Offshore wind turbine yaw prediction control method, system, device and medium

Abstract

The present invention discloses a method, system, device, and medium for predicting and controlling the yaw of an offshore wind turbine. The method comprises: processing historical wind direction data using a multi-time-scale average method to obtain multiple sets of average wind direction data, establishing an Attention-CNN-LSTM short-term wind direction prediction model, and establishing an offshore wind turbine yaw prediction and control system based on the model. When the wind speed exceeds the starting wind speed, the system starts running and monitors the current wind direction. The system predicts the average wind direction angle for the next period based on the current wind direction, and then performs the corresponding yaw action based on the current cabin position. By introducing a self-attention mechanism, the accuracy of the model prediction is improved. At the same time, the yaw prediction and control system is designed in conjunction with the model to enable the system to quickly track wind direction changes, reduce yaw errors, avoid excessive yaw, and improve yaw efficiency.

Description

Technical Field

[0001] The present invention relates to the field of wind direction prediction models, and in particular to a method, system, device and medium for predicting and controlling the yaw of an offshore wind turbine. Background Art

[0002] In line with the dual carbon goals, the installed capacity of renewable energy sources, such as wind power, is steadily expanding. Offshore wind power, as a potential clean energy source, has demonstrated unprecedented vitality and is expected to become a key direction for future new energy development. The primary goal of wind turbine yaw control is to ensure that the turbine consistently follows wind direction changes, maximizing wind energy capture and increasing turbine output power.

[0003] However, the current traditional yaw control method mainly depends on the historical average wind direction data measured by the unit, and its value only reflects the changes in wind direction over the past period of time, and is used to determine whether the current wind direction has changed. Due to the time delay between the yaw system and the wind direction, the traditional yaw control system has insufficient yaw accuracy in wind response, large yaw errors, and frequent yaw control caused by the yaw lagging behind the wind direction, and long yaw times. The traditional yaw control method cannot meet the power requirements of wind turbines and reduces power generation efficiency. With the introduction and improvement of prediction algorithms, current studies have proposed yaw control methods based on wind direction prediction in combination with wind direction prediction. The use of deep learning models such as BP neural networks, convolutional neural networks, and artificial neural networks to predict wind direction has achieved the need to predict future wind direction trends, but the model structure used is often relatively simple, the prediction accuracy is insufficient, and it is difficult to apply to actual yaw control systems. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a method, system, device, and medium for predicting and controlling the yaw of an offshore wind turbine, which improves the accuracy and speed of yaw prediction and is applied to actual offshore wind turbine yaw control systems.

[0005] In a first aspect, an embodiment of the present invention provides a method for predicting and controlling the yaw of an offshore wind turbine, the method comprising:

[0006] Based on multiple sets of average wind direction data, an Attention-CNN-LSTM short-term wind direction prediction model is established;

[0007] Based on the Attention-CNN-LSTM short-term wind direction prediction model, a yaw prediction control system for offshore wind turbines is established;

[0008] When the wind speed exceeds the starting wind speed, the offshore wind turbine yaw prediction control system starts to operate and monitors the current wind direction;

[0009] Predicting the average wind direction angle for the next period based on the current wind direction and the Attention-CNN-LSTM short-term wind direction prediction model;

[0010] According to the average wind direction angle of the next period and the current position of the nacelle, a corresponding yaw action is performed.

[0011] In some embodiments of the present invention, the step of obtaining historical wind direction data and processing the historical wind direction data using a multi-time-scale average method to obtain multiple sets of average wind direction data includes:

[0012] Setting multiple time scales, and averaging the historical wind direction data according to each time scale to obtain multiple groups of average wind direction data corresponding to the multiple time scales;

[0013] Using a sliding window rolling processing method, multiple groups of the average wind direction data are processed separately to obtain multiple groups of training data;

[0014] Normalization is performed on the multiple sets of training data respectively to obtain multiple sets of standardized training data.

[0015] In some embodiments of the present invention, the plurality of groups of average wind direction data include ten-second average wind direction data, twenty-second average wind direction data, and thirty-second average wind direction data;

[0016] The calculation formula of the average wind direction data is:

[0017]

[0018] in, It is expressed as the N-second average wind direction data at the i-th moment, Expressed as the average wind direction j seconds before the i-th moment; i and j are positive integers; N is 10, 20 or 30.

