Wind turbine blade load prediction method and related apparatus

Abstract

The present application discloses a wind turbine blade load prediction method and an apparatus. In the present application, it is not necessary to mount real sensors at all positions for measurement required by a wind turbine blade. The method comprises: acquiring a position for prediction and a mounting position; determining the correlation between the position for prediction and the mounting position; if the correlation is greater than or equal to a correlation threshold, acquiring a regression model, and on the basis of blade load data of the mounting position, performing fitting by means of the regression model to obtain blade load data of the position for prediction; and if the correlation is less than the correlation threshold, acquiring a prediction model, and on the basis of environmental data of the mounting position, performing prediction by means of the prediction model to obtain blade load data of the position for prediction. In different cases of correlation, blade load data of a position for prediction when the correlation is strong is obtained by means of a regression model, and blade load data of a position for prediction when the correlation is weak is obtained by means of a prediction model, thereby improving the accuracy of wind turbine blade load prediction.

Description

A method and related device for predicting wind turbine blade loads Technical Field

[0001] This invention relates to the field of wind power generation technology, and in particular to a method and related apparatus for predicting wind turbine blade loads. Background Technology

[0002] With the continuous development of new energy technologies, the demand for wind power generation is increasing. As the core component for capturing wind energy, excessive load on wind turbine blades can affect their service life, thereby impacting the power conversion efficiency of the wind turbine and the economic benefits of the power plant. Therefore, it is crucial to monitor the load on wind turbine blades.

[0003] In related technologies, real sensors are installed at various locations on the wind turbine blades to detect the load at each location to be tested.

[0004] However, due to issues such as cost and the shape of the wind turbine blades, not all locations to be tested can be equipped with real sensors, resulting in low accuracy in detecting the blade load at each location on the wind turbine blades. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a method and related apparatus for predicting wind turbine blade loads, which solves the problem of low accuracy in blade load data detection.

[0006] Based on this, the following technical solution is disclosed in this application:

[0007] On the one hand, this application provides a method for predicting wind turbine blade loads, the method comprising:

[0008] Obtain the location to be predicted and the location already installed. The location to be predicted is the location on the wind turbine blade where no real sensor is installed, and the location already installed is the location on the wind turbine blade where the real sensor is installed.

[0009] Determine the correlation between the location to be predicted and the already installed location;

[0010] If the correlation is greater than or equal to the correlation threshold, a regression model is obtained. Based on the blade load data of the installed location, the regression model is fitted to obtain the blade load data of the location to be predicted. The regression model is obtained by fitting the simulated blade load data of the location to be predicted and the historical blade load data of the installed location.

[0011] On the other hand, this application provides a wind turbine blade load prediction device, the device comprising: an acquisition unit, a determination unit, a regression unit, and a prediction unit;

[0012] The acquisition unit is used to acquire the position to be predicted and the installed position. The position to be predicted is the position on the wind turbine blade where no real sensor is installed, and the installed position is the position on the wind turbine blade where the real sensor is installed.

[0013] The determining unit is used to determine the correlation between the location to be predicted and the installed location;

[0014] The regression unit is used to obtain a regression model if the correlation is greater than or equal to a correlation threshold. Based on the blade load data at the installed locations, the regression model is fitted to obtain the blade load data at the location to be predicted. The regression model is obtained by fitting the simulated blade load data at the location to be predicted and the historical blade load data at the installed locations.

[0015] On the other hand, embodiments of this application provide a computer device, the computer device including a processor and a memory:

[0016] The memory is used to store computer programs and to transfer the computer programs to the processor;

[0017] The processor is configured to execute the methods described above according to instructions in the computer program.

[0018] On the other hand, embodiments of this application provide a computer-readable storage medium for storing a computer program for performing the methods described above.

[0019] On the other hand, embodiments of this application provide a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods described above.

[0020] As can be seen from the above technical solutions, this application has at least the following beneficial effects:

[0021] This application eliminates the need to install actual sensors at all required detection locations on the wind turbine blades. Instead, it installs actual sensors at the already installed locations on the blades, but not at the locations to be predicted. By analyzing the correlation between the locations to be predicted and the already installed locations, the blade load data at the locations to be predicted can be predicted. This will be explained in detail below.

[0022] The process involves obtaining the location to be predicted and the installed location, and determining the correlation between them. This correlation describes the degree of association between the blade load data at the location to be predicted and the installed location. To improve prediction accuracy, different correlation levels correspond to different models. Specifically, if the correlation is greater than or equal to a correlation threshold, it indicates a strong correlation and a close relationship between the location to be predicted and the installed location. In this case, a corresponding regression model is obtained. The regression model can describe the correlation between the blade load data at the location to be predicted and the installed location through mathematical expressions. Therefore, based on the blade load data at the installed location, the regression model is fitted to obtain the blade load data at the location to be predicted, which is not only highly accurate but also convenient and fast.

[0023] If the correlation is less than the correlation threshold, it indicates a weak correlation between the location to be predicted and the installed location, meaning the relationship is not close enough. In this case, a corresponding prediction model is obtained. This model can utilize multi-layer networks or other complex model structures to capture deep feature relationships between blade load data at different locations and environmental data at the installed location, relationships that are difficult to describe using regression models. Based on the environmental data at the installed location, the prediction model can then predict the blade load data for the location to be predicted. This not only improves the accuracy of predictions under weak correlation but also enhances the adaptability and robustness of the prediction model to changes in different environmental conditions. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 is a schematic diagram of a cross-section of a wind turbine blade provided in this application;

[0026] Figure 2 is a flowchart of a wind turbine blade load prediction method provided in this application;

[0027] Figure 3 is a structural diagram of a wind turbine blade load prediction device provided in this application; and

[0028] Figure 4 is a structural diagram of a computer device provided in this application. Detailed Implementation

[0029] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this application. It should be understood that the drawings and embodiments of this application are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0030] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0031] The wind turbine blade load prediction method provided in this application can be applied to computer equipment with wind turbine blade load prediction capabilities, such as terminal devices and servers. Specifically, terminal devices can be desktop computers, laptops, mobile phones, and tablets; servers can be independent physical servers, server clusters composed of multiple physical servers, or distributed systems. Terminal devices and servers can be directly or indirectly connected via wired or wireless communication, and this application does not impose any restrictions on this connection.

[0032] Blade load data refers to the specific values ​​of various external forces and their effects generated by wind turbine blades due to wind pressure, gravity, centrifugal force and other factors during wind turbine operation. It includes multiple time series values. Blade load data reflects the stress situation of wind turbine blades under different operating conditions and is an important basis for assessing the fatigue life of wind turbine blades, optimizing design and maintenance.

[0033] For example, common blade load data are categorized into aerodynamic load data, centrifugal load data, and gravity load data based on the source of their forces. Aerodynamic load data includes wind pressure and the resulting wind pressure moment; centrifugal load data includes centrifugal force and the resulting bending moment; and gravity load data includes gravity and the resulting gravitational moment.

[0034] In related technologies, due to limitations such as cost and installation difficulty, only a limited number of real sensors can typically be installed at certain locations on the wind turbine blades to be detected. Referring to Figure 1, which is a schematic diagram of a wind turbine blade cross-section provided in this application, 101, 102, and 103 represent the three installed locations of real sensors on the wind turbine blade. The real sensors can acquire blade load data at these installed locations. However, researchers or institutions often want to obtain blade load data at locations other than the installed locations, as shown in Figure 1, where the dashed circles represent possible locations to be predicted. Based on this, the blade load data at the locations to be predicted can be obtained through model prediction using the blade load data at the installed locations.

[0035] In related technologies, blade load data can be predicted by the correlation of various positions of wind turbine blades. However, this method is not very accurate. Specifically, existing prediction methods usually predict the blade load data of the position to be predicted based on a single model when the correlation is strong. Wind turbine blades are of various types, such as short blades and long flexible blades according to their length. Different types of wind turbine blades have varying degrees of correlation. Therefore, the method of using only one model for prediction cannot be applied to all types of wind turbine blades, resulting in a small application range and low prediction accuracy.

[0036] This application predicts blade load data at the location to be predicted by analyzing the correlation between the location to be predicted and the already installed location. This will be explained in detail below.

[0037] Obtain the location to be predicted and the installed location, and determine the correlation between the location to be predicted and the installed location. This correlation is used to describe the degree of association between the load data of the location to be predicted and the load data of the installed location. In order to improve the accuracy of the prediction, different correlation degrees correspond to different models.

