Wind turbine blade damping control method, system and device, storage medium and program product

Abstract

The invention relates to a wind turbine blade damping control method, system and device, a storage medium and a program product. The method comprises the following steps: acquiring blade vibration monitoring data; according to the blade vibration monitoring data, using a pre-trained wind turbine blade damping control model to obtain operation prediction data and damping demand data; wherein the pre-trained wind turbine blade damping control model is obtained by training based on a CNN-LSTM neural network; and according to the operation prediction data and the damping demand data, the effective coincident area between two magnetic poles of an electrorheological damper of the wind turbine blade is adjusted, so that the damping of the electrorheological damper is adjusted by changing the magnetic field intensity of the electrorheological damper. The safety and reliability of operation of the wind turbine can be effectively improved.

Description

Technical Field

[0001] This application relates to the field of wind turbine control technology, and in particular to a wind turbine blade damping control method, system, device, storage medium and program product. Background Technology

[0002] Currently, wind turbine load reduction optimization is mainly achieved through two approaches: one is to adopt electromechanical control systems (such as torque control and pitch control systems), and the other is to integrate passive control devices into the structural design, such as dampers like tuned mass dampers, tuned liquid dampers, and tuned liquid column dampers, to suppress structural vibration response.

[0003] However, the fixed damping value of traditional passive dampers may limit their vibration reduction effect, increasing the risk of wind turbine blade resonance and affecting the safety and reliability of wind turbine operation. Summary of the Invention

[0004] Therefore, it is necessary to provide a wind turbine blade damping control method, system, device, storage medium, and program product that can improve the safety and reliability of wind turbine operation, addressing the aforementioned technical problems.

[0005] In a first aspect, this application provides a method for controlling the damping of wind turbine blades, the method comprising:

[0006] Acquire blade vibration monitoring data;

[0007] Based on blade vibration monitoring data, operational prediction data and damping demand data are obtained using a pre-trained wind turbine blade damping control model; the pre-trained wind turbine blade damping control model is trained based on a CNN-LSTM neural network.

[0008] Based on operational forecast data and damping demand data, the effective overlapping area between the two magnetic poles of the electrorheological damper of the wind turbine blade is adjusted to change the damping of the electrorheological damper by altering its magnetic field strength.

[0009] In one embodiment, the method further includes:

[0010] Obtain blade vibration sample data;

[0011] Preprocessing the blade vibration sample data yields preprocessed blade vibration sample data.

[0012] The preprocessed blade vibration sample data is input into the CNN network of the CNN-LSTM neural network to obtain the spatial features of the blade vibration sample data.

[0013] Spatial features are input into the LSTM network of the CNN-LSTM neural network to train a pre-trained wind turbine blade damping control model.

[0014] In one embodiment, the blade vibration sample data is preprocessed to obtain preprocessed blade vibration sample data, including:

[0015] Denoising of the blade vibration sample data was performed using wavelet transform to obtain the denoised blade vibration sample data.

[0016] The denoised blade vibration sample data was subjected to Min-Max normalization to obtain the preprocessed blade vibration sample data.

[0017] In one embodiment, the CNN network includes a first convolutional layer, a pooling layer, a second convolutional layer, and a flattening layer connected in sequence.

[0018] The convolutional kernels of the first convolutional layer cover 5 time steps and are used to generate 64 feature maps through 64 filters, so that each convolutional kernel in the convolutional layer can detect a local vibration mode.

[0019] Pooling layers are used to reduce the temporal dimension of the output data from the first convolutional layer and suppress noise;

[0020] The second convolutional layer's convolutional kernel covers three time steps and is used to generate 128 feature maps through 128 filters to capture higher-order feature combinations.

[0021] The flattening layer is used to transform the three-dimensional feature tensor output by the second convolutional layer into a two-dimensional tensor for output to the LSTM network.

[0022] In one embodiment, the LSTM network of the CNN-LSTM neural network includes a first LSTM layer and a second LSTM layer;

[0023] The hidden state dimension of the first LSTM layer is 64, which is used to output the state data of each time step of the time sequence based on the temporal sequence corresponding to the spatial features output by the CNN network.

[0024] The second LSTM layer is used to map the state data of the last time step in the state data of each time step into a 32-dimensional vector to obtain a 32-dimensional time sequence code.

[0025] In one embodiment, the method further includes:

[0026] Based on blade vibration monitoring data and a preset reward function, the model parameters corresponding to the pre-trained wind turbine blade damping control model are dynamically updated.

[0027] The preset reward function includes a vibration displacement penalty term, a vibration velocity penalty term, a damping deviation penalty term, and a constant offset term.

[0028] Secondly, this application also provides a wind turbine blade damping control system, the system including a PLC system, a vibration sensor, and an electrorheological damper; the PLC system is connected to the vibration sensor and the electrorheological damper respectively;

[0029] Vibration sensors are used to collect blade vibration monitoring data and transmit the blade vibration monitoring data to the PLC system;

[0030] An electrorheological damper is installed on the wind turbine blade and includes two electromagnetic variable dampers and one tuned mass damper. The electrorheological damper is used to adjust the damping of the electrorheological damper by adjusting the effective area of ​​the overlapping magnetic poles between the two corresponding electromagnetic variable dampers.

