inflow wind speed in front of the wind turbine blades over a future period, more reliable data support can be provided for wind turbine yaw optimization, wake load reduction, and collaborative control.

[0003] Currently, it is usually difficult to directly obtain the actual incoming wind speed in front of each wind turbine blade in a real wind farm. The incoming wind speed is usually approximated by the representative measurement value of the nacelle anemometer or the meteorological tower. However, both of these data have limitations to varying degrees: (1) The nacelle anemometer is usually located on the top of the nacelle or near the rear of the wind turbine. The measurement results are easily affected by the rotor rotation disturbance and the wake of the front wind turbine.

[0004] (2) The data from the wind measurement tower is of high quality, but the number of measurement points is limited. It can only represent the wind conditions in a local area and is difficult to accurately reflect the spatial differences of wind turbines in different locations in a large wind farm and the accurate incoming wind speed of the wind turbines. Summary of the Invention

[0005] This invention provides a method and system for predicting the incoming wind speed of wind turbine blades based on multi-source operational data. It uses wind farm SCADA data as a foundation and a small amount of measured incoming wind speed in front of the blades by nacelle lidar as high-confidence calibration data. The mapping relationship between the nacelle measured wind speed and the actual incoming wind speed of the blades is established through a self-test transfer function. Finally, a CNN-BiGRU-Attention deep learning model is used to predict the incoming wind speed of wind turbine blades over a period of time.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for predicting wind turbine blade inflow based on multi-source operational data includes: (1) Obtain multi-source operation data Collect multi-source operational data within the target wind farm, including SCADA data, nacelle lidar data, and unit power curve data for each wind turbine. (2) Data preprocessing and construction of basic time series features The acquired multi-source operational data is preprocessed, including time alignment, missing value handling, outlier removal, and normalization. The actual wind direction is calculated based on the nacelle location and wind direction, and then converted into sine and cosine components. Based on the preprocessed data, the basic time-series features of all wind turbines are constructed to obtain a time-synchronized effective sample dataset. (3) Constructing input features From the obtained valid sample dataset, select valid data samples of the nacelle lidar data that are in grid-connected status. Set confidence weights based on the data quality and operating status of the samples. Use nacelle wind speed, actual wind direction, wind speed dispersion in the field, and the deviation between the wind speed of a single nacelle and the average nacelle wind speed in the entire field as input features. Use the measured incoming wind speed of the nacelle lidar as the target label. Use weighted ridge regression to solve the transfer function parameters to obtain the self-tested nacelle transfer function. Then use the self-tested nacelle transfer function to obtain the corrected incoming wind features corresponding to the nacelle wind speed. Based on the obtained real wind direction and the coordinates of the wind turbines in the wind farm, the downwind offset distance, lateral offset distance and correlation score between the candidate unit and the target unit are calculated, and the associated unit is selected according to the results; The final input features are: historical wind speed of the target wind turbine and associated units, first-order wind speed change of the historical wind speed sequence, sine and cosine components of the actual wind direction, and normalized downwind and lateral offset distances of the target and candidate units; corresponding output labels are constructed: the incoming wind speed is calculated based on the calibrated nacelle transfer function to obtain the model training dataset; (4) Construct and train the CNN-BiGRU-Attention prediction model The obtained model training dataset is divided into training set and test set according to the time order. A CNN-BiGRU-Attention prediction model including one-dimensional convolutional layer, bidirectional gated recurrent unit layer, temporal attention layer and multi-task output layer is constructed. The prediction model is trained with multi-task loss function to obtain the optimal model parameters and data normalizer. (5) Output prediction results The real-time nacelle wind speed and actual wind direction of the wind turbine units at the wind farm site are used as real-time input features. The real-time input features are standardized using the obtained data normalizer and then input into the trained CNN-BiGRU-Attention prediction model to output the predicted wind speed in front of the target wind turbine blades within the next 15 minutes.

[0007] A further improvement of the present invention is that, in step (1), the SCADA data includes timestamp, nacelle wind speed, nacelle wind direction, nacelle position, active power and operating status; the data of a nacelle lidar includes the incoming wind speed at the distance measuring point in front of the wind turbine, data quality mark and timestamp.

[0008] A further improvement of the present invention is that, in step (2), the effective sample dataset includes timestamp, cabin wind speed, true wind direction, sine component of true wind direction, cosine component of true wind direction, incoming wind speed measured by lidar, data quality indicator of incoming wind speed, active power of wind turbine, first-order wind speed variation, moving average and moving standard deviation of wind speed.