[0019] In some embodiments of the present invention, the Attention-CNN-LSTM short-term wind direction prediction model includes a convolutional neural network, a long short-term memory network, and a self-attention mechanism layer;

[0020] The step of establishing a yaw prediction control system based on the Attention-CNN-LSTM short-term wind direction prediction model includes:

[0021] Inputting each group of the standardized training data into the convolutional neural network respectively;

[0022] The convolutional neural network extracts feature data of each set of the standardized training data, expands the feature data into a one-dimensional sequence, and sends the sequence to the long short-term memory network;

[0023] After the long short-term memory network performs prediction processing on the one-dimensional sequence, the prediction data is output to the self-attention mechanism layer;

[0024] The self-attention mechanism layer obtains an output result based on the predicted data.

[0025] In some embodiments of the present invention, the step of performing a corresponding yaw action based on the average wind direction angle in the next period and the current nacelle position includes:

[0026] Determine the absolute value of the error angle between the average wind direction angle for the next period and the current position of the nacelle;

[0027] When the absolute value of the error angle is greater than a preset angle value within a preset time period, a yaw action is performed.

[0028] In some embodiments of the present invention, the average wind direction angle includes a ten-second average wind direction angle, a twenty-second average wind direction angle, and a thirty-second average wind direction angle; the preset time periods include a first preset time period, a second preset time period, and a third preset time period; and the step of performing a corresponding yaw action based on the average wind direction angle of the next time period and the current cabin position includes:

[0029] When the absolute value of the ten-second average wind direction angle in the next period and the error angle of the nacelle position at the current moment are both greater than the preset angle value within the first preset time period, performing a yaw action; or,

[0030] When the absolute value of the twenty-second average wind direction angle in the next period and the error angle of the nacelle position at the current moment are both greater than the preset angle value within the second preset time period, performing a yaw action; or,

[0031] When the absolute values ​​of the thirty-second average wind direction angle in the next period and the error angle of the nacelle position at the current moment are both greater than the preset angle value within the third preset time period, a yaw action is performed.

[0032] In some embodiments of the present invention, the yawing action includes forward yawing and reverse yawing; when the error angle is greater than or equal to 1 degree, forward yawing is performed; when the error angle is less than or equal to -1 degree, reverse yawing is performed.

[0033] In a second aspect, an embodiment of the present invention provides an offshore wind turbine yaw prediction control system, including the offshore wind turbine yaw prediction control method as described in the first aspect above.

[0034] In a third aspect, an embodiment of the present invention provides a control device, comprising at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed 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 operate the offshore wind turbine yaw prediction control system as described in the second aspect above.

[0035] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the offshore wind turbine yaw prediction control method as described in the first aspect above.

[0036] The yaw prediction control method, system, device, and medium according to the embodiments of the present invention have at least the following beneficial effects:

[0037] By introducing a self-attention mechanism, the advantages of CNN (Convolutional Neural Network) and LSTM (Long Short-Term Memory) in time series prediction are fully utilized. This improves the model's prediction performance and efficiency in learning complex features, meeting the dual requirements of fast and accurate short-term predictions. Furthermore, a yaw prediction control system for offshore wind turbines is designed by combining the Attention-CNN-LSTM (CNN-LSTM network with an attention mechanism) short-term wind direction prediction model. The primary goal is to enable the yaw system to quickly track wind direction changes, minimize yaw error, avoid overuse of the yaw actuator, and improve the yaw efficiency of the wind turbine. This solves the problems of insufficient wind direction accuracy and frequent yaw movements in traditional offshore wind turbine yaw prediction control systems, achieving high-efficiency power generation for the wind turbine. The offshore wind turbine yaw prediction control system predicts multiple average wind direction angles and sets multiple corresponding yaw conditions for each of these average wind direction angles to determine whether to execute a yaw action. This increases the number of judgments, reduces yaw error, effectively avoids frequent yaw caused by short-term wind direction fluctuations, and enhances the stability and reliability of the yaw prediction control system.