[0031] The PLC system is also used to perform the methods described in any of the first aspects.

[0032] Thirdly, this application also provides a wind turbine blade damping control device, the device comprising:

[0033] The data acquisition module is used to acquire blade vibration monitoring data;

[0034] The model processing module is used to obtain operational prediction data and damping demand data based on blade vibration monitoring data and a pre-trained wind turbine blade damping control model; the pre-trained wind turbine blade damping control model is trained based on a CNN-LSTM neural network.

[0035] The data output module is used to adjust the effective area of ​​overlap between the two magnetic poles of the current-routine damper based on the operation prediction data and damping demand data, so as to adjust the damping of the current-routine damper by changing the magnetic field strength of the current-routine damper.

[0036] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the first aspects.

[0037] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in any of the first aspects.

[0038] The aforementioned wind turbine blade damping control method, system, device, storage medium, and program product can obtain operational prediction data and damping demand data based on the collected blade vibration monitoring data and a pre-trained wind turbine blade damping control model based on a CNN-LSTM neural network. Then, based on the obtained operational prediction data and damping demand data, the effective area of ​​overlap between the two magnetic poles of the current-variable damper of the wind turbine blade is adjusted, thereby changing the magnetic field strength of the current-variable damper and adjusting the damping of the current-variable damper. Through the above method, this application can achieve fully automatic control of wind turbine blade damping, thereby significantly improving the safety and reliability of wind turbine operation. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a diagram illustrating the application environment of a wind turbine blade damping control method in one embodiment.

[0041] Figure 2 This is a schematic diagram illustrating the installation effect of an electrorheological damper in a wind turbine blade in one embodiment.

[0042] Figure 3 This is a cross-sectional view of the current-varying damper in one embodiment;

[0043] Figure 4 This is a flowchart illustrating a wind turbine blade damping control method in one embodiment;

[0044] Figure 5 This is a flowchart illustrating the process of training a pre-trained wind turbine blade damping control model in one embodiment.

[0045] Figure 6 This is a schematic diagram of the preprocessing process for blade vibration sample data in one embodiment.

[0046] Figure 7 This is a schematic diagram illustrating the interface process between the output layer of an exemplary CNN-LSTM neural network and a PLC in one embodiment.

[0047] Figure 8 This is a schematic diagram of the online learning process for a wind turbine blade damping control model in one embodiment;

[0048] Figure 9This is a structural block diagram of a wind turbine blade damping control device in one embodiment;

[0049] Figure 10 This is a diagram of the internal structure of an electronic device in one embodiment. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0051] The terms "comprising" and "having," and any variations thereof, as used in this application, are intended to cover non-exclusive inclusion. The term "multiple" as used in this application refers to two or more. The term "and / or" as used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0052] Currently, wind turbine load reduction optimization is mainly achieved through two approaches: one is to adopt electromechanical control systems (such as torque control and pitch control systems), and the other is to integrate passive control devices into the structural design, such as tuned mass dampers (TMD), tuned liquid dampers (TLD), and tuned liquid column dampers (TLCD), to effectively suppress structural vibration response.

[0053] With the continuous development of the wind power industry, wind turbines are becoming larger and larger, and wind turbine blades are also becoming longer, resulting in reduced damping. Therefore, vibration suppression of wind turbine blades has become one of the key technical challenges that the wind power industry urgently needs to solve. However, the damping value of commonly used passive dampers on the market is fixed, which may lead to adverse effects such as limited vibration reduction effect, increased resonance risk, and accelerated structural fatigue.

[0054] Traditional wind turbine blade damping control schemes typically employ shear-mode electrorheological dampers. These dampers are intelligent, adjustable damping devices that utilize the electric field response characteristics of electrorheological fluid (ERF) to rapidly change the damping force through an applied voltage. In shear mode, the ERF is placed between two relatively sliding electrode plates. Users control the fluid's rheological properties by adjusting the electric field strength, thereby dynamically adjusting the damping force.

[0055] However, the above-mentioned shear-mode current-varying dampers still have the following shortcomings in practical applications:

[0056] ① Response Delay and Hysteresis Effect: The theoretical response speed of shear-mode electrorheological dampers is in the millisecond range, but in reality, due to factors such as the particle inertia of electrorheological fluids (ERF) and electrode structure design, a dynamic response delay of 10~100ms may occur. In addition, under high-frequency cyclic loading, ERF may exhibit hysteresis, i.e., the electric field-damping force curves do not coincide, leading to a decrease in the control accuracy of high-frequency vibrations (such as wind turbine blade flutter).

[0057] ② Insufficient yield stress of electrorheological fluids (ERFs): The yield stress of existing ERFs is usually below 10 kPa, which is difficult to meet the high load requirements in practical applications (e.g., large wind turbine blades require damping forces in the tens of kPa range). This forces the damper size to be increased to provide sufficient damping force, increasing weight and cost, and reducing practicality.

[0058] ③ Energy efficiency and power consumption need to be optimized: Although the ER damper has low static power consumption, it needs to maintain high voltage during continuous vibration control, resulting in its overall efficiency still being lower than that of passive dampers, and it will also burden the energy system of off-grid wind turbines. To solve the above problems, this application provides a wind turbine blade damping control method.

