A yaw control method, system and electronic equipment for a wind turbine generator set

Abstract

The present invention relates to a yaw control method, system and electronic equipment for a wind turbine. The method first installs a laser radar on the wind turbine for a period of time to obtain sufficient historical measurement data, and then develops a data-driven machine learning method to obtain a laser radar wind direction estimation model. The laser radar can then be removed or migrated to other wind turbines, and the laser radar wind direction estimation model is used to replace the role of the laser radar in calibrating yaw deviation to improve the accuracy of inflow wind measurement in the wind turbine control process and significantly reduce costs.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine control, and in particular to a yaw control method, system and electronic equipment for a wind turbine. Background Art

[0002] As a typical controller in wind turbines, the yaw control system plays a vital role in the process of wind energy conversion. One of the most critical functions is to drive the yaw motor to ensure that the wind rotor faces the natural inflow wind direction. Ideally, the wind vane measures the inflow wind direction and transmits the direction information to the control module in a digital form. However, in actual situations, the mechanical structure will always have structural deviations due to external reasons. Relevant research results show that even a small deviation will cause a power drop in the unit output power (a drop of about 1.13% for every 4°).

[0003] The deviation of the wind vane from the wind is caused by a variety of reasons, which are mainly divided into static error and dynamic error. There are many reasons for the static yaw error, such as improper human operation, mechanical wear, and harsh environment during the installation and use of the wind vane. This error interacts with the yaw control parameter-yaw deviation threshold, which is the main reason for the long-term misalignment of the unit's impeller to the wind. The dynamic yaw error is mainly due to the fact that the wind vane is installed behind the impeller. The rotation of the impeller will drive the air upstream and downstream near the impeller to rotate, forming a wake on the wind wheel surface and turbulence around the root of the blade, or it may be caused by the yaw controller threshold. The dynamic error will change with the wind conditions.

[0004] With the continuous development of remote sensing wind measurement technology, lidar wind measurement has become a relatively mature wind measurement technology. It can accurately measure multiple incoming wind data in front of the wind rotor. Its measurement range is about 10 to 400 meters, with a measurement accuracy of up to 0.1m / s and a wind direction accuracy of 0.5°. It has become a general consensus that cabin lidar can be used to calibrate yaw deviation or drive the real-time yaw action of wind turbines, thereby improving power generation efficiency. The advantage of using lidar is that the accuracy of incoming wind measurement will be greatly improved, but the disadvantage is that the cost is still very high.

[0005] How to improve the accuracy of wind inflow measurement during wind turbine control and reduce the cost has become a technical problem that needs to be solved urgently. Summary of the invention

[0006] In view of this, the present invention provides a yaw control method, system and electronic equipment for a wind turbine generator set, so as to improve the accuracy of inflow wind measurement during the control process of the wind turbine generator set and reduce the cost.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A yaw control method for a wind turbine generator set, the method comprising the following steps:

[0009] The historical record data of each type of first feature of the wind turbine and the historical measurement data of each type of second feature are obtained; the types of first features include the first wind speed, the first wind direction, the second wind direction, the cabin position, the ambient temperature outside the cabin, the rotor speed, the blade pitch angle, the active power, the power setting value and the status flag of the wind turbine recorded in the SCADA system (Supervisory Control And Data Acquisition) of the wind turbine; the types of second features include the second wind speed, the third wind direction and the status flag of the laser radar measured by the laser radar installed on the wind turbine; the first wind direction and the second wind direction are data obtained by using two traditional main and standby wind vanes respectively.

[0010] The historical record data and the historical measurement data are matched one by one according to time to form a data collection.

[0011] Based on the data collection, a preset number of first features having a greater impact on each type of second features are determined, and an important feature set for each type of second features is constructed.

[0012] For each type of the second feature, the historical record data of each type of the first feature in the important feature set of the second feature is used as input, and the historical measurement data of the second feature is used as output. Based on the XGBoost algorithm, a lidar wind direction estimation model of the second feature is constructed.