[0038] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments with reference to the accompanying drawings, in which:

[0040] Figure 1 A flowchart of a yaw prediction control method provided by one embodiment of the present invention;

[0041] Figure 2 A structural diagram of a control device provided in another embodiment of the present invention;

[0042] Figure 3 This is the 10s average wind direction forecast result graph;

[0043] Figure 4 This is the 20s average wind direction forecast result chart;

[0044] Figure 5 This is the 30s average wind direction forecast result chart;

[0045] Figure 6 This is a comparison chart of the nacelle position for traditional yaw control and yaw control combined with wind direction prediction;

[0046] Figure 7 This is a comparison diagram of the nacelle position without wind direction prediction yaw control and with wind direction prediction yaw control;

[0047] Figure 8 This is the total yaw error result diagram of the three control systems;

[0048] Figure 9 This is the result diagram of the total number of yaw actions of the three control systems;

[0049] Figure 10 This is the total yaw time result diagram of the three control systems. DETAILED DESCRIPTION

[0050] This section will describe in detail the specific embodiments of the present invention. The preferred embodiments of the present invention are shown in the accompanying drawings. The purpose of the accompanying drawings is to supplement the description of the text part of the specification with graphics, so that people can intuitively and vividly understand each technical feature and the overall technical solution of the present invention, but it should not be understood as a limitation on the scope of protection of the present invention.

[0051] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they cannot be understood as limitations on the present invention.

[0052] In the description of the present invention, "several" means one or more, "many" means more than two, "greater than," "less than," and "exceed" are understood to exclude the number itself, while "above," "below," and "within" are understood to include the number itself. The use of "first" and "second" in the description is solely for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, implicitly specifying the number of the indicated technical features, or implicitly specifying the order of the indicated technical features.

[0053] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, and connecting should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0054] The present invention provides a yaw prediction control method that leverages the advantages of CNN and LSTM in time series prediction by introducing a self-attention mechanism. This method improves model prediction performance and the efficiency of learning complex features, meeting the dual requirements of short-term rapid prediction and prediction accuracy. Furthermore, a yaw prediction control system is designed in conjunction with the Attentation-CNN-LSTM short-term wind direction prediction model. The main goals are to enable the yaw system to quickly track wind direction changes, minimize yaw errors, avoid overuse of the yaw actuator, and improve the yaw efficiency of the wind turbine. This method addresses the problems of insufficient wind accuracy and frequent yaw movements in traditional yaw systems, enabling high-efficiency power generation from wind turbines.

[0055] The method of the embodiment of the present invention is further described below based on the accompanying drawings.

[0056] like Figure 1 As shown, the yaw prediction control method of the present invention includes:

[0057] Step S100: acquiring historical wind direction data, and processing the historical wind direction data using a multi-time-scale average method to obtain multiple sets of average wind direction data;

[0058] Step S200: establishing an Attention-CNN-LSTM short-term wind direction prediction model based on multiple sets of average wind direction data;

[0059] Step S300: establishing a yaw prediction control system based on the Attention-CNN-LSTM short-term wind direction prediction model;

[0060] Step S400: When the wind speed exceeds the starting wind speed, the yaw prediction control system starts to operate and monitors the current wind direction;

[0061] Step S500: Predicting the average wind direction angle for the next period based on the current wind direction and the Attention-CNN-LSTM short-term wind direction prediction model;

[0062] Step S600: Execute corresponding yaw action according to the average wind direction angle of the next period and the current nacelle position.

[0063] The execution of the yaw action ensures that the nacelle is always in the appropriate direction to meet the power requirements of the wind turbine.

[0064] It should be noted that multiple time scales, including 10, 20, and 30 seconds, are selected to calculate average wind direction data based on the accuracy of the prediction model. The multi-time-scale averaging method is used to process historical wind direction data, resulting in smoother and more continuous wind direction data, which can improve the accuracy of model predictions. Multiple time scales can be used to set different yaw conditions, increasing the number of judgments and reducing yaw errors.

[0065] It should be noted that the Attention-CNN-LSTM short-term wind direction prediction model includes a convolutional neural network, a long short-term memory network, and a self-attention mechanism layer. Multiple sets of average wind direction data are input into the Attention-CNN-LSTM short-term wind direction prediction model for training.

[0066] It should be noted that the system introduces a self-attention mechanism to give full play to the advantages of CNN and LSTM in time series prediction, improve the performance of model prediction and the efficiency of learning complex features, meet the dual needs of short-term rapid prediction and prediction accuracy, improve the accuracy of yaw prediction, enable the yaw system to quickly track wind direction changes, and minimize yaw errors.

[0067] It should be noted that wind speed is typically measured by a wind speed sensor. When the wind speed exceeds the starting wind speed, the yaw prediction control system begins to operate and monitor the current wind direction. The starting wind speed of the yaw prediction control system is set. When the wind speed reaches the starting wind speed, the yaw prediction control system can quickly detect changes in wind direction and guide the equipment to perform precise wind-facing operations, avoiding energy waste and equipment wear caused by excessive yaw.

