Bending moment load prediction method and device for offshore wind power supporting structure

Abstract

The invention relates to the technical field of wind power generation, in particular to a bending moment load prediction method and device for an offshore wind power supporting structure, and the method comprises the steps: collecting operation state parameters and the actual acceleration at a target position in real time; inputting the data into a trained bending moment load prediction model to obtain bending moment load prediction data of the cross section of the base of the tower drum; the training process of the model is as follows: local feature extraction and fusion of time windows with different lengths are carried out on input data to obtain a first enhanced feature, and the input data is obtained by splicing and arranging simulation acceleration and simulation operation parameters according to the same timestamp; combining the first position feature vector with the first enhanced feature to obtain a second time sequence feature; based on the second time sequence feature, obtaining a predicted bending moment time sequence signal through variable attention calculation and linear mapping; and constructing a loss function to train the model to obtain a trained bending moment load prediction model. According to the technical scheme, the prediction precision of the tower drum base section bending moment load can be improved.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, and in particular to a method and apparatus for predicting the bending moment load of offshore wind power support structures. Background Technology

[0002] Offshore wind turbine support structures are the core force-transmitting components connecting the upper wind turbine to the seabed foundation. During long-term service, they must withstand complex alternating loads from the coupling of multiple physical fields such as wind, waves, and currents. Among these, the cross-section at the connection between the wind turbine tower and the foundation is the area with the most significant stress concentration and is also a critical weak point in structural safety. The bending moment at this cross-section is a core mechanical parameter reflecting the overall stress and dynamic response of the offshore wind turbine support structure. Therefore, long-term and high-precision monitoring of the bending moment at this cross-section is crucial.

[0003] In existing technologies, the bending moment load at the connection between the tower and the foundation is typically obtained using direct measurement. This method requires the installation of physical sensors such as strain gauges at key locations such as the tower base or underwater mud surface to collect strain signals, which are then analyzed and calculated to convert them into bending moment loads. However, due to the high humidity, high salt spray, and strong corrosion characteristics of the marine environment, sensors may corrode and fail, causing measurement signal drift and reduced accuracy. Furthermore, the installation and maintenance of underwater measuring points require specialized offshore equipment and personnel, making the operation difficult, costly, and hindering long-term, stable, and continuous monitoring.

[0004] Therefore, those skilled in the art urgently need to develop a new technical solution to address the above problems. Summary of the Invention

[0005] This invention provides a method and apparatus for predicting the bending moment load of offshore wind power support structures, which can improve the prediction accuracy of the bending moment load of the tower base section.

[0006] In a first aspect, embodiments of the present invention provide a method for predicting the bending moment load of an offshore wind power support structure, including: Real-time acquisition of wind turbine operating status parameters and actual acceleration at target locations on wind power support structures; The operating state parameters and the actual acceleration are input into the trained bending moment load prediction model to obtain the output bending moment load prediction data of the wind power support structure tower base section. The bending moment load prediction model is trained in the following manner: Based on multiple simulated wind condition data and multiple simulated wave data, the simulated operating parameters of the wind turbine, the simulated acceleration at the target location, and the simulated bending moment load of the tower base section are obtained. Local features are extracted from the input data in the time dimension using time windows of different lengths. The extracted parallel local features are then fused to obtain the first enhanced feature. The input data is obtained by concatenating the simulated acceleration and simulated running parameters according to the same timestamp and arranging them in chronological order according to the timestamps. The first position feature vector is combined with the first enhanced feature to obtain the second temporal feature, wherein the first position feature vector is generated based on the dynamic response characteristics of the offshore wind power support structure. Based on the second time series feature, the predicted bending moment time series signal of the tower base section is obtained through variable attention calculation and linear mapping; The model is trained using a loss function constructed based on the difference between the predicted bending moment time series signal and the simulated bending moment load, resulting in a trained bending moment load prediction model.

[0007] Secondly, embodiments of the present invention provide a bending moment load prediction device for offshore wind power support structures, comprising: The data acquisition module collects the operating status parameters of the wind turbine and the actual acceleration at the target location on the wind power support structure in real time. The bending moment prediction module is connected to the data acquisition module. It inputs the operating state parameters and the actual acceleration into the trained bending moment load prediction model to obtain the output bending moment load prediction data of the wind power support structure tower base section. The bending moment load prediction model is trained in the following manner: Based on multiple simulated wind condition data and multiple simulated wave data, the simulated operating parameters of the wind turbine, the simulated acceleration at the target location, and the simulated bending moment load of the tower base section are obtained. Local features are extracted from the input data in the time dimension using time windows of different lengths. The extracted parallel local features are then fused to obtain the first enhanced feature. The input data is obtained by concatenating the simulated acceleration and simulated running parameters according to the same timestamp and arranging them in chronological order according to the timestamps. The first position feature vector is combined with the first enhanced feature to obtain the second temporal feature, wherein the first position feature vector is generated based on the dynamic response characteristics of the offshore wind power support structure. Based on the second time series feature, the predicted bending moment time series signal of the tower base section is obtained through variable attention calculation and linear mapping; The model is trained using a loss function constructed based on the difference between the predicted bending moment time series signal and the simulated bending moment load, resulting in a trained bending moment load prediction model.

