Wind turbine control method and system based on wind farm wind condition prediction

Abstract

The embodiment of the application provides a wind turbine control method and system based on wind farm wind condition prediction, and belongs to the technical field of wind power. The method comprises the following steps: collecting operation condition information of a target wind turbine, and preprocessing the operation condition information; taking the preprocessed operation condition information as a model input parameter, performing wind speed prediction in real time based on a pre-trained wind speed prediction model, and obtaining a time series wind speed prediction value; performing average turbulence intensity prediction in a predetermined time period based on the time series wind speed prediction value; and matching and executing a corresponding wind turbine control scheme based on a wind speed value interval in which the time series wind speed prediction value is located and a turbulence intensity value interval in which the average turbulence intensity is located. Through accurate wind speed and turbulence prediction, the application realizes the intervention of a precise wind power equipment adaptive adjustment strategy, and improves the operation stability and economic benefits of the wind turbine.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind power technology, in particular to a wind turbine control method based on wind farm wind condition prediction and a wind turbine control system based on wind farm wind condition prediction. BACKGROUND

[0002] Currently, with the increasing prominence of environmental and energy issues, developing renewable energy has become one of the core tasks of global economic and technological development. As one of China's main renewable energy sources, wind energy has the advantages of being free, clean, and pollution-free, and has provided important protection for national economic development while meeting energy demand. Wind power, as the main way of utilizing wind energy, plays an important role in China's renewable energy field. However, despite the great potential of wind power, existing wind farms still face many challenges in actual operation.

[0003] Among them, the most significant problem is that the wind speed in the environment where the wind farm is located has high instability and volatility. This volatility leads to instability in power generation during wind power generation, directly affecting the efficiency of power generation and the stability of the power grid. Existing wind measurement devices often have problems of insufficient wind measurement accuracy and reaction lag when facing such complex environmental conditions, especially in the measurement of turbulence intensity. This not only makes wind speed prediction less accurate, but also limits the operation optimization of wind turbines.

[0004] In the prior art, the scheme of adaptive adjustment of wind turbines according to the changes in wind speed and turbulence intensity can theoretically optimize the power generation curve and prolong the service life of the equipment. However, due to the difficulty in accurate prediction of wind speed and turbulence, the adaptive adjustment strategy cannot be accurately intervened. In the absence of accurate prediction data support, the regulation scheme of wind turbines is usually in an inefficient or overly conservative state, and the potential of the equipment cannot be fully utilized. In addition, inaccurate prediction can also cause the protection mechanism of the wind turbine to fail to start in time under extreme weather conditions, increasing the risk of equipment failure. To solve this problem, an adaptive adjustment scheme for the operating state of wind turbines based on accurate wind condition prediction is needed. SUMMARY

[0005] The purpose of the embodiments of the present application is to provide a wind turbine control method and system based on wind farm wind condition prediction to at least solve the problem of insufficient precision of adaptive adjustment of equipment caused by insufficient accuracy of wind condition prediction of existing wind farms.

[0006] In order to achieve the above object, the first aspect of the present application provides a wind turbine control method based on wind farm wind condition prediction, the method comprising: collecting operation condition information of a target wind turbine, and preprocessing the operation condition information; taking the preprocessed operation condition information as a model input parameter, performing wind speed prediction in real time based on a pre-trained wind speed prediction model to obtain a time series wind speed prediction value; performing average turbulence intensity prediction within a predetermined time period based on the time series wind speed prediction value; and matching and executing a corresponding wind turbine control scheme based on a wind speed value interval in which the time series wind speed prediction value is located and a turbulence intensity value interval in which the average turbulence intensity is located.

[0007] Optionally, the preprocessing of the operation condition information comprises: sequentially performing outlier processing, missing value processing and normalization processing on the operation condition information to obtain an initial data set; and performing sliding window processing on the initial data set to obtain operation condition time series data of a preset time length as the data after preprocessing is completed.

[0008] Optionally, the sliding window processing on the initial data set to obtain operation condition time series data of a preset time length comprises: determining a window length of a sliding window based on the preset time length, randomly adjusting a sliding window step, and obtaining a preset number of candidate operation condition time series data in a predetermined historical data length; identifying and comparing data characteristics of each candidate operation condition time series data, and selecting a candidate operation condition time series data with optimal characteristics based on an optimal matching rule as the obtained operation condition time series data; wherein the data characteristics are smoothness and / or volatility.

[0009] Optionally, the pre-training rule of the wind speed prediction model comprises: collecting historical operation condition information of a target wind turbine, and based on the historical condition information, performing wind speed information at each time and corresponding pitch angle information and / or torque at the time; calibrating the wind speed information at each time based on the pitch angle information and / or torque at the corresponding time, constructing a training set and a test set based on the calibration structure; performing model training in a pre-constructed network structure based on the training set to obtain an initial model; verifying the initial model based on the test set, and taking the model that passes the verification as the wind speed prediction model.

[0010] Optionally, the pre-constructed network structure is constructed based on a long short-term memory neural network; each layer of neurons of the long short-term memory neural network comprises sequentially connected forget gate, input gate and output gate, and memory cell.

[0011] Optionally, the forget gate, the input gate and the output gate are respectively represented as:

[0012] f t = σ(ΣW xf x t + ΣW hf xt-1 +ΣW cf x t-1 +b f )

[0013] i t =σ(∑W xi x t +ΣW hi x t-1 +∑W ci x t-1 +b i )

[0014] o t =σ(ΣW xo x t +∑W ho x t-1 +∑W co x t-1 +b o )

[0015] Among them, f t 、i t 、o t are the outputs of the forget gate, input gate, and output gate respectively; σ is the sigmoid activation function; W xf 、W xi 、W xo are the current input x t The weight matrix between the forget gate, input gate and output gate; x t is the input data of the current time step; W hf 、W hi 、W ho are the weight matrices between the hidden state of the previous time step and the forget gate, input gate, and output gate; W cf 、W ci 、W co Divided into the weight matrix between the memory unit state of the previous time step and the forget gate, input gate and output gate; b f 、b i 、b o They are the bias terms of forget gate, input gate and output gate respectively.