[0068] It should be noted that this process extracts wind direction features through the convolutional neural network within the model, then models the time series through the long short-term memory network, and combines the attention mechanism to focus on important features, and finally outputs the prediction result, that is, the average wind direction angle for the next time period.

[0069] It should be noted that the average wind direction angle is multi-time scale, corresponding to the average wind direction data of multiple time scales, that is, the ten-second average wind direction data corresponds to the predicted corresponding ten-second average wind direction angle. At the same time, different yaw conditions are set for each different time scale to increase the number of judgments and reduce the yaw error.

[0070] The technical solution of this embodiment uses multiple sets of average wind direction data at different time scales for model training, improving the model's prediction accuracy. Furthermore, multiple time scales can be used to set multiple yaw conditions, increasing the number of judgments and reducing yaw errors. Furthermore, by introducing the Attention-CNN-LSTM short-term wind direction prediction model, prediction accuracy and speed are improved, enabling the yaw system to quickly track wind direction changes and minimize yaw errors.

[0071] Furthermore, in some embodiments of the present invention, Figure 1 Step S100 in the illustrated embodiment also includes but is not limited to the following steps:

[0072] (1) Setting multiple time scales, averaging the historical wind direction data according to each time scale to obtain multiple groups of average wind direction data corresponding to the multiple time scales;

[0073] Multiple time scales can correspond to setting multiple yaw conditions, increasing the number of judgments and reducing yaw errors; multiple sets of average wind direction data increase the number of model training times and improve model accuracy;

[0074] (2) Using a sliding window rolling processing method, multiple groups of average wind direction data are processed separately to obtain multiple groups of training data;

[0075] The sliding window stride is 1, and multiple sets of average wind direction data are processed into multiple sets of training data. Each set of training data is a one-dimensional matrix, including 40 sampling points of input data and 20 sampling points of output data. This format is conducive to model training. It is understood that the number of sampling points for input data and output data can also be other numbers.

[0076] (3) Normalizing multiple sets of training data respectively to obtain multiple sets of standardized training data;

[0077] Normalization processing makes the data more evenly distributed in the model, improves the training accuracy and speed of the training model, and improves the accuracy of model prediction.

[0078] It should be noted that the multiple time scales include ten seconds, twenty seconds, and thirty seconds, and the average wind direction data include ten-second average wind direction data, twenty-second average wind direction data, and thirty-second average wind direction data; the calculation formula for the average wind direction data is:

[0079]

[0080] in, It is expressed as the N-second average wind direction data at the i-th moment, Expressed as the average wind direction j seconds before the i-th moment; i and j are positive integers; N is 10, 20 or 30.

[0081] The calculation formula for normalization is:

[0082]

[0083] train norm =(train-train mean ) / train std

[0084] Among them, A ij is the data of the pth row and qth column of the training data matrix, R is the total number of rows of the training data matrix, train j-mean is the average value of the average wind direction data in the qth column, train q-std is the standard value of the average wind direction data in the qth column; train norm is the normalized training data.

[0085] Furthermore, in some embodiments of the present invention, Figure 1 Step S200 in the illustrated embodiment also includes but is not limited to the following steps:

[0086] (1) Input each set of standardized training data into the convolutional neural network separately;

[0087] (2) The convolutional neural network extracts the feature data of each set of standardized training data, expands the feature data into a one-dimensional sequence, and sends it to the long short-term memory network;

[0088] (3) After the long short-term memory network predicts the one-dimensional sequence, the output prediction data enters the self-attention mechanism layer;

[0089] (4) The self-attention mechanism layer obtains the output result based on the predicted data.

[0090] The convolutional neural network consists of convolutional and pooling layers. The convolutional layer uses a 2×1 convolution kernel to extract feature data from the standardized training data, generating ten 39×1 output features. The pooling layer uses a 2×1 pooling window with a stride of 1 to pool the output features, outputting the pooled feature data. This pooled feature data is then flattened into a 390×1 one-dimensional sequence via a flattening layer and fed into the long short-term memory network. This helps capture local features and temporal information in the data. The pooling layer further reduces the dimensionality and pools the feature maps, reducing data redundancy while retaining key information. This feature extraction method can improve the model's understanding and analysis of wind direction data. It is understood that the size of the convolution kernel, pooling window, and stride can be set to meet the needs of the prediction model's accuracy.