[0059] Before introducing the specific embodiments of this application, the technical terms involved in this application will be explained:

[0060] PLC (Programmable Logic Controller): A digital electronic control system designed specifically for industrial environments. It implements functions such as logical operations, sequential control, timing and counting, and arithmetic operations through programming, and is widely used in automated production lines, mechanical equipment control, and other fields.

[0061] CNN (Convolutional Neural Network): A deep learning model designed specifically for processing grid-structured data (such as images and time-series signals), which efficiently extracts spatial features through local receptive fields and weight sharing.

[0062] LSTM (Long Short-Term Memory): It is a special type of recurrent neural network (RNN) designed to solve long-term dependency problems and can effectively learn long-term patterns in time-series data (such as vibration signal evolution, language context, etc.).

[0063] TMD (Tuned Mass Damper): A passive vibration control device that absorbs the vibration energy of the main structure through an added mass-spring-damping system. It is widely used in vibration reduction design in fields such as buildings, bridges, and wind turbine blades.

[0064] The wind turbine blade damping control method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, the PLC (Programmable Logic Controller) system 102 communicates with the server 104 via a communication network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated onto the server 104, or it can be located on a cloud or other network server. The server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The displacement sensor 108 can collect blade vibration monitoring data such as blade vibration frequency and amplitude, and transmit the collected blade vibration monitoring data to the PLC system 102.

[0065] The current-routine damper 106 may include two electromagnetic variable dampers and one tuned mass damper. For example... Figure 2 and Figure 3 As shown, Figure 2 This is a schematic diagram showing the installation effect of the electrorheological damper 106 in a wind turbine blade. Figure 3 This is a cross-sectional view of the current-varying damper 106. Specifically, the current-varying damper 106 consists of a tuned mass damper and two electromagnetic variable dampers. The spherical tuned mass damper (TMD) is fixed inside the blade by steel cables. It mainly utilizes the principle of resonance to tune the frequency of the tuned mass damper (TMD) to match the vibration frequency of the main structure of the wind turbine blade. Through anti-phase motion, it absorbs vibration energy, so that the vibration energy of the main structure is transferred to the mass block of the tuned mass damper (TMD) and dissipated (converted into heat energy) through damping.

[0066] Furthermore, two electromagnetic variable dampers are connected at an acute angle below the tuned mass damper (TMD). The electromagnetic variable damper is encased in a cylindrical shell, with one end directly connected to the surface of the TMD sphere and the other end connected to the inner surface of the wind turbine blade. Inside the cylindrical shell are two negative magnetic poles, which are attached to opposite sides and bent at a 90° angle. The shaft of the electromagnetic variable damper, used to connect to the motor of the PLC system, is fixed to the TMD via shims and rolling bearings passing through the sphere. The other end is connected to a positive magnetic pole via six split shafts (three shafts per group, evenly spaced at 90°, i.e., the angle between two adjacent shafts in a group is 45°). This positive magnetic pole is attached to the opposite side and offset from the negative magnetic pole by 90°, positioned inside the negative magnetic pole (meaning the radius corresponding to the positive magnetic pole is smaller than the radius corresponding to the negative magnetic pole).

[0067] Based on such Figure 3 The rotating sector design shown allows the PLC system to control the rotation angle of the electromagnetic variable damper motor, thereby changing the effective coupling area between the two magnetic poles. This alters the magnetic field strength and, based on Lenz's law, adjusts the damping force of the current-variable damper 106. Specifically, the current-variable damper 106 adjusts the damping force by dynamically changing the effective coupling area between the two electromagnets corresponding to the electromagnetic variable damper, as shown in Formula 1 below. With the effective area covered by the magnetic field Direct proportion:

[0068] (Formula 1)

[0069] In one exemplary embodiment, such as Figure 4 As shown, this application provides a wind turbine blade damping control method. Taking the application of this method to a PLC system 102 as an example, the method includes the following steps S202 to S206. Wherein:

[0070] Step S202: Obtain blade vibration monitoring data.

[0071] For example, blade vibration monitoring data can be collected by vibration sensors installed on wind turbine blades.

[0072] In some examples, blade vibration monitoring data may include blade dynamic parameters such as blade vibration frequency, vibration amplitude, phase angle, vibration mode, dynamic strain, vibration direction, and vibration attenuation rate.

[0073] Specifically, after the vibration sensor collects blade vibration monitoring data such as blade vibration frequency, it transmits this blade vibration monitoring data to the PLC system as the data basis for subsequent model prediction.

[0074] Step S204: Based on the blade vibration monitoring data, the pre-trained wind turbine blade damping control model is used to obtain the operation prediction data and damping demand data.

[0075] Among them, the pre-trained wind turbine blade damping control model was obtained by training based on CNN-LSTM neural network.

[0076] For example, the CNN-LSTM neural network of the wind turbine blade damping control model includes a CNN network and an LSTM network. It can be understood that by utilizing CNN and LSTM networks, the wind turbine blade damping control model can be adapted to process sequential data that integrates spatial and temporal information, such as blade vibration monitoring data.