[0013] The currently recorded data of each type of the first feature in the important feature set of each type of the second feature are respectively input into the lidar wind direction estimation model of each type of the second feature to obtain the predicted data of each type of the second feature, and the yaw control of the wind turbine is performed according to the predicted data of each type of the second feature.

[0014] Optionally, determining a preset number of first features having a greater impact on each type of second feature of the laser radar based on the data collection, and constructing an important feature set for each type of second feature, specifically includes:

[0015] Based on the data collection, the Pearson correlation coefficient between the second feature of the kth category and the first feature of each category is calculated using the following formula; k=1, 2, ..., K, K represents the number of second features;

[0016]

[0017] Among them, r jk represents the Pearson correlation coefficient between the second feature of the kth class and the first feature of the jth class, x ijrepresents the i-th historical record data of the first feature of the j-th category, x ik represents the i-th historical measurement data of the second feature of the k-th category, represents the average value of all historical records of the first feature of the jth category, represents the average value of all historical measurement data of the second feature of the kth category, and n represents the number of historical record data or historical measurement data in the data collection;

[0018] The first features are sorted in descending order according to the Pearson correlation coefficient between the second feature of the kth category and the first feature of each category, and a preset number of first features after sorting are selected to form an important feature set of the second feature of the kth category.

[0019] Optionally, the lidar wind direction estimation model is:

[0020]

[0021] Where T is the number of regression trees, is the predicted data of the second feature of the kth category, f t (x k ) is the regression tree obtained at the tth iteration, x k is the recorded data of each category of the first feature in the important feature set of the kth category second feature, and F is the set of regression trees.

[0022] Optionally, the objective function used when constructing the lidar wind direction estimation model of the second feature is:

[0023]

[0024] Among them, obj (t) is the objective function value of the tth iteration, represents the error function, y k represents the historical measurement data of the second feature of the kth category in the data collection, is the predicted data of the second feature of the kth category obtained based on the lidar wind direction estimation model of the tth iteration; ω(f t (x k )) represents the regularization function, f t (x k ) is the regression tree obtained at the tth iteration.

[0025] Optionally, the error function is:

[0026]

[0027] Optionally, the error function is:

[0028]

[0029] Optionally, the regularization function is:

[0030]

[0031] Among them, M represents the number of leaves in the regression tree obtained at the tth iteration, γ is the complexity of each leaf, λ is the parameter to measure the penalty, and w m is the score vector on the mth leaf.

[0032] Optionally, determining a preset number of first features having a greater impact on each type of second features based on the data collection, and constructing an important feature set for each type of second features, further includes:

[0033] The data collection is subjected to fault point elimination and data cleaning.

[0034] A yaw control system for a wind turbine generator set, the system being applied to the above method, the system comprising:

[0035] A data acquisition module is used to acquire historical record data of each type of first feature of the wind turbine generator set and historical measurement data of each type of second feature; the types of first features include the first wind speed, the first wind direction, the second wind direction, the nacelle position, the ambient temperature outside the nacelle, the rotor speed, the blade pitch angle, the active power, the power setting value and the status flag of the wind turbine generator set recorded in the SCADA system of the wind turbine generator set; the types of second features include the second wind speed, the third wind direction and the status flag of the laser radar measured by the laser radar installed on the wind turbine generator set;

[0036] A data correspondence module, used for making one-to-one correspondence between the historical record data and the historical measurement data according to time to form a data collection;

[0037] Based on the data collection, determining a preset number of first features that have a greater impact on each type of second features, and constructing an important feature set for each type of second features;

[0038] a laser radar wind direction estimation model building module, for each type of the second feature, taking the historical record data of each type of the first feature in the important feature set of the second feature as input, taking the historical measurement data of the second feature as output, and building a laser radar wind direction estimation model of the second feature based on the XGBoost algorithm;

[0039] A yaw control module is used to input the current recorded data of each type of the first feature in the important feature set of each type of the second feature into the lidar wind direction estimation model of each type of the second feature, obtain the predicted data of each type of the second feature, and perform yaw control on the wind turbine according to the predicted data of each type of the second feature.

