KR101800217B1 - Correction method for yaw alignment error of wind turbine - Google Patents

Abstract

The present invention relates to a method of correcting a yaw alignment error in a wind turbine generator, capable of obtaining an alignment error close to that of a precision measurement device even if the precision measurement device is not installed in all wind turbines. According to an embodiment of the present invention, the method of correcting a yaw alignment error comprises: calculating a result value with a nacelle wind direction value and additional variable measurement values for each measurement time as input variables, and deriving correlations among the resulting values, the nacelle wind direction value, and the additional variable measurement values through a machine learning algorithm which repeats a process of comparing the result value with a Lidar wind direction value; inputting a current nacelle wind direction value and a current additional variable measurement value from a nacelle sensor and an additional variable sensor; calculating a corrected current wind direction value that reflects the correlations based on the current nacelle wind direction value and the current additional variable measurement value; and controlling, by a control unit, a driving unit in accordance with the corrected current wind direction value.

Description

Technical Field [0001] The present invention relates to a wind turbine generator,

The present invention relates to a method of correcting the yaw alignment error of a wind turbine generator, and more particularly, to a method of correcting yaw alignment error of a wind turbine generator using a learned algorithm, The present invention relates to a method for calibrating a yaw misalignment of a wind turbine generator that can maximize power generation efficiency at a low cost.

Referring to FIG. 1, the yaw misalignment θ in the wind turbine is defined as the difference in relative angle between the wind direction b and the axial direction a of the nacelle. .

Although the wind turbine nacelle is controlled to coincide with the wind direction, it is not always controlled to be zero due to power generation efficiency and mechanical problems. In addition, since the wind direction itself measured by the wind turbine may not be accurate, an alignment error occurs with the actual wind direction even if the direction of the nacelle always agrees with the wind direction.

Therefore, the measurement of the wind direction that approximates the actual wind direction in the wind power generator is most important.

Since the wind direction has a different value depending on the apparatus for measuring the wind direction, the yaw alignment error also has a different value depending on the measuring apparatus.

The wind turbine is usually controlled by the wind direction measured in a nacelle weather vane. Since the nacelle weather vane is located at the rear of the rotor, it is influenced by the rotor wake, and is distorted by the actual wind direction in front of the rotor due to crosswind or topographical influence .

To compensate for this, Lidar, which is a laser-type weather vane, measures wind direction more precisely than a nacelle weather vane because it measures wind direction in front of the rotor as shown in Fig.

In fact, although research has shown that the LIDAR weather vane is able to improve power generation if it is used for yaw alignment error control because it is close to the actual wind direction, permanent installation is difficult due to installation and maintenance costs of LIDAR, It is virtually impossible to install on all wind turbines.

Korean Patent Publication No. 10-2013-0074262 Korean Patent Registration No. 10-1634727

Accordingly, it is an object of the present invention to provide a method and apparatus for calibrating a yaw alignment error of a wind turbine according to a learned algorithm, It is possible to minimize the yaw error and maximize the power generation efficiency.

In addition, according to the method of correcting the yawing error of a wind turbine according to the present invention, the control unit mounted on the wind turbine is not manipulated or replaced, but the parameters inputted to the control unit are adjusted and inputted. There is no need to operate the power generation system and the power generation efficiency can be improved much more easily.

According to the method of correcting the yaw alignment error of the wind turbine according to the present invention, since the machine learning model derives the optimal correction model through repeated training using the information input during the set measurement time, As a result, it is possible to reduce the time and cost required for the analysis. Moreover, since the derived calibration model is improved every time when it is applied to other wind turbine generators, The correction reliability can be continuously improved.

According to an embodiment of the present invention, the above object can be accomplished by a lidar sensor having a sensing position in front of a rotor and a nacelle sensor having a sensing position on one side of the nacelle positioned behind the rotor, Measuring a ladder wind direction value and a nacelle wind direction value at the same time; Performing measurements simultaneously on different additional variables in each additional variable sensor according to the measurement schedule; Calculating a result value by inputting the nacelle wind direction value and the additional parameter measurement value for each measurement time, and comparing the result value with the Lather wind direction value, Deriving a correlation between the Lidar wind direction values; Inputting a current nasal wind direction value and a current additional parameter measurement value in the nacelle sensor and the additional variable sensor; Calculating a corrected current wind direction value reflecting the correlation based on the current nacelle wind direction value and the current additional parameter measurement; And controlling the driving unit according to the corrected current wind direction value in the control unit.