[0077] Furthermore, the CNN network of the wind turbine blade damping control model can be used to extract spatial features from the input sequence data. The LSTM network of the wind turbine blade damping control model can be used to process the temporal dependencies in the sequence, which can extract information about the current state from previous states and has the ability to predict future states.

[0078] Specifically, the PLC system can obtain operational prediction data and damping demand data based on the blade vibration monitoring data and its embedded pre-trained wind turbine blade damping control model, thereby providing a data foundation for the subsequent damping control process.

[0079] Step S206: Based on the operation prediction data and damping demand data, adjust the effective area of ​​the overlapping magnetic poles between the two magnetic poles of the current-variable damper of the wind turbine blade, so as to adjust the damping of the current-variable damper by changing the magnetic field strength of the current-variable damper.

[0080] Among them, the operation forecast data can characterize whether there will be extreme wind conditions or fatigue damage to the wind turbine blades in the future; the damping demand data can characterize the current damping demand of the wind turbine blades.

[0081] In practical applications, operational prediction data can be used to adjust the current-routine damper to a larger damping value in the event of extreme wind conditions or blade fatigue damage in the future; damping demand data can be used to control the rotation angle of the current-routine damper's motor and change the effective overlapping area between the two magnetic poles in the current-routine damper, so as to adjust the current damping magnitude by changing the magnetic field strength.

[0082] Specifically, the PLC system can adjust the effective area of ​​the overlapping magnetic poles between the two magnetic poles of the electrorheological damper of the wind turbine blade based on the operation prediction data and damping demand data output by the wind turbine blade damping control model, so as to meet the current damping demand of the wind turbine blade, and adjust the damping value of the electrorheological damper in a timely manner in the event of extreme wind conditions or fatigue damage to the wind turbine blade in the future.

[0083] The aforementioned wind turbine blade damping control method acquires blade vibration monitoring data and, based on this data, uses a pre-trained wind turbine blade damping control model to obtain operational prediction data and damping requirement data. Then, based on these data, it adjusts the effective overlapping area between the two magnetic poles of the current-variable damper on the wind turbine blade, thereby changing the damping of the current-variable damper by altering its magnetic field strength. In essence, this application, by designing an embedded AI that combines PLC control with deep learning, enables the automatic adjustment of the damping value of the damper on the wind turbine blade based on blade dynamic parameters such as vibration amplitude, significantly improving the safety and reliability of wind turbine operation.

[0084] In one embodiment, such as Figure 5 As shown, the method further includes steps S302 to S308. Wherein:

[0085] Step S302: Obtain blade vibration sample data.

[0086] For example, blade vibration sample data can be collected by vibration sensors installed on the wind turbine blades. Alternatively, blade vibration sample data can also be directly obtained from historical monitoring data of wind turbine blades recorded in a data storage system.

[0087] In some examples, blade vibration sample data may include blade dynamic parameters such as blade vibration frequency, vibration amplitude, phase angle, vibration mode, dynamic strain, vibration direction, and vibration attenuation rate.

[0088] Specifically, the PLC system can collect blade vibration sample data through vibration sensors installed on the wind turbine blades, which can then serve as the data basis for subsequent model training.

[0089] Step S304: Preprocess the blade vibration sample data to obtain preprocessed blade vibration sample data.

[0090] For example, preprocessing the blade vibration sample data may include denoising the blade vibration sample data to improve the training efficiency and model performance of the wind turbine blade damping control model.

[0091] Specifically, the PLC system can preprocess the blade vibration sample data to obtain preprocessed blade vibration sample data.

[0092] Step S306: Input the preprocessed blade vibration sample data into the CNN network of the CNN-LSTM neural network to obtain the spatial features of the blade vibration sample data.

[0093] For example, CNN networks can utilize multiple convolutional layers to capture features at different levels in the input data through convolutional operations similar to those in image processing, and reduce feature dimensionality through pooling layers to preserve salient features in the data.

[0094] Specifically, the PLC system can input the preprocessed blade vibration sample data into the CNN network of the wind turbine blade damping control model to extract the spatial features of the blade vibration sample data. It can be understood that the CNN network in this embodiment is mainly used to extract vibration waveform features as input to the LSTM network.

[0095] Step S308: Input the spatial features into the LSTM network of the CNN-LSTM neural network to train and obtain a pre-trained wind turbine blade damping control model.

[0096] For example, an LSTM network can learn the short-term vibration dependence of a wind turbine blade, as well as the long-term trend of integrated vibration, based on the temporal sequence corresponding to the spatial features extracted from a CNN network.

[0097] In some examples, the LSTM network of the wind turbine blade damping control model has forget gates, input gates, and output gates to form a standard LSTM structure.

[0098] Specifically, the PLC can input the spatial features extracted by the CNN into the LSTM network of the CNN-LSTM neural network to train a pre-trained wind turbine blade damping control model.

[0099] In one possible implementation, to further illustrate the embodiments of this application, this application provides the following exemplary model training process, as detailed below:

[0100] Every 15 minutes, the PLC system captures a set of blade vibration sample data from the environment via displacement sensors. This data includes seven blade dynamic parameters: vibration frequency, vibration amplitude, phase angle, vibration mode, dynamic strain, vibration direction, and vibration attenuation rate. This data is then used to train a pre-designed initial-state wind turbine blade damping control model through deep learning. Based on this blade vibration sample data, the wind turbine blade damping control model, after deep learning via a CNN-LSTM neural network, yields two types of data: the first part is the current damping required by the wind turbine blades, i.e., damping demand data; the second part is the predicted wind speed and blade dynamic parameters within a set future timeframe (e.g., 10 minutes), i.e., operational prediction data.

