Mpc-based offshore wind turbine yaw control method, device, equipment and medium

Abstract

The application provides an offshore wind turbine yaw control method, device, equipment and medium based on MPC. The control method comprises: performing multi-time scale average value processing and variational mode decomposition on the acquired historical wind speed and wind direction data, training the wind speed and wind direction training data based on a convolutional neural network, a long short-term memory network and a self-attention mechanism layer, and constructing a wind speed and wind direction prediction model; predicting the average wind speed and wind direction of the next period from the current wind speed and wind direction and the prediction model; training the predicted average wind speed and wind direction and the current cabin angle to construct an MPC prediction control model; substituting the current angular velocity and angle into the prediction control model for prediction and optimization to obtain a target yaw angular velocity and a target yaw angle; and adjusting the angular velocity to the target yaw angular velocity and adjusting the heading angle to the target yaw angle. The method can accurately predict the wind speed and wind direction, quickly respond to the wind direction change, and improve the precision of the yaw control.

Description

Technical Field

[0001] This invention relates to the field of yaw control technology, and in particular to a yaw control method, device, equipment and medium for offshore wind turbines based on MPC. Background Technology

[0002] In the field of wind power generation, yaw control is a core technology to ensure the efficient and stable operation of wind turbine generators. Wind speed and direction information is captured by wind direction sensors and transmitted to the yaw controller. After receiving the data, the yaw controller comprehensively considers key factors such as the operating status of the wind turbine generator and accurately calculates the required turning angle. Subsequently, the yaw controller issues a turning command to the wind direction actuator, which then adjusts the orientation of the wind turbine's fuselage and rotor according to the command, ensuring that the turbine blades are always aligned with the wind direction.

[0003] Traditional yaw control methods are mainly divided into yaw control methods based on traditional wind direction sensors and yaw control methods based on wind direction prediction. Yaw control methods based on traditional wind direction sensors typically rely on real-time wind direction sensor data, but this data can be affected by environmental noise, sensor accuracy limitations, and data transmission delays, resulting in low accuracy. Yaw control methods based on wind direction prediction, on the other hand, adjust the wind turbine's orientation in advance by predicting future wind direction changes. However, since this method relies on historical data to predict future wind direction changes, this prediction may contain errors. When the actual wind direction deviates from the predicted result, traditional yaw control methods based on wind direction prediction may not be able to make timely adjustments, leading to a discrepancy between the wind turbine's orientation and the actual wind direction. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art. To this end, this invention proposes a yaw control method, device, equipment, and medium for offshore wind turbines based on MPC, which can accurately predict wind speed and direction, quickly respond to changes in wind direction, and improve the accuracy of yaw control.

[0005] In a first aspect, embodiments of the present invention provide a yaw control method for offshore wind turbines based on MPC, comprising:

[0006] Acquire multiple sets of historical wind speed data and multiple sets of historical wind direction data;

[0007] The multiple sets of historical wind speed data and the multiple sets of historical wind direction data are first processed by averaging multiple time scales, and then variational mode decomposition is performed to obtain multiple sets of wind speed training data and multiple sets of wind direction training data corresponding to different frequency terms at different time scales.

[0008] Based on convolutional neural networks, long short-term memory networks, and self-attention mechanisms, multiple sets of wind speed training data and multiple sets of wind direction training data corresponding to different frequency terms at different time scales are used to train and construct models to obtain wind speed and wind direction prediction models.

[0009] Based on the current wind speed and the wind speed and direction prediction model, predict the average wind speed for the next period; based on the current wind direction and the wind speed and direction prediction model, predict the average wind direction angle for the next period.

[0010] The model is trained and constructed based on the average wind speed of the next time period, the average wind direction angle of the next time period, and the orientation angle of the offshore wind turbine nacelle at the current moment, to obtain the MPC predictive control model.

[0011] Based on a preset time step, the current cabin rotation angular velocity and the current cabin orientation angle are substituted into the MPC predictive control model for prediction and optimization to obtain the target yaw angular velocity and the target yaw angle.

[0012] Based on the target yaw rate and the target yaw angle, the angular velocity of the offshore wind turbine is adjusted to the target yaw rate, and the orientation angle of the offshore wind turbine is adjusted to the target yaw angle.

[0013] In some embodiments of the present invention, the step of first performing multi-time-scale averaging on the multiple sets of historical wind speed data and the multiple sets of historical wind direction data, and then performing variational mode decomposition to obtain multiple sets of wind speed training data and multiple sets of wind direction training data corresponding to different frequency terms at different time scales includes:

[0014] The multiple sets of historical wind speed data and the multiple sets of historical wind direction data are respectively processed by multi-time scale averaging to obtain multiple sets of average wind speed data and multiple sets of average wind direction data corresponding to different time scales.

[0015] The multiple sets of average wind speed data and the multiple sets of average wind direction data corresponding to different time scales are calculated based on the variational mode decomposition formula to obtain the wind speed frequency term data and wind direction frequency term data corresponding to different time scales.

[0016] The wind speed frequency data and wind direction frequency data corresponding to different time scales are processed using a sliding window scrolling method to obtain multiple sets of wind speed training data and multiple sets of wind direction training data corresponding to different frequency terms at different time scales.

[0017] In some embodiments of the present invention, the step of performing multi-time-scale averaging processing on the multiple sets of historical wind speed data and the multiple sets of historical wind direction data to obtain multiple sets of average wind speed data and multiple sets of average wind direction data corresponding to different time scales includes:

[0018] The multiple sets of historical wind speed data and the multiple sets of historical wind direction data are divided according to different time scales to obtain multiple sets of first wind speed data and multiple sets of first wind direction data corresponding to different time scales.

[0019] The multiple sets of first wind speed data corresponding to different time scales are averaged to obtain multiple sets of average wind speed data corresponding to different time scales.

[0020] The multiple sets of first wind direction data corresponding to different time scales are averaged to obtain multiple sets of average wind direction data corresponding to different time scales.

[0021] In some embodiments of the present invention, the wind speed frequency term data includes wind speed high-frequency term data, wind speed mid-frequency term data, wind speed low-frequency term data, and wind speed residual term data; the wind direction frequency term data includes wind direction high-frequency term data, wind direction mid-frequency term data, wind direction low-frequency term data, and wind direction residual term data;

[0022] The variational mode decomposition formula is:

[0023]

[0024] Where k is a positive integer, representing the number of categories of frequency data; u k For each mode function, namely the intrinsic mode function of the high-frequency, mid-frequency, and low-frequency wind speed data, and the high-frequency, mid-frequency, and low-frequency wind direction data; ω k δ(t) represents the center frequency of each intrinsic mode function; δ(t) represents the Dirac function; f represents the average wind speed data or average wind direction data; the wind speed residual data is the frequency data remaining after subtracting the extracted high-frequency, mid-frequency, and low-frequency wind speed data from the average wind speed data; the wind direction residual data is the frequency data remaining after subtracting the extracted high-frequency, mid-frequency, and low-frequency wind direction data from the average wind direction data.

