Wind direction prediction method and device for a wind farm

Abstract

The present disclosure provides a wind direction prediction method and device for a wind farm, the wind direction prediction method comprising the following steps: obtaining historical wind direction data of the wind farm; performing wind direction correlation analysis on the obtained historical wind direction data; determining a nonlinear regression model according to the wind direction correlation analysis result, and predicting the wind direction of a wind turbine in the wind farm based on the determined nonlinear regression model. The present disclosure can more accurately predict the wind direction of the wind turbine at different sites, and when the wind vane of the wind turbine fails, the predicted wind direction is used for wind direction compensation to keep the wind turbine at a high wind power efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind power technology, in particular to a wind direction prediction method and device for a wind farm. BACKGROUND

[0002] Since the wind turbine is operated outdoors at high altitude, and the wind direction indicator of the wind turbine is placed outside the cabin cover of the wind turbine, the operating environment of the wind direction indicator of the wind turbine is very complex: vibration, dust, sun exposure, freezing, rain, and various extreme conditions; the installed units of offshore and beach wind farms also have to withstand years of salt spray corrosion, so the probability of failure of the wind direction indicator is relatively high.

[0003] Generally, the emergency control method taken by the prior art when the wind direction indicator of the wind turbine fails is: all wind turbines in the wind farm are centered on themselves, and the three wind turbines with the closest geographical position are formed into a small network through the SCADA system; when the wind turbine is operating normally, the wind direction signal collected by the wind turbine is fed back to the central controller through the SCADA system, and the central controller displays the various data of the wind turbine in real time, including the wind direction; when the wind direction indicator of the wind turbine fails, the central controller displays the failed wind direction indicator in a conspicuous position and sends a warning to the monitoring personnel; at the same time, assuming that the wind directions of the three adjacent wind turbines are similar, the data of the wind direction indicators of the similar wind turbines are transmitted back to the central controller, and after the central controller confirms that the information of 1-2 wind turbines is effective after screening, the SCADA system selects and sends the selected data to the wind turbine with the failed wind direction indicator, so that the wind turbine can obtain wind direction data and continue to operate after the wind direction indicator fails.

[0004] However, since the prior art only assumes that the wind directions of adjacent wind turbines are the same, and does not consider the change in wind direction after a certain distance, the prediction of the wind direction is greatly deviated, and it is difficult to meet the safe operation of the wind turbine, therefore, a more accurate method and device for predicting the wind direction of a certain wind turbine and compensating for the wind direction of the wind turbine with a failed wind direction indicator are needed. SUMMARY

[0005] In order to solve the above problems and / or disadvantages, and at least provide the advantages described below, the present disclosure provides a wind direction prediction method and device for a wind farm.

[0006] An aspect of the present application is to provide a wind direction prediction method for a wind farm, the wind direction prediction method comprising the steps of: obtaining historical wind direction data of the wind farm; performing wind direction correlation analysis on the obtained historical wind direction data; determining a nonlinear regression model according to the wind direction correlation analysis result, and predicting the wind direction of a wind turbine in the wind farm based on the determined nonlinear regression model.

[0007] The step of performing wind direction correlation analysis on the obtained historical wind direction data can comprise: preprocessing the historical wind direction data; dividing the preprocessed data into training tuples and test tuples, respectively calculating the distance between the test tuples and each training tuple; selecting K training tuples with the smallest distance, and setting the label with the highest occurrence frequency in the K training tuples as the category of the test tuple, and dividing the training tuples into at least one region according to the category of the test tuple.

[0008] The step of performing wind direction correlation analysis on the obtained historical wind direction data can further comprise: calculating the error rate, and selecting the K value corresponding to the minimum error rate, wherein the K value corresponding to the minimum error rate represents that the wind directions of K wind turbines in the wind farm have nonlinear correlation.

[0009] The step of determining the nonlinear regression model according to the wind direction correlation analysis result can comprise: performing nonlinear regression analysis on the historical wind direction data in different regions respectively, and establishing a nonlinear regression model according to the result of the nonlinear regression analysis; constructing a cost function according to the nonlinear regression model and the historical wind direction data; introducing a relaxation variable into the constructed cost function to establish an optimization function; and determining the nonlinear regression model by calculating the minimum value of the optimization function.