[0040] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above method when executing the computer program.

[0041] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0042] The present invention discloses a yaw control method, system and electronic equipment for a wind turbine generator set, the method comprising the following steps: obtaining historical record data of each type of first feature of the wind turbine generator set and historical measurement data of each type of second feature; making one-to-one correspondence between the historical record data and the historical measurement data according to time to form a data collection; based on the data collection, determining a preset number of first features that have a greater impact on each type of second feature, and constructing an important feature set for each type of second feature; for each type of the second feature, taking the historical record data of each type of first feature in the important feature set of the second feature as input and the historical measurement data of the second feature as output, and constructing a laser radar wind direction estimation model for the second feature based on an XGBoost algorithm; respectively inputting the current record data of each type of the first feature in the important feature set of each type of the second feature into the laser radar wind direction estimation model for each type of the second feature in one-to-one correspondence, obtaining prediction data for each type of the second feature, and performing yaw control on the wind turbine generator set according to the prediction data for each type of the second feature. The present invention first installs a lidar on a wind turbine for a period of time to obtain sufficient historical measurement data, and then develops a data-driven machine learning method to obtain a lidar wind direction estimation model. The lidar can then be removed or migrated to other wind turbines, and the lidar wind direction estimation model is used to replace the lidar's role in calibrating yaw deviation to improve the accuracy of inflow wind measurement in the wind turbine control process and significantly reduce costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0044] Figure 1 A flow chart of a yaw control method for a wind turbine generator set provided by an embodiment of the present invention;

[0045] Figure 2 A schematic diagram of a data cleaning process provided by an embodiment of the present invention;

[0046] Figure 3 A schematic diagram of the calculation results of the Pearson correlation coefficient provided in an embodiment of the present invention;

[0047] Figure 4 This is a diagram showing the effect of using a laser radar wind direction estimation model to predict wind direction in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] The object of the present invention is to provide a yaw control method, system and electronic equipment for a wind turbine generator set, so as to improve the accuracy of inflow wind measurement during the control process of the wind turbine generator set and reduce the cost.

[0050] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0051] Example 1

[0052] Embodiment 1 of the present invention provides a yaw control method for a wind turbine. The method of the embodiment of the present invention is applicable to a wind turbine. It can eventually replace the role of laser radar in yaw calibration, reduce costs and improve wind energy capture efficiency.

[0053] like Figure 1 As shown, the method comprises the following steps:

[0054] Obtain historical record data of each type of first feature of the wind turbine and historical measurement data of each type of second feature; the types of first features include the first wind speed, first wind direction, second wind direction, cabin position, cabin external ambient temperature, rotor speed, blade pitch angle, active power, power setting value and wind turbine status flag recorded in the SCADA system of the wind turbine; the types of second features include the second wind speed, third wind direction and lidar status flag measured by the lidar installed on the wind turbine.

[0055] The historical record data and the historical measurement data are matched one by one according to time to form a data collection. The data in the data collection is troubleshooted and cleaned using the operating status and key controller parameter information as decision variables, such as Figure 2 As shown in Figure 2, the rated operating condition comprehensive data set is filtered out.

[0056] Based on the data collection, a preset number of first features having a greater impact on each type of second features are determined, and an important feature set for each type of second features is constructed.

[0057] Calculate the Pearson correlation coefficient between the second feature of each category and the first feature of each category. The calculation formula is as follows:

[0058]

[0059] Among them, r jk represents the Pearson correlation coefficient between the second feature of the kth class and the first feature of the jth class, x ij represents the i-th historical record data of the first feature of the j-th category, x ik represents the i-th historical measurement data of the second feature of the k-th category, represents the average value of all historical records of the first feature of the jth category, represents the average value of all historical measurement data of the second feature of the kth category, and n represents the number of historical record data or historical measurement data in the data collection.