[0009] A further improvement of the present invention is that, in step (3), the self-test cabin transfer function is constructed as follows:

[0010]

[0011]

[0012] In the formula, Indicates the incoming air velocity of the fan blades; The vector represents the original measured wind speed in the cabin; D represents the actual wind direction; and S represents the wind speed dispersion within the field. and These represent the 90th and 10th quantiles of the wind speed sample, respectively. This indicates the deviation between the wind speed in a single engine room and the average wind speed in the entire engine room. The specific manifestations are as follows:

[0013] In the formula, The parameters represent the transfer function to be solved. We use weighted ridge regression to solve for the transfer function parameters and minimize the following objective function:

[0014] In the formula, Indicates the first i Confidence weights for each sample; Indicates the first i The measured inflow wind speed corresponding to each sample by lidar; Indicates the first i Input features of each sample; This represents the regularization coefficient.

[0015] A further improvement of the present invention is that, in step (3), the downwind offset distance and the lateral offset distance are:

[0016]

[0017] In the formula, It indicates the downwind offset distance, reflecting the propagation distance from the upstream unit to the target unit along the main direction of the incoming flow; It represents the lateral offset distance, reflecting the degree of lateral offset of the candidate unit relative to the main incoming flow direction; This indicates the distance between the candidate wind turbine and the target wind turbine. ; This indicates the angle between the candidate wind turbine and the current incoming flow. , , This indicates the relative azimuth angle between the candidate wind turbine and the target wind turbine. Represents the plane coordinates of the wind turbine; The aforementioned correlation score is represented as follows:

[0018] according to and Select upstream wind turbines, when and When considering candidate wind turbines as upstream wind turbines, a comprehensive scoring function is set. The historical wind speed sequence and the first-order wind speed change are expressed as follows:

[0019] The historical wind speed sequence is represented as follows:

[0020] In the formula, L = 120; The change in first-order wind speed is expressed as:

[0021] Similarly, by constructing a historical wind direction sequence, the wind speed variable can be replaced with the sine and cosine components of the wind direction.

[0022] A further improvement of this invention is that, in step (4), the constructed CNN-BiGRU-Attention prediction model is as follows: First, a one-dimensional convolutional layer is used to extract local change features at a short time scale. Then, a two-layer GRU network is used to learn the long-term temporal dependencies in the wind speed and direction sequences. Subsequently, a temporal attention mechanism is introduced to adaptively weight the importance of different historical moments, forming a shared feature vector that can characterize the dynamic changes of the wind field. At the output end, the CNN-BiGRU-Attention prediction model adopts a multi-task branch structure to simultaneously predict wind speed and direction results at 30-second and 1-minute scales.

[0023] A wind turbine blade inflow prediction system based on multi-source operational data includes: Multi-source operation data acquisition unit: Collects multi-source operation data within the target wind farm, including SCADA data, nacelle lidar data, and unit power curve data for each wind turbine; Data preprocessing and basic time series feature construction unit: preprocesses the acquired multi-source operation data, including time alignment, missing value handling, outlier removal and normalization, and calculates the true wind direction based on the nacelle location and nacelle wind direction, and converts the true wind direction into sine and cosine components; based on the preprocessed data, constructs the basic time series features of all wind turbines to obtain a time-synchronized effective sample dataset. Constructing input feature units: Select valid and grid-connected data samples from the obtained effective sample dataset of the nacelle lidar data. Set confidence weights according to the data quality and operating status of the samples. Use nacelle wind speed, true wind direction, field wind speed dispersion, and deviation between the wind speed of a single nacelle and the average nacelle wind speed of the entire field as input features. Use the measured incoming wind speed of the nacelle lidar as the target label. Use weighted ridge regression to solve the transfer function parameters to obtain the self-tested nacelle transfer function. Use the self-tested nacelle transfer function to obtain the corrected incoming wind features corresponding to the nacelle wind speed. Based on the obtained real wind direction and the coordinates of the wind turbines in the wind farm, the downwind offset distance, lateral offset distance and correlation score between the candidate unit and the target unit are calculated, and the associated unit is selected according to the results; The final input features are: historical wind speed of the target wind turbine and associated units, first-order wind speed change of the historical wind speed sequence, sine and cosine components of the actual wind direction, and normalized downwind and lateral offset distances of the target and candidate units; corresponding output labels are constructed: the incoming wind speed is calculated based on the calibrated nacelle transfer function to obtain the model training dataset; Construct and train CNN-BiGRU-Attention prediction model units: Divide the obtained model training dataset into training set and test set according to time order, construct a CNN-BiGRU-Attention prediction model including one-dimensional convolutional layer, bidirectional gated recurrent unit layer, temporal attention layer and multi-task output layer, train the prediction model with multi-task loss function to obtain the optimal model parameters and data normalizer. The output prediction result unit takes the real-time nacelle wind speed and actual wind direction of the wind turbine units at the wind farm as real-time input features, and uses the obtained data normalizer to standardize the real-time input features before inputting them into the trained CNN-BiGRU-Attention prediction model to output the predicted wind speed in front of the target wind turbine blades within the next 15 minutes.