[0025] In some embodiments of the present invention, the wind speed frequency data and wind direction frequency data corresponding to different time scales are processed using a sliding window scrolling method to obtain multiple sets of wind speed training data and multiple sets of wind direction training data corresponding to different frequency terms at different time scales, including:

[0026] Set the sliding window step size;

[0027] The sliding window step size is slid over the wind speed frequency data at different time scales until all the wind speed frequency data at all time scales are traversed, resulting in multiple sets of wind speed training data for different frequency terms at different time scales.

[0028] The sliding window step size is moved across the wind direction frequency data at different time scales until all the wind direction frequency data at all time scales are traversed, resulting in multiple sets of wind direction training data for different frequency terms corresponding to different time scales.

[0029] In some embodiments of the present invention, the step of training and constructing a model based on multiple sets of wind speed training data and multiple sets of wind direction training data corresponding to different frequency terms at different time scales, using a convolutional neural network, a long short-term memory network, and a self-attention mechanism layer, to obtain a wind speed and wind direction prediction model includes:

[0030] Normalization processing is performed on the multiple sets of wind speed training data and the multiple sets of wind direction training data corresponding to different frequency terms at different time scales to obtain multiple sets of standard wind speed data and multiple sets of standard wind direction data corresponding to different frequency terms at different time scales.

[0031] The convolutional neural network is used to extract and expand features from the multiple sets of standard wind speed data and the multiple sets of standard wind direction data corresponding to different frequency terms at different time scales, so as to obtain one-dimensional time series data of wind speed and one-dimensional time series data of wind direction corresponding to different frequency terms at different time scales.

[0032] The Long Short-Term Memory network is used to predict the one-dimensional time series data of wind speed and the one-dimensional time series data of wind direction corresponding to different frequency terms at different time scales, respectively, to obtain the first predicted data of wind speed and the first predicted data of wind direction corresponding to different frequency terms at different time scales.

[0033] The self-attention mechanism layer performs feature learning on the first wind speed prediction data and the first wind direction prediction data of different frequencies corresponding to different time scales to obtain the second wind speed prediction data and the second wind direction prediction data of different frequencies corresponding to different time scales.

[0034] The second predicted wind speed data and the second predicted wind direction data corresponding to different frequency terms at different time scales are inversely normalized and then added together to obtain the predicted average wind speed and predicted average wind direction at different time scales.

[0035] A wind speed and wind direction prediction model is constructed based on the predicted average wind speed and the predicted average wind direction at different time scales.

[0036] In some embodiments of the present invention, the step of substituting the current cabin rotation angular velocity and the current cabin orientation angle into the MPC predictive control model for prediction and optimization based on a preset time step to obtain the target yaw angular velocity and the target yaw angle includes:

[0037] Based on a preset time step, the current cabin rotation angular velocity and the current cabin orientation angle are substituted into the MPC predictive control model for prediction to obtain a predicted state sequence; wherein, the predicted state sequence includes multiple predicted state data corresponding to the preset time step, and each predicted state data includes the predicted values ​​of the cabin rotation angular velocity and orientation angle in the next time period under ideal conditions.

[0038] Evolutionary calculations are performed based on the predicted state sequence and the MPC predictive control model to obtain multiple control input sequences corresponding to the predicted state sequence; wherein, the control input sequence contains multiple control input data corresponding to a preset time step, and each control input data includes the rotational angular velocity and orientation angle that the cabin needs to take in the next time period under the actual state;

[0039] Based on the multiple control input sequences and the preset cost function, optimization calculations are performed, and the control input sequence that minimizes the cost function value is determined as the target control input sequence.

[0040] The cabin rotation angular velocity corresponding to the first control input data in the target control input sequence is determined as the target yaw angular velocity, and the cabin orientation angle corresponding to the first control input data in the target control input sequence is determined as the target yaw angle.

[0041] Secondly, embodiments of the present invention provide an MPC-based yaw control device for offshore wind turbines, including at least one control processor and a memory for communicatively connecting with the at least one control processor; the memory stores instructions executable by the at least one control processor, which are executed by the at least one control processor to enable the at least one control processor to perform the MPC-based yaw control method for offshore wind turbines as described in the first aspect above.

[0042] Thirdly, embodiments of the present invention provide an electronic device including an MPC-based yaw control device for offshore wind turbines as described in the second aspect above.

[0043] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer-executable instructions for executing the MPC-based yaw control method for offshore wind turbines as described in the first aspect above.

[0044] The MPC-based yaw control method for offshore wind turbines according to embodiments of the present invention has at least the following beneficial effects:

[0045] Acquire multiple sets of historical wind speed and wind direction data; for each set of historical wind speed and wind direction data, first perform multi-time-scale averaging, then perform variational mode decomposition to obtain multiple sets of wind speed and wind direction training data corresponding to different frequency terms at different time scales; based on convolutional neural networks, long short-term memory networks, and self-attention mechanisms, train and construct models from these multiple sets of wind speed and wind direction training data corresponding to different frequency terms at different time scales to obtain a wind speed and wind direction prediction model; based on the current wind speed and the wind speed and wind direction prediction model, predict the average wind speed for the next time period; based on the current wind direction and... A wind speed and direction prediction model is used to predict the average wind direction angle for the next time period. Based on the average wind speed, average wind direction angle, and the current orientation angle of the offshore wind turbine nacelle for the next time period, the model is trained and constructed to obtain the MPC predictive control model. Based on a preset time step, the current nacelle rotation angular velocity and current nacelle orientation angle are substituted into the MPC predictive control model for prediction and optimization to obtain the target yaw angular velocity and target yaw angle. Based on the target yaw angular velocity and target yaw angle, the angular velocity of the offshore wind turbine is adjusted to the target yaw angular velocity, and the orientation angle of the offshore wind turbine is adjusted to the target yaw angle. According to the technical solution of this embodiment, by constructing a VMD-SA-CNN-LSTM wind speed and direction prediction model and an MPC predictive control model, wind speed and direction can be accurately predicted, wind direction changes can be responded to quickly, and the accuracy of yaw control can be improved. Attached Figure Description

[0046] Figure 1 This is a flowchart of an MPC-based yaw control method for offshore wind turbines provided in one embodiment of the present invention;

[0047] Figure 2 yes Figure 1 Flowchart of step S12;

[0048] Figure 3 yes Figure 2 Flowchart of step S21;

[0049] Figure 4 yes Figure 2 Flowchart of step S23;

[0050] Figure 5 yes Figure 1 Flowchart of step S13;

[0051] Figure 6 This is a diagram of the VMD-SA-CNN-LSTM wind speed and direction prediction model structure;

[0052] Figure 7 yes Figure 1 Flowchart of step S16;

[0053] Figure 8 This is a structural diagram of an MPC-based yaw control device for offshore wind turbines provided in another embodiment of the present invention. Detailed Implementation