[0038] Specifically, if the correlation is greater than or equal to the correlation threshold, it indicates a strong correlation between the location to be predicted and the installed location, suggesting a close relationship. In this case, a corresponding regression model is obtained. The regression model can describe the relationship between the blade load data at the location to be predicted and the blade load data at the installed location through mathematical expressions. Therefore, based on the blade load data at the installed location, the blade load data at the location to be predicted is obtained by fitting the regression model. This method is not only highly accurate but also convenient and fast.

[0039] If the correlation is less than the correlation threshold, it indicates a weak correlation between the location to be predicted and the installed location, meaning the relationship is not close enough. In this case, a corresponding prediction model is obtained. This model can utilize multi-layer networks or other complex model structures to capture deep feature relationships between blade load data at different locations and environmental data at the installed location, relationships that are difficult to describe using regression models. Based on the environmental data at the installed location, the prediction model can then predict the blade load data for the location to be predicted. This not only improves the accuracy of predictions under weak correlation but also enhances the adaptability and robustness of the prediction model to changes in different environmental conditions.

[0040] The following description, with reference to Figure 2, introduces a wind turbine blade load prediction method provided in this application. Referring to Figure 2, this figure is a flowchart of the wind turbine blade load prediction method provided in this application, which may include steps S101-S104.

[0041] S101: Obtain the location to be predicted and the installed location.

[0042] In this context, the locations to be predicted are the positions on the wind turbine blades where no actual sensors are installed, and the locations where actual sensors are installed are the locations on the wind turbine blades where such sensors are installed. For example, continuing as shown in Figure 1, in one possible installation method, 101, 102, and 103 are three already installed locations where actual sensors are installed, and the dashed circles next to 102 and 103 are the locations to be predicted where no actual sensors are installed.

[0043] As one possible implementation, the location to be predicted can be a position on the wind turbine blade where a real sensor is desired to be installed but has not yet been installed. For example, there may be three possible locations on the wind turbine blade where a real sensor could be installed, but currently only two real sensors are installed. Therefore, according to the embodiments of this application, each location can be used as a location to be predicted, and the method provided in the embodiments of this application can be used to select two locations from the three locations to install the real sensor. Examples D1-D4 will be used for illustration later, and will not be elaborated further here.

[0044] Based on factors such as cost and installation difficulty, actual sensors are installed on the wind turbine blades, and the cross-sectional positions of these sensors are recorded as the installed locations. For example, the blade root is the most typical installed location. In addition, actual sensors may also be installed near the section of the blade that bears the maximum load and at one-third of the blade tip, which can also be considered as installed locations.

[0045] S102: Determine the correlation between the location to be predicted and the already installed location.

[0046] Correlation is used to describe the degree of association between blade load data at the location to be predicted and blade load data at the installed location. To ensure the accuracy and comprehensiveness of the correlation, it can be determined under different environmental conditions. For example, the correlation between the location to be predicted and the installed location can be determined under different wind speeds such as 4, 6, 8, 10, ..., 24 m / s.

[0047] In determining the correlation, statistical analysis methods can be used to calculate the known blade load data at the location to be predicted and the known blade load data at the installed location, thereby obtaining the correlation index corresponding to the statistical analysis method and determining the correlation between the location to be predicted and the installed location. The known blade load data can be obtained through real sensors, simulated blade load data, or a combination of real sensors and simulated blade load data. Simulated blade load data is obtained by simulation software through simulating operating conditions.

[0048] The following examples, using two different methods, illustrate the process of determining the correlation between the location to be predicted and the already installed location:

[0049] Method 1: Pearson Coefficient.

[0050] The Pearson coefficient is a statistical indicator that measures the linear correlation between known blade load data at the location to be predicted and blade load data at the installed location. The correlation is determined based on the assumption that the known blade load data conforms to a normal distribution.

[0051] In method one, the determination of relevance must include at least one of the following three cases:

[0052] Scenario 1: Determine the Pearson coefficient for each location to be predicted and an already installed location.

[0053] Scenario 2: Determine the Pearson coefficient for each location to be predicted and for multiple installed locations.

[0054] Scenario 3: Determine the Pearson coefficient for some or all of the locations to be predicted and for one of the already installed locations.

[0055] If there are multiple locations to be predicted, the Pearson coefficient is calculated as follows:

[0056] Where, ρ ppearson Here, X represents the Pearson coefficient, X represents the known blade load data at the installed location, and Y represents the load factor. i The known blade load data for the i-th position to be predicted, where i is a positive integer.

[0057] The Pearson coefficient ranges from [-1, 1]. When the Pearson coefficient is 1, it indicates that there is a perfect linear positive correlation between the location to be predicted and the location that has been installed. The closer the Pearson coefficient is to 1, the stronger the correlation between the location to be predicted and the location that has been installed.

[0058] Method 2: Spearman Rank Correlation Coefficient.

[0059] The Spearman coefficient is a non-parametric correlation measure. Even if the known blade load data does not follow a normal distribution, the Spearman coefficient can still be used to assess the dependence between the known blade load data at the location to be predicted and the known blade load data at the installed location, thereby determining the correlation between the location to be predicted and the installed location.

[0060] In method two, the determination of relevance must include at least one of the following three cases:

[0061] Scenario 1: Determine the Spearman coefficient for each location to be predicted and an already installed location.

[0062] Scenario 2: Determine the Spearman coefficients for each location to be predicted and multiple installed locations.

[0063] Scenario 3: Determine the Spearman coefficients of some or all of the locations to be predicted and one of the already installed locations.

[0064] The formula for calculating the Spearman coefficient is as follows:

[0065] Where, ρ sppearman Here, d represents the Spearman coefficient and d represents the grade difference. The known blade load data for the location to be predicted and the known blade load data for the already installed locations are sorted in ascending order. Grades are assigned based on the sorting. For the two sets of data, the grade values ​​at the same location (i.e., the same grade number) are compared, and the difference between the grade values ​​is calculated; this difference is the grade difference d. j Let represent the j-th grade difference, where j is a positive integer, and n is the number of data points with known blade load data, where n is a positive integer. For example, time series values ​​at the blade root and at a distance of 10 meters from the blade root can be paired, and the Spearman coefficients corresponding to the two values ​​can be obtained by calculating the grade difference, thereby determining the correlation between the blade root and the distance 10 meters from the blade root.

[0066] The Spearman coefficient ranges from [-1, 1]. When the Spearman coefficient is 1, it indicates that there is a perfect positive correlation between the location to be predicted and the location that has been installed. The closer the Spearman coefficient is to 1, the stronger the correlation between the location to be predicted and the location that has been installed.

[0067] By determining the correlation between the location to be predicted and the already installed location, subsequent steps can select the corresponding model to predict the blade load data based on this correlation. The following sections will explain two models respectively.

[0068] S103: If the correlation is greater than or equal to the correlation threshold, then obtain the regression model. Based on the blade load data of the installed location, fit the regression model to obtain the blade load data of the location to be predicted.

[0069] The regression model is a model that fits the correlation between blade load data. It is used to characterize the linear or nonlinear relationship between the blade load data at the installed location and the blade load data at the location to be predicted. For example, regression models include linear regression models and nonlinear regression models. The regression model is obtained by fitting simulated blade load data at the location to be predicted and historical blade load data at the installed location. The fitting process of the regression model will be illustrated with examples E1-E4 later, and will not be repeated here. Based on the number and specific locations of the locations to be predicted, as well as the number and specific locations of the installed locations, the corresponding location distribution is determined. Different location distributions correspond to different regression models.

[0070] The historical blade load data at the installed location is the blade load data collected by the actual sensor at the installed location over a period of time.

[0071] The correlation threshold is used to determine the strength of the correlation. When the correlation is greater than or equal to the correlation threshold, it indicates a strong correlation between the location to be predicted and the installed location, meaning the blade load data at the location to be predicted has a close relationship with the blade load data at the installed location. For example, when the correlation is determined using the Pearson coefficient and the correlation threshold is 0.6, a correlation of 0.7 indicates a strong correlation between the location to be predicted and the installed location.

[0072] In the regression model, it is assumed that there is a strong correlation between the location to be predicted and the location that has been installed. By fitting the correspondence between the blade load data of the location that has been installed and the blade load data of the location to be predicted, the blade load data of the location to be predicted can be predicted based on the correspondence in the regression model and the blade load data of the location that has been installed.