[0101] Understandably, based on the damping demand data output by the model, the PLC system can control the rotation angle of the motor in the variable damper, thereby changing the effective overlapping area between the two magnetic poles in the current-ratio damper. This allows the system to adjust the damping magnitude of the damper by changing the magnetic field strength, achieving targeted vibration reduction. Based on the operational prediction data output by the model, the PLC system can predict whether extreme wind conditions or blade fatigue damage will occur within a set timeframe, and adjust the current-ratio damper to a higher damping level in a timely manner based on the prediction results, thereby improving the system's vibration reduction effect.

[0102] In one embodiment, such as Figure 6 As shown, the blade vibration sample data is preprocessed to obtain preprocessed blade vibration sample data, including the following steps S402 to S404. Wherein:

[0103] Step S402: Denoise the blade vibration sample data using wavelet transform to obtain denoised blade vibration sample data.

[0104] It is understandable that wavelet transform can provide localized information of a signal in both the time and frequency domains, and has a wider range than Fourier transform, making it particularly suitable for analyzing non-stationary signals such as vibration.

[0105] For example, when using wavelet transform to denoise blade vibration sample data, wavelet transform can provide high temporal resolution in the high-frequency band and high frequency resolution in the low-frequency band of the blade vibration sample data by adjusting the size of the variable window, thereby achieving data denoising.

[0106] Specifically, the PLC system can use wavelet transform to denoise the blade vibration sample data to obtain denoised blade vibration sample data.

[0107] Step S404: Perform Min-Max normalization on the denoised blade vibration sample data to obtain preprocessed blade vibration sample data.

[0108] It is understandable that, due to the inconsistent units, large data volume, and wide data range of the acquired blade vibration sample data, this embodiment of the application performs Min-Max normalization processing on these blade vibration sample data with a default range of [0,1]. Min-Max normalization processing can maintain the relative relationship of the original data, and if there are abnormal sample data values, other normal data may be compressed into a narrow range after normalization processing, while abnormal sample data values ​​are significantly distinguished by other ranges.

[0109] It should be noted that Min-Max normalization is a linear data scaling method, as shown in Equation 2, which can map the original data to a specified range:

[0110] (Formula 2)

[0111] in, The sample data after normalization. For the sample data to be processed, The smallest sample data in the sample dataset. It is the largest sample data in the sample dataset.

[0112] Specifically, the PLC system can perform Min-Max normalization on the denoised blade vibration sample data to obtain preprocessed blade vibration sample data.

[0113] In one embodiment, the CNN network includes a first convolutional layer, a pooling layer, a second convolutional layer, and a flattening layer connected in sequence.

[0114] The convolutional kernels of the first convolutional layer cover 5 time steps and are used to generate 64 feature maps through 64 filters, so that each convolutional kernel in the convolutional layer can detect a local vibration mode.

[0115] Pooling layers are used to reduce the temporal dimension of the output data from the first convolutional layer and suppress noise;

[0116] The second convolutional layer's convolutional kernel covers three time steps and is used to generate 128 feature maps through 128 filters to capture higher-order feature combinations.

[0117] The flattening layer is used to transform the three-dimensional feature tensor output by the second convolutional layer into a two-dimensional tensor for output to the LSTM network.

[0118] Specifically, the first convolutional layer (the first layer of the CNN network) generates 64 different feature maps, with its convolutional kernels covering 5 time steps and activating the ReLU function to introduce non-linearity (e.g., suppressing negative values ​​through ReLU:max(0,x)). It can be understood that the first convolutional layer enables each convolutional kernel to detect a local vibrational pattern in the sample data, and through 64 filters, 64 different feature maps can be generated.

[0119] Pooling layers (the second layer of a CNN network) reduce the temporal dimension of feature data by taking the maximum value of the data every two time steps, based on the output data of the first convolutional layer. This preserves the significant features in the data, effectively reduces the computational cost of the model, and suppresses noise.

[0120] The second convolutional layer (the third layer of the CNN network) increases to 128 filters to capture higher-order feature combinations based on the pooling layer output data. It's worth noting that in this second convolutional layer, the kernel time step is reduced to 3, thus making the temporal patterns more accurate.

[0121] The unfolded layer (the fourth layer of the CNN network) transforms the three-dimensional feature tensor output by the second convolutional layer into a two-dimensional tensor, which is then used as input to the subsequent LSTM network. It can be understood that in the CNN-LSTM neural network, the main role of the CNN network is to extract the vibration waveform (i.e., spatial features) of the wind turbine blades from the sample data, using it as input to the LSTM network.

[0122] In one embodiment, the LSTM network of the CNN-LSTM neural network includes a first LSTM layer and a second LSTM layer;

[0123] The hidden state dimension of the first LSTM layer is 64, which is used to output the state data of each time step of the time sequence based on the temporal sequence corresponding to the spatial features output by the CNN network.