[0054] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0055] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0056] In the description of this invention, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0057] In the description of this invention, unless otherwise explicitly defined, terms such as "set up," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0058] This invention provides a yaw control method, a yaw control device, and a computer-readable storage medium for offshore wind turbines based on MPC (Multi-Mode Control). The MPC-based yaw control method includes: processing historical wind speed and direction data using multi-timescale averaging and variational mode decomposition; training the wind speed and direction data using a convolutional neural network, long short-term memory network, and self-attention mechanism layer to construct a wind speed and direction prediction model; predicting the average wind speed and direction for the next time period using the current wind speed and direction and the prediction model; training the model with the predicted average wind speed and direction and the current nacelle angle to construct an MPC predictive control model; substituting the current angular velocity and angle into the predictive control model for prediction and optimization to obtain the target yaw angular velocity and target yaw angle; adjusting the angular velocity to the target yaw angular velocity and the orientation angle to the target yaw angle. This method can accurately predict wind speed and direction, quickly respond to wind direction changes, and improve the accuracy of yaw control.

[0059] The following description, based on the accompanying drawings, further elaborates on the MPC-based yaw control method for offshore wind turbines according to embodiments of the present invention.

[0060] Reference Figure 1 , Figure 1 The flowchart illustrates a yaw control method for offshore wind turbines based on MPC, provided in an embodiment of the present invention. This yaw control method for offshore wind turbines based on MPC includes, but is not limited to, the following steps:

[0061] S11, acquire multiple sets of historical wind speed data and multiple sets of historical wind direction data;

[0062] S12, For multiple sets of historical wind speed data and multiple sets of historical wind direction data, first perform multi-time scale averaging processing, and then perform variational mode decomposition to obtain multiple sets of wind speed training data and multiple sets of wind direction training data with different frequency terms corresponding to different time scales.

[0063] S13, based on convolutional neural network, long short-term memory network and self-attention mechanism layer, performs model training and construction on multiple sets of wind speed training data and multiple sets of wind direction training data corresponding to different frequency terms at different time scales, and obtains wind speed and wind direction prediction model.

[0064] S14. Based on the current wind speed and wind speed-direction prediction model, predict the average wind speed for the next period; based on the current wind direction and wind speed-direction prediction model, predict the average wind direction angle for the next period.

[0065] S15. The model is trained and constructed based on the average wind speed of the next time period, the average wind direction angle of the next time period, and the orientation angle of the offshore wind turbine nacelle at the current moment, so as to obtain the MPC predictive control model.

[0066] S16, based on the preset time step, substitute the current cabin rotation angular velocity and the current cabin orientation angle into the MPC predictive control model for prediction and optimization, and obtain the target yaw angular velocity and the target yaw angle;

[0067] S17, based on the target yaw rate and target yaw angle, adjust the angular velocity of the offshore wind turbine to the target yaw rate, and adjust the orientation angle of the offshore wind turbine to the target yaw angle.

[0068] In step S11, multiple sets of historical wind speed data and multiple sets of historical wind direction data are acquired. The acquired data needs to be preprocessed. The preprocessing steps include, but are not limited to, removing duplicate data, abnormal data or missing data, converting the data into a format suitable for subsequent analysis and processing, and standardizing the data.

[0069] In step S12, multi-timescale averaging refers to dividing the data according to different time scales and then calculating the average of all data at each time scale. Specifically, for multiple sets of historical wind speed and historical wind direction data, the data are divided into 10-second, 20-second, and 30-second time scales and the average is calculated.

[0070] More specifically, the calculation formula for averaging multiple sets of historical wind speed data and multiple sets of historical wind direction data over multiple time scales is as follows:

[0071]

[0072] in, This represents the average wind speed or average wind direction over N seconds at time i. This represents the average wind speed or average wind direction data j seconds before time i; i and j are positive integers; the average wind speed N is 10; the average wind direction N is 10, 20, or 30.

[0073] In step S12, Variational Modal Decomposition (VMD) can decompose a multi-component signal into multiple single-component amplitude-modulated and frequency-modulated signals in one step. In a specific embodiment, VMD is used to process multiple sets of average wind speed data and average wind direction data at different time scales, decomposing them into multiple component data, namely wind speed frequency term data and wind direction frequency term data.

[0074] Specifically, for each time scale (e.g., 10 seconds), multiple sets of average wind speed data are decomposed into three frequency components using VMD: high-frequency wind speed data, mid-frequency wind speed data, and low-frequency wind speed data. The wind speed residual data is the remaining frequency data after extracting the high-frequency, mid-frequency, and low-frequency components from the average wind speed data; that is, the average wind speed data minus its high-frequency, mid-frequency, and low-frequency components. Similarly, multiple sets of average wind direction data at the same time scale are also decomposed into three frequency components using VMD: high-frequency wind direction data, mid-frequency wind direction data, and low-frequency wind direction data. The wind direction residual data is the remaining portion of the average wind direction data after extracting the high-frequency, mid-frequency, and low-frequency components; that is, the average wind direction data minus its high-frequency, mid-frequency, and low-frequency components.

[0075] More specifically, the variational mode decomposition process decomposes multiple sets of average wind speed data and multiple sets of average wind direction data into k eigenmode functions, respectively. The variational mode decomposition formula is as follows:

[0076]

[0077] Where k is a positive integer, representing the number of types of frequency data; in one embodiment, k is 3. k For each mode function, namely the intrinsic mode function of the high-frequency, mid-frequency, and low-frequency wind speed data, and the high-frequency, mid-frequency, and low-frequency wind direction data; ω k δ(t) represents the center frequency of each intrinsic mode function; δ(t) represents the Dirac function; f represents the average wind speed data or average wind direction data; the wind speed residual data is the frequency data remaining after subtracting the extracted high-frequency, mid-frequency, and low-frequency wind speed data from the average wind speed data; the wind direction residual data is the frequency data remaining after subtracting the extracted high-frequency, mid-frequency, and low-frequency wind direction data from the average wind direction data.

[0078] To solve the variational mode decomposition formula, an augmented Lagrangian function is introduced to transform the constrained variational problem into an unconstrained variational problem. The calculation expression is as follows:

[0079]

[0080] Where α is the penalty factor and λ is the Lagrange multiplier. The optimal solution u is obtained by iteratively updating the Lagrange multiplier λ according to the calculated expression. k w k This allows us to obtain the high-frequency, mid-frequency, low-frequency, and residual data after decomposing the average wind speed and average wind direction data, respectively.

[0081] In step S13, wind speed and wind direction frequency data are extracted and separated through variational mode decomposition. After normalization, features are extracted from each frequency data using a convolutional neural network (CNN). The time series is then predicted using a long short-term memory (LSTM) network and combined with a self-attention (SA) layer for dynamic feature learning. Finally, the predicted frequency terms and residual terms are output sequentially. After inverse normalization, they are summed to obtain the predicted average wind speed and predicted average wind direction corresponding to each frequency term.