[0010] The step of determining the nonlinear regression model by calculating the minimum value of the optimization function can comprise: constructing a dual problem function by introducing a Lagrange operator into the established optimization function; obtaining equations for each variable in the dual problem function by calculating the dual problem function; substituting the obtained equations for each variable into the established optimization function to obtain a new optimization function; and determining the nonlinear regression model by calculating the minimum value of the new optimization function.

[0011] The wind direction prediction method can further comprise: performing reinforcement learning on the determined nonlinear regression model.

[0012] The wind direction prediction method can further comprise: when a wind vane of a specific wind turbine fails, predicting the wind direction of the specific wind turbine based on the determined nonlinear regression model, and using the predicted wind direction to compensate for the wind direction of the specific wind turbine; and monitoring the wind power curve of the specific wind turbine to determine whether the specific wind turbine continues to operate.

[0013] The step of monitoring the wind power curve of the specific wind turbine to determine whether the specific wind turbine continues to operate can comprise: if an abnormality is detected in the wind power curve of the specific wind turbine, stopping the specific wind turbine; and if no abnormality is detected in the wind power curve of the specific wind turbine, maintaining the operation of the specific wind turbine using the predicted wind direction.

[0014] Another aspect of the present application is to provide a wind direction prediction device for a wind farm, comprising: a data acquisition module configured to acquire historical wind direction data of the wind farm; a data analysis module configured to perform wind direction correlation analysis on the acquired historical wind direction data; a wind direction prediction module configured to determine a nonlinear regression model according to the wind direction correlation analysis result, and predict the wind direction of a specific wind turbine in the wind farm based on the determined nonlinear regression model.

[0015] The data analysis module can be configured to: preprocess the historical wind direction data acquired by the data acquisition module; divide the preprocessed data into training tuples and test tuples, and calculate the distance between the test tuples and each training tuple respectively; select K training tuples with the smallest distance, and set the label with the highest occurrence frequency in the K training tuples as the class of the test tuple, and divide the training tuples into at least one region according to the class of the test tuple.

[0016] The data analysis module can be further configured to: calculate the error rate, and select the K value corresponding to the minimum error rate, wherein the K value corresponding to the minimum error rate represents that the wind directions of K wind turbines in the wind farm have nonlinear correlation.

[0017] The wind direction prediction module can be configured to: perform nonlinear regression analysis on the historical wind direction data in different regions respectively, and establish a nonlinear regression model according to the result of the nonlinear regression analysis; construct a cost function according to the nonlinear regression model and the historical wind direction data; establish an optimization function by introducing a slack variable to the constructed cost function; and determine the nonlinear regression model by calculating the minimum value of the optimization function.

[0018] The wind direction prediction module can be further configured to: construct a dual problem function by introducing a Lagrange operator to the established optimization function; obtain the equations for each variable in the dual problem function by calculating the dual problem function; obtain a new optimization function by substituting the obtained equations for each variable into the established optimization function; and determine the nonlinear regression model by calculating the minimum value of the new optimization function.

[0019] The wind direction prediction device can further comprise a reinforcement learning module configured to perform reinforcement learning on the determined nonlinear regression model.

[0020] The wind direction prediction device can further comprise a wind direction compensation module configured to: when a wind vane of a specific wind turbine fails, perform wind direction compensation on the specific wind turbine using the wind direction of the specific wind turbine predicted by the determined nonlinear regression model; and monitor the wind power curve of the specific wind turbine to determine whether the specific wind turbine continues to operate.

[0021] The wind direction compensation module can also be configured to: if it is monitored that the wind power curve of the specific wind turbine is abnormal, shut down the specific wind turbine; and if it is monitored that the wind power curve of the specific wind turbine is not abnormal, keep the specific wind turbine running using the predicted wind direction.

[0022] Another aspect of the present application provides a computer readable storage medium storing a computer program, wherein the computer program, when executed by a processor, causes the processor to perform the wind direction prediction method described above.

[0023] Another aspect of the present application provides a computer device comprising a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, causes the processor to perform the wind direction prediction method described above.