[0008] Thirdly, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method described in the first aspect of the present invention.

[0009] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method described in the first aspect of the present invention.

[0010] The present invention provides a method and apparatus for predicting the bending moment load of an offshore wind power support structure. This method analyzes the operating state parameters of the wind turbine and acceleration data at easily measurable target locations using a trained bending moment load prediction model to obtain the bending moment load at the tower base section. It eliminates the need to install sensors in highly corrosive environments such as the tower base or underwater mud surfaces, saving labor costs associated with underwater measurement point installation and maintenance. In this bending moment load prediction model, structural physical constraints are applied through data learning, establishing a reasonable mapping between acceleration and bending moment. Operating state parameters related to wind turbine operating data are also used as input data for the acceleration-based bending moment reconstruction model. The dynamic response characteristics of the offshore wind power support structure are characterized by location feature vectors, establishing an end-to-end mapping from multi-source time-series signals to structural bending moment. This eliminates the need for precise physical models and complex parameter inversion, achieving long-term, stable, low-cost, and high-precision real-time monitoring and prediction of the bending moment at key sections of the offshore wind power support structure. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of a method for predicting the bending moment load of an offshore wind power support structure according to an embodiment of the present invention; Figure 2 This is a hardware architecture diagram of an electronic device provided in an embodiment of the present invention; Figure 3 This is a structural diagram of a bending moment load prediction device for an offshore wind power support structure provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of an offshore wind power support structure provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of simulated acceleration and simulated bending moment loads output by OpenFAST under any working condition, provided by an embodiment of the present invention; Figure 6 This is a trend chart of the change of loss value and coefficient of determination R² during the training process provided by an embodiment of the present invention; Figure 7 This is a schematic diagram of the target position acceleration and the bending moment of the tower base section provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the target position acceleration and the bending moment of the tower base section provided in another embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0015] Please refer to Figure 1 This invention provides a method for predicting the bending moment load of an offshore wind power support structure, the method comprising: Step 100: Real-time acquisition of the operating status parameters of the wind turbine and the actual acceleration at the target location on the wind power support structure; Step 102: Input the operating state parameters and actual acceleration into the trained bending moment load prediction model to obtain the output bending moment load prediction data of the wind power support structure tower base section.

[0016] The bending moment load prediction model is trained in the following way: Based on multiple simulated wind condition data and multiple simulated wave data, the simulated operating parameters of the wind turbine, the simulated acceleration at the target location, and the simulated bending moment load of the tower base section are obtained. Local features of different time windows are extracted from the input data in the time dimension. The extracted parallel local features are fused to obtain the first enhanced feature. The input data is obtained by splicing the simulated acceleration and simulated running parameters according to the same timestamp and arranging them according to the order of the timestamps. The first position feature vector is combined with the first enhanced feature to obtain the second temporal feature. The first position feature vector is generated based on the dynamic response characteristics of the offshore wind power support structure. Based on the second time series characteristics, the predicted bending moment time series signal of the tower base section is obtained through variable attention calculation and linear mapping; The model is trained using a loss function constructed based on the difference between the predicted bending moment time series signal and the simulated bending moment load, resulting in a trained bending moment load prediction model.

[0017] In this embodiment of the invention, the offshore wind power support structure includes a wind turbine tower and a wind turbine foundation. The tower base section, located at the connection between the wind turbine tower and the wind turbine foundation, is a critical area of ​​stress concentration. To address the difficulty in directly measuring the bending moment load at the tower base section, easily measurable physical quantities such as operating state parameters and actual acceleration are obtained. These operating state parameters and actual acceleration are then input into a bending moment load prediction model to obtain predicted bending moment load data for the tower base section. The target location can be one or more, representing the specific location where an acceleration sensor is installed. For example, the target location can be the apex of the wind turbine tower or any predetermined height above sea level on the wind turbine tower. The target location can be determined based on the specific scenario and the installation requirements of the acceleration sensor.