[0082] It should be noted that each set of wind speed frequency data and each set of wind direction frequency data should be trained using a convolutional neural network, a long short-term memory network, and a self-attention mechanism layer to output the prediction results of the corresponding frequency terms and construct a wind speed and wind direction prediction model. That is to say, the high-frequency wind speed data, mid-frequency wind speed data, low-frequency wind speed data, and wind speed residual data, as well as the high-frequency wind direction data, mid-frequency wind direction data, low-frequency wind direction data, and wind direction residual data should be sequentially input into the convolutional neural network, the long short-term memory network, and the self-attention mechanism layer for data training to obtain the predicted average wind speed and the predicted average wind direction.

[0083] It should be noted that this wind speed and direction prediction model is based on variational mode decomposition (VMD), convolutional neural network (CNN), long short-term memory network (LSTM), and self-attention mechanism layer (SA), namely the VMD-SA-CNN-LSTM wind speed and direction prediction model.

[0084] In step S14, the current wind speed and current wind direction are generally measured by physical sensors such as wind vanes. The measured current wind speed is substituted into the wind speed and direction prediction model to predict the average wind speed for the next time period; the measured current wind direction is substituted into the wind speed and direction prediction model to predict the average wind direction angle for the next time period.

[0085] In steps S15 and S16, the average wind speed and average wind direction angle for the next time period are predicted by the VMD-SA-CNN-LSTM wind speed and direction prediction model; the orientation angle of the offshore wind turbine nacelle at the current moment is given by the nacelle sensor; yaw error refers to the angle between the current orientation of the nacelle and the ideal orientation. Yaw speed refers to the angular velocity of the nacelle's rotation; yaw direction refers to the orientation angle of the nacelle.

[0086] The predicted state sequence refers to the prediction of the yaw state of the wind turbine over a future period of time. It includes multiple predicted state data within a preset time step. Each predicted state data includes the predicted values ​​of the nacelle's rotational angular velocity and orientation angle in the next period under ideal conditions.

[0087] The control input sequence refers to a series of commands or signals planned to adjust the yaw angle based on the current nacelle orientation angle and the predicted yaw state of the wind turbine over a future period. It includes multiple control input data points corresponding to a preset time step, with each control input data point including the rotational angular velocity and orientation angle that the nacelle needs to adopt in the next time period under the actual conditions.

[0088] The preset time step refers to the length of time predicted by the MPC (Model Predictive Control) predictive control model. In one embodiment, the MPC predicts the target yaw rate and yaw angle for the next 10 seconds, i.e., the target yaw rate and yaw angle within the next 10 seconds. That is, the predicted state sequence contains 10 predicted state data points for the next 10 seconds, each containing the predicted values ​​of the cabin's rotational angular velocity and heading angle under ideal conditions within that 1 second; the control input sequence contains 10 control input data points for the next 10 seconds, each containing the required rotational angular velocity and heading angle for the cabin under actual conditions within that 1 second. It should be noted that the time step can be set according to actual needs.

[0089] In the MPC predictive control model, the average wind speed and average wind direction angle for the next time period are predicted by the VMD-SA-CNN-LSTM wind speed and direction prediction model, along with the current orientation angle of the offshore wind turbine nacelle. This yields a predicted state sequence, i.e., the ideal rotational angular velocity and orientation angle of the nacelle for the next time period, and the yaw error is then calculated. Based on this yaw error, the MPC predictive control model considers multiple influencing factors for optimization, including the square of the yaw error, average wind speed, yaw distance, yaw time, number of yaw actions, and the derivative of the yaw speed. It evolves based on these multiple influencing factors, resulting in multiple control input sequences. Finally, the control input sequence that satisfies the constraints is selected from these sequences as the target control sequence.

[0090] In the MPC predictive control model, the predicted cabin orientation angle at the next moment under ideal conditions is expressed as:

[0091]

[0092] Where, θ np(k+1) represents the predicted cabin orientation angle at the next moment; θ np (k) represents the current cabin orientation angle; To predict the cabin's rotational angular velocity for the next time period; T c For time; when T c When the value is 1 second, it means that the prediction is the cabin's orientation angle for the next second.

[0093] When T c When the yaw error is 1 second, the yaw error in the next second is expressed as:

[0094]

[0095] Where, θ ye (k+1|k) represents the yaw error for the next second; θ wd (k+1|k) represents the predicted average wind direction for the next second; θ np (k) represents the current cabin orientation angle; To predict the cabin's rotational angular velocity in the next time period.

[0096] Specifically, when the preset time step is 10, and each time step is 1 second, the MPC predictive control model is expressed as:

[0097]

[0098] The time step of the MPC predictive control model is 10.

[0099] (k+1|k) represents the cabin state predicted at time k+1 based on time k.

[0100] (k+2|k) represents the cabin state predicted at time k+2 based on time k.

[0101] (k+3|k) represents the cabin state predicted at time k+3 from time k.

[0102] (k+4|k) represents the cabin state predicted at time k+4 based on time k.

[0103] (k+5|k) represents the cabin state predicted at time k+5 from time k.

[0104] (k+6|k) represents the cabin state predicted at time k+6 from time k.

[0105] (k+7|k) represents the cabin state predicted at time k+7 from time k.

[0106] (k+8|k) represents the cabin state predicted at time k+8 from time k.

[0107] (k+9|k) represents the cabin state predicted at time k+9 based on time k.

[0108] (k+10|k) represents the cabin state predicted at time k+10 from time k.

[0109] It should be noted that the process of finding the control input sequence with the minimum cost function value from multiple control input sequences as the target control sequence, and then applying the target yaw rate and target yaw angle represented by the target input sequence to the next prediction process, is the MPC predictive control model optimization process.

[0110] The constraints, i.e., the cost function, of the MPC predictive control model are expressed as:

[0111]

[0112] QF(2)=ω2v wd QF(1);

[0113]

[0114]

[0115] It should be noted that QF(1) considers the impact of yaw error on the output power of the wind turbine. Since the yaw error directly affects the output power in the form of the square of the cosine, QF(1) is expressed in the form of the square of the yaw error.

[0116] It should be noted that when considering the impact of wind speed on the yaw error of wind turbines, QF(2) divides the operating state of the wind turbine into three different intervals: interval one covers the wind speed range of [3,5] or [12,24], interval two covers the wind speed range of [5,9], and interval three covers the wind speed range of [9,12]. Based on these wind speed intervals, the parameter ω2 in QF(2) will be dynamically adjusted accordingly to ω 21 ω 22 or ω 23 To adapt to different wind speed conditions; in addition, each wind speed range corresponds to a different yaw control target: when the wind speed exceeds the rated wind speed by 12m / s, reduce yaw action to limit power; when the wind speed is 9m / s to 12m / s, reduce yaw error to optimize power output; when the wind speed is 5m / s to 9m / s, balance yaw accuracy and number of yaws; when the wind speed is 3m / s to 5m / s, pay attention to the use of yaw actuators; when the wind speed is 3m / s lower than the cut-in wind speed or 24m / s higher than the cut-out wind speed, the wind turbine is shut down.