[0024] A further improvement of this invention is that, in the acquisition of multi-source operational data units, SCADA data includes timestamps, nacelle wind speed, nacelle wind direction, nacelle position, active power, and operating status; and data from a single nacelle lidar includes the incoming wind speed at the distance measuring point in front of the wind turbine, data quality indicators, and timestamps.

[0025] A further improvement of this invention is that, in the data preprocessing and basic time series feature construction unit, the effective sample dataset includes timestamps, cabin wind speed, true wind direction, the sine component of the true wind direction, the cosine component of the true wind direction, the incoming wind speed measured by lidar, the data quality indicator of the incoming wind speed, the active power of the wind turbine, the first-order wind speed variation, the moving average and moving standard deviation of the wind speed.

[0026] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the wind turbine blade inflow prediction method based on multi-source operating data.

[0027] Compared with the prior art, the present invention has at least the following beneficial technical effects: This invention provides a method and system for predicting wind turbine blade inflow based on multi-source operational data. Firstly, it overcomes the shortcomings of traditional nacelle anemometers, which involve indirect wind measurement and large data errors. It uses measured data from the nacelle lidar as a high-confidence benchmark to accurately obtain the true inflow wind speed in front of the blade. Secondly, it fully considers the impact of wind direction changes on the blade inflow wind speed. Based on real-time wind direction, it dynamically identifies upstream associated units of the target wind turbine, constructing multi-dimensional spatial correlation features such as along-wind propagation distance and lateral offset distance between units. This integrates the spatial coupling relationship of multiple wind turbine units and wake distribution characteristics into the prediction model. The model can accurately learn the complex airflow evolution patterns within the wind farm, effectively solving the problem of low wind speed prediction accuracy caused by multi-turbine wake interference, and significantly improving the reliability of wind speed prediction across the entire wind farm. Finally, based on a nacelle transfer function autonomously constructed from lidar measured wind speed data, it accurately establishes a mapping relationship between the nacelle-observed wind speed and the true inflow wind speed in front of the blade. By independently calibrating and solving the transfer function, it eliminates the reliance on manufacturer-secret parameters, resulting in greater adaptability, practicality, and portability. It can also perform personalized wind speed correction for different units.

[0028] This invention organically combines physical mechanisms with data-driven deep learning algorithms. It calculates a self-calibrated transfer function to map the observed wind speed in the nacelle to the actual incoming wind speed ahead of the blades. This data is then input into a CNN-BiGRU-Attention hybrid prediction model to predict the incoming wind speed to the blades. CNN extracts local fluctuation features from historical wind speed and direction sequences, BiGRU learns long-term temporal dependencies between time points, and the Attention mechanism adaptively highlights historical time slices more important for future predictions. Through the synergistic effect of these three mechanisms, the model's ability to characterize short-term wind speed changes, wind direction shifts, and complex wind condition evolution is enhanced. This improves the accuracy and stability of the predicted incoming wind speed and direction for the next 15 minutes, preserving the actual operating mechanisms and physical laws of wind turbines while fully leveraging the advantages of deep learning models in extracting complex nonlinear temporal and spatial features. It balances the model's physical interpretability with prediction accuracy, achieving a two-way empowerment of mechanism-driven and data-driven approaches. Attached Figure Description

[0029] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0030] Figure 1 This is a flowchart of a wind turbine blade inflow prediction method based on multi-source operational data.

[0031] Figure 2 This is a flowchart of the data preprocessing process.

[0032] Figure 3 This is a schematic diagram of the CNN-GRU-Attention prediction model structure constructed in this invention.

[0033] Figure 4 This is a result block diagram of a wind turbine blade inflow prediction system based on multi-source operational data. Detailed Implementation

[0034] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0035] In the description of this invention, it should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0036] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0037] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0038] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0039] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0040] Example 1 like Figure 1 As shown, this invention provides a method for predicting the incoming flow of wind turbine blades based on multi-source operational data. The overall flowchart describes the implementation process of wind turbine blade incoming flow prediction: First, collect wind turbine SCADA operation data, lidar wind measurement data, and turbine power curve data, and perform data processing such as time alignment and anomaly removal; second, obtain the self-tested nacelle transfer function based on lidar measured data and power curve verification; then, use the transfer function, historical wind speed and wind direction change characteristics of the target turbine, and dynamic upstream turbine spatial correlation information as input features, and input them into the CNN-BiGRU-Attention prediction model to predict the incoming wind speed of wind turbine blades for the next 15 minutes. The specific process is as follows: Step 1: Data Collection and Preprocessing Step 1.1: Data Collection The data required for this invention includes SCADA data, cabin lidar data, and power curve data. SCADA data includes timestamps. ts Cabin wind speed Cabin wind direction WD_NAC Cabin position NAC_POS Active power P Operating status ST The nacelle lidar data includes the incoming wind speed at the ranging point in front of the wind turbine. Data quality indicators q_flag timestamp ts .