[0124] The second LSTM layer is used to map the state data of the last time step in the state data of each time step into a 32-dimensional vector to obtain a 32-dimensional time sequence code.

[0125] For example, 32-dimensional time-series coding can be used to characterize the feature abstraction of damping demand data and the feature abstraction of running prediction data.

[0126] In practical applications, before inputting the spatial features output by the CNN network into the first LSTM layer, the feature dimension of the spatial features output by the CNN network is split into a preset number of segments, thereby transforming the spatial local features extracted by the CNN into a temporal sequence. Optionally, the preset number of segments can be set to 10 segments.

[0127] Specifically, the PLC system outputs the state at each time step through a first-layer LSTM with a hidden state dimension of 64, and then further reduces the feature dimension through a second-layer LSTM to extract a higher-level temporal abstraction, finally outputting a 32-dimensional temporal code.

[0128] In one possible implementation, the process of docking the output layer of the CNN-LSTM neural network with the PLC is as follows: Figure 7 As shown. It can be understood that after the neural network outputs a 32-dimensional timing code, it needs to be converted into physical parameters executable by the PLC through output range mapping. For example, assuming the current control current range of the damper is 0-5A, the output range mapping process is shown in Formula 3 below:

[0129] (Formula 3)

[0130] in, This refers to the current command output by the PLC system to the current-varying damper.

[0131] In some examples, it is also necessary to convert the model's floating-point output to an integer format supported by the PLC before data matching. This data matching specifically includes: ① Matching data types with register ranges: matching the floating-point output of the model with the PLC's holding registers, and matching the PLC's integer instructions with the input range of the PLC's analog output modules; ② Matching memory addresses: matching the variables in the model program (usually a Python program) with the absolute addresses of the holding registers.

[0132] In practical applications, after the wind turbine blade damping control model is calculated using CNN-LSTM, the output data can be converted into damping parameters (0-10000) via Modbus TCP communication and written into the holding register (also known as the Modbus register) of the PLC system. The damping parameters are a standardized instruction value used in communication between the model program (i.e., the wind turbine blade damping control model) and the PLC instruction execution program; these parameters correspond to the pitch, yaw, and speed parameters used for actively suppressing vibration.

[0133] For example, when the wind turbine blade damping control model calculates that a moderately high damping is required, a damping parameter of 7500 is generated and written to the Modbus register. The PLC instruction executor then reads this value of 7500 from the Modbus register and performs a matching conversion. Specifically, the PLC instruction executor linearly maps this damping parameter, which has a value range of 0-10000, to a current command (e.g., a control current of 0-5A) that the electrorheological damper can execute. For example, the calculation process for the damping parameter mapping is shown in Formula 4 below:

[0134] (Formula 4)

[0135] Where I represents the control current output by the PLC system to the current-varying damper. Taking a damping parameter of 7500 as an example, the corresponding control current output by the PLC system is 3.75A. Optionally, the PLC system can be a Siemens S7-1200.

[0136] In one embodiment, the method further includes the following steps:

[0137] Based on blade vibration monitoring data and a preset reward function, the model parameters of the pre-trained wind turbine blade damping control model are dynamically updated.

[0138] The preset reward function may include a vibration displacement penalty term, a vibration velocity penalty term, a damping deviation penalty term, and a constant offset term.

[0139] For example, the reward function can be used to define what goal the reinforcement learning model is expected to achieve. For the smart damper control problem, the objective of the reward function could be set as adaptively adjusting the system's damping value under dynamic wind loads.

[0140] Understandable, such as Figure 8 As shown, the pre-trained wind turbine blade damping control model can also be learned online. It can continuously and dynamically update the model parameters based on the actual vibration suppression effect fed back by the displacement sensor, thereby making the model more accurate.

[0141] Specifically, as shown in Formula 5 below, the reward function It is the "feedback" signal obtained by the agent after taking an action at each time step t, based on the reward function. The parameters of the wind turbine blade damping control model can be continuously adjusted to obtain good vibration reduction feedback, and the aforementioned reward effect can be accumulated.

[0142] (Formula 5)

[0143] in, This represents the weight of the displacement penalty term (usually a negative value). This represents the structural displacement response at time step t. This represents the weight of the speed penalty term (usually a negative value). This represents the structural velocity response at time step t. This represents the weight of the damping deviation penalty term (usually a negative value). This represents the damping ratio of the actual output of the reinforcement learning model at time step t. This represents the weight of the constant offset or the excitation term (which may be positive). This indicates the nominal damping ratio or the reference damping parameter. It should be noted that... This can be a pre-set fixed damping value that performs reasonably well under normal wind loads, or a safe and energy-efficient baseline operating point. Optionally, the current damping value can be set to 0.05 when initializing the model.

[0144] Furthermore, the above reward function It can be broken down into the following parts:

[0145] ① Vibration displacement penalty term: This penalty term characterizes the degree to which the structure deviates from its original equilibrium position. The larger the term, the greater the structural amplitude, and the more necessary the penalty becomes. It can be understood that the vibration displacement penalty term is the most fundamental control objective of this reward function and the primary target for suppression.