[0024] The wind direction prediction method for a wind farm described above can predict the wind direction of each wind turbine at different sites according to the environment of the wind turbine at each site. Using correlation analysis and regression analysis modeling based on machine learning, the wind direction of a wind turbine can be more accurately predicted, so that the wind turbine can still maintain high wind power efficiency after the anemometer fails, effectively reduce downtime as an emergency system, improve the availability of wind turbines, increase the unit output of wind turbines, and significantly improve economic benefits. BRIEF DESCRIPTION OF DRAWINGS

[0025] The above and other aspects, features, and advantages of the exemplary embodiments of the present application will be more apparent from the following description taken in conjunction with the accompanying drawings, in which:

[0026] Figure 1 A flowchart showing a wind direction prediction method for a wind farm according to an embodiment of the present application is shown;

[0027] Figure 2 A diagram showing correlation analysis using a K-Nearest Neighbor (KNN) algorithm according to an embodiment of the present application is shown;

[0028] Figure 3 A flowchart showing a method of determining a nonlinear regression model according to an embodiment of the present application is shown;

[0029] Figure 4 A flowchart showing a method for wind direction compensation for a wind farm according to an embodiment of the present application is shown;

[0030] Figure 5 A block diagram showing a wind direction prediction device for a wind farm according to an embodiment of the present application is shown.

[0031] In the drawings, like reference numerals will be understood to refer to like elements, features, and structures. DETAILED DESCRIPTION

[0032] The following description with reference to the accompanying drawings is provided to assist in a comprehensive understanding of exemplary embodiments of the present application as defined by the claims and their equivalents. The following description includes various specific details to assist in that understanding but these details are to be regarded as merely illustrative. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the application. In addition, descriptions of well-known functions and constructions can be omitted for clarity and conciseness.

[0033] Hereinafter, exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings. It should be understood that the wind direction prediction method for a wind farm and the wind direction prediction apparatus according to the exemplary embodiments of the present application can be applied to various wind turbines, and by analyzing historical wind direction data and modeling using correlation analysis and regression analysis based on machine learning, the wind direction of a certain wind turbine can be more accurately predicted, so that the wind turbine can maintain a high wind power efficiency even after a failure of a wind vane, effectively reduce downtime as an emergency system, improve the availability of wind turbines, increase the unit output of wind turbines, and significantly improve economic efficiency.

[0034] Figure 1 A flowchart of a wind direction prediction method for a wind farm according to an embodiment of the present application is shown.

[0035] Referring to Figure 1 In step S101, historical wind direction data of a wind farm is acquired. According to an embodiment of the present application, the acquired historical wind direction data can include historical wind direction data of a wind farm read from a SCADA system.

[0036] In step S102, wind direction correlation analysis is performed on the acquired historical wind direction data. First, the historical wind direction data acquired in step S101 is preprocessed, and then wind direction correlation analysis is performed on the preprocessed data.

[0037] According to the embodiment of the present disclosure, the correlation analysis can be performed using the KNN algorithm. Generally, the wind farm where the wind turbine is located has a large area, and not all wind turbines in the wind farm have correlation. Some wind turbines in the wind farm can have correlation in wind direction due to terrain reasons. It is difficult to calculate the air fluid mechanics by establishing a physical model, and the calculation can only be performed one by one for different wind farms. In the embodiment, the KNN algorithm is used to perform correlation analysis, and a mathematical model is constructed to predict the wind turbines in the wind farm that are relatively close in wind direction. The KNN algorithm is mainly applied to the identification of unknown things, that is, to determine which category the unknown thing belongs to. The KNN algorithm judgment idea is based on Euclidean theorem, and the characteristics of the unknown thing are determined to be closest to the characteristics of the known things in which category.

[0038] After preprocessing the obtained historical wind direction data, the preprocessed data is divided into training tuples and test tuples. Based on Euclidean theorem, the distance between the test tuples and each training tuple is calculated, wherein when calculating the similarity between two points, the Euclidean distance is usually used, and its equation can be expressed as:

[0039]

[0040] wherein x j represents the test tuple, y j represents the training tuple, d euc (x, y) represents the distance between the two points.