[0018] The training data for the bending moment load prediction model includes simulated operating parameters of the wind turbine, simulated acceleration at the target location, and simulated bending moment load at the tower base section. This training data is calculated based on simulated wind and wave data. Simulated operating parameters and simulated acceleration serve as input data, while simulated bending moment load serves as output data, with a one-to-one correspondence between them. Simulated acceleration and simulated operating parameters are concatenated according to timestamps, with data from the same timestamp combined into a single data point. These data points are then arranged chronologically according to their timestamps to obtain a multivariate time series, which is the input data. This input data is then fed into a multi-scale feature extraction layer (which includes multiple parallel time windows of varying lengths, i.e., different convolutional kernel sizes). Local features of different durations are extracted from the input data through these time windows, and the extracted local features are then fused to obtain the first enhanced feature. Considering the phased dynamic response of offshore wind power support structures to wind and waves, a first position feature vector is assigned to each first enhanced feature. This first position feature vector reflects the relationship between the first enhanced feature and the operational phase of the offshore wind power support structure, resulting in a second time-series feature. The second time-series feature is then processed through variable attention modeling in the variable attention mechanism module and linear mapping in the fully connected layer to obtain the predicted bending moment time-series signal of the tower base section. A loss function is constructed based on the difference between the predicted bending moment time-series signal and the simulated bending moment load. Training is continuously performed with the goal of minimizing the value of the loss function (or reducing it to a preset threshold) until a well-trained bending moment load prediction model is obtained. Preferably, the mean squared error (MSE) is used as the loss function, and the Adam optimizer is used for parameter optimization during training.

[0019] As can be seen, in this embodiment of the invention, structural physical constraints are implicitly applied through data learning, a reasonable mapping between acceleration and bending moment is established, and the operating parameter data in SCADA (Supervisory Control and Data Acquisition) related to wind turbine operation data is added as input data for the acceleration-based bending moment reconstruction model. This can not only adaptively learn the dynamic laws of time series data, but also strengthen the feature interaction between multiple variables, deeply explore the data features between different channels, and improve the collaborative prediction effect between acceleration and SCADA data.

[0020] In one embodiment of the present invention, the bending moment load prediction data includes the bending moment load in the x-direction and the bending moment load in the y-direction of the wind power support structure tower base section; The simulated bending moment load includes the bending moment load in the x-direction and the bending moment load in the y-direction of the wind turbine support structure tower base section.

[0021] In this embodiment, the origin is set at the intersection of the centerline of the undeformed tower and the mean sea level (the boundary between the tower base and the sea level). The x-axis is defined by the 0° wind direction (directly in front of the wind direction), and the y-axis is defined by the left side of the 0° wind direction (i.e., 90° counterclockwise in the top view). The bending moment load prediction data, simulated bending moment load, and predicted bending moment time series signals all include bending moment load data in both the x and y directions.

[0022] In one embodiment of the present invention, the operating state parameters include: wind speed in the x-direction, wind speed in the y-direction, yaw angle, rotational speed, and propeller pitch angle.

[0023] In this embodiment, the operating status parameters include five SCADA parameters: wind speed in the x-direction, wind speed in the y-direction, yaw angle, rotational speed, and pitch angle. These parameters are used to characterize the main environmental loads (wind speed), the wind turbine's windward state (yaw), and the operating control state (rotational speed and pitch) acting on the structure, and together constitute the operating status parameters that affect the bending moment of the tower base section.

[0024] In one embodiment of the present invention, based on multiple simulated wind condition data and multiple simulated wave data, simulated operating parameters of the wind turbine, simulated acceleration at the target location, and simulated bending moment load of the tower base section are obtained, including: Simulated wind data was generated using the IEC Kaimal spectral model, and corresponding simulated wave data was generated using the JONSWAP spectral model. Based on simulated wind and wave data, the simulated operating parameters of the wind turbine, the simulated acceleration at the target location, and the simulated bending moment load of the tower base section are obtained using the OpenFAST model.

[0025] In this embodiment, the simulation steps for environmental load data include the simulation of wind condition data and the simulation of random wave data. The dataset construction steps include the acquisition steps for dynamic response data and wind power data. Specifically, the acquisition step for dynamic response data includes simulating the dynamic response of the wind power support structure based on wind condition data and random wave data. The dynamic response includes simulated acceleration data and simulated bending moment load.