[0073] The following explanation uses a linear regression model as an example to illustrate the regression model, which can be represented by the following formula: y = θ0 + θ1x1 + θ1x1 + ... + θ p x p

[0074] Where y represents the blade load data at the location to be predicted, x1, x2, ..., xp For the blade load data at each installed location, θ0, θ1, ..., θ p The parameters are obtained by fitting the linear regression model, where θ0 is the offset, θ1, θ2, ..., θ p The weighting coefficients for the blade load data at each installed location to the blade load data y at the location to be predicted are: for example, θ1 represents the weighting coefficient of x1 to y, and p is a positive integer representing the sequence number of the installed location.

[0075] After obtaining a linear regression model based on simulated blade load data at the location to be predicted and historical blade load data at existing installation locations, this model describes the linear correlation between the simulated blade load data at the location to be predicted and the historical blade load data at existing installation locations. This linear correlation accurately reflects how the blade load data at each existing installation location corresponds to the blade load data at the location to be predicted, under conditions of strong correlation. Therefore, by inputting the blade load data from existing installation locations into the linear regression model, and based on the correlation described by this linear regression model, the blade load data at the location to be predicted can be predicted.

[0076] A strong correlation indicates a strong linear or non-linear relationship between the blade load data at the location to be predicted and the blade load data at the installed locations. In this case, the blade load data at each location is highly consistent, and the blade load data at the installed locations can well represent the blade load data at the location to be predicted. The parameters obtained by the regression model are more accurate, making the correlation obtained by the regression model during training more reliable. The correlation obtained by the regression model can be described by mathematical formulas, and correspondingly, the training and prediction processes of the regression model are simpler and more direct. Therefore, in the case of strong correlation, using a regression model to predict the blade load data at the location to be predicted is not only highly accurate but also convenient and fast.

[0077] S104: If the correlation is less than the correlation threshold, then obtain the prediction model, and make a prediction based on the environmental data of the installed location to obtain the blade load data of the location to be predicted.

[0078] The prediction model is a complex machine learning model used to characterize the intricate relationship between environmental data at the installed location and blade load data at the location to be predicted. It is trained based on historical environmental data from the installed location and simulated blade load data from the location to be predicted. The prediction model can capture deep-seated features in the data through multi-layer nonlinear transformations, thereby improving the accuracy and robustness of the prediction.

[0079] Environmental data refers to the various external and internal factors that affect blade load data during wind turbine operation. For example, environmental data can include environmental variables such as wind speed, wind direction, and turbulence intensity, as well as one or more combinations of unit operating variables such as impeller speed and pitch angle. Historical environmental data is environmental data collected at a specific location on the wind turbine blade over a period of time. For example, historical environmental data could include wind speed, wind direction, turbulence intensity, impeller speed, and pitch angle at the blade root over the past hour.

[0080] When the correlation is less than the correlation threshold, it indicates that the correlation between the location to be predicted and the installed location is weak, the consistency of the blade load data at each location is low, the accuracy of the parameters obtained by the regression model is low, and the reliability of the correlation obtained by the regression model during training is low. Therefore, it is difficult to accurately predict the blade load data at the location to be predicted through the regression model. In the case of weak correlation, the blade load data at the location to be predicted can be predicted by the prediction model.

[0081] For example, when the correlation is determined by the Pearson coefficient and the correlation threshold is 0.6, if the correlation is 0.3, which is less than the correlation threshold, it means that the correlation between the location to be predicted and the installed location is weak, and prediction can be made using the model to be predicted.

[0082] Predictive models possess high complexity and powerful learning capabilities. Through multi-layered networks or other complex model structures, they can capture intricate relationships in blade load data that are difficult to accurately fit using regression models, thus compensating for the low prediction accuracy of regression models when correlations are weak. During the training process, feature extraction and fusion are typically performed. The fusion of multi-source data (historical environmental data, simulated blade load data) improves the adaptability of the predictive model, helping it to more comprehensively understand the distribution patterns of blade load data at different locations under different environments. This also reduces noise interference in the data, thereby improving prediction accuracy.

[0083] Predictive models can also be used to predict blade load data at the desired location when the correlation is greater than or equal to the correlation threshold. While regression models can provide sufficient accuracy when the correlation is greater than or equal to the threshold, considering the complexity of training predictive models, it is more convenient and faster to use predictive models to predict blade load data at the desired location when the correlation is less than the threshold and the accuracy of regression models is low.

[0084] According to the wind turbine blade load prediction method provided in the above embodiments, this application no longer installs real sensors at all required detection locations on the wind turbine blade. Instead, it installs real sensors at the already installed locations on the wind turbine blade, but does not install real sensors at the locations to be predicted. By analyzing the correlation between the locations to be predicted and the already installed locations, the blade load data at the locations to be predicted can be predicted. This will be explained in detail below.

[0085] Obtain the location to be predicted and the installed location, and determine the correlation between the location to be predicted and the installed location. This correlation is used to describe the degree of association between the load data of the location to be predicted and the load data of the installed location. In order to improve the accuracy of the prediction, different correlation degrees correspond to different models.

[0086] Specifically, if the correlation is greater than or equal to the correlation threshold, it indicates a strong correlation between the location to be predicted and the installed location, suggesting a close relationship. In this case, a corresponding regression model is obtained. The regression model can describe the relationship between the blade load data at the location to be predicted and the blade load data at the installed location through mathematical expressions. Therefore, based on the blade load data at the installed location, the blade load data at the location to be predicted is obtained by fitting the regression model. This method is not only highly accurate but also convenient and fast.

[0087] If the correlation is less than the correlation threshold, it indicates a weak correlation between the location to be predicted and the installed location, meaning the relationship is not close enough. In this case, a corresponding prediction model is obtained. This model can utilize multi-layer networks or other complex model structures to capture deep feature relationships between blade load data at different locations and environmental data at the installed location, relationships that are difficult to describe using regression models. Based on the environmental data at the installed location, the prediction model can then predict the blade load data for the location to be predicted. This not only improves the accuracy of predictions under weak correlation but also enhances the adaptability and robustness of the prediction model to changes in different environmental conditions.

[0088] Wind turbine blades are prone to significant deformation and vibration under wind conditions, such as long, flexible blades. This deformation and vibration may result in a weak correlation between the predicted location and the installed location, leading to inaccurate correlations in the regression model. This application provides a method for predicting wind turbine blade loads, see A1-A3:

[0089] A1: Determine the correlation between each location to be predicted and the already installed locations.

[0090] The locations to be predicted can include multiple locations, and each location to be predicted and the installed location must each fall under at least one of the following three conditions:

[0091] Scenario 1: Determine the correlation between each location to be predicted and an already installed location.

[0092] Scenario 2: Determine the correlation between each location to be predicted and multiple installed locations.

[0093] Scenario 3: Determine the correlation between some or all of the locations to be predicted and an already installed location.

[0094] A2: If all correlations are greater than or equal to the correlation threshold, then obtain the regression model.

[0095] If all correlations are greater than or equal to the correlation threshold, it indicates that the correlations determined at each predicted location of the wind turbine blade are strong, the wind turbine blade as a whole has strong correlations, and there is no weakening of the correlations at some predicted locations due to deformation and vibration. The blade load data at each location are highly consistent, and the blade load data at the installed locations can well represent the blade load data at the predicted locations. The parameters obtained by the regression model fitting are relatively accurate in this case, the regression model has high prediction accuracy, and training is convenient and quick. Therefore, when all correlations are greater than or equal to the correlation threshold, the regression model is obtained.

[0096] A3: If at least one of the multiple correlations is less than the correlation threshold, then the prediction model is obtained.

[0097] If one of the multiple correlations is less than the correlation threshold, it indicates that the correlation between the blades is weak in a local area. This means that the correlation between the location to be predicted and the installed location is weak, and the consistency of the blade load data at the location to be predicted with other locations is weak. If a regression model is used for prediction, the weak local correlation will make the correlation obtained by the regression model inaccurate, and the prediction at the location to be predicted where the correlation is weak will also be inaccurate.

[0098] In this case, a prediction model is obtained, and the blade load data at each position to be predicted on the wind turbine blade is predicted by the prediction model. The prediction model can capture the deep features and complex relationships in the blade load data through multi-layer networks or other complex model structures. It can overcome the problem that the regression model cannot accurately fit the correlation relationship, so that the prediction of blade load data at the position to be predicted corresponding to the weak correlation is more accurate.