[0041] Then, the K training tuples with the smallest distance calculated according to equation (1) are selected, and the label with the highest frequency in the K training tuples is set as the category of the test tuple. The training tuples are divided into at least one region according to the category of the test tuple. Finally, the error rate is calculated, and the K value corresponding to the minimum error rate is selected, wherein the K value corresponding to the minimum error rate represents that the wind directions of K wind turbines in the wind farm have nonlinear correlation. The process of performing correlation analysis using the KNN algorithm according to the embodiment of the present disclosure will be described in detail below with reference to Figure 2 .

[0042] Figure 2 A diagram showing the process of performing correlation analysis using the KNN algorithm according to the embodiment of the present disclosure is shown.

[0043] In Figure 2In this embodiment, 16 wind turbines are selected as training data, and are numbered respectively, and the test data is inputted, the Euler distance between the test data and each training data is calculated, the K points with the smallest distance are selected, the frequency of the category of the first K points is determined, and the category with the highest frequency in the first K points is returned as the predicted classification of the test data. From Figure 2 It can be seen that the training data is divided into four regions after correlation analysis by the KNN algorithm, wherein the wind turbines with correlation numbered 01, 02, 03, 04, 05 and 08 are classified into the same region, the wind turbines with correlation numbered 06 and 07 are classified into the same region, the wind turbines with correlation numbered 10, 11, 12 and 13 are classified into the same region, and the wind turbines with correlation numbered 09, 14, 15 and 16 are classified into the same region. In this embodiment, the KNN algorithm is used for correlation analysis, but the present disclosure is not limited to this algorithm.

[0044] Returning to Figure 1 After step S102, the wind farm has been divided into different regions. The wind direction changes of the wind turbines in each divided region have correlation. In step S103, a nonlinear regression model is determined according to the wind direction correlation analysis result, and the wind direction of a specific wind turbine in the wind farm is predicted based on the determined nonlinear regression model. The determination of the nonlinear regression model for predicting the wind direction of a specific wind turbine will be described in detail below with reference to Figure 3 .

[0045] Figure 3 A flowchart of a method for determining a nonlinear regression model according to an embodiment of the present application is shown. According to an embodiment of the present disclosure, a support vector regression machine (SVR) can be used to establish a nonlinear regression model.

[0046] In step S301, nonlinear regression analysis is performed on the historical wind direction data in different regions respectively, and a nonlinear regression model is established according to the results of the nonlinear regression analysis. According to an embodiment of the present application, nonlinear regression analysis is performed on the wind direction data in the four different regions divided in Figure 2 , for example, nonlinear regression analysis is performed on the wind direction data in the region composed of wind turbines numbered 01, 02, 03, 04, 05 and 08. According to the regression analysis result, a nonlinear regression model function d(x) is constructed. d(x) can be expressed in the form of d(x) = (ω·x) + b, wherein x represents historical wind direction data, ω represents a multi-dimensional space weight vector, b represents an intercept, and d(x) represents a predicted wind direction. In the SVR algorithm, the purpose of SVR is to determine the regression function by estimating ω and b.

[0047] At step S302, a cost function is constructed according to the established nonlinear regression model and historical wind direction data. In the present embodiment, according to the SVR algorithm, the cost function is represented as the distance of a set of data to the fitted curve and the maximum value in the distance values is found. The cost function of SVR can be represented as:

[0048] cost(x) = max(0, |d(x) - g(x)| - ε) (2)

[0049] where |d(x) - g(x)| represents the distance of the wind direction data of a certain wind turbine to the regression plane, d(x) represents the predicted wind direction, g(x) represents the actual wind direction value, and ε represents a tolerance value. Since the data cannot all be on the regression plane and the sum of the distances is relatively large, a tolerance value ε can be given to the distance of all data to the regression plane to prevent overfitting. This parameter is an empirical parameter and needs to be given manually. If the distance of the data to the regression plane is less than ε, the cost is 0, otherwise it is |d(x) - g(x)| - ε.