[0091] The LSTM network uses 32 hidden units in its layers. It predicts the one-dimensional sequence and outputs multiple 32x1 matrices X, representing the predicted data, which are then fed into the self-attention mechanism layer. The introduction of the LSTM network enables the model to handle long-term dependencies in time series data. The LSTM network predicts the one-dimensional sequence extracted by the convolutional neural network, capturing the temporal variations in wind direction data and improving the model's time series prediction capabilities. The number of hidden units can be adjusted to meet the needs of the prediction model's accuracy.

[0092] The self-attention layer header value is set to 1, and the key value is set to 2. The batch size is 50, meaning that 50 samples are selected from the final state for training during each training cycle, i.e., the matrix X with a batch size of 50 is selected for training. The epoch number is 100, meaning that all samples are trained 100 times. The initial learning rate is 0.001. The historical average wind direction data input is 40 data points, and the prediction time step is 20 data points. It is understood that the parameters for batch size, epoch number, initial learning rate, historical average wind direction data input, and prediction time step can be set to other appropriate parameters.

[0093] In this embodiment, each matrix X will pass through three different weight matrices W Q 、W K and W V Converted into vectors Q, K and V respectively; through dot product and scaling operations, the attention score matrix a is obtained n , use the Softmax function to calculate the above attention score a n Normalize and get the weight coefficient Finally, the weight coefficient The weighted summation with the value vector V is used to obtain the final output I;

[0094] Q=XW Q +bias Q

[0095] K=XW K +bias K

[0096] V=XW v +bias V

[0097]

[0098] Among them, n and N represent the attention score matrix a n Dimension; W Q 、W K and W V is the corresponding randomly initialized weight matrix, which is updated during the training process and has a size of 2×32; X is the 32x1 input matrix; I is the 32×1 output matrix; K T is the transposed matrix of K, d k is a scaling factor, the size of which is the dimension of the key of the layer head of the self-attention mechanism; bias Q 、bias K and bias V is the bias matrix, both of size 2×1.

[0099] Furthermore, in some embodiments of the present invention, Figure 1 Step S600 in the illustrated embodiment also includes but is not limited to the following steps:

[0100] (1) Determine the average wind direction angle for the next period and the absolute value of the error angle of the cabin position at the current moment;

[0101] (2) When the absolute value of the error angle is greater than the preset angle value within a preset time period, the yaw action is performed.

[0102] It should be noted that the average wind direction angle includes a ten-second average wind direction angle, a twenty-second average wind direction angle, and a thirty-second average wind direction angle; the preset time period includes a first preset time period, a second preset time period, and a third preset time period;

[0103] Furthermore, in some embodiments of the present invention, step S620 further includes:

[0104] Set the preset time period to T and the counter to T1=0;

[0105] When the absolute value of the average wind direction angle at the last moment of the next period and the error angle of the cabin position at the current moment is greater than the preset angle, T1=T1+1, that is, T1 increases by 1; continue to predict the average wind direction angle at the next moment, when the absolute value of the predicted average wind direction angle and the error angle of the cabin position at the current moment is less than or equal to the preset angle, T1=0; when T1 and T are equal, it means that the absolute value of the average wind direction angle of the next period and the error angle of the cabin position at the current moment are both greater than the preset angle within the preset time period, and the yaw action is executed at this time.

[0106] Among them, the average wind direction angle is second-level data, that is, when the preset time T = N seconds is set (N is a positive integer), the average wind direction angle of the next time period needs to be predicted once every second within N seconds, and the N consecutive predicted average wind direction angles within N seconds need to be greater than the preset angle in order to perform the yaw action.

[0107] Further, in some embodiments of the present invention, the first preset time period is 30 seconds, the second preset time period is 20 seconds, the third preset time period is 10 seconds, and the preset angle value is 10 degrees;

[0108] When the absolute value of the error angle between the ten-second average wind direction angle at the 5th second of the next period and the current cabin position is greater than 10 degrees within 30 seconds, the yaw action is performed; or,

[0109] When the absolute value of the error angle between the twenty-second average wind direction angle at the 10th second of the next period and the current cabin position is greater than 10 degrees within 20 seconds, the yaw action is performed; or,

[0110] When the absolute value of the error angle between the thirty-second average wind direction angle at the 15th second of the next period and the current nacelle position is greater than 10 degrees within 10 seconds, the yaw action is performed.