[0041] Step 1.2: Data Preprocessing like Figure 2 As shown, the present invention processes the collected data as follows.

[0042] 1) Time synchronization The timestamps of SCADA data, LiDAR data, and power data are uniformly converted to datetime format and time-aligned according to a 30-second sampling interval.

[0043] 2) Data cleaning First, missing values ​​are filled in and duplicate and obvious outliers are removed. Historical forward filling is used to fill in the missing values, and the label data still retains the true and valid mark, thereby avoiding future information leakage. Second, the wind speed is limited to a reasonable physical range. Finally, samples that are unsuitable for training, such as those that are out of service, have malfunctions, communication abnormalities, or severe power limitations, are removed.

[0044] 3) Data normalization Scaling the data to [0,1] facilitates subsequent calculations and improves model inference speed. The formula is as follows:

[0045] In the formula, For the first One data point, The minimum value of this set of data. The maximum value of this set of data. For the first The data is normalized.

[0046] The actual wind direction is the superposition of the cabin position and the wind direction. After calculation, the actual wind direction is normalized. Since wind direction is a circular variable, it is converted into sine and cosine components, as shown in the following formula:

[0047]

[0048]

[0049] In the formula, The sine component representing wind direction. The cosine component representing wind direction. D This represents the original wind direction angle.

[0050] 4) Sample Construction Training samples are generated using a sliding window method, with 80% of the data being training data and 20% being test data.

[0051] Step 2: Calculate the self-test transfer function A mapping relationship between the wind speed at the nacelle and the incoming wind speed of the wind turbine blades is constructed to obtain the self-test nacelle transfer function applicable to the target wind farm unit, providing high-confidence input features and supervision targets for the subsequent CNN-BiGRU-Attention prediction model.

[0052] Step 2.1: Data Collection and Processing 1) Data Collection Collect the original wind speed of the nacelle, nacelle position, relative wind direction, active power, wind turbine operating status, and the incoming wind speed in front of the impeller measured by the lidar at the same time stamp when the lidar is installed.

[0053] 2) Data processing The data was screened for quality according to the data processing method proposed in step 1.2, and data samples with low lidar signal quality, abnormal measurement distance, rain and fog interference, wind turbine shutdown, fault, power limitation, abnormal yaw, and timestamp synchronization were removed.

[0054] Step 2.2: Self-testing the construction of the transfer function 1) Set different confidence weights Samples with valid lidar data, normal grid-connected wind turbine operation, no power limitation or reduction, small yaw error, power curves in a monotonically increasing range, and no significant jumps in wind speed and power are designated as high-confidence samples, and their weights can be set as follows: =1.

[0055] Samples that are shut down, malfunctioning, severely limited in power, have abnormal active power, have anemometer malfunctions, have invalid lidar signals, or have out-of-sync timestamps are discarded as invalid samples and will not participate in the training of the self-test nacelle transfer function.

[0056] 2) Transfer function The transfer function is constructed as follows:

[0057]

[0058]

[0059] In the formula, Indicates the incoming air velocity of the fan blades; The vector represents the original measured wind speed in the cabin; D represents the actual wind direction; and S represents the wind speed dispersion within the field. and These represent the 90th and 10th quantiles of the wind speed sample, respectively. This indicates the deviation between the wind speed in a single engine room and the average wind speed in the entire engine room. The specific manifestations are as follows:

[0060] In the formula, This represents the parameters of the transfer function to be solved.

[0061] We use weighted ridge regression to solve for the transfer function parameters and minimize the following objective function:

[0062] In the formula, Indicates the first i Confidence weights for each sample; Indicates the first i The measured inflow wind speed corresponding to each sample by lidar; Indicates the first i Input features of each sample; This represents the regularization coefficient.

[0063] Step 3: CNN-BiGRU-Attention Prediction Model Step 3.1: Constructing Input Features 1) Transfer function The self-test function that maps the nacelle wind speed to the equivalent incoming airflow speed in front of the impeller, as shown in step 2.2, is added to the model input as a physics enhancement module.