[0099] For example, if a location to be predicted is located on the tip third of a wind turbine blade, and the correlation between this location and the already installed locations is less than a correlation threshold, it indicates that at least one of the correlations is below the threshold, suggesting a weak correlation. In this case, the association obtained through regression model fitting is inaccurate, making it impossible to accurately predict blade load data for all locations to be predicted. In such situations, a prediction model is needed. This prediction model avoids the problem of inaccurate correlation fitting by the regression model, ensuring the accuracy of the prediction.

[0100] For wind turbine blades with multiple locations to be predicted, multiple correlations are determined based on these locations. A regression model is obtained when all correlations are strong, and a prediction model is obtained when at least one correlation is weak. This ensures that relatively accurate blade load data can be predicted for each location regardless of whether all locations satisfy a strong correlation.

[0101] Regression models include linear regression models and nonlinear regression models. Generally, linear regression models are accurate in prediction; however, for various types of wind turbine blades, linear regression models often fail to predict the load accurately for specific types of blades. To address the problem that linear regression models cannot accurately predict the load for multiple types of wind turbine blades, this application also provides a method for predicting wind turbine blade loads, see B1-B3:

[0102] B1: Obtain the length of the wind turbine blades.

[0103] The length of a wind turbine blade is the distance from the root to the tip of the blade. Based on the blade length, a length threshold can be used to distinguish whether a wind turbine blade is a long flexible blade or a short blade. For example, if the length threshold is set to 40 meters, then wind turbine blades with a length greater than or equal to 40 meters are classified as long flexible blades, while wind turbine blades with a length less than 40 meters are classified as short blades.

[0104] B2: If the correlation is greater than or equal to the correlation threshold and the length is less than the length threshold, then obtain the linear regression model.

[0105] If the correlation is greater than or equal to the correlation threshold, it indicates a strong correlation. A regression model can be used to fit the relationship between blade load data. Wind turbine blades shorter than the length threshold are considered short blades. Short blades typically have higher rigidity and are less prone to significant bending and torsion. Due to their shorter length, they are less affected by environmental factors such as wind speed and direction. Therefore, the blade load data at the location to be predicted and the blade load data at the installed location can be approximated as a linear correlation. A linear regression model is obtained for short blades with strong correlation. The correlation fitted by the linear regression model closely approximates the linear correlation between actual blade load data, making predictions more accurate.

[0106] B3: If the correlation is greater than or equal to the correlation threshold and the length is less than the length threshold, then a nonlinear regression model is obtained.

[0107] If the correlation is greater than or equal to the correlation threshold, it indicates a strong correlation. The correlation between the blade load data can be obtained by fitting a regression model. Blades with a length greater than or equal to the length threshold are long and flexible blades. Due to their longer length, long and flexible blades have a certain degree of flexibility and are prone to bending and twisting under the action of wind, resulting in a more complex correlation between the blade load data at different locations. Long and flexible blades are more susceptible to the influence of environmental factors such as wind speed and wind direction. The correlation between the blade load data at the location to be predicted and the blade load data at the installed location is significantly different from the linear correlation, making the prediction by the linear regression model inaccurate.

[0108] Nonlinear regression models may include other complex nonlinear forms such as exponential functions and logarithmic functions, which can better fit the correlation between nonlinear blade load data. For long and flexible blades with strong correlation, a nonlinear regression model is obtained, and a nonlinear correlation is fitted based on the nonlinear regression model. Compared with linear regression models, it can more accurately predict the blade load data at the predicted position on long and flexible blades with weak correlation.

[0109] By obtaining regression models corresponding to wind turbine blade types under conditions of strong correlation, linear regression models are used to more accurately fit the linear correlation of short blades, while nonlinear regression models are used to fit the nonlinear correlation of long and flexible blades, thus ensuring high prediction accuracy for both long and flexible blades and short blades.

[0110] After obtaining the blade load data at the location to be predicted through model prediction, this application embodiment also provides two possible subsequent application methods, wherein application method one refers to C1-C3 and application method two refers to D1-D4.

[0111] The application method one will be explained below, see C1-C3 for details:

[0112] C1: Obtain the first blade load data of the position to be predicted at the first time, the second blade load data of the position to be predicted at the second time, and the load threshold of the position to be predicted.

[0113] The second time is later than the first time. The load threshold at the location to be predicted is used to indicate the life risk level of the location to be predicted. The higher the blade load data at the location to be predicted, the higher the corresponding life risk level. The higher the life risk level, the greater the risk of damage to the wind turbine blades.

[0114] C2: If the load data of the first blade is less than the load threshold and the load data of the second blade is greater than or equal to the load threshold, then a warning message is sent for the location to be predicted.

[0115] If the load data of the first blade is less than the load threshold, it indicates that the life risk at the predicted location at the first time is of a low level. If the load data of the second blade is greater than or equal to the load threshold, it indicates that the life risk at the predicted location at the second time is of a high level. When the risk level jumps from low to high, it indicates that the risk of damage to the wind turbine blades has increased. In response to this increase in risk level, an early warning message is sent.

[0116] Early warning information is used to remind relevant personnel or organizations that the wind turbine blade load corresponding to the early warning information needs to be paid attention to and countermeasures should be taken. Countermeasures include reducing wind turbine power, continuously monitoring the location to be predicted, shutting down, and maintaining wind turbine blades. Early warning information includes the warning content, location information of the location to be predicted, blade load data, risk level data, etc.

[0117] Based on the relationship between the first load data, the second load data, and the load threshold, if it is determined that the risk level of the location to be predicted increases between the first and second time periods, an early warning message is sent to the location to be predicted.

[0118] The risk level classification includes the following two scenarios:

[0119] Scenario 1: Two-tier division (low-tier and high-tier)

[0120] In Scenario 1, the life risk level is divided into low and high levels, which is determined by a load threshold.

[0121] Low level: If the blade load data is less than the load threshold, it indicates that the risk of damage at the location to be predicted is low.

[0122] High level: If the blade load data is greater than or equal to the load threshold, it indicates that the risk of damage at the location to be predicted is high.

[0123] In Scenario 1, when the risk level jumps from low to high, it indicates that the risk of damage to the wind turbine blade at the predicted location has increased. In this case, an early warning message is sent to the predicted location.

[0124] Compared to scenario one, scenario two has a more detailed classification of levels.

[0125] Scenario 2: Three-level division (low level, medium level, and high level)

[0126] In Scenario 2, the life risk level is divided into low, medium and high levels, which are determined by two load thresholds (first load threshold and second load threshold).

[0127] Low level: If the blade load data is less than the first load threshold, it indicates that the risk of damage at the location to be predicted is low.

[0128] Medium level: If the blade load data is greater than or equal to the first load threshold and less than the second load threshold, it indicates that the damage risk at the location to be predicted is at a medium level.

[0129] High Level: If the blade load data is greater than or equal to the second load threshold, it indicates a high risk of damage at the predicted location. In Scenario 2, different levels of warning information can be sent based on the degree of risk increase. When the risk level jumps from low to medium, or from medium to high, it indicates an increased risk of damage at the predicted location of the wind turbine blade, and a low-level warning information is sent. When the risk level jumps from low to high, it indicates a significant increase in the risk of damage at the predicted location of the wind turbine blade, and in this case, a high-level warning information is sent.

[0130] C3: If the load data of the first blade is greater than or equal to the load threshold, or the load data of the second blade is greater than or equal to the load threshold, then a warning message is sent for the location to be predicted.

[0131] If the load data of the first blade is greater than or equal to the load threshold, it means that the location to be predicted belongs to a higher level at the first time. If the load data of the second blade is greater than or equal to the load threshold, it means that the location to be predicted belongs to a higher level at the second time.

[0132] If the risk level of the location to be predicted is high, it indicates that the risk of damage to the location is high, and an early warning message needs to be sent to the location.

[0133] By monitoring the load data of the first and second blades, and based on load thresholds, it is possible to promptly detect whether the life risk level of the predicted location has increased or whether the predicted location is at a higher level, and issue early warnings for both situations. This effectively avoids wind turbine blade damage or accidents caused by excessive load, ensures the timeliness of early warning information feedback, and enhances the reliability and stability of the wind turbine system.

[0134] The following explains application method two; please refer to D1-D4 for details:

[0135] D1: Get multiple installed locations.