[0146] ② Vibration velocity penalty term: From an energy perspective, velocity is directly related to kinetic energy. Especially when the system is about to enter the resonance region, a rapid increase in velocity is a warning signal that needs attention. Therefore, setting a vibration velocity penalty term helps to tighten the damping of the system in advance.

[0147] ③ Damping deviation penalty term (i.e., damping adjustment overhead): This penalty term represents the amount of change between the current action and the previous action. It can be used to suppress unnecessary frequent adjustments, thereby ensuring smooth controller behavior and reducing system "oscillation" or wear of mechanical parts.

[0148] ④ Constant offset or excitation term (i.e., penalty for deviating from the "normal damping" range): Damping cannot be too large or too small for a long time. In actual application scenarios, the heat dissipation capacity of the damper in a real system is limited. Therefore, an ideal operating range can be set, and the damper can be kept as close to this range as possible to improve system reliability.

[0149] It is understandable that in the above reward function, , This determines the level of attention users pay to "vibrational energy"; The larger the size, the more effort is put into smooth control; The larger the value, the greater the emphasis on long-term reliability rather than immediate performance. Optionally, in practical applications, since displacement is the dominant factor, velocity is the secondary factor, and the adjustment amplitude penalty and deviation penalty are less important, the initial values ​​for each weight in the reward function can be set as follows: =1.0, =0.5, =0.1, =0.1.

[0150] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0151] In one exemplary embodiment, this application also provides a wind turbine blade damping control system, the system including a PLC system, a vibration sensor, and an electrorheological damper; the PLC system is connected to the vibration sensor and the electrorheological damper respectively.

[0152] The vibration sensor is used to collect blade vibration monitoring data and transmit the blade vibration monitoring data to the PLC system; the current-varying damper is installed on the wind turbine blade and includes two electromagnetic variable dampers and one tuned mass damper; the current-varying damper is used to adjust the damping of the current-varying damper by adjusting the effective area of ​​the overlapping magnetic poles between the two corresponding electromagnetic variable dampers; the PLC system is also used to execute the wind turbine blade damping control method as described in any of the above method embodiments.

[0153] The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations in the embodiments of this application can be found in the limitations of the wind turbine blade damping control method described above, and will not be repeated here.

[0154] It is understood that, compared to traditional damping control systems, this application possesses at least the following beneficial technical effects:

[0155] ① This application has a faster response speed, and by relying on the real-time control of the PLC system, the timeliness of the system response is significantly improved.

[0156] ②This application no longer uses ERF, and since there is no liquid, it avoids the liquid viscosity characteristics and does not have the problem of insufficient yield stress of electrorheological fluids.

[0157] ③ This application does not require maintaining a high-pressure state, and due to the absence of energy loss caused by liquid damping, the overall control efficiency of the system is high.

[0158] Based on the same inventive concept, this application also provides a wind turbine blade damping control device for implementing the wind turbine blade damping control method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more wind turbine blade damping control device embodiments provided below can be found in the limitations of the wind turbine blade damping control method described above, and will not be repeated here.

[0159] In one exemplary embodiment, such as Figure 9 As shown, a wind turbine blade damping control device 900 is provided, comprising:

[0160] Data acquisition module 902 is used to acquire blade vibration monitoring data;

[0161] The model processing module 904 is used to obtain operation prediction data and damping demand data based on the blade vibration monitoring data and a pre-trained wind turbine blade damping control model; wherein, the pre-trained wind turbine blade damping control model is trained based on a CNN-LSTM neural network.

[0162] The data output module 906 is used to adjust the effective area of ​​overlap between the two magnetic poles of the current-routine damper based on the operation prediction data and damping demand data, so as to adjust the damping of the current-routine damper by changing the magnetic field strength of the current-routine damper.

[0163] In one embodiment, the device 900 further includes:

[0164] The model training module is used to acquire blade vibration sample data; preprocess the blade vibration sample data to obtain preprocessed blade vibration sample data; input the preprocessed blade vibration sample data into the CNN network of the CNN-LSTM neural network to obtain spatial features for the blade vibration sample data; input the spatial features into the LSTM network of the CNN-LSTM neural network to train a pre-trained wind turbine blade damping control model.

[0165] In one embodiment, the model training module is further configured to denoise the blade vibration sample data using wavelet transform to obtain denoised blade vibration sample data; and to perform Min-Max normalization on the denoised blade vibration sample data to obtain preprocessed blade vibration sample data.

[0166] In one embodiment, the CNN network includes a first convolutional layer, a pooling layer, a second convolutional layer, and a flattening layer connected in sequence. The convolutional kernels of the first convolutional layer cover 5 time steps and are used to generate 64 feature maps through 64 filters, so that each convolutional kernel in the convolutional layer detects a local vibrational pattern. The pooling layer is used to reduce the temporal dimension of the output data of the first convolutional layer and suppress noise. The convolutional kernels of the second convolutional layer cover 3 time steps and are used to generate 128 feature maps through 128 filters to capture higher-order feature combinations. The flattening layer (also known as the unfolding layer) is used to transform the three-dimensional feature tensor output by the second convolutional layer into a two-dimensional tensor for output to the LSTM network.