[0026] Specifically, the dataset construction step uses OpenFAST software to simulate the dynamic response of the wind turbine support structure. A finite element model of the monitored offshore wind turbine structure is established using OpenFAST software to numerically generate acceleration-bending moment signal pairs, which serve as the model training dataset. In the environmental load data simulation step, the IEC Kaimal model is used to simulate wind conditions, and the JONSWAP spectrum is used to generate random wave data. Wind and wave forces are the main environmental loads acting on offshore wind turbines. To comprehensively cover the actual operating scenarios of offshore wind power, turbulent wind conditions are simulated using the IEC Kaimal model, and random wave loads are generated based on the JONSWAP spectrum. It is assumed that wave height and wave period follow normal distributions N(3.8, 0.5) and N(5.5, 0.2), respectively, while wind direction and wave direction are randomly generated between 0° and 360°.

[0027] In one embodiment, to achieve deep learning training and testing, a total of 103 sea state simulations were conducted. 80 sea state simulations were used for training, 20 for validation, and the remaining 3 sea state simulations served as a test set to verify the model's load reconstruction performance. The environmental parameters for the three simulations are shown in Table 1. Wind speed is a core control factor affecting the dynamic response of wind turbine structures. As shown in the table, the wind speeds differ significantly across the three simulations, covering a wide range of sea state conditions. This provides data support for the reliability and generalization of subsequent research.

[0028] Table 1: Test Sea State Parameters In one embodiment of the present invention, the first positional feature vector is combined with the first enhanced feature to obtain the second temporal feature, including: The first enhanced feature is encoded into a learnable location to generate a first location feature vector. The learnable location encoding is iteratively optimized during the training of the bending moment load prediction model with the goal of minimizing the loss function. The learnable location encoding is used as a location feature vector to characterize the dynamic response characteristics of the wind power support structure under different operating stages. The different operating stages include: the preset start-up stage, the rated operation stage, the turbulent condition stage, and the shutdown stage. The first positional feature vector is combined with the first enhanced feature to obtain the second temporal feature.

[0029] In this embodiment, a learnable position encoding module is introduced. A set of trainable parameter vectors is used to assign a unique positional representation to each temporal position of the input sequence. This parameter vector (i.e., the first positional feature vector) is then superimposed on the feature-extracted input data to obtain the second temporal feature. This enables the model to perceive the position in the second temporal feature, accurately identifying which operational stage of the wind power support structure the corresponding data belongs to, thereby effectively capturing global temporal correlations. The parameters in the learnable position encoding are continuously optimized during model training with the goal of minimizing the objective function value, until the optimal learnable position encoding parameters are obtained. Among them, the startup stage is the stage when the wind turbine starts to rotate from a stationary state; the rated operation stage is the stage when the wind turbine reaches its design speed and power and is in a stable power generation state; the turbulent condition stage is the stage when the wind turbine encounters high-intensity turbulent winds during rated operation, and the wind speed and direction change rapidly and randomly; and the shutdown stage is the stage when the wind turbine transitions from the operating state to the stationary state (including normal shutdown and emergency shutdown due to fault).

[0030] In one embodiment of the present invention, based on the second time-series characteristics, the predicted bending moment time-series signal of the tower base section is obtained through variable attention calculation and linear mapping, including: The second time-series features are mapped to query vector Q, key vector K, and value vector V; Based on the correlation between query vector Q and key vector K, an attention weight matrix between query vector Q and key vector K is generated. Deep features are obtained by weighting and fusing the value vector V based on the attention weight matrix; By performing a linear mapping of the deep features to a fully connected layer, the predicted bending moment time series signal of the tower base section is obtained.

[0031] In this embodiment, the second temporal feature is input into the variable attention mechanism module and mapped to three types of feature vectors: Query (query vector Q), Key (key vector K), and Value (value vector V). An attention weight matrix is ​​generated by calculating the degree of correlation between variables. Then, the Value feature is weighted and fused based on the attention weight matrix to strengthen the dependency relationship of the input variable dimension and realize dual-dimensional modeling of temporal dimension and variable dimension.

[0032] In one embodiment of the present invention, local features are extracted from the input data in the time dimension using time windows of different lengths. The extracted parallel local features are then fused to obtain a first enhanced feature, including: Local features are extracted from the input data using multiple parallel time windows of different lengths. The local features extracted in each time window are subjected to nonlinear activation and random deactivation regularization to obtain multiple regularized local features; The first enhanced feature is obtained by concatenating multiple regularized local features.

[0033] In this embodiment, a parallel convolutional layer module is used to extract local features from the multi-channel input data. For example, this parallel convolutional layer module consists of three independent one-dimensional convolutional branches, each configured with convolutional kernels of sizes 3, 5, and 7, respectively, to cover local correlations at different temporal ranges. Each one-dimensional convolutional branch is followed by a ReLU activation function and a Dropout layer, ultimately fusing the output features of the three branches. The ReLU activation function enhances the model's non-linear expressive power and alleviates the gradient vanishing problem, effectively capturing short, medium, and long-scale local features. The Dropout layer randomly deactivates some neurons (regularization), reducing the risk of overfitting.