[0136] If there are multiple installed locations for the wind turbine blades, obtain all installed locations.

[0137] D2: For the target installed location among multiple installed locations, determine the offset installation location based on the target installed location.

[0138] The target installed location is the installed location among multiple installed locations for which the offset installation location needs to be determined.

[0139] Offset mounting locations are determined around the target mounting location. There may be one or more offset mounting locations. The distance between the offset mounting location and the target mounting location is less than a distance threshold. In this embodiment, it is desirable to evaluate which location, between the offset mounting location and the target mounting location, is more suitable for mounting the actual sensor.

[0140] D3: Replace the offset installation position with the target installed position, execute the steps of obtaining the position to be predicted and the installed position, and subsequent steps, to obtain the blade load data of the position to be predicted under the offset installation position.

[0141] Here, the offset installation position is used to replace the target installed position, and steps S102-S104 are executed. For the target installed position, this can include: determining a first target correlation between the position to be predicted and the target installed position; if the first target correlation is greater than or equal to a correlation threshold, a regression model is obtained, and the blade load data of the target installed position is fitted using the regression model to obtain the blade load data of the position to be predicted under the target installed position as the first target blade load data; and determining a second target correlation between the position to be predicted and the offset installation position, if the second target correlation is greater than or equal to the correlation threshold, a regression model is obtained, and the blade load data of the target installed position is fitted using the regression model to obtain the blade load data of the position to be predicted under the offset installation position as the second target blade load data. This is equivalent to predicting the impact of the offset installation position on the position to be predicted based on the offset installation position, other installed positions besides the target installed position, and the position to be predicted, i.e., the blade load data of the position to be predicted under the offset installation position. Similarly, the impact of the target installed position on the position to be predicted can be the blade load data of the position to be predicted under the target installed position.

[0142] D4: Based on the first target blade load data and the second target blade load data, determine the location of the actual sensor to be installed on the wind turbine blade from the offset installation location and the target installed location.

[0143] For example, if the blade load data at the predicted position under the target installation position is greater than the blade load data at the predicted position under the offset installation position, it indicates that the various external forces and their effects at the offset installation position are smaller. Therefore, the real sensor can be installed at the predicted position instead of the target installation position.

[0144] By installing the actual sensor at either an offset mounting location or the target mounting location, different installation layouts are obtained for each location. Based on these different layouts, prediction results are generated for each. The prediction results for the offset mounting location are compared with those for the target mounting location, and the location with the more accurate prediction is selected as the position for installing the actual sensor on the wind turbine blade.

[0145] For example, suppose there are three installed positions A, B, and C on a wind turbine blade, where A is the target installed position. Based on position A, an offset installed position A' is determined, where the distance between A' and A is less than a preset threshold. Next, A, B, and C are determined as installed positions, and the blade load data X-data1 for the position to be predicted is predicted. Then, A', B, and C are determined as installed positions, and the blade load data X-data2 for the position to be predicted is predicted.

[0146] Then, the accuracy of X-data1 and X-data2 is compared, which can be obtained by comparing them with the simulated blade load data on X.

[0147] If the blade load data at position A' is found to be more accurate, then A' will be replaced with the new position for installing the actual sensor. By adjusting the position of the actual sensor, the installation layout will be made more reasonable, thereby making the prediction of the blade load data at the position to be predicted corresponding to this installation layout more accurate; otherwise, A will remain the position for installing the actual sensor.

[0148] By comparing the predicted blade load data from different locations, a more reasonable location for installing the actual sensor can be found between the offset installation location and the target installation location, resulting in a more reasonable installation layout and thus improving the prediction accuracy of the blade load data. Furthermore, the position of the actual sensor can be adjusted based on the newly determined installation location. Subsequently, the regression model can be updated based on the blade load data obtained at the newly determined installation location to obtain a more accurate regression model. This is beneficial for subsequent sensor installation and adjustment, and can further improve the prediction accuracy of the blade load data.

[0149] The regression model provided in this application can be obtained in the following ways, see E1-E4:

[0150] E1: Obtain simulated blade load data and historical load data.

[0151] Historical load data can be actual blade load data, or it can be calibrated based on simulated blade load data and actual blade load data.

[0152] E2: Obtain the initial regression model, which includes the regression model parameters to be adjusted.

[0153] The initial regression model has the same model structure as the regression model. The initial regression model includes regression model parameters to be adjusted. After the fitting process, the regression model parameters are adjusted, and the initial regression model is transformed into a regression model.

[0154] E3: Based on historical load data, the first predicted blade load data for the position to be predicted is obtained by fitting an initial regression model.

[0155] Using historical load data as input, the initial regression model is fitted to obtain the first predicted blade load data for the location to be predicted.

[0156] E4: Based on the difference between the first predicted blade load data and the simulated blade load data, adjust the regression model parameters of the initial regression model to obtain the regression model.

[0157] In the process of fitting the regression model, the fitting objective is to minimize the difference between the first predicted blade load data and the simulated blade load data, so that the regression model can more accurately predict the blade load data at the position to be predicted.

[0158] During the fitting process, the difference between the first predicted blade load data and the simulated blade load data can be continuously compared. Based on this difference, optimization algorithms (such as the least squares method) can be used to adjust the parameters of the initial regression model to minimize the prediction error.

[0159] The optimization process of the least squares method is explained below:

[0160] The optimization process of least squares is a method to find the optimal function fit by minimizing the sum of squared errors. Taking linear regression as an example, the optimization process of least squares is as follows:

[0161] (1) Data collection: Obtain a series of observations (x1, y1), (x2, y2), ..., (x n y n ), where x n For the nth historical payload data, y n This represents the load data for the nth simulated blade, where n is a positive integer.

[0162] (2) Setting up the model: (Taking the simplest linear regression model as an example) Set up the linear regression model y = ax + b, where a and b are the regression model parameters of the initial model.

[0163] (3) Construct the objective function: Calculate the sum of squared errors SSE represents the sum of squared errors.

[0164] (4) Find the partial derivatives: Take the partial derivatives with respect to a and b respectively, and obtain and

[0165] (5) Set the partial derivatives to zero: Solve the system of equations and

[0166] (6) Solve the system of equations: The optimal estimates a′ and b′ of a and b are obtained by solving the system of equations. a′ and b′ are the regression model parameters of the regression model.

[0167] (7) Evaluate the model: Use the obtained model to make predictions and evaluate the model's performance.

[0168] By continuously adjusting the regression model parameters of the initial regression model through optimization algorithms, a regression model including the optimal regression model parameters is obtained. This allows the model to more accurately predict the blade load data at the location to be predicted on the wind turbine blade, thereby improving prediction accuracy.

[0169] The prediction model provided in this application can be obtained in the following ways, see F1-F4:

[0170] F1: Acquire historical environmental data and simulated blade load data.

[0171] F2: Get the initial prediction model, which includes the prediction model parameters to be adjusted.

[0172] The initial prediction model has the same model structure as the prediction model. The prediction model structure includes the prediction model parameters to be adjusted. After the training process, the prediction model parameters are adjusted, and the initial prediction model will be transformed into the prediction model.

[0173] F3: Based on historical environmental data, the second predicted blade load data for the location to be predicted is obtained through the initial prediction model.

[0174] The data input to the model (historical environmental data and simulated blade data) is normalized to eliminate the influence of dimensions, and the data is divided into training set, validation set and test set. The data set is divided in the following ways: random partitioning, time series partitioning, stratified sampling (the data is divided into different layers according to features, and a part of the data is extracted from each layer for the partitioning of the dataset) and cross-validation.

[0175] The model is trained based on an initial prediction model. In each training round, the second predicted blade load data for the location to be predicted is calculated. The training objective is to minimize the difference between the second predicted blade load data and the simulated blade load data, enabling the prediction model to more accurately predict the blade load data for the location to be predicted.

[0176] F4: Based on the difference between the second predicted blade load data and the simulated blade load data, adjust the prediction model parameters of the initial prediction model to obtain the prediction model.

[0177] The difference between the second predicted blade load data and the simulated blade load data can be obtained through a loss function.

[0178] Based on the prediction error, appropriate optimization algorithms (such as gradient descent, Adam, SGD, etc.) are used to adjust the parameters of the initial prediction model to minimize the difference between the second predicted blade load data and the simulated blade load data. At the same time, regularization techniques (such as L1, L2, Dropout, etc.) are used to prevent overfitting during training.