[0016] Optionally, the memory unit is used to perform memory unit state update at each time step, and the corresponding update rule is:

[0017] c t =f t c t-1 +i t tanh(∑W xc x t +∑W hc x t-c +bt )

[0018] wherein c t is the memory cell state at the current time step; f t , i t are the outputs of the forget gate and the input gate, respectively; c t-1 is the memory cell state at the previous time step; tanh is the hyperbolic tangent function; W xc , W hc are the weight matrices between the current input x1 and the memory cell state at the previous time step x t-1 ; b c is the bias term of the memory cell.

[0019] Optionally, the output gate is configured to perform the hidden state update at each time step, and the corresponding update rule is:

[0020] h t = o t tanh(c t )

[0021] wherein h t is the hidden state at the current time step; o t is the output of the output gate; tanh is the hyperbolic tangent function; c t is the memory cell state at the current time step.

[0022] Optionally, the average turbulence intensity prediction in the predetermined time period is performed based on the time series wind speed prediction value, including: taking the time series wind speed prediction value as an input parameter of a preset mapping function, performing the average turbulence intensity prediction in the corresponding predetermined time period to obtain the average turbulence intensity prediction value in the predetermined time period; wherein the mapping function is used to represent the mapping relationship between the time series wind speed prediction value and the average turbulence intensity prediction value in the corresponding time period.

[0023] Optionally, the method further includes: performing mapping function construction, including: collecting environmental state information of the target wind turbine in a fixed time length; wherein the environmental state information includes wind speed information, and any one or more of temperature information, air pressure information and humidity information; fitting turbulence information in the corresponding time length based on the environmental state information; fitting the mapping relationship between the wind speed information and the turbulence information based on the wind speed information in the fixed time length and the turbulence information in the corresponding time length.

[0024] Optionally, the mapping relationship between the wind speed information and the turbulence information based on the wind speed information in the fixed time length and the turbulence information in the corresponding time length includes: calculating the average wind speed in the corresponding time length based on the wind speed information in the fixed time length, and constructing a turbulence intensity fitting process reference variable based on the average wind speed, represented as:

[0025]

[0026] wherein, ReV is a turbulence intensity fitting process reference variable; T m is an average difference between the average wind speed in the corresponding time length and a preset expected average wind speed; T rated is a preset expected average wind speed in the time length; β is a fitting process parameter; a mapping relationship fitting between the wind speed information and the turbulence information is performed based on the turbulence intensity fitting process reference variable, which is expressed as:

[0027] I real = K x ReV + b

[0028] wherein, I real is turbulence information in the corresponding time length; K and b are fitting process variables.

[0029] Optionally, based on the wind speed numerical interval where the time series wind speed prediction value is located and the turbulence intensity numerical interval where the average turbulence intensity is located, a corresponding wind turbine control scheme is matched and executed, including: if the time series wind speed prediction value is located in a first wind speed numerical interval and the turbulence intensity is located in a first turbulence intensity numerical interval, it is determined as a first matching relationship, and then a variable pitch action pre-adjustment scheme is matched and executed; if the time series wind speed prediction value is located in a second wind speed numerical interval and the turbulence intensity is located in the first turbulence intensity numerical interval, it is determined as a second matching relationship, and then a power limiting adjustment scheme is matched and executed; if the time series wind speed prediction value is located in the second wind speed numerical interval and the turbulence intensity is located in a second turbulence intensity numerical interval, it is determined as a third matching relationship, and then a power increasing adjustment scheme is matched and executed; wherein the first wind speed numerical interval is smaller than the second wind speed numerical interval, and the first turbulence intensity numerical interval is larger than the second turbulence intensity numerical interval; otherwise, if it does not belong to any one of the first matching relationship, the second matching relationship and the third matching relationship, the wind turbine control is not executed.

[0030] The second aspect of the application provides a wind turbine control system based on wind farm wind condition prediction, the system comprising: a collection unit for collecting operating condition information of a target wind turbine; a preprocessing unit for preprocessing the operating condition information; a wind speed prediction unit for taking the preprocessed operating condition information as a model input parameter, performing real-time wind speed prediction based on a pre-trained wind speed prediction model, and obtaining a time series wind speed prediction value; a turbulence mapping unit for performing average turbulence intensity prediction in a predetermined time period based on the time series wind speed prediction value; and a control unit for matching and executing a corresponding wind turbine control scheme based on the wind speed numerical interval where the time series wind speed prediction value is located and the turbulence intensity numerical interval where the average turbulence intensity is located.

[0031] In another aspect, the present application provides a computer readable storage medium, which stores instructions that, when executed on a computer, cause the computer to perform the wind turbine control method based on wind farm wind condition prediction described above.

[0032] Through the above technical solution, the present application ensures the accuracy and consistency of input data by collecting and preprocessing the operating condition information of the target wind turbine. Then, the pre-trained wind speed prediction model is used to obtain real-time time series wind speed prediction values, ensuring that the wind speed prediction can respond to environmental changes in a timely manner. Based on this, the average turbulence intensity in the predetermined time period is further predicted, thereby providing dual reference basis for wind speed and turbulence. Next, according to the numerical interval of the time series wind speed and turbulence intensity, the most suitable wind turbine control scheme is matched. This strategy not only optimizes the power generation efficiency of the wind turbine and ensures accurate control under different wind conditions, but also prolongs the service life of the equipment and avoids the risk of equipment failure caused by extreme weather or wind speed fluctuations. In summary, this scheme improves the operation stability and economic benefits of the wind turbine through accurate wind speed and turbulence prediction.