[0034] In one specific embodiment, the model's input dimension is batch size × time length × number of channels (B × L × N, where B represents the batch size of a single training run, L corresponds to the time step length of the input time series data, and N is the number of feature channels in the input data); the mean squared error (MSE) is used as the loss function. The Adam optimizer is employed for parameter optimization, and the coefficient of determination R0 is selected. 2 The predictive performance of the model is quantified using the core evaluation index. The dynamic response (simulated acceleration signal and bending moment) of the wind power support structure is simulated using OpenFAST software. The simulation time for a single working condition is 30 minutes, the sampling frequency is 50Hz, and a total of 103 channels and 90,000 data points of time series are generated.

[0035] Since the connection between the tower and the pile is typically the primary deformation location for onshore and offshore wind turbine collapses, the bending moment at this location is chosen for reconstruction. Figure 4 As shown, assuming that a biaxial acceleration sensor is installed at the top of the tower and at node 5 of the tower, the bending moment reconstruction process is to reconstruct the bending moment of the tower base section using the acceleration at these two easily measurable locations. Figure 5This paper presents the acceleration data of the tower top and tower node 5 in the x and y directions, as well as the bending moment of the tower base section, output by OpenFAST under a certain working condition. A comparison of the six signal segments in the figure clearly shows that the x-direction acceleration signal at the tower top and the y-direction bending moment signal at the tower base section exhibit similar fluctuation patterns, while the y-direction acceleration at the tower top and the x-direction bending moment signal at the tower base section also show significant synchronicity. This is because the bending moment of the tower base section is closely related to the structural deformation. For wind turbine towers, the vibration is often dominated by the first mode, and the tower top is the location of the maximum deformation in the first mode. Therefore, the bending moment of the tower base section is closely related to the acceleration at the tower top. This reflects the transmission effect of structural deformation (acceleration) to the foundation load (bending moment), which is consistent with the laws of structural dynamics. It also suggests that the bending moment of the tower base section can be directly reconstructed from the tower top acceleration.

[0036] Compared to the tower top, the acceleration at node 5 of the tower contains higher frequency signal components. This is because the second-order modal deformation in the middle of the tower may be greater than that at the top. Even though the bending moment at the tower base section is dominated by the first-order mode, the acceleration at node 5, which contains more second-order modes, may still play an important role in the reconstruction of the bending moment at the tower base section. Therefore, exploring the reconstruction effect of "using only the tower top acceleration" and "using both the tower top and node 5 acceleration" is of great significance. The working conditions are shown in Table 2, where YawBrTAxp, YawBrTAyp, TwHt5Alxt, and TwHt5Alyt represent the accelerations in the x and y directions at the tower top and node 5, respectively, and TwrBsMxt and TwrBsMyt represent the bending moments in the x and y directions at the tower base section. The definitions of these variable names are consistent with the output files of the OpenFAST software.

[0037] Table 2 During model training, acceleration is used as input and bending moment at the tower base section is used as output. To improve training performance, all generated accelerations and bending moments are normalized to the [0,1] interval, and the mean and standard deviation of each data segment are recorded. After the model generates bending moment data, it is then reverse-normalized to reconstruct the bending moment time history at the actual physical scale. Figure 6 This demonstrates the loss and coefficient of determination R during training. 2 The changing trend indicates that the iTransformer model achieves satisfactory training results after 500 epochs, with a maximum R0. 2 The value is 0.9725.

[0038] After model training is complete, the acceleration responses of three sets of test sea states are input into the trained model for bending moment reconstruction. The bending moment reconstruction effect in the x-direction is as follows: Figure 7 As shown. Figure 7The reconstruction effect using only the single acceleration at the top of the tower is represented by the corresponding reconstruction bending moment R in the x-direction. 2 The values ​​are 0.8767, -0.0861, and -0.2447. The comparison shows that under low wind speed conditions, the model can reconstruct the bending moment value relatively well; however, under high wind speed conditions, due to the instability of structural vibration caused by the load, the reconstruction effect significantly decreases. It should be noted that R0 in this case... 2 A negative value does not mean the bending moment cannot be reconstructed at all, but rather that the reconstructed signal shows a certain degree of overall offset, which has some impact on the accurate assessment of strength or fatigue. After adding the acceleration at node 5 of the tower, the reconstructed bending moment R in the x-direction... 2 All of these results showed a certain degree of improvement, indicating that multi-point response can supplement the spatial distribution information of structural deformation and improve the integrity of bending moment load mapping.