[0064] 2) Constructing input features Construct historical wind speed sequences and first-order wind speed variations. The characteristics of historical wind speed variations are represented as follows:

[0065] The historical wind speed sequence is represented as follows:

[0066] In the formula, L=120.

[0067] The change in first-order wind speed is expressed as:

[0068] 3) Shifts in historical trends Historical Wind Direction Change Characteristics Table

[0069]

[0070] 4) Dynamic upstream unit spatial correlation information The upstream turbine is dynamically selected based on the current wind direction of the target turbine.

[0071] First, convert the latitude and longitude coordinates of the wind turbine into planar coordinates. Calculate the distance between the candidate wind turbine and the target wind turbine. relative azimuth and the angle between the candidate wind turbine and the current incoming flow :

[0072]

[0073]

[0074] According to the and Select upstream wind turbines, when and When considering candidate wind turbines as upstream wind turbines, a comprehensive scoring function is set:

[0075] The top 8 wind turbines were selected based on the comprehensive scoring function. The spatial projection characteristics of the selected 8 wind turbines were then analyzed.

[0076]

[0077] in, It reflects the propagation distance from the upstream unit to the target unit along the main direction of the incoming flow. It reflects the degree of lateral offset of the candidate unit relative to the main incoming flow direction.

[0078] Finally, after normalizing the spatial projection, the following input features are obtained for the structure of each wind turbine participating in the modeling:

[0079] The target unit plus the units participating in the modeling, therefore the model input is:

[0080] Step 3.2: Construct the CNN-BiGRU-Attention prediction model The specific structure of the model used in this invention is as follows: Figure 3 As shown, firstly, local variation features at short timescales are extracted using a one-dimensional convolutional layer, and then a two-layer GRU network is used to learn the long-term temporal dependencies in the wind speed and direction sequences. Subsequently, a temporal attention mechanism is introduced to adaptively weight the importance of different historical moments, forming a shared feature vector that can characterize the dynamic changes of the wind field. At the output end, the CNN-BiGRU-Attention prediction model adopts a multi-task branch structure, simultaneously predicting wind speed and direction results at 30-second and 1-minute timescales.

[0081] Step 3.3: Construction of Evaluation Indicators The multi-task loss function is used as the evaluation metric, and its formula is shown below:

[0082] In the formula, This is the wind speed prediction loss, used to constrain the error between the predicted wind speed and the target wind speed; This is the trend loss, used to constrain the predicted wind speed change trend to be consistent with the actual trend. This refers to relative wind speed loss; Loss due to wind direction; For the measured incoming flow monitoring loss using lidar; This represents the loss in power curve consistency.

[0083] Step 3.4: Output of Prediction Results First, the CNN-BiGRU-Attention prediction model from step 3.1 is trained based on the evaluation metrics from step 3.2, and the optimal model parameters and normalizer are saved. Second, the latest data is read in real time, and the predicted wind speed and direction of the blade inflow for the next 15 minutes are output. The prediction results can be further used for yaw optimization.

[0084] Example 2 like Figure 1 As shown, the present invention provides a wind turbine blade inflow prediction method based on multi-source operating data. It uses measured inflow data from the nacelle lidar and historical SCADA data to calibrate the mapping relationship between the nacelle wind speed and the inflow wind speed in front of the rotor. It also combines the historical time-series operating characteristics and spatial projection relationship of the target wind turbine and the dynamically associated upstream wind turbines to construct a wind turbine blade inflow prediction model that integrates physical correction and data-driven approach. This method is based on wind farm SCADA operation data, nacelle lidar wind measurement data, and turbine power curve data. First, it performs time synchronization, outlier removal, missing value processing, operation status filtering, and normalization on the multi-source data. Then, based on the lidar measured inflow wind speed in front of the blades, combined with nacelle wind speed, actual wind direction, yaw status, power status, and the discrete characteristics of wind conditions in the field, a self-tested nacelle transfer function is established. Finally, the inflow characteristics after the self-tested transfer function is corrected, the historical wind speed and direction change characteristics of the target turbine, and the dynamic spatial correlation characteristics of the upstream turbine are input into the CNN-BiGRU-Attention prediction model to obtain the predicted inflow wind speed and direction in front of the wind turbine blades within the next 15 minutes.

[0085] The prediction steps using the above method are as follows: (1) Obtain multi-source operation data.

[0086] Collect SCADA data, nacelle lidar data, and turbine power curve data for each wind turbine within the target wind farm. SCADA data includes timestamps, nacelle wind speed, nacelle wind direction, nacelle location, active power, and operating status; nacelle lidar data includes the incoming wind speed at the distance measurement point in front of the rotor, data quality indicators, and timestamps.