[0033] Other features and advantages of the present application will be described in detail in the following detailed description. BRIEF DESCRIPTION OF DRAWINGS

[0034] The accompanying drawings are included to provide a further understanding of the present application and constitute a part of the specification, which together with the following detailed description, serve to explain the present application. In the drawings:

[0035] Figure 1 is a step flow chart of the wind turbine control method based on wind farm wind condition prediction provided by an embodiment of the present application;

[0036] Figure 2 is an internal module structure diagram of the LSTM network provided by an embodiment of the present application;

[0037] Figure 3 is an actual wind speed time series diagram provided by an embodiment of the present application;

[0038] Figure 4 is a predicted wind speed time series diagram provided by an embodiment of the present application;

[0039] Figure 5 is a system structure diagram of the wind turbine control system based on wind farm wind condition prediction provided by an embodiment of the present application. DETAILED DESCRIPTION

[0040] The specific embodiments of the present application are described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely intended for illustration and explanation of the present application and are not intended to limit the present application.

[0041] Figure 1 is a method flowchart of a wind turbine control method based on wind farm wind condition prediction provided by an embodiment of the present application. As shown in Figure 1 , the embodiment of the present application provides a wind turbine control method based on wind farm wind condition prediction, which comprises the following steps:

[0042] Step S10: Collecting operation condition information of a target wind turbine and pre-processing the operation condition information.

[0043] Specifically, the operation condition information is real-time pitch angle information and / or real-time torque of the wind turbine.

[0044] In the embodiment of the present application, although the conventional anemometer can provide wind speed data, due to the rapidity and complexity of wind speed change, the measurement result often has a lag, which cannot reflect the current wind condition change in time, thereby affecting the real-time regulation of the wind turbine. By directly selecting the measurable variables of the wind turbine which have high correlation with wind speed and turbulence intensity, especially the pitch angle and torque value, this scheme can more quickly obtain the operation state information related to wind speed and turbulence. As the key parameters of the wind turbine operation state, the pitch angle and torque can indirectly reflect the changes of wind speed and turbulence without direct measurement of the anemometer. This prediction method based on the operation information of the wind turbine itself can greatly reduce the feedback lag of wind speed measurement and improve the real-time and accuracy of wind speed prediction.

[0045] Based on the scheme of the present application, the scheme of the present application avoids the delay feedback problem caused by the conventional anemometer and significantly improves the response speed of the wind turbine. Through real-time collection and analysis of the pitch angle and torque value, the system can more quickly make a prediction of wind speed and turbulence intensity, and then more accurately adjust the operation state of the wind turbine. In this way, not only the power generation efficiency of the wind farm can be improved, but also the service life of the wind turbine can be effectively prolonged, and the loss and failure risk of the equipment due to the reaction lag can be reduced.

[0046] Further, the pre-processing of the operation condition information comprises: sequentially performing outlier processing, missing value processing and normalization processing on the operation condition information to obtain an initial data set; and performing sliding window processing on the initial data set to obtain condition time series data of a preset time length as the data after the pre-processing is completed.

[0047] In the embodiments of the present application, the preprocessing process includes outlier processing, missing value processing and normalization processing on the operating condition information. The main purpose of this step is to eliminate noise data, repair potential errors caused by data missing, and unify different dimensional values to the same scale through normalization, so as to facilitate subsequent model processing. Outlier processing can remove extreme data points caused by equipment failure or other incidental factors, ensuring the reliability of model input data; missing value processing can make up for discontinuity in the data acquisition process through interpolation or filling techniques; and normalization processing can reduce the adverse effects of different orders of magnitude data on model training by mapping the data to a uniform range (such as 0 to 1).

[0048] Further, after the preprocessing is completed, the scheme further performs sliding window processing on the initial data set. The sliding window technique is a common means for processing time series data, which can ensure that continuous time series data of a fixed time length can be obtained each time. By setting the window length and step size, the sliding window can capture the dynamic change trend of the wind turbine in a certain time period, thereby better reflecting the time series characteristics of wind speed and turbulence. Time series data of a preset time length can provide stable input for subsequent models, thereby improving the prediction accuracy.

[0049] Specifically, the sliding window processing is performed on the initial data set to obtain operating condition time series data of a preset time length, including: determining the window length of the sliding window based on the preset time length, randomly adjusting the step size of the sliding window, and obtaining a preset number of candidate operating condition time series data in a predetermined historical data length; identifying and comparing the data characteristics of each candidate operating condition time series data, and selecting the candidate operating condition time series data with the optimal characteristics based on an optimization matching rule as the obtained operating condition time series data; wherein the data characteristics are smoothness and / or volatility.

[0050] In the embodiments of the present application, the time series data of the latest window length is directly used as the collected data. This passive collection method is simple and has low implementation cost, and is suitable for scenarios with high real-time requirements. However, since it is based on the latest time series data without further optimization selection, it may not guarantee that the collected data has optimal characteristics, for example, the smoothness or volatility of the data may be poor, thereby affecting the training effect of the subsequent model.

[0051] Based on this, the scheme of the present application can ensure that the collected data is more in line with the needs of subsequent models by actively selecting time series data with better features. Specifically, the active acquisition scheme first sets the window length of the sliding window and allows the step of the sliding window to be adjusted randomly, thereby generating multiple candidate operating condition time series data in the most recent period of time. These candidate time series data cover the operating characteristics of different time periods, and therefore can more comprehensively reflect the historical operating state of the wind turbine. Next, the system identifies and compares the features of these candidate data, mainly investigating the smoothness and volatility of the data. These two data features can reflect the stability and consistency of the time series data, and are key indicators for optimizing the accuracy of subsequent model training. Finally, the system selects the candidate time series data with the best features based on the optimization matching rule as the input data for model training.