[0039] like Figure 8 The image shown is a reconstruction result obtained by using the acceleration data of the target location and five SCADA parameters as inputs to the iTransformer model in this embodiment. Five SCADA parameters were selected as input variables: wind speed in the x-direction (Wind1VelX), wind speed in the y-direction (Wind1VelY), yaw (NacYaw), rotational speed (RotSpeed), and pitch angle (BldPitch1). Combined with four acceleration data points from the target location: x-direction acceleration at the tower top (YawBrTAxp), y-direction acceleration at the tower top (YawBrTAyp), x-direction acceleration at node 5 of the tower (TwHt5Alxt), and y-direction acceleration at node 5 of the tower (TwHt5Alyt), a total of nine input variables were obtained.

[0040] Figure 8 This demonstrates the moment reconstruction effect in the x-direction. From Figure 8 It can be seen that after further integrating SCADA data such as wind speed and engine speed, R 2 The significant improvement confirms that operating conditions and environmental parameters can provide source information on load generation, effectively compensating for the lack of information in structural response; when using all nine variables, the bending moment reconstruction effect R in the x-direction under sea state 1 condition is significantly improved. 2 The highest value can reach 0.9926. It should be noted that the proposed method adopts the iTransformer model. This model, through the collaborative optimization of parallel convolutional layers, learnable positional encoding, and variable attention mechanism, can not only adaptively learn the dynamic patterns of time series data, but also enhance the feature interaction between multiple variables. Therefore, it can deeply mine the data features between different channels and significantly improve the collaborative prediction effect between acceleration and SCADA data.

[0041] like Figure 2 , Figure 3As shown in the figure, this specification provides a bending moment load prediction device for offshore wind power support structures. The device can be implemented through software, hardware, or a combination of both. From a hardware perspective, such as... Figure 2 The diagram shown is a hardware architecture diagram of an electronic device for predicting the bending moment load of an offshore wind power support structure, as provided in an embodiment of this specification. Except for... Figure 2 In addition to the processor, memory, network interface, and non-volatile memory shown, the electronic device in the embodiment may also include other hardware, such as a forwarding chip responsible for processing packets. Taking software implementation as an example, such as... Figure 3 As shown, a device in a logical sense is formed by the CPU of the electronic device in which it is located reading the corresponding computer program from the non-volatile memory into the memory for execution.

[0042] like Figure 3 As shown in the figure, this embodiment provides a bending moment load prediction device for an offshore wind power support structure, comprising: The data acquisition module 300 collects the operating status parameters of the wind turbine and the actual acceleration at the target location on the wind power support structure in real time. The bending moment prediction module 302 is connected to the data acquisition module. It inputs the operating state parameters and the actual acceleration into the trained bending moment load prediction model to obtain the output bending moment load prediction data of the wind power support structure tower base section.

[0043] The bending moment load prediction model is trained in the following manner: Based on multiple simulated wind condition data and multiple simulated wave data, the simulated operating parameters of the wind turbine, the simulated acceleration at the target location, and the simulated bending moment load of the tower base section are obtained. Local features are extracted from the input data in the time dimension using time windows of different lengths. The extracted parallel local features are then fused to obtain the first enhanced feature. The input data is obtained by concatenating the simulated acceleration and simulated running parameters according to the same timestamp and arranging them in chronological order according to the timestamps. The first position feature vector is combined with the first enhanced feature to obtain the second temporal feature, wherein the first position feature vector is generated based on the dynamic response characteristics of the offshore wind power support structure. Based on the second time series feature, the predicted bending moment time series signal of the tower base section is obtained through variable attention calculation and linear mapping; The model is trained using a loss function constructed based on the difference between the predicted bending moment time series signal and the simulated bending moment load, resulting in a trained bending moment load prediction model.

[0044] In the embodiments described in this specification, the data acquisition module 300 can be used to execute step 100 in the above method embodiments, and the bending moment prediction module 302 can be used to execute step 102 in the above method embodiments.

[0045] In one embodiment of this specification, combining the first positional feature vector with the first enhanced feature to obtain the second temporal feature includes: The first enhanced feature is subjected to learnable position encoding to generate the first position feature vector. The learnable position encoding is iteratively optimized during the training of the bending moment load prediction model with the goal of minimizing the loss function. The learnable position encoding is used as a position feature vector to characterize the dynamic response characteristics of the wind power support structure under different operating stages. The different operating stages include: a preset start-up stage, a rated operation stage, a turbulent condition stage, and a shutdown stage. The first positional feature vector is combined with the first enhanced feature to obtain the second temporal feature.