Here, the deriving the correlation may include: a function-type machine learning algorithm in which the nacelle wind direction value and the additional parameter measurement values are independent variables and have independent weights for the respective independent variables; And a step of determining each of the weights so that a difference between the ladder wind direction value and the ladder wind direction value is within a set error, using the nacelle wind direction value and the additional parameter measurement value as input, Calculating the current wind direction value includes replacing the current wind direction value with the corrected wind direction value by inputting the current nacelle wind direction value and the current additional parameter measurement value into a machine learning algorithm having the determined weight value can do.

Alternatively, the step of deriving the correlation may comprise: a function-type machine learning algorithm having the narkel wind direction value and the additional parameter measurement values as independent variables and having independent weights for each independent variable; Determining the respective weights so that an error correction value corresponding to the difference between the nacelle wind direction value and the Lada wind direction value is calculated as the result value by using the nacelle wind direction value and the additional variable measurement value as inputs, Calculating the corrected current wind direction value comprises inputting the current nacelle wind direction value and the current additional parameter measurement value into a machine learning algorithm having the determined weight value to calculate the error correction value, And calculating the corrected current wind direction value by reflecting the wind direction value.

In each of the embodiments, the additional parameter may include at least one of the wind speed at the nacelle sensor, the amount of power generation, the rotational speed of the rotor, or the absolute wind direction at the nacelle sensor measured at the measurement time.

For example, the machine learning algorithm may be provided with an algorithm that obtains each weight that minimizes the difference between the wind direction prediction value and the LADIR wind direction value.

Meanwhile, when the nacelle wind direction value is an absolute wind direction, the controller receives the yaw direction data and compares the yaw direction data with the corrected current wind direction value to calculate a yaw alignment error, thereby driving the drive unit.

Alternatively, the method may further include storing the Ladia wind direction value, the nacelle wind direction value, and the additional parameter measurement values corresponding to each measurement time.

Therefore, according to the present invention, it is possible to realize an alignment error equal to or close to that of the precision measuring device even if the precision measuring device, which is expensive and not easy to maintain, is not installed in all the wind turbines, , Yaw error can be minimized and the power generation efficiency can be maximized.

In addition, according to the method of correcting the yawing error of a wind turbine according to the present invention, the control unit mounted on the wind turbine is not manipulated or replaced, but the parameters inputted to the control unit are adjusted and inputted. There is no need to operate the power generation system and the power generation efficiency can be improved much more easily.

According to the method of correcting the yaw alignment error of the wind turbine according to the present invention, since the machine learning model derives the optimal correction model through repeated training using the information input during the set measurement time, As a result, it is possible to reduce the time and cost required for the analysis. Moreover, since the derived calibration model is improved every time when it is applied to other wind turbine generators, The correction reliability can be continuously improved.

FIG. 1 is a conceptual diagram for explaining a misalignment error,
2 is a perspective view of a wind turbine according to an embodiment of the present invention,
FIG. 3 is a configuration diagram of an apparatus for correcting a urine alignment error for a wind turbine according to an embodiment of the present invention,
4 is a flow chart of a method for correcting a yaw alignment error for a wind turbine according to an embodiment of the present invention;
5 to 8 are graphs showing the relationship diagrams of the LADIA wind direction values for the additional variables.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

Referring to FIG. 4, a method for correcting yaw misalignment of a wind turbine according to an embodiment of the present invention includes the steps of (S10) measuring a Lahira wind direction value, a nacelle wind direction value, and an additional parameter, (S13) of deriving a correlation between a result value, a nacelle wind direction value, and a ladder wind direction value through a machine learning algorithm (S13); and a step (S17) calculating a corrected current wind direction value reflecting a correlation with the current nacelle wind direction value, and calculating a yaw alignment error based on the corrected current wind direction value (S19) do.

Prior to explanation, the wind turbine generator 100 and the yaw drive control structure to which the present invention is applied will be described with reference to Figs.

Referring to FIG. 1, the yaw alignment error (θ) is an angle formed by the wind direction (b) and the axial direction (a) of the nacelle.