[0087] (2) Data preprocessing and feature construction.

[0088] like Figure 2 As shown, the timestamps of the acquired data are uniformly converted to a standard time format and time-aligned according to a preset sampling interval. Missing values ​​are filled in, outliers are removed, and normalization is performed. After wind direction normalization, it is converted into sine and cosine components.

[0089] (3) Construct input features.

[0090] The transfer function parameters are solved by weighted ridge regression to obtain the self-tested nacelle transfer function applicable to the target wind farm. Several upstream wind turbines with high correlation are selected, and their historical wind speed, wind direction, power and spatial projection characteristics are used as model inputs.

[0091] (4) Construct and train the CNN-BiGRU-Attention prediction model.

[0092] The data is divided into training and testing sets, which are then input into the prediction model for training. This yields the optimal model parameters and a data normalizer. The internal structure of the prediction model is as follows: Figure 3 As shown.

[0093] (5) Output prediction results Real-time data is input into the trained optimal model parameters and data normalizer, and the predicted wind speed and direction in front of the target wind turbine blades for the next 15 minutes are output.

[0094] Example 3 like Figure 4 As shown, the present invention provides a wind turbine blade inflow prediction system based on multi-source operational data, comprising: Multi-source operation data acquisition unit: Collects multi-source operation data within the target wind farm, including SCADA data, nacelle lidar data, and unit power curve data for each wind turbine; Data preprocessing and basic time series feature construction unit: preprocesses the acquired multi-source operation data, including time alignment, missing value handling, outlier removal and normalization, and calculates the true wind direction based on the nacelle location and nacelle wind direction, and converts the true wind direction into sine and cosine components; based on the preprocessed data, constructs the basic time series features of all wind turbines to obtain a time-synchronized effective sample dataset. Constructing input feature units: Select valid and grid-connected data samples from the obtained effective sample dataset of the nacelle lidar data. Set confidence weights according to the data quality and operating status of the samples. Use nacelle wind speed, true wind direction, field wind speed dispersion, and deviation between the wind speed of a single nacelle and the average nacelle wind speed of the entire field as input features. Use the measured incoming wind speed of the nacelle lidar as the target label. Use weighted ridge regression to solve the transfer function parameters to obtain the self-tested nacelle transfer function. Use the self-tested nacelle transfer function to obtain the corrected incoming wind features corresponding to the nacelle wind speed. Based on the obtained real wind direction and the coordinates of the wind turbines in the wind farm, the downwind offset distance, lateral offset distance and correlation score between the candidate unit and the target unit are calculated, and the associated unit is selected according to the results; The final input features are: historical wind speed of the target wind turbine and associated units, first-order wind speed change of the historical wind speed sequence, sine and cosine components of the actual wind direction, and normalized downwind and lateral offset distances of the target and candidate units; corresponding output labels are constructed: the incoming wind speed is calculated based on the calibrated nacelle transfer function to obtain the model training dataset; Construct and train CNN-BiGRU-Attention prediction model units: Divide the obtained model training dataset into training set and test set according to time order, construct a CNN-BiGRU-Attention prediction model including one-dimensional convolutional layer, bidirectional gated recurrent unit layer, temporal attention layer and multi-task output layer, train the prediction model with multi-task loss function to obtain the optimal model parameters and data normalizer. The output prediction result unit takes the real-time nacelle wind speed and actual wind direction of the wind turbine units at the wind farm as real-time input features, and uses the obtained data normalizer to standardize the real-time input features before inputting them into the trained CNN-BiGRU-Attention prediction model to output the predicted wind speed in front of the target wind turbine blades within the next 15 minutes.

[0095] In the multi-source operation data acquisition unit of this embodiment, SCADA data includes timestamp, nacelle wind speed, nacelle wind direction, nacelle position, active power, and operating status; data from a single nacelle lidar includes the incoming wind speed at the ranging point in front of the wind turbine, data quality flag, and timestamp.

[0096] In the data preprocessing and basic time series feature construction unit of this embodiment, the effective sample dataset includes timestamps, cabin wind speed, true wind direction, sine component of true wind direction, cosine component of true wind direction, incoming wind speed measured by lidar, data quality flags of incoming wind speed, active power of wind turbine, first-order wind speed variation, moving average and moving standard deviation of wind speed.

[0097] Example 4 The present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of a wind turbine blade inflow prediction method based on multi-source operating data.