[0046] In one embodiment of this specification, obtaining the predicted bending moment time series signal of the tower base section based on the second time series characteristics, through variable attention calculation and linear mapping, includes: The second time-series feature is mapped to a query vector Q, a key vector K, and a value vector V; Based on the correlation between the query vector Q and the key vector K, an attention weight matrix is ​​generated between the query vector Q and the key vector K; The value vector V is weighted and fused based on the attention weight matrix to obtain deep features; By performing a linear mapping of the deep features to a fully connected layer, the predicted bending moment time series signal of the tower base section is obtained.

[0047] In one embodiment of this specification, the step of extracting local features from the input data over time windows of different lengths, and fusing the extracted parallel local features to obtain a first enhanced feature, includes: The input data is processed by extracting local features using multiple parallel time windows of different lengths. The local features extracted in each time window are subjected to nonlinear activation and random deactivation regularization to obtain multiple regularized local features; The multiple regularized local features are concatenated to obtain the first enhanced feature.

[0048] In one embodiment of this specification, the bending moment load prediction data includes the bending moment load in the x-direction and the bending moment load in the y-direction of the wind power support structure tower base section; The simulated bending moment load includes the bending moment load in the x-direction and the bending moment load in the y-direction of the base section of the wind power support structure tower.

[0049] In one embodiment of this specification, the operating state parameters include: wind speed in the x-direction, wind speed in the y-direction, yaw angle, rotational speed, and propeller pitch angle.

[0050] In one embodiment of this specification, obtaining the simulated operating parameters of the wind turbine, the simulated acceleration at the target location, and the simulated bending moment load on the tower base section based on multiple simulated wind condition data and multiple simulated wave data includes: Simulated wind data was generated using the IEC Kaimal spectral model, and corresponding simulated wave data was generated using the JONSWAP spectral model. Based on the simulated wind and wave data, the simulated operating parameters of the wind turbine, the simulated acceleration at the target location, and the simulated bending moment load of the tower base section are obtained using the OpenFAST model.

[0051] It is understood that the structures illustrated in the embodiments of this specification do not constitute a specific limitation on a bending moment load prediction device for an offshore wind power support structure. In other embodiments of this specification, a bending moment load prediction device for an offshore wind power support structure may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0052] The information interaction and execution process between the modules in the above-mentioned device are based on the same concept as the method embodiments in this specification, and the specific details can be found in the descriptions in the method embodiments in this specification, so they will not be repeated here.

[0053] This specification also provides an electronic device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements a method for predicting the bending moment load of an offshore wind power support structure according to any embodiment of this specification.

[0054] This specification also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform a method for predicting the bending moment load of an offshore wind power support structure according to any embodiment of this specification.

[0055] Specifically, a system or apparatus equipped with a storage medium may be provided, on which software program code implementing the functions of any of the embodiments described above is stored, and the computer (or CPU or MPU) of the system or apparatus may read and execute the program code stored in the storage medium.

[0056] In this case, the program code read from the storage medium can itself implement the function of any of the above embodiments, and therefore the program code and the storage medium storing the program code constitute a part of this specification.

[0057] Storage media embodiments for providing program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer via a communication network.

[0058] Furthermore, it should be clear that not only can the program code read by the computer be executed, but also the operating system or other components operating on the computer can be instructed based on the program code to perform some or all of the actual operations, thereby realizing the function of any of the embodiments described above.

[0059] Furthermore, it is understood that the program code read from the storage medium is written to the memory set in the expansion board inserted into the computer or to the memory set in the expansion module connected to the computer. Then, based on the instructions of the program code, the CPU or other components installed on the expansion board or expansion module execute some and all of the actual operations, thereby realizing the function of any of the above embodiments.

[0060] It should 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.

[0061] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as ROM, RAM, magnetic disk, or optical disk.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this specification, and are not intended to limit them. Although this specification has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this specification.

Claims

1. A method for predicting the bending moment load of an offshore wind power support structure, characterized in that, include: Real-time acquisition of wind turbine operating status parameters and actual acceleration at target locations on wind power support structures; The operating state parameters and the actual acceleration are input into the trained bending moment load prediction model to obtain the output bending moment load prediction data of the wind power support structure tower base section. The bending moment load prediction model is trained in the following manner: Based on multiple simulated wind condition data and multiple simulated wave data, the simulated operating parameters of the wind turbine, the simulated acceleration at the target location, and the simulated bending moment load of the tower base section are obtained. Local features are extracted from the input data in the time dimension using time windows of different lengths. The extracted parallel local features are then fused to obtain the first enhanced feature. The input data is obtained by concatenating the simulated acceleration and simulated running parameters according to the same timestamp and arranging them in chronological order according to the timestamps. The first position feature vector is combined with the first enhanced feature to obtain the second temporal feature, wherein the first position feature vector is generated based on the dynamic response characteristics of the offshore wind power support structure. Based on the second time series feature, the predicted bending moment time series signal of the tower base section is obtained through variable attention calculation and linear mapping; The model is trained using a loss function constructed based on the difference between the predicted bending moment time series signal and the simulated bending moment load, resulting in a trained bending moment load prediction model.