[0052] Based on the scheme of the present application, through the active acquisition scheme, the feature excellent data that best meets the model requirements can be selected from among numerous candidate time series data, thereby greatly improving the accuracy and stability of model training. Compared with passive acquisition, active acquisition not only ensures the smoothness and consistency of time series data, but also reduces noise interference caused by excessive data volatility. This optimization method ensures that the model can obtain high-quality training data, thereby providing more accurate results in wind speed and turbulence intensity prediction and helping the wind turbine to achieve more stable operation and higher power generation performance. In addition, through the random adjustment of the step of the sliding window and the comparison of the features of multiple candidate data, the generalization ability of the model can be enhanced, making it better adapt to different wind conditions and complex working conditions.

[0053] Step S20: The preprocessed operating condition information is used as a model input parameter, and real-time wind speed prediction is performed based on the pre-trained wind speed prediction model to obtain a time series wind speed prediction value.

[0054] Specifically, before performing prediction, corresponding wind speed prediction model training needs to be performed, historical operating condition information of the target wind turbine is collected, and wind speed information at each time and corresponding time blade angle information and / or torque are obtained based on the historical operating condition information; the wind speed information at each time is calibrated based on the corresponding time blade angle information and / or torque, and a training set and a test set are constructed based on the calibration structure; the training set is used to train the model in a pre-constructed network structure to obtain an initial model; the test set is used to verify the initial model, and the model that passes the verification is used as the wind speed prediction model.

[0055] In the embodiments of the present application, the scheme requires collecting historical operating condition information of the target wind turbine, including pitch angle information and / or torque information of the wind turbine at different times, which reflects the actual operating state of the wind turbine. By correlating the historical operating condition information with the wind speed information at the corresponding time, the system can provide accurate input data for the subsequent wind speed prediction model. In particular, the pitch angle and torque, which have a direct correlation with the wind speed, can reflect the trend of the change of the wind speed to some extent. After collecting the data, the system will calibrate the data. The calibration process is to one-to-one correspond the pitch angle information and / or torque information at each time with the wind speed data, establishing a mapping relationship between the two. This process is equivalent to providing a mapping reference between the input and output of the model, so that the model can understand the correlation between the operating condition parameters of the wind turbine and the change of the wind speed. After calibration, the system will construct a training set and a test set based on the calibration structure. The training set is used for model training, providing a large amount of historical data to help the model learn the complex relationship between wind speed and operating condition information; and the test set is used for subsequent model verification to ensure that the model not only performs well on the training set, but also has strong generalization ability on unknown data.

[0056] Further, the model is trained in the pre-constructed network structure. By using the training set, the model can gradually optimize its parameters, learn the non-linear relationship between the pitch angle information, torque information and wind speed, and finally obtain an initial wind speed prediction model. Finally, the initial model is verified based on the test set. If the model performs as expected on the test set, i.e. the predicted wind speed is highly consistent with the true wind speed, the model is considered to pass the verification and is put into use as the formal wind speed prediction model.

[0057] Based on the scheme of the present application, the wind speed prediction model can accurately learn the change rule of the wind speed based on the historical operating condition information of the wind turbine, especially the key variables such as pitch angle and torque. The calibration process ensures the correlation between the data input and output, and the construction of the training set and the test set ensures the scientificity and integrity of the model training and verification process. Through the training and verification of the model, the final wind speed prediction model not only has high precision, but also has good generalization ability in the face of different wind conditions. The core advantage of this scheme is that it can improve the accuracy of wind speed prediction, help the wind turbine make timely and accurate adjustments under complex and variable wind conditions, and thus improve the power generation efficiency and prolong the service life of the equipment. In addition, the process of model training and verification ensures the reliability and stability of the wind speed prediction model in actual application, further optimizing the operating effect of the wind turbine.

[0058] Further, the pre-constructed network structure is constructed based on a long short-term memory neural network; each layer of neurons of the long short-term memory neural network comprises sequentially connected forgetting gates, input gates and output gates, and memory cells.

[0059] In the embodiment of the present application, the pre-constructed network structure adopts a long short-term memory neural network (LSTM), which is a deep learning model specially designed for processing long time series data. The traditional recurrent neural network (RNN) often encounters the problem of gradient disappearance or gradient explosion when processing long sequences, especially in the training process of long time series data, the network cannot effectively retain the information of early input, thereby causing poor prediction effect of the model. In order to solve this problem, the LSTM network is selected, and the core advantage of which is that it can well handle long time dependence and effectively solve the problem of gradient disappearance.

[0060] Figure 2 The internal module structure diagram of the LSTM network is shown in the figure, the internal structure of the LSTM is composed of three important gate units: forgetting gate (f t ), input gate (i t ) and output gate (o t ). They control the input to adjust the state of the memory cell to retain or forget the past information, and ensure that the model can effectively handle long-range dependence problems in the processing of long sequence data. The specific process is represented as:

[0061] 1) Forgetting gate (f t ): control whether to retain the memory state h t-1 at the last time. The cross multiplication symbol in the figure represents an element-level multiplication operation, combined with f t and h t-1 in the above figure, it is explained that the gate unit retains or forgets part of the past memory information.

[0062] 2) Input gate (i t ): decide whether to update the memory cell after processing the current input x t and the previous output y t-1 . The input gate uses the tanh function to perform nonlinear transformation on the input, and then determines how to update the memory cell through the gate unit.

[0063] 3) Output gate (o t ): the output gate decides the output at the current time t, which combines the current memory state and the data processed by the tanh function to generate the current hidden state h t and output y t .

[0064] Further, the LSTM network increases the unit control mechanism on the basis of the RNN, ensures that the error can be transmitted in a long time step, and avoids the gradual loss of information over time. Each LSTM unit contains three key parts: a forgetting gate, an input gate, and an output gate, which control the preservation, update, and output of information, respectively. Specifically, the forgetting gate decides whether to discard the previous memory at the current time, the input gate is responsible for writing new information to the memory unit, and the output gate controls the output of the information at the current time to the next layer of neurons.