2, a wind turbine generator 100 according to the present invention includes a tower 2, a nacelle 2, a rotor 4, and a blade 5 do.

2, since the RDA sensor 20 has the sensing position P in front of the rotor 4 even if it is provided on the side of the nacelle 2, the flow sensor 20 can minimize flow distortion and noise, And the nacelle sensor 10 is located on the nacelle side and its sensing position is also located behind the rotor, so that it includes flow distortion and noise due to various causes.

Referring to FIG. 3, a configuration for performing a method of calibrating a yaw alignment error of a wind turbine according to an embodiment of the present invention includes a nacelle sensor 10 for measuring a nacelle wind direction value, An optimization module 30 for deriving a correlation between the wind direction predicted value and the Lidar wind direction value, an axial direction sensing section 60, and a wind direction determination section 40 for calculating a yaw alignment error according to the corrected current wind direction value A control unit 40, and a driving unit 50 for rotationally driving the nacelle according to the yaw alignment error.

In this case, when the wind direction value measured by the nacelle sensor 10 is a relative wind direction with respect to the position of the nacelle 2, the yaw axis direction data in the axial direction sensing unit 60 is unnecessary, The yaw direction data is received from the axial direction sensing unit 60. The control unit 40 calculates the alignment error by the driving unit 50, (50).

Hereinafter, each step will be described in detail.

First, in the nacelle sensor 10 having the sensing position at one side of the ladder sensor 20 having the sensing position in front of the rotor and the nacelle 2 located at the rear of the rotor 4, Is measured (S10).

Here, the measurement schedule may be a means necessary for deriving a statistically significant correlation between a wind direction prediction value and a Lahira wind direction value, which will be described later. Specifically, the measurement schedule may be a measurement number or a measurement time.

On the other hand, according to the measurement schedule, the additional variable sensor 70 simultaneously measures different additional variables (S10).

In the method of correcting the yaw misalignment error of the wind turbine according to the present invention, the use of a plurality of additional variables in addition to the wind direction value is performed because the difference between the Rather wind direction value and the nacelle wind direction value is affected by various other variables, In order to approximate the wind direction value as much as possible, it is desirable to introduce various additional parameters and to take the weight into consideration.

For example, even if the wind direction is the same, the wind direction will have different characteristics depending on the terrain and the orientation. That is, as shown in FIG. 6, if the azimuth direction is divided into 10-degree units and the average value of the difference between the wind direction value and the nacelle wind direction value is calculated, it can be seen that the value is different for each azimuth Green and orange indicate ± 1 SD.)

On the other hand, in the case of the wind speed, as shown in FIG. 7, since the influence of the rotor wake is different when the wind speed is changed, it can be seen that the Lidia wind direction value and the Nacelle wind direction value are distributed differently according to the wind speed.

In addition, even in the case of the rotational speed of the rotor, it affects the rotor wake, and the difference between the wind direction values of the ladder and the nacelle according to the rotor rotational speed is as shown in FIG.

For example, the additional variables may be various types, for example, the wind speed, the power generation amount, the rotation speed of the rotor, or the absolute wind direction of the nacelle sensor 10 measured at the measurement time Any one or two or more.

Meanwhile, the Lahore wind direction value, the nacelle wind direction value, and the additional parameter measurement values are stored correspondingly to each measurement time.

Next, the optimization module 30 calculates a wind direction prediction value by inputting the nacelle wind direction value and the additional parameter measurement values for each measurement time, and repeats the process of comparing the wind direction prediction value with the Lather wind direction value. A correlation between the resultant value and the nadir wind direction value and the Lidu wind direction value is derived (S13).

The optimization module 30 refers to a physical configuration in which a machine learning algorithm is built in. The optimization module 30 performs a learning process and a step in which the learned algorithm processes the nacelle wind direction value may be performed in one physical device However, the optimization module 30 is not limited to such a physical classification, but may have a function of deriving an optimal correlation and a function 2 for processing the current wind direction value in the learned algorithm. Should be understood as a concept that includes both.

The step of deriving the correlation may be provided in various ways. For example, the result output through the machine learning algorithm may be approximate to the Raid wind direction value, Value and the Lydia wind direction value.