2. The method according to claim 1, characterized in that, The step of combining the first positional feature vector with the first enhanced feature to obtain the second temporal feature includes: The first enhanced feature is subjected to learnable position encoding to generate the first position feature vector. The learnable position encoding is iteratively optimized during the training of the bending moment load prediction model with the goal of minimizing the loss function. The learnable position encoding is used as a position feature vector to characterize the dynamic response characteristics of the wind power support structure under different operating stages. The different operating stages include: a preset start-up stage, a rated operation stage, a turbulent condition stage, and a shutdown stage. The first positional feature vector is combined with the first enhanced feature to obtain the second temporal feature.

3. The method according to claim 1, characterized in that, The process of obtaining the predicted bending moment time series signal of the tower base section based on the second time series feature, through variable attention calculation and linear mapping, includes: The second time-series feature is mapped to a query vector Q, a key vector K, and a value vector V; Based on the correlation between the query vector Q and the key vector K, an attention weight matrix is ​​generated between the query vector Q and the key vector K; The value vector V is weighted and fused based on the attention weight matrix to obtain deep features; By performing a linear mapping of the deep features to a fully connected layer, the predicted bending moment time series signal of the tower base section is obtained.

4. The method according to claim 1, characterized in that, The step involves extracting local features from the input data over time windows of different lengths, and then fusing the extracted parallel local features to obtain the first enhanced feature, including: The input data is processed by extracting local features using multiple parallel time windows of different lengths. The local features extracted in each time window are subjected to nonlinear activation and random deactivation regularization to obtain multiple regularized local features; The multiple regularized local features are concatenated to obtain the first enhanced feature.

5. The method according to claim 1, characterized in that, The bending moment load prediction data includes the bending moment load in the x-direction and the bending moment load in the y-direction of the base section of the wind power support structure tower. The simulated bending moment load includes the bending moment load in the x-direction and the bending moment load in the y-direction of the base section of the wind power support structure tower.

6. The method according to claim 5, characterized in that, The operating status parameters include: wind speed in the x-direction, wind speed in the y-direction, yaw angle, rotational speed, and propeller pitch angle.

7. The method according to claim 1, characterized in that, The process of obtaining simulated operating parameters of the wind turbine, simulated acceleration at the target location, and simulated bending moment load on the tower base section based on multiple simulated wind condition data and multiple simulated wave data includes: Simulated wind data was generated using the IEC Kaimal spectral model, and corresponding simulated wave data was generated using the JONSWAP spectral model. Based on the simulated wind and wave data, the simulated operating parameters of the wind turbine, the simulated acceleration at the target location, and the simulated bending moment load of the tower base section are obtained using the OpenFAST model.

8. A bending moment load prediction device for an offshore wind power support structure, characterized in that, include: The data acquisition module collects the operating status parameters of the wind turbine and the actual acceleration at the target location on the wind power support structure in real time. The bending moment prediction module is connected to the data acquisition module. It inputs the operating state parameters and the actual acceleration into the trained bending moment load prediction model to obtain the output bending moment load prediction data of the wind power support structure tower base section. The bending moment load prediction model is trained in the following manner: Based on multiple simulated wind condition data and multiple simulated wave data, the simulated operating parameters of the wind turbine, the simulated acceleration at the target location, and the simulated bending moment load of the tower base section are obtained. Local features are extracted from the input data in the time dimension using time windows of different lengths. The extracted parallel local features are then fused to obtain the first enhanced feature. The input data is obtained by concatenating the simulated acceleration and simulated running parameters according to the same timestamp and arranging them in chronological order according to the timestamps. The first position feature vector is combined with the first enhanced feature to obtain the second temporal feature, wherein the first position feature vector is generated based on the dynamic response characteristics of the offshore wind power support structure. Based on the second time series feature, the predicted bending moment time series signal of the tower base section is obtained through variable attention calculation and linear mapping; The model is trained using a loss function constructed based on the difference between the predicted bending moment time series signal and the simulated bending moment load, resulting in a trained bending moment load prediction model.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor, when executing the computer program, implements the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of any one of claims 1-7.

Images