[0060] Based on this, the influence of laser radar wind direction on each SCADA feature is sorted, and the important SCADA feature set is screened. That is, the first features are sorted in descending order according to the Pearson correlation coefficient between the second feature of the kth category and the first feature of each category, and the first features with the first preset number after sorting are selected to form the important feature set of the second feature of the kth category.

[0061] The Pearson correlation coefficient calculation between the first feature obtained in the SCADA system and the second feature measured by the LiDAR is: Figure 3 For its effect diagram, Figure 3 In the above method, the influence of lidar wind direction on each first feature of SCADA can be sorted to obtain the important feature set.

[0062] For each type of the second feature, taking the historical record data of each type of the first feature in the important feature set of the second feature as input and the historical measurement data of the second feature as output, a lidar wind direction estimation model of the second feature is constructed based on the XGBoost algorithm;

[0063] Assuming that the model has T decision trees, the lidar wind direction estimation model can be expressed as follows:

[0064]

[0065] Where: T is the number of regression trees, is the predicted data of the second feature of the kth category, f t (x k ) is the regression tree obtained at the tth iteration, x kis the recorded data of each category of the first feature in the important feature set of the kth category second feature, and F is the set of regression trees.

[0066] The main idea of ​​XGBoost is that each update is based on the previous model prediction, adding a new tree f to fit the residual between the previous tree's prediction result and the true value to form a new model, and the new model is used as the basis for the next model learning. The specific expansion is as follows:

[0067]

[0068]

[0069]

[0070] …

[0071]

[0072] Where: Represents the predicted data of the tth iteration; represents the prediction data of the previous round (t-1); f t (x k ) indicates that the newly added regression tree is fitted with residuals.

[0073] Obviously, the goal of prediction is to make the predicted data of the population As close as possible to the real data y i , and it is required to have the greatest possible generalization ability. Therefore, from a mathematical point of view, this is a problem of optimal general function, so the objective function is simplified as follows:

[0074]

[0075] In the formula: the first term Represents the error function, also called the loss function, which is usually a square loss or logistic loss. The error function fits the sample to the maximum extent through continuous learning of the model, and finally obtains the minimum difference between the prediction result and the prediction target. The error function calculation formula is as follows:

[0076]

[0077] or

[0078] The second term ω(f t (x k )) represents the regularization function. The regularization function defines the complexity of the tree and increases the stability of the model by continuously simplifying the model. The specific formula is as follows:

[0079]

[0080] Among them, M represents the number of leaves in the regression tree obtained at the tth iteration, γ is the complexity of each leaf, λ is the parameter to measure the penalty, and w m is the score vector on the mth leaf. L2 regularization of w is equivalent to adding L2 smoothing to the score of each leaf node to avoid overfitting.

[0081] To calculate the error between the lidar wind direction estimate and the actual lidar wind direction measurement, several error-related performance indicators were used, including mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE). The corresponding values ​​of MAE, MAPE, and RMSE are calculated as follows:

[0082]

[0083]

[0084]

[0085] The wind direction estimation accuracy is evaluated based on the results, and the machine learning model parameters are optimized to obtain the lidar wind direction estimation model.

[0086] The currently recorded data of each type of the first feature in the important feature set of each type of the second feature are input into the laser radar wind direction estimation model of each type of the second feature in a one-to-one correspondence, and the prediction data of each type of the second feature is obtained, and the wind turbine is yaw controlled according to the prediction data of each type of the second feature. Specifically, the laser radar wind direction estimation model obtained after optimization is stored locally, which can be used for real-time yaw control of the wind turbine.