[0117] It should be noted that QF(3) considers the yaw distance of the wind turbine, and the sampling time T cOne second; by minimizing the yaw distance, the wind turbine is ensured to always stay in the optimal direction, improving energy efficiency and performance.

[0118] It should be noted that QF(4) considers the yaw time of the wind turbine. When the yaw speed of the wind turbine is not zero, the yaw time is incremented by one second each time. Increasing the yaw time will lead to increased energy consumption and wear on the nacelle. Minimizing the yaw time helps to improve the long-term stability of the wind turbine.

[0119] It should be noted that QF(5) considers the number of wind turbine start-ups and shutdowns. When the yaw speed changes from zero to non-zero or from non-zero to zero, the number of wind turbine start-ups and shutdowns increases by one. Frequent start-ups and shutdowns can lead to premature wear of the wind turbine and increase maintenance costs. By minimizing the number of start-ups and shutdowns, the service life of the wind turbine can be extended and economic benefits can be improved.

[0120] It should be noted that QF(6) takes into account the magnitude of the derivative of the yaw speed of the wind turbine to avoid the yaw speed of the wind turbine being too large; by controlling the rate of change of the yaw speed, the wind turbine can avoid violent movement, thereby reducing the load and energy consumption of the equipment.

[0121] In step S17, a yaw action is performed based on the target yaw angular velocity and the target yaw angle, that is, the angular velocity of the offshore wind turbine is adjusted to the target yaw angular velocity, and the orientation angle of the offshore wind turbine is adjusted to the target yaw angle.

[0122] Through steps S11 to S17, a VMD-SA-CNN-LSTM wind speed and direction prediction model is constructed to predict the average wind speed and average wind direction angle for the next time period. Based on the predicted average wind speed and average wind direction angle for the next time period and the current orientation angle of the offshore wind turbine nacelle, the model is trained and constructed to build an MPC predictive control model. The MPC predictive control model is used to predict the nacelle's rotational angular velocity for the next time period and calculate the yaw error. Considering various influencing factors such as the yaw error, the control input sequence is optimized to obtain the target yaw angular velocity and target yaw angle in the target control input sequence. Based on the target yaw angular velocity and target yaw angle, the angular velocity of the offshore wind turbine is adjusted to the target yaw angular velocity, and the orientation angle of the offshore wind turbine is adjusted to the target yaw angle. This method can accurately predict wind speed and direction, quickly respond to wind direction changes, and improve the accuracy of yaw control.

[0123] According to some embodiments of the present invention, with reference to Figure 2 ,exist Figure 1 Step S12 of the illustrated embodiment also includes, but is not limited to, the following steps:

[0124] S21. Perform multi-time-scale averaging on multiple sets of historical wind speed data and multiple sets of historical wind direction data to obtain multiple sets of average wind speed data and multiple sets of average wind direction data corresponding to different time scales.

[0125] S22, calculate the wind speed frequency term data and wind direction frequency term data corresponding to different time scales based on the variational mode decomposition formula, respectively;

[0126] S23. For wind speed frequency data and wind direction frequency data corresponding to different time scales, a sliding window scrolling method is used to process them respectively, so as to obtain multiple sets of wind speed training data and multiple sets of wind direction training data corresponding to different frequency terms at different time scales.

[0127] Through steps S21 to S23, multi-timescale averaging can remove some high-frequency noise or fluctuations in the data, improving the stability and accuracy of the wind speed and direction prediction model. Variational mode decomposition can capture multiple frequency components from multiple sets of average wind speed and wind direction data, obtaining complex features in wind speed and direction, and improving the prediction accuracy of the wind speed and direction prediction model.

[0128] According to some embodiments of the present invention, with reference to Figure 3 ,exist Figure 2 Step S21 in the illustrated embodiment also includes, but is not limited to, the following steps:

[0129] S31, divide multiple sets of historical wind speed data and multiple sets of historical wind direction data into different time scales to obtain multiple sets of first wind speed data and multiple sets of first wind direction data corresponding to different time scales;

[0130] S32, average the first wind speed data corresponding to different time scales to obtain multiple sets of average wind speed data corresponding to different time scales.

[0131] S33, averages multiple sets of first wind direction data at different time scales to obtain multiple sets of average wind direction data at different time scales.

[0132] By dividing the data into different time scales through steps S31 to S33, we can better understand the changing trends and periodicity of wind speed and direction in different time ranges. Averaging can remove some high-frequency noise or fluctuations in the data, thus improving data quality.

[0133] According to some embodiments of the present invention, with reference to Figure 4 ,exist Figure 2 Step S23 of the illustrated embodiment also includes, but is not limited to, the following steps:

[0134] S41, Set the sliding window step size;

[0135] S42, slide the sliding window step size on the wind speed frequency data at different time scales until all time scales of wind speed frequency data have been traversed to obtain multiple sets of wind speed training data corresponding to different frequency items at different time scales.

[0136] S43, slide the sliding window step size on the wind direction frequency data at different time scales until all time scales of wind direction frequency data have been traversed, and obtain multiple sets of wind direction training data for different frequency items corresponding to different time scales.

[0137] It should be noted that, in one embodiment, the step size of the sliding window is set to 1. This sliding window moves sequentially across wind speed frequency data at various time scales until it covers and processes wind speed frequency data at all time scales. This process yields multiple sets of wind speed training data, each corresponding to a different time scale and frequency term. Similarly, the sliding window moves across wind direction frequency data with the same step size until it traverses all time scales of wind direction frequency data, resulting in multiple sets of wind direction training data corresponding to different time scales and frequency terms. Each set of wind speed training data and each set of wind direction training data is organized into a one-dimensional matrix, containing input data at 40 sampling points. For wind speed training data, each one-dimensional array contains wind speed measurements at 40 consecutive time points; similarly, for wind direction training data, each one-dimensional array contains wind direction measurements at 40 consecutive time points.

[0138] By employing the sliding window technique in steps S41 to S43, the information in the wind speed frequency data and wind direction frequency data can be fully utilized to extract the wind speed and wind direction features at different time scales. At the same time, the data features of different time scales and frequency terms are considered, which can more accurately capture the complexity of wind speed and wind direction.

[0139] According to some embodiments of the present invention, with reference to Figures 5 to 6 ,exist Figure 1 Step S13 of the illustrated embodiment also includes, but is not limited to, the following steps:

[0140] S51, normalize the multiple sets of wind speed training data and multiple sets of wind direction training data corresponding to different frequency items at different time scales to obtain multiple sets of standard wind speed data and multiple sets of standard wind direction data corresponding to different frequency items at different time scales.