[0111] The 5th second of the next period refers to a sequence of average wind direction angles for the next 5 seconds predicted based on the wind direction data at the current moment, and the average wind direction angle at the 5th second is taken as the predicted data.

[0112] In this embodiment, the yaw action includes positive yaw and negative yaw. When the error angle is greater than or equal to 1 degree, positive yaw is performed; when the error angle is less than or equal to -1 degree, negative yaw is performed. When the absolute value of the error angle is less than 1 degree, the yaw action is completed.

[0113] The yaw prediction control system predicts multiple average wind direction angles and sets multiple corresponding yaw conditions for these average wind direction angles to determine whether to execute a yaw action. Furthermore, by introducing the concepts of a preset time period and a preset angle value, it ensures that a yaw action is only executed when the absolute value of the error angle is greater than the preset angle value for a continuous period of time. This can effectively avoid frequent yaws caused by short-term fluctuations in wind direction and enhance the stability and reliability of the yaw prediction control system. At the same time, the embodiment defines two actions, positive yaw and reverse yaw, respectively, based on the positive and negative values ​​of the error angle. This yaw strategy based on the direction of the error angle is more accurate.

[0114] In order to further evaluate the performance of the yaw prediction control system, three aspects will be evaluated: the accuracy of wind direction tracking, the use of yaw actuators, and power generation.

[0115] The accuracy of wind direction tracking is indicated by two performance indicators: mean absolute error and root mean square error. The calculation formula is as follows:

[0116] θ ye =θ wd -θ np

[0117]

[0118] Among them, θ ye is the yaw error, θ wd and θ np are wind direction angle and nacelle position angle respectively, MAE(θ ye ) is the mean absolute error of the yaw error, RMSE(θ ye ) is the root mean square error of the yaw error;

[0119] The usage of the yaw actuator is indicated by two performance indicators: yaw action time and yaw action times. The calculation formula is as follows:

[0120]

[0121] Among them, t yaw is the activation time of the yaw actuator, is the nacelle yaw speed. This formula indicates that when the yaw speed is not 0, the yaw action time begins to accumulate. yaw is the number of activations of the yaw actuator. This formula indicates that when the current yaw speed is different from the previous time, the number of activations of the yaw actuator increases by 1.

[0122] The power increase is indicated by the power reduction factor indicator, and the calculation formula is as follows:

[0123]

[0124] Among them, P a is the output power of the wind turbine when there is a yaw error. ξ is the power reduction factor of the wind turbine. P ideal and P red They represent the output power of wind turbines under ideal conditions and actual conditions respectively. is the yaw error at the jth moment, f j Its probability distribution.

[0125] Five hours of second-level wind direction data from Guishan Island were used to conduct performance evaluation and verify the superiority of the present invention in yaw prediction control.

[0126] In Table 1, the CNN and LSTM wind direction prediction models show similar prediction performance, but their structures are simple and their prediction performance is insufficient compared to the CNN-LSTM hybrid neural network. Based on the CNN-LSTM model, the self-attention mechanism is introduced to show the determination coefficient R 2 The accuracy of wind direction prediction increased by 0.05589, the mean absolute error (MAE) decreased by 0.2103, and the root mean square error (RMSE) decreased by 0.3469, further improving the accuracy of wind direction prediction.

[0127] Table 2 shows the statistical results of the evaluation indicators of three yaw control systems. C1 represents traditional yaw control (referred to as control system 1), C2 represents yaw control without wind direction prediction (referred to as control system 2), and C3 represents yaw control combined with wind direction prediction using Attention-CNN-LSTM (referred to as control system 3). System 3 exhibits the best performance in Table 2, with the smallest power reduction (3.267%). Considering the installed capacity of the offshore wind farm is 500MW and the effective power generation time is 3500 hours per year, the implementation of System 3 will generate an additional 2848.3MWh of power. The wind farm's power generation capacity can be increased by 0.15%.

[0128]

[0129] Table 1 Performance statistics of 30-second average wind direction prediction for four deep learning models

[0130]

[0131] Table 2 Statistical results of evaluation indicators of three yaw control systems

[0132] It is understandable that when comparing the performance of the four deep learning models of CNN, LSTM, CNN-LSTM, and Attention-CNN-LSTM, each deep learning model involves the same training parameter settings.

[0133] Figure 6 The middle red line represents the traditional yaw control system (referred to as control system 1). Figure 7 The middle red line represents the yaw control system without wind direction prediction (referred to as control system 2). Figure 6 and Figure 7 The yellow line in the middle represents the yaw control system (referred to as control system 3) that combines Attention-CNN-LSTM wind direction prediction.