In any case, the correlation, that is, the resultant value is not a fixed number for all input values (such as the nacelle wind direction value), but rather a measure of the moment measured at the nacelle sensor 10 and each additional variable sensor 70 It appears as a function that takes a value as input.

Here, the machine learning algorithm does not distinguish between linear and nonlinear, and various known algorithms can be used without limitation.

For example, the machine learning algorithm may be provided by an algorithm that obtains each weight that minimizes the difference between the wind direction predicted value based on the resultant value and the Lidar wind direction value. For example, an RSS (Residual Sum of Squares) And a weighting factor for the weighting factor.

More specifically, after a total of N samples are measured, it is possible to obtain the weight (W) of the following algorithm for minimizing RSSfmf using the sample data. These algorithms reflect how the difference between the measured values of the Lidar sensor and the nacelle sensor changes with various additional parameters.

(f : Machine Learning Algorithm

: i번째 시간에 측정된 독립변수에 상응하는 요 정렬 오차 보정값

: the correction value of the urine alignment error corresponding to the independent variable measured at the i-th time

W: weight of machine learning algorithm to learn

X: independent variable. This is the time series value measured at the same time. The following values are used for the independent variables.

i번째 시간에 나셀 풍속계에 의해 측정된 풍속값

the wind speed value measured by the nacelle anemometer at the i-th time

i번째 시간에 나셀 풍향계와 나셀 위치에 의해 측정된 절대 풍향

Absolute wind direction measured by nacelle weather vane and nacelle position at time i

i번째 시간에 측정된 발전량

The power generation amount measured at the i-th time

i번째 시간에 측정된 로터회전속도)

the rotor rotational speed measured at the ith time)

By using the learned machine learning model, it is possible to calculate the correction error of yaw alignment from each independent variable.

: j번째 시간에 측정된 독립변수에 상응하는 요 정렬 오차 보정값 (인덱스 j는 학습에 사용되지 않은 데이터를 말함)(

: a correction value of the urine alignment error corresponding to the independent variable measured at the j-th time (index j refers to data not used for learning)

: 학습된 가중치 벡터)

: Learned weight vector)

는 풍속, 나셀의 위치 등에 의해 매 시점마다 비선형적으로 변하는 값이다. The corrected value thus obtained is added to the nacelle wind direction value to obtain a corrected current wind direction value approximating to the Ladeau wind direction value. here,

Is a value that changes nonlinearly at every point of time due to the wind speed, the position of the nacelle, and the like.

보정된 풍향값)(

Corrected wind direction value)

As described above, deriving the correlation may be performed by using a function learning machine learning algorithm in which the nasal wind direction value and the additional variable measurement values are independent variables and each independent variable has independent weights , The nacelle wind direction value and the additional parameter measurement value stored for each measurement time are inputted and the error correction value corresponding to the difference between the nacelle wind direction value and the Lada wind direction value is calculated as the result value, As shown in FIG.

In this case, the step of calculating the corrected current wind direction value may include inputting the current nasal wind direction value and the current additional parameter measurement value into the machine learning algorithm having the determined weight value, And calculating the corrected current wind direction value.

Alternatively, the step of deriving the correlation may comprise: a function-type machine learning algorithm having the narkel wind direction value and the additional parameter measurement values as independent variables and having independent weights for each independent variable; Determining the respective weights so that the difference between the ladder wind direction value and the ladder wind direction value is within the set error, using the nacelle wind direction value and the additional parameter measurement value as input.

That is, in the present embodiment, unlike the above-described embodiment, the result value output from the machine learning algorithm is the corrected current wind direction value. Therefore, in this case, the step of calculating the corrected current wind direction value may include inputting the current nasal wind direction value and the current additional parameter measurement value into the machine learning algorithm having the determined weight value, With a wind direction value.

The above-described steps S10 to S13 assume that the Lahida wind direction value is an actual wind direction value, correspondingly input the measured nacelle wind direction value and additional variables, and obtain a result approximating the Lahida wind direction value as much as possible This is a process of deriving a correlation by learning through a machine learning algorithm to have

The steps S15 to S19 described below are steps of inputting the current nacelle wind direction value measured in real time and the measurement value of the additional variable into the learned algorithm in the normal wind turbine yaw control drive in which the Lidar sensor is removed And outputting the resultant value and controlling the yaw alignment error in the controller 40. [

The current nacelle wind direction value and the current additional parameter measurement value are input in the nacelle sensor 10 and the additional parameter sensor 70 (S15).