[0065] Further, the memory unit of the LSTM can retain important historical information between multiple time steps through these gate mechanisms. Unlike traditional RNNs, each neuron in the LSTM network not only processes the input at the current time, but also effectively saves and transmits key information from previous time steps, allowing long-term sequence dependencies to be maintained. This design enables the LSTM model to effectively handle long-term dependencies in the wind speed prediction task, as wind speed data has obvious time series characteristics, and the future operating conditions of the wind turbine are largely dependent on the historical wind speed trends.

[0066] Based on the scheme, by introducing the LSTM neural network, the gradient disappearance problem of the RNN in long sequence training can be effectively solved, and the model has good prediction accuracy when processing long time series. The design of the LSTM network enables the error to be transmitted in a long time step, ensuring that important information is not lost in the network layer-by-layer transmission. Therefore, the wind speed model can make full use of historical wind speed data to make more accurate predictions. In addition, the gating mechanism in the LSTM allows the model to finely control the input information, so that the model can not only retain valuable long-time information, but also forget useless information in time, further improving the training efficiency and prediction effect of the model. Through this scheme, the wind turbine can achieve more stable and accurate wind speed prediction, thereby optimizing the power generation efficiency and equipment operating state.

[0067] Specifically, the forgetting gate, the input gate, and the output gate are represented as:

[0068] f t =σ(∑W xf x t +∑W hf x t-1 +∑W cf x t-1 +b f )

[0069] i t =σ(∑W xi x t +∑W hi x t-1+∑W ci x t-1 +b i )

[0070] o t =σ(∑W xo x t +∑W ho x t-1 +∑W co x t-1 +b o )

[0071] where f t , i t , o t are the outputs of the forget gate, input gate and output gate respectively; σ is the sigmoid activation function; W xf , W xi , W xo are the weight matrices between the current input x t and the forget gate, input gate and output gate respectively; x t is the input data of the current time step; W hf , W hi , W ho are the weight matrices between the hidden state of the previous time step and the forget gate, input gate and output gate respectively; W cf , W ci , W co are the weight matrices between the memory cell state of the previous time step and the forget gate, input gate and output gate respectively; b f , b i , b o are the bias terms of the forget gate, input gate and output gate respectively.

[0072] Specifically, the memory cell is configured to perform memory cell state update at each time step, and the corresponding update rule is:

[0073] c t = f t c t-1 + i t tanh(∑W xc x t +∑W hc x t-1 +b c )

[0074] where c t is the memory cell state of the current time step; f t , i t are the outputs of the forget gate and input gate respectively; c t-1 is the memory cell state of the previous time step; tanh is the hyperbolic tangent function; Wxc , W hc are the current input x t and the previous time step input x t-1 and the weight matrix between the memory cell state; b c is the bias term of the memory cell.

[0075] Specifically, the output gate is used to perform the hidden state update of each time step, and the corresponding update rule is:

[0076] h t = o t tanh(c t )

[0077] where h t is the hidden state of the current time step; o t is the output of the output gate; tanh is the hyperbolic tangent function; c t is the memory cell state of the current time step.

[0078] Step S30: performing average turbulence intensity prediction in a predetermined time period based on the time series wind speed prediction value.

[0079] Specifically, the time series wind speed prediction value is taken as the input parameter of the preset mapping function, and the average turbulence intensity prediction in the corresponding predetermined time period is performed to obtain the average turbulence intensity prediction value in the predetermined time period; wherein the mapping function is used to represent the mapping relationship between the time series wind speed prediction value and the average turbulence intensity prediction value in the corresponding time period.

[0080] In the embodiment of the present application, the predicted time series wind speed is taken as the input of the preset mapping function, and the mapping relationship between the wind speed and the turbulence intensity is realized through this function. The function of the mapping function is to convert the wind speed change characteristics in a period of time into a turbulence intensity prediction value, so as to derive the turbulence intensity based on the wind speed data without directly measuring the turbulence. This method can effectively utilize the wind speed prediction data to generate a more comprehensive environmental condition evaluation. The construction of the mapping function is based on the complex physical relationship between the wind speed and the turbulence intensity. Through the fitting of historical data and the optimization of the model, the mapping function can accurately predict the average turbulence intensity in a period of time under different wind speed fluctuation conditions. The application of this mapping relationship enables the wind turbine to reasonably evaluate the turbulence intensity that will occur according to the current wind speed prediction, so as to make optimized operation adjustment.

[0081] Specifically, the method further comprises: performing mapping function construction, including: collecting environmental state information of the target wind turbine within a fixed time length; wherein the environmental state information includes wind speed information, and any one or more of temperature information, air pressure information, and humidity information; fitting turbulence information within a corresponding time length based on the environmental state information; fitting a mapping relationship between wind speed information and turbulence information based on wind speed information within a fixed time length and turbulence information within a corresponding time length.

[0082] In the embodiments of the present application, the environmental state information of the target wind turbine within a fixed time length is collected, which includes not only the core wind speed information, but also various factors such as temperature, air pressure, and humidity. By introducing these environmental factors, the actual operating environment of the wind turbine can be more comprehensively reflected, and the problem of inaccurate prediction caused by single dependence on wind speed information can be avoided. Temperature, air pressure, and humidity have a significant impact on wind speed and turbulence intensity, and by integrating these environmental information, the relationship between wind speed and turbulence intensity can be more accurately captured.

[0083] Specifically, based on the collected environmental state information, turbulence information within a corresponding time length is fitted. This process involves combining wind speed information with other environmental state data to generate turbulence information within a period of time through model fitting. The fitting process can be based on analysis of historical data to establish a suitable model to predict turbulence intensity. The fitted turbulence information can provide more reliable reference for the operation and regulation of the wind turbine.

[0084] In one possible implementation, it is assumed that the following information is collected in a certain wind farm:

[0085] Time period: 600 seconds.

[0086] Wind speed information: In this period, the average wind speed is 7.5 m / s, and the standard deviation is 2.1 m / s.

[0087] Temperature information: The temperature in this period remains around 15℃, with small changes.

[0088] Air pressure information: The air pressure is 1012 hPa, with slight fluctuations.