[0087] After optimizing the parameters of the built lidar wind direction estimation model, Figure 4 This is a comparison diagram between the predicted data of the optimized lidar wind direction estimation model and the actual measurement value of the lidar. It can be observed that the error between the two is small. The lidar wind direction estimation model can play a substitute role for lidar in real-time yaw calibration, improve power generation efficiency and increase the reliability of unit operation.

[0088] Example 2

[0089] Embodiment 2 of the present invention provides a yaw control system for a wind turbine, characterized in that the system is applied to the method described in embodiment 1, and the system comprises:

[0090] The data acquisition module is used to obtain the historical record data of each type of first feature of the wind turbine and the historical measurement data of each type of second feature; the types of first features include the first wind speed, the first wind direction, the second wind direction, the cabin position, the ambient temperature outside the cabin, the rotor speed, the blade pitch angle, the active power, the power setting value and the status flag of the wind turbine recorded in the SCADA system of the wind turbine; the types of second features include the second wind speed, the third wind direction and the status flag of the laser radar measured by the laser radar installed on the wind turbine.

[0091] The data correspondence module is used to make one-to-one correspondence between the historical record data and the historical measurement data according to time to form a data collection.

[0092] Based on the data collection, a preset number of first features having a greater impact on each type of second features are determined, and an important feature set for each type of second features is constructed.

[0093] A laser radar wind direction estimation model building module is used to build a laser radar wind direction estimation model for each type of the second feature based on the XGBoost algorithm, using the historical record data of each type of the first feature in the important feature set of the second feature as input and the historical measurement data of the second feature as output.

[0094] A yaw control module is used to input the current recorded data of each type of the first feature in the important feature set of each type of the second feature into the lidar wind direction estimation model of each type of the second feature, obtain the predicted data of each type of the second feature, and perform yaw control on the wind turbine according to the predicted data of each type of the second feature.

[0095] The specific implementation method of the functions of each module in Example 2 of the present invention is the same as the specific steps of the method in Example 1, and will not be repeated here.

[0096] Example 3

[0097] Embodiment 3 of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in Embodiment 1 when executing the computer program.

[0098] The present invention first installs a lidar on a wind turbine for a period of time to obtain sufficient historical measurement data, and then develops a data-driven machine learning method to obtain a lidar wind direction estimation model. The lidar can then be removed or migrated to other wind turbines, and the lidar wind direction estimation model is used to replace the lidar's role in calibrating yaw deviation to improve the accuracy of inflow wind measurement in the wind turbine control process and significantly reduce costs.