[0141] S52, through convolutional neural networks, features are extracted and expanded from multiple sets of standard wind speed data and multiple sets of standard wind direction data corresponding to different frequency terms at different time scales, to obtain one-dimensional time series data of wind speed and one-dimensional time series data of wind direction corresponding to different frequency terms at different time scales.

[0142] S53, using a long short-term memory network, predicts one-dimensional time series data of wind speed and one-dimensional time series data of wind direction corresponding to different frequency terms at different time scales, and obtains the first predicted data of wind speed and the first predicted data of wind direction corresponding to different frequency terms at different time scales.

[0143] S54. By using the self-attention mechanism layer, feature learning is performed on the first predicted wind speed data and the first predicted wind direction data of different frequencies corresponding to different time scales to obtain the second predicted wind speed data and the second predicted wind direction data of different frequencies corresponding to different time scales.

[0144] S55. After inverse normalization of the second prediction data of wind speed and the second prediction data of wind direction corresponding to different frequency terms at different time scales, the data are added together to obtain the predicted average wind speed and predicted average wind direction at different time scales.

[0145] S56. Based on the predicted average wind speed and predicted average wind direction at different time scales, a wind speed and wind direction prediction model is constructed.

[0146] In step S51, the normalization calculation formula is as follows:

[0147]

[0148] train norm =(train-train) mean ) / train std ;

[0149] Among them, A pq Let R be the data in the p-th row and q-th column of the training data matrix, and let R be the total number of rows in the training data matrix. q-mean The average wind direction data in column q is the training value. q-std The standard value for the average wind direction data in column q; train norm This is the normalized standard wind speed data.

[0150] In step S52, the convolutional layer includes convolution and pooling operations. In the convolution operation, each convolutional kernel slides across the input data (such as standard wind speed data or standard wind direction data), covering a small portion of the data (local region) each time, and calculates the weighted sum of that region and the convolutional kernel to extract the features of that region. The extracted features are represented as follows:

[0151]

[0152] Where Y(i,j) is the feature extracted after convolution; X is the input time series; W is the convolution kernel weight matrix; k is the convolution kernel size; m and n are the weight positions inside the convolution kernel; and b is the bias term.

[0153] Pooling reduces the size of the convolutional feature map by selecting the maximum value within the pooling window, while preserving the salient features of the data, as shown below:

[0154]

[0155] Where P(i,j) is the output feature after pooling; A is the feature map after convolution activation function; s is the stride, representing the step size of each sliding of the pooling window; m and n are the sizes of the pooling window.

[0156] In one embodiment, features of the input standard wind speed and standard wind direction data are first extracted using 10 convolutional kernels of size 2×1 through convolutional layers. This process generates 10 feature maps, each 39x1 in size, which effectively capture the local features of the data. Subsequently, pooling layers employ max pooling, sampling these feature maps using a 2×1 pooling window (with a stride of 1). After processing by the pooling layers, the feature data is further passed to a flattening layer and transformed into a 390×1 one-dimensional time series dataset, which is then input into a Long Short-Term Memory (LSTM) network.

[0157] In step S53, the Long Short-Term Memory Network receives one-dimensional time series data from the flattened layer, learns long-term dependencies in the data using the gating mechanism and cell state update formula, uses its time series modeling capability to predict the input time series, and outputs the prediction results as input to the self-attention mechanism layer.

[0158] Long Short-Term Memory (LSTM) networks consist of three gate mechanisms (input gate, forget gate, and output gate) and a cell state, enabling LSM units to retain useful information for extended periods and capture long-term dependencies. The forget gate determines the information deleted from the cell state in the previous time step, the input gate controls the information received by the cell state, and the output gate controls the information output by the cell state. The formulas for the output information of each gate mechanism are as follows:

[0159] f t =σ(W f [h t-1 ,x t ]+b f );

[0160] i t =σ(W i [ht-1 ,x t ]+b i );

[0161] o t =σ(W o [h t-1 ,x t ]+b o );

[0162] Among them, W f Forget gate and input x t The previous time step hidden state h t-1 The weight matrix between them; W i For the input gate and input x t The previous time step hidden state h t-1 The weight matrix between them; W o For the output gate and input x t The previous time step hidden state h t-1 The weight matrix between; b f b i b o These are the corresponding biases; f t Output for the forget gate; i t For input gate output; o t σ is the output gate output; σ is the sigmoid activation function, which transforms the output to the interval [0,1].

[0163] In one embodiment, the Long Short-Term Memory (LSTM) network layer consists of 32 LSM units. For each time step of the input, this layer generates a 32-dimensional output vector. The LSM network performs deep processing on the input one-dimensional time-series data through its unique gating mechanisms (including input gates, forget gates, and output gates) and cell states, thereby making predictions. The prediction results are represented as multiple 32×1 matrices X, which contain the LSM network's understanding and prediction information of the time-series data; subsequently, these prediction results are fed into a self-attention mechanism layer for further analysis and processing.

[0164] In step S54, the self-attention mechanism layer outputs predicted data based on the long short-term memory network, utilizes the correlation between keys, queries and values, and calculates the importance of each part through a weighted allocation principle, thereby dynamically adjusting the weights to achieve precise attention and efficient processing of the input information.

[0165] In one embodiment, the self-attention layer comprises a single-head self-attention mechanism and a fully connected layer. Specifically, the head value of the self-attention mechanism layer is set to 1, and the key value is set to 2. The self-attention mechanism utilizes the correlation between keys, queries, and values, calculating the importance of each part through a weighted allocation principle. The attention weight matrix assigns a weight to each element of the value matrix; the higher the weight, the more important the element. The size of the fully connected layer depends on its averaging time scale: 10 neurons for a 10-second average wind direction, 20 neurons for a 20-second average wind direction, and 30 neurons for a 30-second average wind direction. The fully connected layer integrates all learned features and reshapes the output matrix. Finally, a regression layer is connected to return the predicted output data, the size of which depends on its averaging time scale.

[0166] Q = XW Q +b Q ;

[0167] K = XW K +b K ;

[0168] V = XW V +b V ;

[0169] Where Q, K, and V are the query matrix, key matrix, and value matrix, respectively; W Q W K W V The corresponding weight matrices are all 2×32; X is the input time series, which is a 32x1 input matrix. Q b K b V These are the corresponding bias matrices, all of which are 2×1 in size.

[0170] The attention score matrix a is obtained through dot product and scaling operations. n The attention score a above was calculated using the Softmax function. n Normalization is performed to obtain the weight coefficients. Finally, the weighting coefficients are used. The final output I is obtained by weighted summation of the value vector V;

[0171]

[0172] Where n and N represent attention score matrices. The dimension of K; I is a 32×1 output matrix; T It is the transpose of K, d k It is the scaling factor.

[0173] It should be noted that by introducing R 2 The coefficients, MAE, and RMSE evaluation indicators are used to quantitatively measure the prediction accuracy of the VMD-SA-CNN-LSTM wind speed and direction prediction model.