[0050] At step S303, an optimization function is established by introducing a slack variable to the constructed cost function. In this process, the slack variable ξ i 、 The constraint condition is obtained as follows:

[0051]

[0052] As can be seen, after the introduction of the slack variable, the above problem is converted into a problem of minimizing the slack variable. Here, it is assumed that the calculation result of ω satisfies a normal distribution, and according to the Bayesian linear regression model, ω has an L2 norm constraint, so that the optimization function is established as follows:

[0053]

[0054] where C represents a penalty factor, and the value of C is an empirical parameter given manually.

[0055] where equation (4) can be transformed into equation (5):

[0056]

[0057] At step S304, the nonlinear regression model is determined by calculating the minimum value of the optimization function. After the optimization function is established at step S303, a Lagrange operator is introduced to the established optimization function to construct a dual problem function. Specifically, the Lagrange operators α, β * 、 * are introduced into the optimization function, and the optimization problem is converted into a dual problem, where the dual problem function can be represented by the following equation (6):

[0058]

[0059] The partial derivatives of the dual problem function with respect to the variables ω, b, ξ i 、 C are calculated, and then the calculated partial derivatives are set to 0 to obtain the following equations (7), (8), (9), (10), and (11):

[0060]

[0061]

[0062]

[0063]

[0064]

[0065] The obtained equations with respect to the variables ω, b, ξ i 、 C are substituted into the established optimization function (5) to obtain a new optimization function, which can be expressed by the following equation (12):

[0066]

[0067] It should be noted that in the process of substituting the equations with respect to the variables obtained according to the partial derivatives into equation (5), the variables ξ i 、 β, β * have been mutually cancelled in the calculation process.

[0068] In addition, if a kernel function is used, equation (12) can be transformed into equation (13):

[0069]

[0070] After obtaining the new optimization function, the nonlinear regression model is determined by calculating the minimum value of the new optimization function (i.e., equation (12) or equation (13)). Finally, the wind direction of the wind farm is predicted by the determined nonlinear regression model function. However, the present disclosure is not limited to using the SVR algorithm to establish the nonlinear regression model.

[0071] In addition, according to the embodiment of the present application, the present application also comprises reinforcement learning on the determined nonlinear regression model. Reinforcement learning is an important machine learning method, which has many applications in intelligent control and analysis prediction. Specifically, in the case that the wind direction indicator of the wind turbine is in normal operation, the wind direction data is continuously collected and accumulated, the wind direction is continuously predicted according to the wind direction prediction method of the present application, and the predicted value is compared with the actual value. With the continuous increase of the data amount, the loss function value of the fitting curve will be continuously reduced, so that the wind direction prediction model continuously enhances the algorithm learning in the running process to be closer to the actual value, and increases the prediction accuracy.

[0072] Figure 4 A flow chart of a method for wind direction compensation of a wind farm according to an embodiment of the present application is shown.

[0073] Referring to Figure 4 In step S401, the historical wind direction data of the wind farm is acquired. In step S402, the acquired historical wind direction data is subjected to wind direction correlation analysis. In the present embodiment, the KNN algorithm is used to perform the wind direction correlation analysis. In the KNN algorithm, the selected neighbor samples are all objects that have been correctly classified, and the class of the sample to be classified is determined only according to the class of the nearest neighbor sample or samples in the decision-making. For example, in Figure 2 In the present embodiment, the wind turbines numbered 01, 02, 03, 04, 05 and 08 having correlation are classified into the same region, i.e., these wind turbines are determined as one category.

[0074] In step S403, the nonlinear regression model is determined according to the wind direction correlation analysis result. Here, the nonlinear regression model is determined according to the SVR algorithm. The process of step S403 is the same as that of step S103, and will not be described again in the present embodiment.

[0075] In step S404, when the wind direction indicator of a certain wind turbine fails, step S405 is entered. Otherwise, the wind turbine remains in the current running state to continue running.

[0076] In step S405, the wind direction of the wind turbine with the failed wind direction indicator is predicted based on the determined nonlinear regression model, and the wind direction compensation is performed on the wind turbine with the failed wind direction indicator using the predicted wind direction, so that the wind turbine remains running. For example, if the wind direction indicator of a certain wind turbine fails, the wind direction of the wind turbine with the failed wind direction indicator is predicted according to the wind direction correlation analysis result in step S402 and the determined nonlinear regression model in step S403, the predicted wind direction is taken as the wind direction in the current state to maintain the running of the wind turbine, i.e., the wind direction compensation is performed.