[0167] In one embodiment, the LSTM network of the CNN-LSTM neural network includes a first LSTM layer and a second LSTM layer; the hidden state dimension of the first LSTM layer is 64, which is used to output the state data of each time step corresponding to the time sequence according to the time sequence corresponding to the spatial features output by the CNN network; the second LSTM layer is used to map the state data of the last time step in the state data of each time step to a 32-dimensional vector to obtain a 32-dimensional time sequence code.

[0168] In one embodiment, the device 900 further includes:

[0169] The reward module is used to dynamically update the model parameters corresponding to the pre-trained wind turbine blade damping control model based on the blade vibration monitoring data and a preset reward function. The preset reward function includes a vibration displacement penalty term, a vibration velocity penalty term, a damping deviation penalty term, and a constant offset term.

[0170] Each module in the aforementioned wind turbine blade damping control device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0171] In one exemplary embodiment, an electronic device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 10As shown, this electronic device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores blade vibration monitoring data and blade vibration sample data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a wind turbine blade damping control method.

[0172] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0173] In one embodiment, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0174] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0175] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0176] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0177] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0178] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method of damping control of a wind turbine blade, characterized by, The method comprises: obtaining blade vibration monitoring data; obtaining operation prediction data and damping demand data by using a pre-trained wind turbine blade damping control model based on the blade vibration monitoring data; wherein the pre-trained wind turbine blade damping control model is trained based on a CNN-LSTM neural network; adjusting the effective area of overlap between the two magnetic poles of the current variable damper of the wind turbine blade to adjust the damping of the current variable damper by changing the magnetic field strength of the current variable damper according to the operation prediction data and the damping demand data.

2. The method of claim 1, wherein, The method further comprises: obtaining blade vibration sample data; preprocessing the blade vibration sample data to obtain preprocessed blade vibration sample data; inputting the preprocessed blade vibration sample data into the CNN network of the CNN-LSTM neural network to obtain spatial features for the blade vibration sample data; training the pre-trained wind turbine blade damping control model by inputting the spatial features into the LSTM network of the CNN-LSTM neural network.

3. The method of claim 2, wherein, The preprocessing of the blade vibration sample data to obtain preprocessed blade vibration sample data comprises: denoising the blade vibration sample data using wavelet transform to obtain denoised blade vibration sample data; performing Min-Max normalization on the denoised blade vibration sample data to obtain the preprocessed blade vibration sample data.

4. The method of claim 2, wherein, The CNN network comprises a first convolutional layer, a pooling layer, a second convolutional layer and a flattening layer connected in sequence; the convolutional kernel of the first convolutional layer covers 5 time steps, and is used to generate 64 feature maps through 64 filters, so that each convolutional kernel in the convolutional layer detects one local vibration mode; the pooling layer is used to reduce the time dimension of the output data of the first convolutional layer and suppress noise; the convolutional kernel of the second convolutional layer covers 3 time steps, and is used to generate 128 feature maps through 128 filters to capture higher-order feature combinations; the flattening layer is used to convert the three-dimensional feature tensor output by the second convolutional layer into a two-dimensional tensor for output to the LSTM network.

5. The method of claim 2, wherein, The LSTM network of the CNN-LSTM neural network comprises a first LSTM layer and a second LSTM layer; the hidden state dimension of the first LSTM layer is 64, and is used to output state data of each time step corresponding to a time sequence corresponding to the spatial features output by the CNN network; the second LSTM layer is used to map the state data of the last time step in the state data of each time step to a 32-dimensional vector to obtain a 32-dimensional time sequence encoding.

6. The method according to any one of claims 1 to 5, characterized in that, The method further comprises: dynamically updating the model parameters corresponding to the pre-trained wind turbine blade damping control model based on a preset reward function according to the blade vibration monitoring data; wherein the preset reward function comprises a vibration displacement penalty term, a vibration speed penalty term, a damping deviation penalty term and a constant offset term.

7. A wind turbine blade damping control system, characterised in that, The system comprises a PLC system, a vibration sensor, and an electro-rheological damper; the PLC system is connected to the vibration sensor and the electro-rheological damper respectively; The vibration sensor is used to collect blade vibration monitoring data and transmit the blade vibration monitoring data to the PLC system; The electro-rheological damper is arranged on a wind turbine blade and comprises two electromagnetic variable dampers and a tuned mass damper; the electro-rheological damper is used to adjust the damping of the electro-rheological damper by adjusting the effective area of overlap between two magnetic poles corresponding to the two electromagnetic variable dampers; The PLC system is also used to perform the method of any one of claims 1 to 6.

8. A wind turbine blade damping control device, characterised in that, The device comprises: a data acquisition module configured to obtain blade vibration monitoring data; a model processing module configured to obtain operation prediction data and damping requirement data by using a pre-trained wind turbine blade damping control model based on the blade vibration monitoring data; wherein the pre-trained wind turbine blade damping control model is trained based on a CNN-LSTM neural network; a data output module configured to adjust the effective area of overlap between two magnetic poles of an electro-rheological damper based on the operation prediction data and the damping requirement data, so as to adjust the damping of the electro-rheological damper by changing the magnetic field strength of the electro-rheological damper.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the steps of the method of any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program, when executed by a processor, implements the steps of the method of any one of claims 1 to 6.

Images