[0179] Repeat steps F3 and F4 above until the prediction model parameters reach their optimal state, thus obtaining the final prediction model.

[0180] By introducing historical environmental data and continuously optimizing and iterating to obtain the optimal prediction model parameters, the prediction model is trained and obtained. This allows the prediction model to fit the relationship parameters between the blade load data at the location to be predicted and the blade load data at the installed location to the greatest extent. Moreover, the relationship parameters incorporate the characteristics of historical environmental data, making them closer to the actual operating conditions. Even when the correlation is not strong, the model can accurately predict the blade load data at the location to be predicted.

[0181] In one possible implementation, the prediction model is a multi-layer neural network, including an input layer, a one-dimensional convolutional layer, other convolutional layers, and an output layer. This prediction model can be obtained in the following ways, see G1-G4:

[0182] G1: Input environmental data into the input layer, and then pass the environmental data to the one-dimensional convolutional layer through the input layer.

[0183] The input layer is the first layer of a multi-layer neural network, used to receive and preprocess environmental data, historical environmental data of the installed location, and simulated blade load data of the location to be predicted.

[0184] G2: Extracts features from environmental data through a one-dimensional convolutional layer to obtain multiple feature vectors.

[0185] One-dimensional convolutional layers are used to process time series data and can capture local patterns in time series payload data.

[0186] Other convolutional layers can be deeper convolutional layers, used to further extract and combine features.

[0187] G3: Based on multiple feature vectors, it makes predictions through other convolutional layers and uses an attention mechanism to obtain the prediction results.

[0188] Attention mechanisms are techniques that allow predictive models to dynamically focus on important parts of data when processing it. Applied after convolutional layers, they can improve the model's attention to key information. Specifically, attention mechanisms adaptively assign weights to different parts of the input data, allowing the predictive model to focus more on features that have a greater impact on blade load data. For example, wind speed and turbulence intensity may have a greater impact on blade load data than other environmental data; the attention mechanism will assign them a larger attention score, thus more accurately predicting the blade load data at the location to be predicted.

[0189] G4: Based on the prediction results, the output layer is used to obtain the blade load data.

[0190] The output layer is the last layer of the multi-layer neural network. It is responsible for converting the final prediction results (feature vectors) obtained from the previous multi-layer neural networks into blade load data for the position to be predicted on the wind turbine blade. The output layer includes activation functions corresponding to the prediction results, which convert the feature vectors into blade load data.

[0191] One-dimensional convolutional layers in the prediction model effectively capture local features in time-series environmental data, while other convolutional layers further extract and combine these features, enabling the model to learn deeper and more complex feature representations. The attention mechanism allows the prediction model to dynamically focus on important features when processing data, thereby improving its tolerance to noise and outliers and enhancing its overall robustness. Through a multi-layered neural network structure, particularly combining one-dimensional convolutional layers and the attention mechanism, the model can more accurately capture the complex relationship between environmental data and blade load data, thus improving the accuracy of predicting blade load data at the predicted location.

[0192] In studying wind turbine blades, researchers are particularly interested in obtaining blade load data at key cross-sectional locations on the blades. For example, this could be the location with the highest blade load, or the location where the blade load data is close to the theoretical load limit. To obtain blade load data at key cross-sectional locations through the installation of sensors using an optimized location distribution, this application provides a method for determining the location distribution, as detailed in H1-H3:

[0193] H1: Get the selection model.

[0194] The selection model is a machine learning model with selection properties used to determine a combination of installation locations among multiple positions on a wind turbine blade. This combination of installation locations indicates the preferred location for installing the sensor among several undetermined installation positions. For example, the combination of installation locations may include three locations: the blade root, the blade tip, and the blade center.

[0195] The undetermined installation location is the position on the wind turbine blade where a sensor needs to be installed. The sensor can include real sensors and virtual sensors. If the blade load data of a position on the wind turbine blade is obtained through a regression model or a prediction model, it can be regarded as a virtual sensor being installed at that position.

[0196] H2: Based on environmental data from multiple undetermined installation locations, a combination of installation locations is obtained by selecting a model for prediction.

[0197] Environmental data for multiple potential installation locations typically includes various parameters such as wind speed, wind direction, temperature, humidity, and air pressure. To prevent overfitting and reduce computational complexity, feature selection can be performed on these environmental data to select the desired combination of key cross-sectional locations, i.e., installation locations, predicted by the model. For example, assuming there are 10 types of environmental data, including wind speed, wind direction, temperature, humidity, and air pressure, feature selection removes redundant features, selecting the three environmental data that have the greatest impact on predicting the key cross-sectional locations: wind speed, wind direction, and humidity.

[0198] Feature selection methods are generally classified into three categories: filtering, wrapping, and embedded. Filtering methods establish statistical relationships and select features from environmental data based on these relationships. For example, the statistical relationship could be correlation, selecting three environmental data points with high correlation as the primary environmental data. Wrapping methods use specific machine learning algorithms to evaluate the quality of environmental data and select the best ones. For example, Recursive Feature Elimination (RFE) combined with a linear regression model can be used to recursively remove unimportant environmental data, ultimately selecting the three environmental data points that have the greatest impact on predicting the critical section location. Embedded methods typically perform feature selection during model training, embedding the process of selecting environmental data features into the model training process. For example, a LASSO regression model can be used, adjusting the regularization parameter to make the weights of certain environmental data points zero, thereby selecting the three environmental data points that have the greatest impact on predicting the critical section location.

[0199] Different environmental data correspond to different combinations of installation locations. By inputting the environmental data into a trained selection model, the corresponding combinations of installation locations can be obtained.

[0200] H3: Determine the location of the actual sensor or the location of the virtual sensor on the wind turbine blades based on the installation location combination.

[0201] Depending on the multiple installation locations in the installation location combination, all locations can be equipped with real sensors, all locations can be equipped with virtual sensors, or, based on factors such as the installation difficulty of real sensors and the existing number of real sensors, real sensors can be installed in some locations to obtain the installed locations, and virtual sensors can be installed in some locations.

[0202] In one possible implementation, the method for selecting the model's training method is described in H4-H7:

[0203] H4: Obtain historical environmental data for multiple historical pending installation locations and the tags corresponding to the multiple historical pending installation locations.

[0204] Among them, the historical pending installation locations are the pending installation locations used to train the selection model, and the labels are used to describe the preferred combination of installation locations among multiple historical pending installation locations.

[0205] During the training of the selection model, historical environmental data of batches of pending installation locations are acquired, along with batch labels corresponding to these locations. The number of preferred installation location combinations indicated by the batch labels may vary. For example, a batch of installation location combinations may include combination A and combination B, where combination A has 6 installation locations and combination B has 3.

[0206] H5: Obtain the initial selection model, which includes the selection model parameters to be adjusted.

[0207] The initial selection model has the same model structure as the selection model. The selection model's model structure includes the selection model parameters to be adjusted. After the training process, the selection model parameters are adjusted, and the initial selection model will be transformed into the selection model.

[0208] H6: Based on historical environmental data of multiple historical undetermined installation locations, predictions are made through an initial selection model to obtain a combination of predicted installation locations for multiple historical undetermined installation locations.

[0209] The predicted installation location combination is obtained from the output of the initial selection model.

[0210] The historical environmental data of the acquired undetermined installation locations are input into the initial selection model. Based on the current selection model parameters, the initial selection model processes this historical environmental data and outputs a combination of predicted installation locations corresponding to the undetermined historical installation locations. For example, for a set of historical environmental data containing 5 undetermined historical installation locations, the initial selection model may output a combination of predicted installation locations containing 3 preferred installation locations (such as the blade root, blade tip, and wind turbine blade center).

[0211] H7: Based on the difference between the predicted installation location combination and the label, adjust the selection model parameters of the initial selection model to obtain the selection model.

[0212] The difference between the predicted installation location combinations and the labels is a metric for measuring the accuracy of the initial selection model; this difference can be quantified, for example, using a loss function. By adjusting the selection model parameters of the initial selection model and minimizing the difference between the predicted installation location combinations and the labels, the accuracy of the predicted installation location combinations obtained by the initial selection model is improved, ultimately resulting in a selection model with more accurate output installation location combinations.

[0213] Based on the training-obtained selection model, and according to environmental data from multiple undetermined installation locations, the optimal installation location distribution corresponding to the environmental data is determined. This optimal installation location distribution covers key cross-sectional locations, improving the relevance and representativeness of the wind turbine blade load data, and enhancing its adaptability and robustness to changes in different environmental conditions.