The corrected current wind direction value reflecting the correlation is calculated based on the current nacelle wind direction value and the current additional parameter measurement value (S17).

As described above, the corrected current wind direction value may be an error correction value for the difference of the nacelle wind direction value measured by the optimization module 30 from the RDA sensor value as the target value. The output value of the optimization module itself may be the corrected current wind direction value.

The control unit 40 controls the driving unit 50 according to the corrected current wind direction value.

Here, when the wind direction value measured by the nacelle sensor 10 and the Lidar sensor 20 is the absolute wind direction, the corrected current wind direction value received by the control unit 40 is also the absolute wind direction. In this case, Data is additionally received, the yaw alignment error is calculated therefrom, and the driving unit 50 can be driven and controlled accordingly.

On the other hand, when the wind direction measured by the nacelle sensor 10 and the lidar sensor 20 is a relative wind direction, that is, a relative angle with respect to the current nacelle direction, the controller 40 controls the drive unit according to the corrected current wind direction value. .

2: Nacelle 3: Tower
4: rotor 5: blade
10: first sensing unit 20: second sensing unit
30: Optimization module 40:
50: driving part 60: axial direction sensing part
100: Wind generator

Claims (7)

A lidar sensor having a sensing position in front of the rotor and a nacelle sensor having a sensing position at one side of the nacelle located behind the rotor measure the ladder direction value and the nacelle wind direction value simultaneously according to the set measurement schedule ;
Performing measurements simultaneously on different additional variables in each additional variable sensor according to the measurement schedule;
Calculating a result value by inputting the nacelle wind direction value and the additional parameter measurement value for each measurement time, and comparing the result value with the Lather wind direction value, Deriving a correlation between the Lidar wind direction values;
Inputting a current nasal wind direction value and a current additional parameter measurement value in the nacelle sensor and the additional variable sensor;
Calculating a corrected current wind direction value reflecting the correlation based on the current nacelle wind direction value and the current additional parameter measurement;
And controlling the driving unit according to the corrected current wind direction value in the control unit. The method according to claim 1,
Wherein deriving the correlation comprises:
Wherein the nell field wind direction value and the additional parameter measurement values are set as independent variables and each of the independent variables has an independent weight. And determining each of the weights so as to calculate a result value whose difference from the Lavada wind direction value is within a set error,
Calculating the corrected current wind direction value comprises: inputting the current nasal wind direction value and current additional parameter measurement value into a machine learning algorithm having the determined weight value, and replacing the calculated wind direction prediction value with the corrected current wind direction value Wherein the yaw alignment error correction method of the wind power generator includes: The method according to claim 1,
Wherein deriving the correlation comprises:
Wherein the nell field wind direction value and the additional parameter measurement values are set as independent variables and each of the independent variables has an independent weight. And determining each of the weights so that an error correction value corresponding to a difference between the nacelle wind direction value and the Lather wind direction value is calculated as the result value,
The calculating of the corrected current wind direction value may include inputting the current nacelle wind direction value and the current additional parameter measurement value into a machine learning algorithm having the determined weight and reflecting the calculated error correction value to the current nacelle wind direction value And calculating the corrected current wind direction value. The method according to claim 1,
Wherein the additional parameter includes at least one of a wind speed in the nacelle sensor, a power generation amount, a rotation speed of the rotor, or an absolute wind direction in the nacelle sensor measured at the measurement time. 3. The method of claim 2,
Wherein the machine learning algorithm is provided by an algorithm that obtains each weight that minimizes the difference between the wind direction prediction value and the LADIR wind direction value. The method according to claim 1,
The yaw direction data is received and compared with the corrected current wind direction value to calculate a yaw misalignment error and the yaw misalignment correction of the wind turbine driving the drive unit accordingly, Way. The method according to claim 1,
And storing the Ladia wind direction value, the nacelle wind direction value, and the additional parameter measurement values corresponding to each measurement time and storing the ladder wind direction value, the ladder wind direction value, the nacelle wind direction value, and the additional parameter measurement values.

Images