[0089] Humidity information: The humidity is 60%, which is relatively stable.

[0090] Based on these data, the specific process of fitting turbulence intensity is as follows:

[0091] 1) Data preprocessing: The collected wind speed, temperature, air pressure, and humidity information is standardized, for example, the wind speed is normalized to the interval of 0-1, to ensure that the scales of the variables are consistent.

[0092] 2) Model Selection: Choose a linear regression model as the fitting model, or, depending on the task requirements, a deep learning model such as LSTM for complex time series fitting. Assuming a linear regression model is selected, the model input variables are wind speed, temperature, air pressure, and humidity, and the output is turbulence intensity.

[0093] 3) Model training: Use historical wind speed and environmental status data to train a linear regression model to understand the impact of wind speed and environmental conditions on turbulence intensity.

[0094] 4) Turbulence Fitting: The model uses the wind speed (7.5 m / s, standard deviation 2.1 m / s), temperature (15°C), pressure (1012 hPa), and humidity (60%) data collected over the past 600 seconds to fit the turbulence intensity for that period. For example, the model predicts a turbulence intensity of 0.25, indicating relatively weak turbulence during that period.

[0095] Output result: The fitted turbulence intensity of 0.25 is passed to the control system of the wind turbine. The wind turbine can adjust its operating strategy based on the turbulence prediction value, such as adjusting the blade angle to cope with turbulence changes.

[0096] Furthermore, after obtaining wind speed and turbulence information, the system further establishes a precise mapping model between the two by fitting the mapping relationship between wind speed and turbulence information. This mapping model can predict turbulence intensity in real time based on changes in wind speed, forming a correlation between wind speed and turbulence. This process is trained and optimized based on previously collected data to ensure that the mapping relationship is accurately applicable under different wind conditions. By using wind speed and turbulence data over a fixed time period, the mapping model can effectively learn the complex nonlinear relationship between the two, thereby providing highly accurate turbulence predictions.

[0097] Specifically, the mapping relationship fitting between the wind speed information and the turbulence information based on the wind speed information within a fixed time length and the turbulence information within the corresponding time length includes: calculating the wind speed mean within the corresponding time length based on the wind speed information within the fixed time length, and constructing a turbulence intensity fitting process reference variable based on the wind speed mean, which is expressed as:

[0098]

[0099] Among them, ReV is the reference variable of the turbulence intensity fitting process; T m is the average difference between the mean wind speed within the corresponding time length and the preset expected wind speed mean; T rated is the preset expected wind speed mean within the time length; β is the fitting process parameter; the mapping relationship between wind speed information and turbulence information is fitted based on the reference variable of the turbulence intensity fitting process, which is expressed as:

[0100] I real = K x Rev + b

[0101] where I real is the turbulence information within the corresponding time length; K and b are fitting process variables.

[0102] Step S40: Based on the wind speed value interval where the time series wind speed prediction value is located and the turbulence intensity value interval where the average turbulence intensity is located, the corresponding wind turbine control scheme is matched and executed.

[0103] Specifically, if the time series wind speed prediction value is in the first wind speed value interval and the turbulence intensity is in the first turbulence intensity value interval, it is determined as the first matching relationship, and the wind turbine control scheme matched and executed is the pitch action pre-adjustment scheme; if the time series wind speed prediction value is in the second wind speed value interval and the turbulence intensity is in the first turbulence intensity value interval, it is determined as the second matching relationship, and the wind turbine control scheme matched and executed is the power limit adjustment scheme; if the time series wind speed prediction value is in the second wind speed value interval and the turbulence intensity is in the second turbulence intensity value interval, it is determined as the third matching relationship, and the wind turbine control scheme matched and executed is the power increase adjustment scheme; wherein the first wind speed value interval is smaller than the second wind speed value interval, and the first turbulence intensity value interval is greater than the second turbulence intensity value interval; otherwise, if it does not belong to any one of the first matching relationship, the second matching relationship and the third matching relationship, the wind turbine control is not executed.

[0104] In the embodiments of the present application, the scheme dynamically adjusts the control strategy of the wind turbine through the matching relationship between the wind speed prediction value and the turbulence intensity, to cope with the operating state of the wind turbine under different wind conditions. The scheme is mainly divided into four types of wind conditions, each of which is classified according to the average wind speed and turbulence intensity value, and different control strategies are executed for different matching conditions, such as pitch pre-adjustment, power limit adjustment or power increase operation. Specifically, it includes:

[0105] 1) The first matching relationship (rated wind speed high turbulence wind condition):

[0106] Condition: When the average wind speed is between 8-12 m / s (first wind speed value interval) and the turbulence intensity is between 0.12-0.2 (first turbulence intensity value interval).

[0107] Control strategy: It is determined as the rated wind speed high turbulence wind condition. At this time, the wind turbine may face large wind speed fluctuations, therefore, in order to reduce the impact of turbulence on the unit, the controller will execute the pitch action pre-adjustment scheme, that is, by adjusting the pitch angle in advance to cope with possible wind speed changes, thereby reducing the impact on the wind turbine and optimizing the running stability.

[0108] 2) The second matching relationship (high wind speed high turbulence wind condition):

[0109] Condition: When the average wind speed is between 12-20 m / s (second wind speed numerical range) and the turbulence intensity is between 0-0.08 (second turbulence intensity numerical range).

[0110] Control strategy: It is determined as high wind speed and low turbulence wind condition. In this case, although the wind speed is high, the turbulence intensity is low, and the environment is relatively stable, the wind turbine can fully utilize the high wind speed for power generation. At this time, the controller will execute the power increase adjustment scheme, that is, increase the power output and improve the power generation efficiency under the premise of ensuring the safety of the equipment.

[0111] 3) Third matching relationship (high wind speed and low turbulence wind condition):

[0112] Condition: When the average wind speed is between 12-20 m / s (second wind speed numerical range) and the turbulence intensity is between 0-0.08 (second turbulence intensity numerical range).