[0134] Figures 6 and 7 This is a comparison of the cabin positions of three yaw control systems under the same wind conditions. In the time period from 17000 to 19000, it can be seen that control system 2 is the first to make a judgment on the wind direction change, followed by control system 3, and control system 1 is the last. There is a significant lag between the wind direction and the cabin position in control system 1. After a period of time when the wind direction changes, the yaw action adjusts the cabin to the specified position. Figure 2 、 Figure 3 It can be seen that control system 2 has more sensitive parameters than control system 1, but while it quickly responds to wind direction changes, it also yaws frequently. In contrast, control system 3 has smoother yaw movements and a more accurate yaw response to the wind.

[0135] Figures 8-10 The blue line in the middle represents the traditional yaw control system (referred to as control system 1), the red line represents the yaw control system without wind direction prediction (referred to as control system 2), and the yellow line represents the yaw control system combined with Attention-CNN-LSTM wind direction prediction (referred to as control system 3).

[0136] Figures 8-10The results of the three control systems are shown in the following graphs: total error, total number of yaw movements, and total yaw time. The proposed control system 3 achieves the lowest total error, total number of yaw movements, and total yaw time of the three control systems, at 159,525°, 44 movements, and 390 seconds, respectively, significantly outperforming control system 1. Control system 1 has the largest yaw error of 166,457° and the lowest number of yaw movements of 44 and 452 seconds. It uses larger control parameters, sacrificing a certain amount of yaw error in exchange for a lower number of yaw movements. Control system 2 reduces the yaw error to 163,721, but increases the number of yaw movements and yaw time to 100 and 824 seconds, resulting in low yaw efficiency. In contrast, the proposed control system 3 meets the yaw accuracy requirements for wind resistance, reducing the yaw error by 2.6% compared to control system 1, while also improving yaw efficiency. The number of yaw movements and time are reduced by 56% and 53.8% compared to control system 2, avoiding frequent yaw.

[0137] In summary, it can be seen that under yaw control using historical wind direction, System 2 has smaller control parameters than System 1. Frequent yaw control only slightly improves wind accuracy, reducing yaw error by 1.64%, which is unacceptable. However, under the same strategy, Control System 3, which incorporates wind direction prediction, not only achieves a smaller yaw error than Control System 2, but also simultaneously meets the dual requirements of fewer yaws and less time, reducing yaw error by 2.6% compared to Control System 1. Therefore, the proposed Control System 3, which incorporates wind direction prediction, effectively improves the wind accuracy of traditional yaw control systems and increases the power generation efficiency of wind turbines.

[0138] In a second aspect, an embodiment of the present invention further provides an offshore wind turbine yaw prediction control system, including the offshore wind turbine yaw prediction control method as described above.

[0139] In a third aspect, the present invention also provides a yaw prediction control device for an offshore wind turbine, such as Figure 2 As shown, including:

[0140] The processor 701 may be implemented as a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is configured to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0141] The memory 702 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 702 can store an 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 702 and is called by the processor 701 to run the offshore wind turbine yaw prediction control system of the embodiments of this application.

[0142] Input / output interface 703, used to implement information input and output;

[0143] Communication interface 704, used to implement communication interaction between the apparatus and other devices, which can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.);

[0144] Bus 705 , which transmits information between various components of the device (e.g., processor 801 , memory 802 , input / output interface 703 , and communication interface 704 );

[0145] The processor 701 , the memory 702 , the input / output interface 703 and the communication interface 704 are connected to each other in communication within the device via a bus 705 .

[0146] In a fourth aspect, an embodiment of the present invention further provides a storage medium, which is a computer-readable storage medium and stores a computer program. When the computer program is executed by a processor, the above-mentioned offshore wind turbine yaw prediction control method is implemented.

[0147] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory optionally includes a memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of the above-mentioned networks include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof. The device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and are located in one place, or may be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment.