[0077] Meanwhile, the wind power curve of the specific wind turbine is monitored to determine whether the specific wind turbine continues to operate. If the monitored wind power curve is abnormal, the wind turbine is shut down to prevent a safety accident; if the monitored wind power curve is not abnormal, the predicted wind direction is used to maintain the operation of the wind turbine, i.e., the wind turbine can continue to operate in the case of failure of the wind vane. In this way, the power generation and economic losses caused by directly shutting down the wind turbine when the wind vane is damaged are avoided.

[0078] Figure 5 A block diagram of a wind direction prediction device for a wind farm according to an embodiment of the present application is shown.

[0079] Referring to Figure 5 The wind direction prediction device 500 includes a data acquisition module 501 configured to acquire historical wind direction data of the wind farm; a data analysis module 502 configured to perform wind direction correlation analysis on the acquired historical wind direction data; a wind direction prediction module 503 configured to determine a nonlinear regression model according to the wind direction correlation analysis result and predict the wind direction of a wind turbine in the wind farm based on the determined nonlinear regression model; a reinforcement learning module 504 configured to perform reinforcement learning on the determined nonlinear regression model; and a wind direction compensation module 505 configured to, when a wind vane of a certain wind turbine fails, perform wind direction compensation on the wind turbine using the wind direction predicted by the determined nonlinear regression model for the wind turbine to keep the wind turbine operating and monitor the wind power curve of the wind turbine to determine whether the wind turbine continues to operate.

[0080] In the process of performing wind direction correlation analysis on the acquired data, the data analysis module 502 pre-processes the historical wind direction data acquired by the data acquisition module 501; divides the pre-processed data into training tuples and test tuples, respectively calculates the distance between the test tuples and each training tuple; selects K training tuples with the smallest distance, sets the label with the most occurrences in the K training tuples as the category of the test tuple, and divides the training tuples into at least one region according to the category of the test tuple.

[0081] After the test tuple set is tested, the data analysis module 502 calculates the error rate and selects the K value corresponding to the minimum error rate, wherein the K value corresponding to the minimum error rate indicates that the wind directions of K wind turbines in the wind farm have nonlinear correlation. It should be noted that this process can use the method of step S102 to perform correlation analysis of the wind direction, which will not be described here again, but the present disclosure is not limited to using the method of step S102.

[0082] After the wind direction correlation analysis, the wind direction prediction module 503 respectively performs nonlinear regression analysis on the historical wind direction data in different regions divided by the data analysis module 502, and determines a nonlinear regression model according to the result of the nonlinear regression analysis, constructs a cost function according to the nonlinear regression model and the historical wind direction data, introduces a slack variable into the constructed cost function to establish an optimization function, and determines the nonlinear regression model by calculating the minimum value of the optimization function. In the determination of the nonlinear regression model, the wind direction prediction module 503 introduces a Lagrange operator into the established optimization function to construct a dual problem function, obtains the equations of each variable in the dual problem function by calculating the dual problem function, substitutes the obtained equations for each variable into the established optimization function to obtain a new optimization function, and determines the nonlinear regression model by calculating the minimum value of the new optimization function. The process is the same as that in step S103, and will not be described in detail here.

[0083] When the wind direction indicator of a certain wind turbine fails, the wind direction compensation module 505 uses the predicted wind direction for the wind turbine as the current wind direction of the wind turbine, compensates the wind direction of the wind turbine, and detects whether the wind power curve of the wind turbine is normal. If it is detected that the wind power curve of the wind turbine is abnormal, the wind turbine is stopped; if it is detected that the wind power curve of the wind turbine is normal, the predicted wind direction is used to keep the wind turbine running.

[0084] In addition, according to the embodiment of the present application, when the wind turbine is running normally, the determined nonlinear regression model can be continuously reinforced by the reinforcement learning module 504. The nonlinear regression model for predicting the wind direction continuously enhances the algorithm learning during the running process to make the predicted wind direction closer to the actual value.