[0099] In this specification, various embodiments are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0100] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A yaw control method for a wind turbine generator set, It is characterized in that The method comprises the following steps: Obtain historical record data of each type of first feature of the wind turbine and historical measurement data of each type of second feature; the types of first features include the first wind speed, first wind direction, second wind direction, cabin position, cabin external ambient temperature, rotor speed, blade pitch angle, active power, power setting value and wind turbine status flag recorded in the SCADA system of the wind turbine; the types of second features include the second wind speed, third wind direction and status flag of the laser radar measured by the laser radar installed on the wind turbine; Matching the historical record data and the historical measurement data one by one according to time to form a data collection; Based on the data collection, determining a preset number of first features that have a greater impact on each type of second features, and constructing an important feature set for each type of second features; For each type of the second feature, taking the historical record data of each type of the first feature in the important feature set of the second feature as input and the historical measurement data of the second feature as output, a lidar wind direction estimation model of the second feature is constructed based on the XGBoost algorithm; Inputting the current recorded data of each type of the first feature in the important feature set of each type of the second feature into the lidar wind direction estimation model of each type of the second feature in a one-to-one correspondence, obtaining the predicted data of each type of the second feature, and performing yaw control on the wind turbine according to the predicted data of each type of the second feature; The step of determining a preset number of first features having a greater impact on each type of second feature of the laser radar based on the data collection, and constructing an important feature set for each type of second feature specifically includes: Based on the data collection, the Pearson correlation coefficient between the second feature of the kth category and the first feature of each category is calculated using the following formula; k=1, 2, ..., K, K represents the number of second features; Among them, r jk represents the Pearson correlation coefficient between the second feature of the kth class and the first feature of the jth class, x ij represents the i-th historical record data of the first feature of the j-th category, x ik represents the i-th historical measurement data of the second feature of the k-th category, represents the average value of all historical records of the first feature of the jth category, represents the average value of all historical measurement data of the second feature of the kth category, and n represents the number of historical record data or historical measurement data in the data collection; Sort the first features according to the Pearson correlation coefficient between the second feature of the kth class and the first feature of each class from large to small; Select a preset number of first features after sorting to form an important feature set of the k-th category second features; The laser radar wind direction estimation model is: Where T is the number of regression trees, is the predicted data of the second feature of the kth category, f t (x k ) is the regression tree obtained at the tth iteration, x k is the recorded data of the first feature of each category in the important feature set of the second feature of the kth category, and F is the set of regression trees; The objective function used to construct the lidar wind direction estimation model of the second feature is: Among them, obj (t) is the objective function value of the t-th iteration, represents the error function, y k represents the historical measurement data of the second feature of the k-th class in the data set, is the predicted data of the second feature of the k-th class obtained based on the lidar wind direction estimation model of the t-th iteration; ω(f t (x k )) represents the regularization term function, f t (x k ) is the regression tree obtained from the t-th iteration.

2. The yaw control method of a wind turbine according to claim 1, It is characterized in that The error function is:

3. The yaw control method of a wind turbine according to claim 1, It is characterized in that The error function is:

4. The yaw control method of a wind turbine according to claim 1, It is characterized in that The regularization function is: where M represents the number of leaves in the regression tree obtained at the t-th iteration, γ is the complexity of each leaf, λ is the parameter measuring the penalty, and w m is the score vector on the m-th leaf.

5. The yaw control method of a wind turbine according to claim 1, It is characterized in that The step of determining a preset number of first features having a greater impact on each type of second features based on the data collection, and constructing an important feature set for each type of second features, further includes: The data collection is subjected to fault point elimination and data cleaning.

6. A yaw control system for a wind turbine. It is characterized in that The yaw control system of the wind turbine generator set is applied to the yaw control method of the wind turbine generator set according to any one of claims 1 to 5, and the yaw control system of the wind turbine generator set comprises: A data acquisition module is used to acquire historical record data of each type of first feature of the wind turbine generator set and historical measurement data of each type of second feature; the types of first features include the first wind speed, the first wind direction, the second wind direction, the nacelle position, the ambient temperature outside the nacelle, the rotor speed, the blade pitch angle, the active power, the power setting value and the status flag of the wind turbine generator set recorded in the SCADA system of the wind turbine generator set; the types of second features include the second wind speed, the third wind direction and the status flag of the laser radar measured by the laser radar installed on the wind turbine generator set; A data correspondence module, used for making one-to-one correspondence between the historical record data and the historical measurement data according to time to form a data collection; Based on the data collection, determining a preset number of first features that have a greater impact on each type of second features, and constructing an important feature set for each type of second features; a laser radar wind direction estimation model building module, for each type of the second feature, taking the historical record data of each type of the first feature in the important feature set of the second feature as input, taking the historical measurement data of the second feature as output, and building a laser radar wind direction estimation model of the second feature based on the XGBoost algorithm; A yaw control module is used to input the current recorded data of each type of the first feature in the important feature set of each type of the second feature into the lidar wind direction estimation model of each type of the second feature, obtain the predicted data of each type of the second feature, and perform yaw control on the wind turbine according to the predicted data of each type of the second feature.

7. An electronic device, It is characterized in that The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the yaw control method of a wind turbine set according to any one of claims 1 to 5 when executing the computer program.

Images