[0214] In addition to the wind turbine blade load prediction method provided in this application embodiment, a wind turbine blade load prediction device is also provided, as shown in FIG3. The device 200 includes: an acquisition unit 201, a determination unit 202, and a regression unit 203.

[0215] The acquisition unit 201 is used to acquire the position to be predicted and the installed position. The position to be predicted is the position on the wind turbine blade where no real sensor is installed, and the installed position is the position on the wind turbine blade where the real sensor is installed.

[0216] The determining unit 202 is used to determine the correlation between the location to be predicted and the installed location;

[0217] The regression unit 203 is used to obtain a regression model if the correlation is greater than or equal to the correlation threshold, and to obtain the blade load data of the position to be predicted by fitting the regression model according to the blade load data of the installed position. The regression model is obtained by fitting the simulated blade load data of the position to be predicted and the historical blade load data of the installed position.

[0218] As one possible implementation, the device 200 further includes a prediction unit 204, which is used to obtain a prediction model if the correlation is less than the correlation threshold, and to make a prediction based on the environmental data of the installed location to obtain the blade load data of the location to be predicted. The prediction model is trained based on the historical environmental data of the installed location and the simulated blade load data of the location to be predicted.

[0219] As can be seen from the above technical solution, the wind turbine blade load prediction device provided in this application includes an acquisition unit, a determination unit, a regression unit, and a prediction unit. The acquisition unit acquires the location to be predicted and the installed location. The regression unit determines the correlation between the location to be predicted and the installed location. This correlation describes the degree of association between the blade load data at the location to be predicted and the blade load data at the installed location. To improve prediction accuracy, different correlation levels correspond to different models. Specifically, if the correlation is greater than or equal to a correlation threshold, it indicates a strong correlation between the location to be predicted and the installed location, indicating a close relationship. In this case, the regression unit acquires the corresponding regression model. The regression model can describe the correlation between the blade load data at the location to be predicted and the blade load data at the installed location through mathematical relationships. Therefore, based on the blade load data at the installed location, the regression model is used to fit the data to obtain the blade load data at the location to be predicted, which is not only highly accurate but also convenient and fast.

[0220] If the correlation is less than the correlation threshold, it indicates that the correlation between the location to be predicted and the installed location is weak, and the relationship is not close enough. In this case, the prediction unit acquires the corresponding prediction model. The prediction model can use multi-layer networks or other complex model structures to capture deep feature relationships between blade load data at different locations and environmental data at the installed location that are difficult to describe by regression models. Thus, based on the environmental data at the installed location, the prediction model makes predictions to obtain the blade load data at the location to be predicted. This not only improves the accuracy of prediction under weak correlation but also enhances the adaptability and robustness of the prediction model to changes in different environmental conditions.

[0221] As one possible implementation, if the length of the wind turbine blade is greater than a length threshold, and the predicted position includes multiple locations, then the determining unit is specifically used for:

[0222] Determine the correlation between each of the locations to be predicted and the already installed locations;

[0223] The regression unit is specifically used for:

[0224] If each of the aforementioned correlations is greater than or equal to the correlation threshold, then the regression model is obtained;

[0225] The prediction unit is specifically used for:

[0226] If at least one of the multiple correlations is less than the correlation threshold, then the prediction model is obtained.

[0227] As one possible implementation, if the regression model includes both linear and nonlinear regression models, then the regression unit is specifically used for:

[0228] Obtain the length of the wind turbine blades;

[0229] If the correlation is greater than or equal to the correlation threshold and the length is less than the length threshold, then the linear regression model is obtained;

[0230] If the correlation is greater than or equal to the correlation threshold, and the length is greater than or equal to the length threshold, then the nonlinear regression model is obtained.

[0231] As one possible implementation, the device further includes a warning unit, the warning unit being used for:

[0232] The first blade load data of the location to be predicted at a first time, the second blade load data of the location to be predicted at a second time, and the load threshold of the location to be predicted are obtained, wherein the second time is later than the first time.

[0233] If the first blade load data is less than the load threshold, and the second blade load data is greater than or equal to the load threshold, then an early warning message is sent for the location to be predicted.

[0234] If the load data of the first blade is greater than or equal to the load threshold, or the load data of the second blade is greater than or equal to the load threshold, then an early warning message is sent for the location to be predicted.

[0235] As one possible implementation, the acquisition unit 201 is used to acquire the location to be predicted and multiple installed locations;

[0236] The regression unit 203 is used to target the installed location among the plurality of installed locations.

[0237] An offset installation position is determined based on the target installation position, wherein the distance between the offset installation position and the target installation position is less than a distance threshold;

[0238] Determine the first target correlation between the location to be predicted and the target installation location; if the first target correlation is greater than or equal to the correlation threshold, obtain the regression model, and fit it to the blade load data of the target installation location to obtain the blade load data of the location to be predicted at the target installation location as the first target blade load data; and

[0239] Determine the second target correlation between the position to be predicted and the offset installation position. If the second target correlation is greater than or equal to the correlation threshold, obtain the regression model. Based on the blade load data of the target installation position, fit the regression model to obtain the blade load data of the position to be predicted under the offset installation position as the second target blade load data.

[0240] The device further includes a position determination unit, the position determination unit being used for:

[0241] Based on the first target blade load data and the second target blade load data, the position where the actual sensor is installed on the wind turbine blade is determined from the offset installation position and the target installed position.

[0242] As one possible implementation, the apparatus further includes a regression model fitting unit, which is used for:

[0243] Obtain the simulated blade load data and the historical load data;

[0244] Obtain an initial regression model, which includes the regression model parameters to be adjusted;

[0245] Based on the historical load data, the first predicted blade load data for the position to be predicted is obtained by fitting the initial regression model.

[0246] Based on the difference between the first predicted blade load data and the simulated blade load data, the regression model parameters of the initial regression model are adjusted to obtain the regression model.

[0247] As one possible implementation, the apparatus further includes a prediction model training unit, which is used for:

[0248] Acquire the historical environmental data and the simulated blade load data;

[0249] Obtain an initial prediction model, which includes prediction model parameters to be adjusted;

[0250] Based on the historical environmental data, the initial prediction model is used to predict the second predicted blade load data at the location to be predicted.

[0251] Based on the difference between the second predicted blade load data and the simulated blade load data, the prediction model parameters of the initial prediction model are adjusted to obtain the prediction model.

[0252] As one possible implementation, if the prediction model includes an input layer, a one-dimensional convolutional layer, other convolutional layers, and an output layer, then the prediction unit is specifically used for:

[0253] The environmental data is input into the input layer, and then passed to the one-dimensional convolutional layer through the input layer.

[0254] The environmental data is used to extract features through the one-dimensional convolutional layer to obtain multiple feature vectors;

[0255] Based on the multiple feature vectors, prediction is performed through other convolutional layers, and an attention mechanism is used to obtain the prediction result.

[0256] Based on the prediction results, the output layer is used to output the blade load data.

[0257] As one possible implementation, the device further includes an installation location selection unit, the installation location selection unit being used for:

[0258] Obtain the selection model;

[0259] Based on environmental data from multiple undetermined installation locations, a combination of installation locations is obtained through prediction using the selection model. This combination of installation locations is used to indicate the preferred location among the multiple undetermined installation locations for installing the sensor.

[0260] Based on the installation location combination, determine the location where the real sensor is installed on the wind turbine blade or the location where the virtual sensor is installed on the wind turbine blade;

[0261] The training method for the selected model is as follows:

[0262] Obtain historical environmental data for multiple historical undetermined installation locations and tags corresponding to the multiple historical undetermined installation locations. The tags are used to describe the preferred combination of installation locations among the multiple historical undetermined installation locations.

[0263] Obtain an initial selection model, which includes selection model parameters to be adjusted;

[0264] Based on the historical environmental data of the multiple historical undetermined installation locations, the predicted installation location combination of the multiple historical undetermined installation locations is obtained through the initial selection model.

[0265] Based on the difference between the predicted installation location combination and the label, the selection model parameters of the initial selection model are adjusted to obtain the selection model.

[0266] This application also provides a computer device. Referring to Figure 4, which shows a structural diagram of a computer device provided in this application embodiment, the device includes a memory 310 and a processor 320.