[0085] According to the wind direction prediction method and device for a wind farm, the historical wind direction data is analyzed and the wind direction of the wind farm is modeled, the correlation analysis and the regression analysis modeling based on machine learning are used, the wind direction of a certain wind turbine can be more accurately predicted, the wind turbine can still maintain high wind power efficiency after the wind direction indicator fails, the downtime can be effectively reduced as an emergency system, the availability of the wind turbine can be improved, the unit output of the wind turbine can be increased, and the economic benefits are obvious.

[0086] A wind direction prediction method for a wind farm according to an embodiment of the present application can be implemented as computer readable codes on a computer readable recording medium, or can be transmitted through a transmission medium. The computer readable recording medium is any data storage device that can store data which can be afterward read by a computer system. The computer readable recording medium stores a computer program which, when executed by a processor, causes the processor to perform the wind direction prediction method for a wind farm in the present disclosure. Examples of the computer readable recording medium include a read-only memory (ROM), a random access memory (RAM), a compact disc (CD)-ROM, a digital versatile disc (DVD), a magnetic tape, a floppy disk, and an optical data storage device, but are not limited thereto. The transmission medium can include a carrier wave which is transmitted through a network or various types of communication channels. The computer readable recording medium can also be distributed over a network coupled computer system so that the computer readable codes are stored and executed in a distributed manner.

[0087] Another embodiment of the present application provides a computer device including a processor and a memory storing a computer program which, when executed by the processor, causes the processor to perform the wind direction prediction method for a wind farm in the present disclosure.

[0088] Although the present application has been shown and described with respect to particular exemplary embodiments thereof, it will be apparent to those skilled in the art that various changes in form and detail can be made without departing from the spirit and scope of the application as defined by the following claims and their equivalents.

Claims

1. A wind direction prediction method for a wind farm, characterized in that, The wind direction prediction method comprises the following steps: acquiring historical wind direction data of a wind farm; performing wind direction correlation analysis on the acquired historical wind direction data; determining, according to the wind direction correlation analysis result, a plurality of wind turbine generators in the wind farm having nonlinear correlation with the wind direction of a specific wind turbine generator; determining a nonlinear regression model for the specific wind turbine generator based on the historical wind direction data of the plurality of wind turbine generators, and predicting the wind direction of the specific wind turbine generator based on the determined nonlinear regression model, wherein the nonlinear regression model is determined in the following manner: performing nonlinear regression analysis on the historical wind direction data of the plurality of wind turbine generators, and establishing the nonlinear regression model according to the result of the nonlinear regression analysis; constructing a cost function according to the nonlinear regression model and the historical wind direction data of the plurality of wind turbine generators; establishing an optimization function by introducing a slack variable to the constructed cost function; constructing a dual problem function by introducing a Lagrange operator to the established optimization function; obtaining equations of each variable in the dual problem function by calculating the dual problem function; substituting the obtained equations of each variable into the established optimization function to obtain a new optimization function; and determining the nonlinear regression model by calculating the minimum value of the new optimization function, wherein the step of performing wind direction correlation analysis on the acquired historical wind direction data comprises: preprocessing the historical wind direction data; dividing the preprocessed data into training tuples and test tuples, and calculating the distance between the test tuple and each training tuple respectively; selecting K training tuples with the smallest distance, setting the label with the highest occurrence in the K training tuples as the class of the test tuple, and dividing the training tuples into at least one region according to the class of the test tuple, wherein the at least one region includes a region containing the plurality of wind turbine generators.

2. The wind direction prediction method of claim 1, wherein, The step of performing wind direction correlation analysis on the acquired historical wind direction data further comprises: calculating the error rate and selecting the K value corresponding to the minimum error rate, wherein the K value corresponding to the minimum error rate indicates that the wind directions of K wind turbine generators in the wind farm have nonlinear correlation.

3. The wind direction prediction method of claim 1, wherein, Further comprising: performing reinforcement learning on the determined nonlinear regression model.