[0113] Control strategy: It is determined as high wind speed and low turbulence wind condition. In this case, although the wind speed is high, the turbulence intensity is low, and the environment is relatively stable, the wind turbine can fully utilize the high wind speed for power generation. At this time, the controller will execute the power increase adjustment scheme, that is, increase the power output and improve the power generation efficiency under the premise of ensuring the safety of the equipment.

[0114] 4) Other wind conditions (not matching any condition):

[0115] Condition: If the current wind speed and turbulence intensity do not meet the above three matching relationships, the system is determined as other wind conditions.

[0116] Control strategy: At this time, the controller will not perform additional operations, maintain the current running state, avoid unnecessary adjustments, and reduce unnecessary energy consumption and regulation delay.

[0117] Based on the scheme of the present application, the scheme of the present application can flexibly cope with different wind conditions according to different combinations of wind speed and turbulence intensity. Especially in extreme conditions such as high turbulence and high wind speed, the advance execution of pitch and power limiting control can effectively avoid equipment overload and wind turbine loss, and protect the running safety of the equipment. At the same time, in the case of relatively stable wind condition and low turbulence, the system can timely increase the power, fully utilize the wind energy, and improve the power generation efficiency. The core of the whole control scheme is to closely combine the running state of the wind turbine with the changes of wind speed, turbulence and other environmental parameters, so that the wind turbine can realize intelligent adjustment and maintain high efficiency in complex wind conditions, and significantly improve the service life and operation reliability of the equipment. Through this dynamic matching adjustment scheme, the wind farm can maintain the optimal running state in different wind conditions, and improve the overall energy utilization rate.

[0118] In one possible embodiment, the wind speed is predicted based on the wind farm wind condition prediction-based wind turbine control method proposed in the present application scheme, and compared with the measured wind speed. Figure 3 is a measured wind speed time series diagram, Figure 4 is a predicted wind speed time series diagram. The horizontal axis represents time in seconds (s), and the vertical axis represents wind speed in meters per second (m / s). The measured wind speed curve shows the fluctuations of wind speed in the actual environment, while the LSTM predicted wind speed curve reflects the model's prediction of future wind speed after neural network training. As can be seen, Figure 3 and Figure 4 demonstrates the prediction accuracy of the LSTM model, helping to verify its effectiveness in practical applications. As can be seen from the figure, although the measured wind speed fluctuates to some extent, the LSTM model can accurately predict the trend of wind speed changes, and the time series trends of the two are highly consistent. Through comparative analysis, the advantages of the LSTM model in wind condition prediction are further confirmed, especially in dealing with the instability and volatility of wind speed.

[0119] Figure 5 is a system structure diagram of a wind farm wind condition prediction-based wind turbine control system provided by an embodiment of the present application. As Figure 5 shown, the present application embodiment provides a wind farm wind condition prediction-based wind turbine control system, which comprises: a collection unit for collecting the operating condition information of a target wind turbine and preprocessing the operating condition information; a wind speed prediction unit for taking the preprocessed operating condition information as a model input parameter, performing real-time wind speed prediction based on a pre-trained wind speed prediction model, and obtaining a time series wind speed prediction value; a turbulence mapping unit for performing average turbulence intensity prediction within a predetermined time period based on the time series wind speed prediction value; and a control unit for matching and executing a corresponding wind turbine control scheme based on the wind speed value range in which the time series wind speed prediction value is located and the turbulence intensity value range in which the average turbulence intensity is located.

[0120] The present application embodiment further provides a computer-readable storage medium having instructions stored thereon, which, when executed on a computer, causes the computer to perform the above-mentioned wind farm wind condition prediction-based wind turbine control method.

[0121] Those skilled in the art can understand that all or part of the steps of the method for implementing the above-mentioned embodiments can be completed by programs instructing relevant hardware, the programs are stored in a storage medium, and the programs include a plurality of instructions for enabling a single-chip microcomputer, a chip or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media capable of storing program codes.

[0122] The optional embodiments of the present application are described in detail above in combination with the drawings, but the embodiments of the present application are not limited to the specific details in the above-described embodiments. Within the technical concept scope of the embodiments of the present application, various simple modifications can be made to the technical solutions of the embodiments of the present application, and these simple modifications all belong to the protection scope of the embodiments of the present application. In addition, it should be noted that each specific technical feature described in the above-described specific embodiments can be combined in any appropriate manner without contradiction. In order to avoid unnecessary repetition, the embodiments of the present application will not be described again for various possible combinations.

[0123] In addition, various different embodiments of the present application can also be combined in any manner, as long as it does not deviate from the idea of the embodiments of the present application, and it should also be considered as disclosed by the embodiments of the present application.

Claims

1. A wind turbine control method based on wind farm wind condition prediction, characterized in that: The method comprises: Collecting operating condition information of the target wind turbine generator set and preprocessing the operating condition information; The pre-processed operating condition information is used as the model input parameter, and the wind speed prediction is performed in real time based on the pre-trained wind speed prediction model to obtain the time series wind speed prediction value; The pre-training rules of the wind speed prediction model include: collecting historical operating condition information of the target wind turbine, the historical operating condition information including wind speed information at each moment and blade angle information and / or torque at the corresponding moment; calibrating the wind speed information at each moment based on the blade angle information and / or torque at the corresponding moment, and constructing a training set and a test set based on the calibration structure; performing model training in a pre-constructed network structure based on the training set to obtain an initial model; verifying the initial model based on the test set, and using the verified model as the wind speed prediction model; Performing average turbulence intensity prediction within a predetermined time period based on the time series wind speed prediction value; Based on the wind speed value interval of the time series wind speed prediction value and the turbulence intensity value interval of the average turbulence intensity, the corresponding wind turbine control scheme is matched and executed; wherein, Based on the wind speed numerical interval in which the time-series wind speed prediction value is located and the turbulence intensity numerical interval in which the average turbulence intensity is located, a corresponding wind turbine control scheme is matched and executed, including: if the time-series wind speed prediction value is in a first wind speed numerical interval and the turbulence intensity is in the first turbulence intensity numerical interval, it is determined that a first matching relationship exists, then the wind turbine control scheme matched and executed is a pitch action pre-adjustment scheme; if the time-series wind speed prediction value is in a second wind speed numerical interval and the turbulence intensity is in the first turbulence intensity numerical interval, it is determined that a second matching relationship exists, then the wind turbine control scheme matched and executed is a power limiting adjustment scheme; if the time-series wind speed prediction value is in the second wind speed numerical interval and the turbulence intensity is in the second turbulence intensity numerical interval, it is determined that a third matching relationship exists, then the wind turbine control scheme matched and executed is a power increasing adjustment scheme; wherein the first wind speed numerical interval is smaller than the second wind speed numerical interval, and the first turbulence intensity numerical interval is larger than the second turbulence intensity numerical interval; otherwise, if it does not belong to any of the first matching relationship, the second matching relationship, and the third matching relationship, the wind turbine control is not executed.