[0267] The memory 310 is used to store program code and transmit the program code to the processor;

[0268] The processor 320 is used to execute any of the wind turbine blade load prediction methods provided in the above embodiments according to the instructions in the program code.

[0269] This application also provides a computer-readable storage medium for storing a computer program that executes any of the wind turbine blade load prediction methods provided in the above embodiments.

[0270] This application also provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the wind turbine blade load prediction method provided in the various optional implementations of the above aspects.

[0271] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0272] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0273] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0274] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0275] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting wind turbine blade loads, characterized in that, The method includes: Obtain the location to be predicted and the location already installed. The location to be predicted is the location on the wind turbine blade where no real sensor is installed, and the location already installed is the location on the wind turbine blade where the real sensor is installed. Determine the correlation between the location to be predicted and the installed location; and If the correlation is greater than or equal to the correlation threshold, a regression model is obtained. Based on the blade load data of the installed location, the regression model is fitted to obtain the blade load data of the location to be predicted. The regression model is obtained by fitting the simulated blade load data of the location to be predicted and the historical blade load data of the installed location.

2. The method according to claim 1, characterized in that, After determining the correlation between the location to be predicted and the installed location, the method further includes: If the correlation is less than the correlation threshold, a prediction model is obtained. Based on the environmental data of the installed location, the prediction model is used to predict the blade load data of the location to be predicted. The prediction model is trained based on the historical environmental data of the installed location and the simulated blade load data of the location to be predicted.

3. The method according to claim 1, characterized in that, If the length of the wind turbine blade is greater than a length threshold, and the predicted location includes multiple locations, then determining the correlation between the predicted location and the installed location includes: Determine the correlation between each of the locations to be predicted and the already installed locations; If the correlation is greater than or equal to the correlation threshold, then a regression model is obtained, including: If each of the aforementioned correlations is greater than or equal to the correlation threshold, then the regression model is obtained.

4. The method according to claim 2, characterized in that, If the length of the wind turbine blade is greater than a length threshold, and the predicted location includes multiple locations, then determining the correlation between the predicted location and the installed location includes: Determine the correlation between each of the locations to be predicted and the already installed locations; If the correlation is less than the correlation threshold, then obtaining the prediction model includes: If at least one of the multiple correlations is less than the correlation threshold, then the prediction model is obtained.

5. The method according to claim 1, characterized in that, If the regression model includes a linear regression model and a nonlinear regression model, then if the correlation is greater than or equal to a correlation threshold, obtaining the regression model includes: Obtain the length of the wind turbine blades; If the correlation is greater than or equal to the correlation threshold, and the length is less than the length threshold, then the linear regression model is obtained; and If the correlation is greater than or equal to the correlation threshold, and the length is greater than or equal to the length threshold, then the nonlinear regression model is obtained.

6. The method according to claim 1, characterized in that, The method further includes: The first blade load data of the location to be predicted at a first time, the second blade load data of the location to be predicted at a second time, and the load threshold of the location to be predicted are obtained, wherein the second time is later than the first time. If the first blade load data is less than the load threshold, and the second blade load data is greater than or equal to the load threshold, then a warning message is sent for the location to be predicted; and If the load data of the first blade is greater than or equal to the load threshold, or the load data of the second blade is greater than or equal to the load threshold, then an early warning message is sent for the location to be predicted.

7. The method according to claim 1, characterized in that, The process of obtaining the location to be predicted and the installed locations includes: obtaining the location to be predicted and multiple installed locations; and The process of determining the correlation between the location to be predicted and the installed location; and if the correlation is greater than or equal to a correlation threshold, obtaining a regression model, and fitting the regression model to obtain the blade load data for the location to be predicted based on the blade load data of the installed location, includes: For the target installation location among the multiple installation locations, An offset installation position is determined based on the target installation position, wherein the distance between the offset installation position and the target installation position is less than a distance threshold; Determine the first target correlation between the location to be predicted and the target installation location; if the first target correlation is greater than or equal to the correlation threshold, obtain the regression model, and fit it to the blade load data of the target installation location to obtain the blade load data of the location to be predicted at the target installation location as the first target blade load data; and Determine the second target correlation between the position to be predicted and the offset installation position. If the second target correlation is greater than or equal to the correlation threshold, obtain the regression model. Based on the blade load data of the target installation position, fit the regression model to obtain the blade load data of the position to be predicted under the offset installation position as the second target blade load data. The method further includes: determining the position of the actual sensor on the wind turbine blade from the offset installation position and the target installed position based on the first target blade load data and the second target blade load data.

8. The method according to claim 1, characterized in that, The method further includes: Obtain the simulated blade load data and the historical load data; Obtain an initial regression model, which includes the regression model parameters to be adjusted; Based on the historical load data, the initial regression model is fitted to obtain the first predicted blade load data for the location to be predicted; and Based on the difference between the first predicted blade load data and the simulated blade load data, the regression model parameters of the initial regression model are adjusted to obtain the regression model.

9. The method according to claim 2, characterized in that, The method further includes: Acquire the historical environmental data and the simulated blade load data; Obtain an initial prediction model, which includes prediction model parameters to be adjusted; Based on the historical environmental data, the initial prediction model is used to predict the second predicted blade load data for the location to be predicted; and Based on the difference between the second predicted blade load data and the simulated blade load data, the prediction model parameters of the initial prediction model are adjusted to obtain the prediction model.

10. The method according to claim 2, characterized in that, If the prediction model includes an input layer, a one-dimensional convolutional layer, other convolutional layers, and an output layer, then the step of predicting the blade load data at the location to be predicted based on the environmental data of the installed location using the prediction model includes: The environmental data is input into the input layer, and then passed to the one-dimensional convolutional layer through the input layer. The environmental data is used to extract features through the one-dimensional convolutional layer to obtain multiple feature vectors; Based on the multiple feature vectors, prediction is performed through other convolutional layers, and an attention mechanism is used to obtain the prediction result; and Based on the prediction results, the output layer is used to output the blade load data.

11. The method according to claim 1, characterized in that, The method further includes: Obtain the selection model; Based on environmental data from multiple potential installation locations, a selection model is used to predict and obtain a combination of installation locations. This combination of installation locations indicates the preferred location among the multiple potential installation locations for installing the sensor. Based on the installation location combination, determine the location where the real sensor is installed on the wind turbine blade or the location where the virtual sensor is installed on the wind turbine blade; The training method for the selected model is as follows: Obtain historical environmental data for multiple historical undetermined installation locations and tags corresponding to the multiple historical undetermined installation locations. The tags are used to describe the preferred combination of installation locations among the multiple historical undetermined installation locations. Obtain an initial selection model, which includes selection model parameters to be adjusted; Based on historical environmental data of the multiple historical undetermined installation locations, a predicted combination of installation locations is obtained through the initial selection model; and Based on the difference between the predicted installation location combination and the label, the selection model parameters of the initial selection model are adjusted to obtain the selection model.

12. [Corrected according to Rule 91, 21.11.2025] A wind turbine blade load prediction device, characterized in that, The device includes: an acquisition unit, a determination unit, and a regression unit; The acquisition unit is used to acquire the position to be predicted and the installed position. The position to be predicted is the position on the wind turbine blade where no real sensor is installed, and the installed position is the position on the wind turbine blade where the real sensor is installed. The determining unit is used to determine the correlation between the location to be predicted and the installed location; and The regression unit is used to obtain a regression model if the correlation is greater than or equal to the correlation threshold. Based on the blade load data of the installed location, the regression model is fitted to obtain the blade load data of the location to be predicted. The regression model is obtained by fitting the simulated blade load data of the location to be predicted and the historical blade load data of the installed location.

13. The apparatus according to claim 11, characterized in that, The device further includes a prediction unit, which is used to obtain a prediction model if the correlation is less than the correlation threshold, and to make a prediction based on the environmental data of the installed location to obtain the blade load data of the location to be predicted. The prediction model is trained based on the historical environmental data of the installed location and the simulated blade load data of the location to be predicted.

14. A computer device, characterized in that, The computer device includes a processor and memory: The memory is used to store computer programs and to transfer the computer programs to the processor; The processor is configured to perform the method according to any one of claims 1-11 according to the computer program.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program for performing the method according to any one of claims 1-11.

16. A computer program product comprising a computer program, characterized in that, When it is run on a computer device, it causes the computer device to perform the method described in any one of claims 1-11.

Images