[0148] Those skilled in the art will appreciate that all or some of the steps and systems in the method disclosed above can be implemented as software, firmware, hardware, and appropriate combinations thereof. Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, and the computer-readable medium can include computer storage media (or non-transitory media) and communication media (or temporary media). As known to those skilled in the art, the term computer storage media is included in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data) and is volatile and non-volatile, removable, and non-removable. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory, or other memory technology, CD-ROM, digital versatile disks (DVD), or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage, or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, as is well known to those skilled in the art, communication media typically includes computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0149] Throughout this specification, references to terms such as "one embodiment," "further embodiments," "some specific embodiments," or "some examples" indicate that the specific features, structures, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0150] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

Claims

1. A method for predicting and controlling the yaw of an offshore wind turbine, characterized in that: The method comprises: Acquire historical wind direction data, and process the historical wind direction data using a multi-time-scale average method to obtain multiple sets of average wind direction data; Based on multiple sets of average wind direction data, an Attention-CNN-LSTM short-term wind direction prediction model is established; Based on the Attention-CNN-LSTM short-term wind direction prediction model, a yaw prediction control system is established; When the wind speed exceeds the starting wind speed, the yaw prediction control system starts to operate and monitors the current wind direction; Predicting the average wind direction angle for the next period based on the current wind direction and the Attention-CNN-LSTM short-term wind direction prediction model; Executing corresponding yaw actions according to the average wind direction angle in the next period and the current position of the nacelle; The step of obtaining historical wind direction data and processing the historical wind direction data using a multi-time scale average method to obtain multiple groups of average wind direction data includes: setting multiple time scales, and averaging the historical wind direction data according to each time scale to obtain multiple groups of average wind direction data corresponding to the multiple time scales; using a sliding window rolling processing method to process the multiple groups of average wind direction data to obtain multiple groups of training data; and normalizing the multiple groups of training data to obtain multiple groups of standardized training data. The plurality of groups of average wind direction data include ten-second average wind direction data, twenty-second average wind direction data, and thirty-second average wind direction data; The calculation formula of the average wind direction data is: in, It is expressed as the N-second average wind direction data at the i-th moment, It is expressed as the average wind direction j seconds before the i-th moment; i and j are positive integers; N is 10, 20 or 30; The Attention-CNN-LSTM short-term wind direction prediction model includes a convolutional neural network, a long short-term memory network, and a self-attention mechanism layer. The step of establishing the Attention-CNN-LSTM short-term wind direction prediction model based on multiple groups of average wind direction data includes: inputting each group of standardized training data into the convolutional neural network respectively; extracting feature data of each group of standardized training data by the convolutional neural network, and expanding the feature data into a one-dimensional sequence, and sending it to the long short-term memory network; after the long short-term memory network performs prediction processing on the one-dimensional sequence, outputting the predicted data to the self-attention mechanism layer; and the self-attention mechanism layer obtains an output result based on the predicted data. The step of performing a corresponding yaw action based on the average wind direction angle in the next time period and the cabin position at the current moment includes: determining the absolute value of the error angle between the average wind direction angle in the next time period and the cabin position at the current moment; when the absolute value of the error angle is greater than a preset angle value within a preset time period, performing a yaw action.

2. The offshore wind turbine yaw prediction control method according to claim 1, characterized in that: The average wind direction angle includes a ten-second average wind direction angle, a twenty-second average wind direction angle, and a thirty-second average wind direction angle; the preset time period includes a first preset time period, a second preset time period, and a third preset time period; The step of performing a corresponding yaw action according to the average wind direction angle in the next period and the current position of the nacelle includes: When the absolute value of the ten-second average wind direction angle in the next period and the error angle of the nacelle position at the current moment are both greater than the preset angle value within the first preset time period, performing a yaw action; or, When the absolute value of the twenty-second average wind direction angle in the next period and the error angle of the nacelle position at the current moment are both greater than the preset angle value within the second preset time period, performing a yaw action; or, When the absolute values ​​of the thirty-second average wind direction angle in the next period and the error angle of the nacelle position at the current moment are both greater than the preset angle value within the third preset time period, a yaw action is performed.

3. The offshore wind turbine yaw prediction control method according to claim 1, characterized in that: The yaw action includes forward yaw and reverse yaw; when the error angle is greater than or equal to 1 degree, forward yaw is performed; when the error angle is less than or equal to -1 degree, reverse yaw is performed.

4. A yaw prediction control system for an offshore wind turbine, characterized in that: The method comprises the offshore wind turbine yaw prediction control method according to any one of claims 1 to 3.

5. A yaw prediction control device for an offshore wind turbine, characterized in that: The invention comprises at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed 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 operate the offshore wind turbine yaw prediction control system as claimed in claim 4.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the offshore wind turbine yaw prediction control method according to any one of claims 1 to 3.

Images