2. The method according to claim 1, characterized in that Preprocessing the operating condition information includes: Performing outlier processing, missing value processing, and normalization processing on the operating condition information in sequence to obtain an initial data set; Sliding window processing is performed on the initial data set to obtain operating condition time series data of a preset time length as data after preprocessing is completed.

3. The method according to claim 2, characterized in that Performing sliding window processing on the initial data set to obtain operating condition time series data of a preset time length includes: Determine the window length of the sliding window based on a preset time length, randomly adjust the sliding window step size, and obtain a preset number of candidate working condition time series data in a predetermined historical data length; Identify and compare the data features of each candidate operating condition time series data, and select the candidate operating condition time series data with the best features based on the optimization matching rule as the obtained operating condition time series data; wherein, The data are characterized by smoothness and / or volatility.

4. The method according to claim 1, wherein The pre-built network structure is constructed based on a long short-term memory neural network; Each layer of neurons in the long short-term memory neural network includes a forget gate, an input gate, an output gate, and a memory unit that are connected in sequence.

5. The method according to claim 4, characterized in that The forget gate, input gate and output gate are respectively expressed as: in, 、 、 are the outputs of the forget gate, input gate, and output gate respectively; is the sigmoid activation function; 、 、 The current input The weight matrix between the forget gate, input gate and output gate; is the input data of the current time step; 、 、 are the weight matrices between the hidden state of the previous time step and the forget gate, input gate, and output gate respectively; 、 、 It is divided into the weight matrix between the memory unit state of the previous time step and the forget gate, input gate and output gate; 、 、 They are the bias terms of forget gate, input gate and output gate respectively.

6. The method according to claim 4, characterized in that The memory unit is used to perform the memory unit state update at each time step, and the corresponding update rule is: ； in, is the memory unit state at the current time step; 、 are the outputs of the forget gate and input gate respectively; is the memory unit state at the previous time step; is the hyperbolic tangent function; 、 The current input and the previous time step input The weight matrix between the memory cell state and the memory cell state; is the bias term of the memory unit.

7. The method according to claim 4, characterized in that The output gate is used to perform hidden state updates at each time step, and the corresponding update rule is: ； in, is the hidden state of the current time step; is the output of the output gate; is the hyperbolic tangent function; is the memory cell state at the current time step.

8. The method according to claim 1, characterized in that The average turbulence intensity forecast for a predetermined time period is performed based on the time series wind speed forecast value, including: The time series wind speed prediction value is used as the input parameter of the preset mapping function, and the average turbulence intensity prediction within the corresponding predetermined time period is performed to obtain the average turbulence intensity prediction value within the predetermined time period; wherein, The mapping function is used to characterize the mapping relationship between the time series wind speed prediction value and the average turbulence intensity prediction value in the corresponding time period.

9. The method according to claim 8, characterized in that The method further comprises: Execute mapping function construction, including: Collect environmental status information of target wind turbines within a fixed time period; The environmental status information includes wind speed information, and any one or more of temperature information, air pressure information, and humidity information; Performing turbulence information fitting within a corresponding time length based on the environmental state information; The mapping relationship between the wind speed information and the turbulence information is fitted based on the wind speed information within a fixed time length and the turbulence information within a corresponding time length.

10. The method according to claim 9, characterized in that The step of fitting the mapping relationship between the wind speed information and the turbulence information based on the wind speed information within a fixed time length and the turbulence information within a corresponding time length includes: Based on the wind speed information within a fixed time length, the wind speed mean within the corresponding time length is calculated, and the reference variable of the turbulence intensity fitting process is constructed based on the wind speed mean, which is expressed as: ； in, is the reference variable for the turbulence intensity fitting process; is the average difference between the mean wind speed within the corresponding time length and the preset expected wind speed mean; is the preset expected wind speed mean within the time length; is the fitting process parameter; The mapping relationship between wind speed information and turbulence information is fitted based on the reference variables in the turbulence intensity fitting process, which is expressed as: ； in, is the turbulence information within the corresponding time length; K and b are fitting process variables.

11. A wind turbine control system based on wind farm wind condition prediction, characterized in that: The system is used to execute the wind turbine control method based on wind farm wind condition prediction according to any one of claims 1 to 10, and the system includes: A collection unit, used to collect operating condition information of the target wind turbine; A preprocessing unit, configured to preprocess the operating condition information; The wind speed prediction unit is used to use the pre-processed operating condition information as a model input parameter, perform wind speed prediction based on the pre-trained wind speed prediction model in real time, and obtain a time series wind speed prediction value; a turbulence mapping unit, configured to perform an average turbulence intensity prediction within a predetermined time period based on the time series wind speed prediction value; The control unit is used to match and execute the corresponding wind turbine control scheme based on the wind speed numerical interval of the time series wind speed prediction value and the turbulence intensity numerical interval of the average turbulence intensity.

Images