[0098] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0099] This application is described with reference to flowchart illustrations and / or block diagrams of methods, systems, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0100] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0101] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0102] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0103] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for predicting the incoming flow of wind turbine blades based on multi-source operational data, characterized in that, include: (1) Obtain multi-source operation data Collect multi-source operational data within the target wind farm, including SCADA data, nacelle lidar data, and unit power curve data for each wind turbine. (2) Data preprocessing and construction of basic time series features The acquired multi-source operational data is preprocessed, including time alignment, missing value handling, outlier removal and normalization. The actual wind direction is calculated based on the cabin location and cabin wind direction, and the actual wind direction is converted into sine and cosine components. Based on the preprocessed data, the basic time-series characteristics of all wind turbines are constructed to obtain a time-synchronized effective sample dataset. (3) Constructing input features From the obtained valid sample dataset, select valid data samples of the nacelle lidar data that are in grid-connected status. Set confidence weights based on the data quality and operating status of the samples. Use nacelle wind speed, actual wind direction, wind speed dispersion in the field, and the deviation between the wind speed of a single nacelle and the average nacelle wind speed in the entire field as input features. Use the measured incoming wind speed of the nacelle lidar as the target label. Use weighted ridge regression to solve the transfer function parameters to obtain the self-tested nacelle transfer function. Then use the self-tested nacelle transfer function to obtain the corrected incoming wind features corresponding to the nacelle wind speed. Based on the obtained real wind direction and the coordinates of the wind turbines in the wind farm, the downwind offset distance, lateral offset distance and correlation score between the candidate unit and the target unit are calculated, and the associated unit is selected according to the results; The final input features are: historical wind speed of the target wind turbine and associated units, first-order wind speed change of the historical wind speed sequence, sine and cosine components of the actual wind direction, and normalized downwind and lateral offset distances of the target and candidate units; corresponding output labels are constructed: the incoming wind speed is calculated based on the calibrated nacelle transfer function to obtain the model training dataset; (4) Construct and train the CNN-BiGRU-Attention prediction model The obtained model training dataset is divided into training set and test set according to the time order. A CNN-BiGRU-Attention prediction model including one-dimensional convolutional layer, bidirectional gated recurrent unit layer, temporal attention layer and multi-task output layer is constructed. The prediction model is trained with multi-task loss function to obtain the optimal model parameters and data normalizer. (5) Output prediction results The real-time nacelle wind speed and actual wind direction of the wind turbine units at the wind farm site are used as real-time input features. The real-time input features are standardized using the obtained data normalizer and then input into the trained CNN-BiGRU-Attention prediction model to output the predicted wind speed in front of the target wind turbine blades within the next 15 minutes.

2. The wind turbine blade inflow prediction method based on multi-source operational data according to claim 1, characterized in that, In step (1), the SCADA data includes timestamp, nacelle wind speed, nacelle wind direction, nacelle position, active power and operating status; the data of a nacelle lidar includes the incoming wind speed at the distance measuring point in front of the wind turbine, data quality flag and timestamp.

3. The wind turbine blade inflow prediction method based on multi-source operational data according to claim 1, characterized in that, In step (2), the effective sample dataset includes timestamps, cabin wind speed, true wind direction, sine component of true wind direction, cosine component of true wind direction, incoming wind speed measured by lidar, data quality indicators of incoming wind speed, active power of wind turbine, first-order wind speed variation, moving average and moving standard deviation of wind speed.

4. The wind turbine blade inflow prediction method based on multi-source operational data according to claim 1, characterized in that, In step (3), the self-test cabin transfer function is constructed as follows: In the formula, Indicates the incoming air velocity of the fan blades; D represents the original measured wind speed in the cabin; D represents the actual wind direction. S represents the wind speed dispersion within the field. and These represent the 90th and 10th quantiles of the wind speed sample, respectively. This indicates the deviation between the wind speed in a single engine room and the average wind speed in the entire engine room. The specific manifestations are as follows: In the formula, The parameters represent the transfer function to be solved. We use weighted ridge regression to solve for the transfer function parameters and minimize the following objective function: In the formula, Indicates the first i Confidence weights for each sample; Indicates the first i The measured inflow wind speed corresponding to each sample by lidar; Indicates the first i Input features of each sample; This represents the regularization coefficient.

5. The wind turbine blade inflow prediction method based on multi-source operational data according to claim 1, characterized in that, In step (3), the downwind offset distance and the lateral offset distance are: In the formula, It indicates the downwind offset distance, reflecting the propagation distance from the upstream unit to the target unit along the main direction of the incoming flow; This indicates the lateral offset distance, reflecting the degree of lateral offset of the candidate unit relative to the main incoming flow direction; This indicates the distance between the candidate wind turbine and the target wind turbine. ; This indicates the angle between the candidate wind turbine and the current incoming flow. , , This indicates the relative azimuth angle between the candidate wind turbine and the target wind turbine. Represents the plane coordinates of the wind turbine; The aforementioned correlation score is represented as follows: according to and Select upstream wind turbines, when and When considering candidate wind turbines as upstream wind turbines, a comprehensive scoring function is set. The historical wind speed sequence and the first-order wind speed change are expressed as follows: The historical wind speed sequence is represented as follows: In the formula, L = 120; The change in first-order wind speed is expressed as: Similarly, by constructing a historical wind direction sequence, the wind speed variable can be replaced with the sine and cosine components of the wind direction.