[0174] R 2 The coefficients are used to measure the goodness of fit of the model, R0 2 A value close to 1 indicates a good model fit, meaning the predicted data fits the actual data well. R0 2 A value close to 0 indicates weak predictive ability of the model; the calculation expression is:

[0175]

[0176] Among them, y i The actual value; This is a predicted value; is the average of the true values; n is the total number of data points.

[0177] MAE is the average of the absolute differences between the true and predicted values. It measures the magnitude of the model's prediction error. The smaller the MAE, the higher the model's prediction accuracy. The calculation expression is:

[0178]

[0179] Among them, y i The actual value; is the predicted value; n is the total number of data points.

[0180] RMSE is the square root of the average of the squared errors between predicted and actual values. It measures the prediction error of a model and is particularly sensitive to larger errors. The calculation expression is:

[0181]

[0182] In one embodiment, to further evaluate the performance of the yaw control method, the evaluation will focus on three aspects: the accuracy of wind direction tracking, the use of yaw actuators, and power enhancement. The accuracy of wind direction tracking is quantitatively analyzed using yaw error accumulation, average yaw error, and root mean square yaw error indices, calculated as follows:

[0183] θ ye =θ wd -θ np ;

[0184]

[0185] Where, θ ye The yaw error is determined by the wind angle θ. wd Subtract the cabin orientation angle θ np It is derived that TE(θ)ye MAE(θ) represents the total yaw error, which measures the cumulative magnitude of the yaw error. ye RMSE(θ) represents the mean absolute error, reflecting the average level of yaw error; ye The root mean square error is used to monitor the fluctuation of yaw error, especially when the yaw error is large.

[0186] Through steps S51 to S56, the VMD-SA-CNN-LSTM wind speed and direction prediction model fully leverages the advantages of CNN in local feature extraction, LSTM in time-dependent modeling, and SA in dynamic selection to achieve high-precision ultra-short-term wind speed and direction prediction, providing accurate wind speed and direction prediction data for yaw control.

[0187] According to some embodiments of the present invention, with reference to Figure 6 ,exist Figure 1 Step S16 of the illustrated embodiment also includes, but is not limited to, the following steps:

[0188] S61, based on a preset time step, substitute the current cabin rotation angular velocity and the current cabin orientation angle into the MPC predictive control model for prediction, and obtain a predicted state sequence; wherein, the predicted state sequence includes multiple predicted state data corresponding to the preset time step, and each predicted state data includes the predicted values ​​of the cabin rotation angular velocity and orientation angle in the next time period under ideal conditions.

[0189] S62, based on the predicted state sequence and the MPC predictive control model, perform evolution calculations to obtain multiple control input sequences corresponding to the predicted state sequence; wherein, the control input sequence contains multiple control input data corresponding to a preset time step, and each control input data includes the rotational angular velocity and orientation angle that the cabin needs to take in the next time period under the actual state;

[0190] S63, perform optimization calculations based on multiple control input sequences and a preset cost function, and determine the control input sequence that minimizes the cost function value as the target control input sequence;

[0191] S64, the cabin rotation angular velocity corresponding to the first control input data in the target control input sequence is determined as the target yaw angular velocity, and the cabin orientation angle corresponding to the first control input data in the target control input sequence is determined as the target yaw angle.

[0192] It should be noted that in the target control sequence corresponding to the minimum cost function value, only the first control input data will be applied. The cabin rotation angular velocity corresponding to the first control input data in the target control input sequence is determined as the target yaw angular velocity, and the cabin orientation angle corresponding to the first control input data in the target control input sequence is determined as the target yaw angle. Then, based on the target yaw angular velocity, the target yaw angle, and the MPC predictive control model, the target control sequence for the next time period is obtained, and this process is used for iterative optimization.

[0193] In steps S61 to S64, the MPC predictive control model predicts the cabin's state over a future period based on the current angular velocity and orientation angle of the cabin. This allows for advance awareness of the cabin's dynamic changes and enables more accurate control decisions. Based on the predicted state sequence, multiple possible control input sequences are obtained through evolutionary calculations. These control input sequences represent the possible future states of the cabin under different control strategies. By selecting the control input sequence that minimizes the cost function as the target control input sequence, this process optimizes the control strategy, ensuring the cabin operates smoothly and efficiently under actual conditions.

[0194] like Figure 8 As shown, Figure 8 This is a structural diagram of an MPC-based yaw control device for offshore wind turbines according to an embodiment of the present invention. The present invention also provides an MPC-based yaw control device for offshore wind turbines, comprising:

[0195] The processor 801 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0196] The memory 802 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 802 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 802 and called and executed by the processor 801 to execute the MPC-based yaw control method for offshore wind turbines according to the embodiments of this application.

[0197] The 803 input / output interface is used to implement information input and output.

[0198] The communication interface 804 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0199] Bus 805 transmits information between various components of the device (e.g., processor 801, memory 802, input / output interface 803, and communication interface 804);

[0200] The processor 801, memory 802, input / output interface 803, and communication interface 804 are connected to each other within the device via bus 805.

[0201] This application also provides an electronic device, including the MPC-based yaw control device for offshore wind turbines as described above.

[0202] This application embodiment also provides a storage medium, which is a computer-readable storage medium, storing a computer program that, when executed by a processor, implements the above-described MPC-based yaw control method for offshore wind turbines.

[0203] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof. The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate, and may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0204] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically include computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0205] The above provides a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A method for MPC-based yaw control of an offshore wind turbine generator unit, characterized in that, The method comprises the following steps: acquiring a plurality of sets of historical wind speed data and a plurality of sets of historical wind direction data; respectively performing multi-time scale average value processing on the plurality of sets of historical wind speed data and the plurality of sets of historical wind direction data, and then performing variational mode decomposition to obtain a plurality of sets of wind speed training data and a plurality of sets of wind direction training data corresponding to different time scales and different frequency items; based on a convolutional neural network, a long short-term memory network and a self-attention mechanism layer, performing model training and construction on the plurality of sets of wind speed training data and the plurality of sets of wind direction training data corresponding to different time scales and different frequency items to obtain a wind speed and wind direction prediction model; predicting an average wind speed size of a next time period according to a current wind speed and the wind speed and wind direction prediction model, and predicting an average wind direction angle of the next time period according to a current wind direction and the wind speed and wind direction prediction model; performing model training and construction according to the average wind speed size of the next time period, the average wind direction angle of the next time period and a current heading angle of a machine cabin of a marine wind turbine generator to obtain an MPC prediction control model; based on a preset time step, substituting a current cabin rotation angular velocity and a current cabin heading angle into the MPC prediction control model to perform prediction and optimization to obtain a target yaw angular velocity and a target yaw angle; adjusting an angular velocity of the marine wind turbine generator to the target yaw angular velocity and adjusting a heading angle of the marine wind turbine generator to the target yaw angle according to the target yaw angular velocity and the target yaw angle. The method comprises the following steps: respectively performing multi-time scale average value processing on the plurality of sets of historical wind speed data and the plurality of sets of historical wind direction data to obtain a plurality of sets of average wind speed data and a plurality of sets of average wind direction data corresponding to different time scales; and based on a variational mode decomposition formula, calculating the plurality of sets of average wind speed data and the plurality of sets of average wind direction data corresponding to different time scales to obtain wind speed frequency item data and wind direction frequency item data corresponding to different time scales; and wherein k is a positive integer, representing the number of frequency data types; u k are the modal functions, i.e. the eigenmodal functions of the wind speed high frequency data, the wind speed medium frequency data, the wind speed low frequency data, the wind direction high frequency data, the wind direction medium frequency data, and the wind direction low frequency data; ω k are the center frequencies of the respective eigenmodal functions; δ(t) is the Dirac function; f is the average wind speed data or the average wind direction data; the wind speed residual data is the frequency data of the residual part after the average wind speed data is subtracted by the extracted wind speed high frequency data, the wind speed medium frequency data, and the wind speed low frequency data; and the wind direction residual data is the frequency data of the residual part after the average wind direction data is subtracted by the extracted wind direction high frequency data, the wind direction medium frequency data, and the wind direction low frequency data.