4. The wind direction prediction method of claim 1, wherein, Further comprising: when the wind direction instrument of the specific wind turbine generator fails, predicting the wind direction of the specific wind turbine generator based on the determined nonlinear regression model, and using the predicted wind direction to compensate the wind direction of the specific wind turbine generator; and monitoring the wind power curve of the specific wind turbine generator to determine whether the specific wind turbine generator continues to remain in operation.

5. The wind direction prediction method of claim 4, wherein, The step of monitoring the wind power curve of the specific wind turbine generator to determine whether the specific wind turbine generator continues to remain in operation comprises: if it is monitored that the wind power curve of the specific wind turbine generator is abnormal, stopping the specific wind turbine generator; if it is monitored that the wind power curve of the specific wind turbine generator is normal, using the predicted wind direction to keep the specific wind turbine generator in operation.

6. A wind direction prediction device for a wind farm, characterized in that The wind direction prediction device comprises: a data acquisition module configured to acquire historical wind direction data of a wind farm; a data analysis module configured to perform wind direction correlation analysis on the acquired historical wind direction data, and determine a plurality of wind turbine generators in the wind farm having nonlinear correlation with the wind direction of a specific wind turbine generator according to the wind direction correlation analysis result; a wind direction prediction module configured to determine a nonlinear regression model for the specific wind turbine generator according to the historical wind direction data of the plurality of wind turbine generators, and predict the wind direction of the specific wind turbine generator based on the determined nonlinear regression model, wherein the nonlinear regression model is determined by performing nonlinear regression analysis on the historical wind direction data of the plurality of wind turbine generators, and establishing the nonlinear regression model according to the nonlinear regression analysis result; constructing a cost function according to the nonlinear regression model and the historical wind direction data of the plurality of wind turbine generators; establishing an optimization function by introducing a slack variable to the constructed cost function; constructing a dual problem function by introducing a Lagrange operator to the established optimization function; obtaining equations of each variable in the dual problem function by calculating the dual problem function; obtaining a new optimization function by substituting the obtained equations of each variable into the established optimization function; and determining the nonlinear regression model by calculating the minimum value of the new optimization function, wherein the data analysis module is configured to: preprocess the historical wind direction data acquired by the data acquisition module; divide the preprocessed data into training tuples and a test tuple, and calculate the distance between the test tuple and each training tuple respectively; select K training tuples with the smallest distance, set the label with the highest occurrence frequency in the K training tuples as the class of the test tuple, and divide the training tuples into at least one region according to the class of the test tuple, wherein the at least one region includes a region containing the plurality of wind turbine generators.

7. The wind direction prediction apparatus of claim 6, wherein The data analysis module is further configured to: calculate the error rate, and select the K value corresponding to the minimum error rate, wherein the K value corresponding to the minimum error rate indicates that the wind directions of K wind turbine generators in the wind farm have nonlinear correlation.

8. The wind direction prediction apparatus of claim 6, wherein further comprising a reinforcement learning module configured to perform reinforcement learning on the determined nonlinear regression model.

9. The wind direction prediction apparatus of claim 6, wherein further comprising a wind direction compensation module configured to: when the wind direction instrument of the specific wind turbine generator fails, perform wind direction compensation on the specific wind turbine generator using the predicted wind direction of the specific wind turbine generator by the determined nonlinear regression model; monitor the wind power curve of the specific wind turbine generator to determine whether the specific wind turbine generator continues to operate.

10. The wind direction prediction apparatus of claim 9, wherein The wind direction compensation module is further configured to: if the wind power curve of the specific wind turbine generator is monitored to be abnormal, perform shutdown processing on the specific wind turbine generator; if the wind power curve of the specific wind turbine generator is monitored to be normal, keep the specific wind turbine generator operating using the predicted wind direction.

11. A computer readable storage medium storing a computer program, wherein the computer program comprises program instructions configured to cause a processor to perform the method according to any one of claims 1 to 10. The computer program is run by the processor, and the processor executes the wind direction prediction method of any one of claims 1-5.

12. A computer device comprising a processor and a memory storing a computer program, characterized in that, The computer program is run by the processor, and the processor executes the wind direction prediction method of any one of claims 1-5.

Images