6. The wind turbine blade inflow prediction method based on multi-source operational data according to claim 1, characterized in that, In step (4), the constructed CNN-BiGRU-Attention prediction model is as follows: First, local change features at a short time scale are extracted through a one-dimensional convolutional layer, and then a two-layer GRU network is used to learn the long-term temporal dependencies in the wind speed and wind direction sequences. Subsequently, a time attention mechanism is introduced to adaptively weight the importance of different historical moments, forming a shared feature vector that can characterize the dynamic changes of the wind field; On the output side, the CNN-BiGRU-Attention prediction model adopts a multi-task branch structure to simultaneously predict wind speed and direction at 30-second and 1-minute scales.

7. A wind turbine blade inflow prediction system based on multi-source operational data, characterized in that, include: Multi-source operation data acquisition unit: Collects multi-source operation data within the target wind farm, including SCADA data, nacelle lidar data, and unit power curve data for each wind turbine; Data preprocessing and basic time series feature construction unit: preprocesses the acquired multi-source operational data, including time alignment, missing value handling, outlier removal and normalization, and calculates the true wind direction based on the cabin location and cabin wind direction, and converts the true wind direction into sine and cosine components. Based on the preprocessed data, the basic time-series characteristics of all wind turbines are constructed to obtain a time-synchronized effective sample dataset. Constructing input feature units: Select valid and grid-connected data samples from the obtained effective sample dataset of the nacelle lidar data. Set confidence weights according to the data quality and operating status of the samples. Use nacelle wind speed, true wind direction, field wind speed dispersion, and deviation between the wind speed of a single nacelle and the average nacelle wind speed of the entire field as input features. Use the measured incoming wind speed of the nacelle lidar as the target label. Use weighted ridge regression to solve the transfer function parameters to obtain the self-tested nacelle transfer function. Use the self-tested nacelle transfer function to obtain the corrected incoming wind features corresponding to the nacelle wind speed. Based on the obtained real wind direction and the coordinates of the wind turbines in the wind farm, the downwind offset distance, lateral offset distance and correlation score between the candidate unit and the target unit are calculated, and the associated unit is selected according to the results; The final input features are: historical wind speed of the target wind turbine and associated units, first-order wind speed change of the historical wind speed sequence, sine and cosine components of the actual wind direction, and normalized downwind and lateral offset distances of the target and candidate units; corresponding output labels are constructed: the incoming wind speed is calculated based on the calibrated nacelle transfer function to obtain the model training dataset; Construct and train CNN-BiGRU-Attention prediction model units: Divide the obtained model training dataset into training set and test set according to time order, construct a CNN-BiGRU-Attention prediction model including one-dimensional convolutional layer, bidirectional gated recurrent unit layer, temporal attention layer and multi-task output layer, train the prediction model with multi-task loss function to obtain the optimal model parameters and data normalizer. The output prediction result unit takes the real-time nacelle wind speed and actual wind direction of the wind turbine units at the wind farm as real-time input features, and uses the obtained data normalizer to standardize the real-time input features before inputting them into the trained CNN-BiGRU-Attention prediction model to output the predicted wind speed in front of the target wind turbine blades within the next 15 minutes.

8. The wind turbine blade inflow prediction system based on multi-source operational data according to claim 7, characterized in that, In the acquisition of multi-source operational data units, SCADA data includes timestamps, nacelle wind speed, nacelle wind direction, nacelle position, active power, and operating status; data from a single nacelle lidar includes the incoming wind speed at the ranging point in front of the wind turbine, data quality indicators, and timestamps.

9. A wind turbine blade inflow prediction system based on multi-source operational data according to claim 7, characterized in that, In the data preprocessing and basic time series feature construction unit, the effective sample dataset includes timestamps, cabin wind speed, true wind direction, sine component of true wind direction, cosine component of true wind direction, incoming wind speed measured by lidar, data quality indicators of incoming wind speed, active power of wind turbine, first-order wind speed variation, moving average and moving standard deviation of wind speed.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the wind turbine blade inflow prediction method based on multi-source operating data according to any one of claims 1-6.

Images