2. The MPC-based offshore wind turbine yaw control method of claim 1, wherein, respectively adopting a sliding window rolling manner to process the wind speed frequency item data and the wind direction frequency item data corresponding to different time scales to obtain a plurality of sets of wind speed training data and a plurality of sets of wind direction training data corresponding to different time scales and different frequency items. The wind speed frequency item data comprises wind speed high frequency item data, wind speed medium frequency item data, wind speed low frequency item data and wind speed residual item data; and the wind direction frequency item data comprises wind direction high frequency item data, wind direction medium frequency item data, wind direction low frequency item data and wind direction residual item data. The variational mode decomposition formula is as follows: The method comprises the following steps: respectively performing multi-time scale average value processing on the plurality of sets of historical wind speed data and the plurality of sets of historical wind direction data to obtain a plurality of sets of average wind speed data and a plurality of sets of average wind direction data corresponding to different time scales; and The multiple sets of historical wind speed data and the multiple sets of historical wind direction data are divided according to different time scales respectively to obtain corresponding multiple sets of first wind speed data and multiple sets of first wind direction data under different time scales; The multiple sets of first wind speed data corresponding to different time scales are averaged to obtain multiple sets of average wind speed data corresponding to different time scales; The multiple sets of first wind direction data corresponding to different time scales are averaged to obtain multiple sets of average wind direction data corresponding to different time scales.

3. The MPC-based offshore wind turbine yaw control method of claim 1, wherein, The wind speed frequency term data and the wind direction frequency term data corresponding to different time scales are processed by a sliding window rolling method to obtain multiple sets of wind speed training data and multiple sets of wind direction training data of different frequencies corresponding to different time scales, including: setting a sliding window step; sliding the sliding window step on the wind speed frequency term data under different time scales until all the wind speed frequency term data under all time scales are traversed to obtain multiple sets of wind speed training data of different frequencies corresponding to different time scales; sliding the sliding window step on the wind direction frequency term data under different time scales until all the wind direction frequency term data under all time scales are traversed to obtain multiple sets of wind direction training data of different frequencies corresponding to different time scales.

4. The MPC-based offshore wind turbine yaw control method of claim 1, wherein, The multiple sets of wind speed training data and the multiple sets of wind direction training data of different frequencies corresponding to different time scales are trained and constructed based on a convolutional neural network, a long short-term memory network and a self-attention mechanism layer to obtain a wind speed and wind direction prediction model, including: The multiple sets of wind speed training data and the multiple sets of wind direction training data of different frequencies corresponding to different time scales are normalized to obtain multiple sets of standard wind speed data and multiple sets of standard wind direction data of different frequencies corresponding to different time scales; The multiple sets of standard wind speed data and the multiple sets of standard wind direction data of different frequencies corresponding to different time scales are respectively extracted and unfolded by the convolutional neural network to obtain wind speed one-dimensional time series data and wind direction one-dimensional time series data of different frequencies corresponding to different time scales; The wind speed one-dimensional time series data and the wind direction one-dimensional time series data of different frequencies corresponding to different time scales are respectively predicted by the long short-term memory network to obtain wind speed first prediction data and wind direction first prediction data of different frequencies corresponding to different time scales; The wind speed first prediction data and the wind direction first prediction data of different frequencies corresponding to different time scales are learned by the self-attention mechanism layer to obtain wind speed second prediction data and wind direction second prediction data of different frequencies corresponding to different time scales; The wind speed second prediction data and the wind direction second prediction data of different frequencies corresponding to different time scales are added after being de-normalized to obtain predicted average wind speed and predicted average wind direction of different time scales; The wind speed and wind direction prediction model is constructed according to the predicted average wind speed and the predicted average wind direction of different time scales.

5. The MPC-based offshore wind turbine yaw control method of claim 1, wherein, The current nacelle rotation angular velocity and the current nacelle orientation angle are substituted into the MPC prediction control model based on a preset time step to perform prediction and optimization, so as to obtain a target yaw angular velocity and a target yaw angle, including: The current nacelle rotation angular velocity and the current nacelle orientation angle are substituted into the MPC prediction control model based on a preset time step to perform prediction, so as to obtain a prediction state sequence; wherein the prediction state sequence includes a plurality of prediction state data corresponding to a preset time step, and each prediction state data includes a predicted value of the nacelle rotation angular velocity and the nacelle orientation angle in the next time period under an ideal state; Evolution calculation is performed according to the prediction state sequence and the MPC prediction control model, so as to obtain a plurality of control input sequences corresponding to the prediction state sequence; wherein the control input sequence includes a plurality of control input data corresponding to a preset time step, and each control input data includes the nacelle rotation angular velocity and the nacelle orientation angle to be taken by the nacelle in the next time period under an actual state; Optimization calculation is performed according to the plurality of control input sequences and a preset cost function, and the control input sequence that makes the cost function value minimum is determined as a target control input sequence; The nacelle rotation angular velocity corresponding to the first control input data in the target control input sequence is determined as the target yaw angular velocity, and the nacelle orientation angle corresponding to the first control input data in the target control input sequence is determined as the target yaw angle.

6. An MPC-based offshore wind turbine yaw control apparatus, characterized in that, The offshore wind turbine yaw control device based on MPC includes at least one control processor and a memory connected in communication with the at least one control processor; the memory stores instructions executable by the at least one control processor, and the instructions are executed by the at least one control processor to enable the at least one control processor to perform the offshore wind turbine yaw control method based on MPC according to any one of claims 1 to 5.

7. An electronic device, comprising: The offshore wind turbine yaw control device based on MPC according to claim 6.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer executable instructions for enabling a computer to perform the offshore wind turbine yaw control method based on MPC according to any one of claims 1 to 5.

Images