Patent Grant
9664176
This application is the U.S. National Phase under 35 U.S.C. §371 of International Application No. PCT/KR2013/001874, filed on Mar. 8, 2013, which in turn claims the benefit of Korean Patent Application No. 10-2012-0103396, filed on Sep. 18, 2012, the disclosures of which are incorporated by reference herein.
The present invention relates to a method of automatically calculating power curve limits for power curve monitoring of a wind turbine, and more particularly, to a method of automatically calculating power curve limits for power curve monitoring of a wind turbine which can exclude much faulty data included in wind speed-power data measured in a wind turbine and automatically calculate optimal power curve limits for power curve monitoring.
Generally, wind farms have to be efficiently and stably operated. To this end, a supervisory control and data acquisition (SCADA) system and a condition monitoring system (CMS) are used as representative key operating technologies.
A SCADA system for operating a wind farm is a computer-based system remotely controlling wind turbines in conjunction with a controller thereof and acquiring data for analyzing and reporting operating performance of the wind turbines, and performs control, monitoring, analysis, and reporting functions.
The SCADA system which has been developed with emphasis on overall operation of a wind farm is focused on monitoring of current operating conditions of individual wind turbines. That is, in order to monitor the individual wind turbines, the SCADA system is focused on acquiring and analyzing representative characteristic values of components, particularly, temperature and pressure values in addition to information concerning turbine operation.
As described above, the SCADA system is focused on monitoring the current operating conditions of the wind turbines, whereas a condition monitoring system (CMS) is aimed at diagnosing malfunctions of wind turbines at an early stage and preventing breakdown thereof by more carefully monitoring, analyzing, and predicting conditions of wind turbine components, thereby enhancing reliability and economic feasibility of the wind turbines.
The condition monitoring system is mainly classified into a blade condition monitoring system, a component vibration condition monitoring system, and an oil condition monitoring system according to monitoring regions and diagnoses malfunctions in advance using a variety of advanced analysis techniques.
Like the SCADA system, the condition monitoring system performs monitoring, analysis, and reporting functions. However, the condition monitoring system is differentiated from the SCADA system in terms of monitoring target regions and analysis and prediction techniques.
As described above, the SCADA system and the condition monitoring system monitor malfunctions of individual wind turbine components, such as a blade, a gear box, and a generator and generate alarm messages by differentiating alarm levels (e.g., caution, warning, and alarm) according to a predicted degree of influence of the malfunctions.
However, a technique of determining malfunctions in terms of an entire wind turbine system instead of individual components is not yet applied to the SCADA system or the condition monitoring system.
A power curve showing wind turbine power versus wind speed in a graph form is a representative indicator representing turbine performance in terms of an entire wind turbine system, as an official performance assurance indicator of a wind turbine that has to be ensured by a turbine manufacturer.
Referring to
Due to the observation results, monitoring techniques based on a power curve attract the same attention as the condition monitoring system in the condition monitoring of a wind turbine. However, in spite of importance thereof, studies on the power curve monitoring techniques are not being actively conducted yet.
Referring to
However, in the method proposed by the ISET, a distance between upper and lower limits is increased with increasing wind speed higher than the rated wind speed to fit a stall-control turbine selected as an application target, thereby making it difficult to apply the method to mainly used pitch control in large-scale wind turbines.
Referring to
The Intelligent Systems Lab tested various data mining algorithms for power curve estimation. Thereamong, k-NN model exhibited the most excellent performance. In addition, the intelligent systems lab proposed the residual control chart technique for setting power curve limits.
However, accuracy of the k-NN model of the data mining technique proposed by the Intelligent Systems Lab can be significantly reduced by faulty data. In addition, each time new data is given, k pieces of adjacent data have to be selected by calculating distances between the new data and all learning data. Therefore, it takes more time to calculate the distances therebetween as the amount of learning data increases.
The related art is disclosed in Korean Patent Publication No. 10-2010-0031897 (entitled “Device for Monitoring Wind Power Generator”).
In the application of the power curve monitoring techniques, there is a common problem of using only normal data, from which faulty data is excluded, as an input for a power curve limit setting algorithm. In the case of an irregular energy source such as wind power, there is no guarantee that only normal data is obtained for all wind speeds during a test run or for a year after installation. Therefore, faulty data must be excluded from measured data in order to actually apply the power curve limit setting algorithm.
However, in the related art, the power curve monitoring techniques are not easy to apply to actual fields since a significant amount of effort and time is required to exclude faulty data for the application of the power curve limit setting algorithm.
The present invention has been conceived to solve such problems in the related art and it is an aspect of the present invention to provide a method of automatically calculating power curve limits for the power curve monitoring of a wind turbine which can set optimal power curve limits by automatically calculating power curve limits for power curve monitoring even if wind speed-power data measured in a wind turbine includes a large amount of faulty data in addition to normal data.
In accordance with one aspect of the present invention, a method of automatically calculating power curve limits for power curve monitoring of a wind turbine includes: classifying input data of wind power generation through variable speed bins by an input data classification unit; calculating a mean value and a standard deviation of power for each bin for the classified input data by a power calculation unit; estimating a power curve for the calculated mean value of power for each bin by a power curve estimation unit; searching for accurate power curve limits while shifting the estimated power curve by a limit search unit; and setting the input data between the accurate power curve limits as new input data by a data extraction unit.
The input data classification unit may determine wind speed-power data of the input data based on the variable speed bins according to a bin width given by equation below:
Bin Width=1/Number of repetition of entire algorithm loop(m/sec)
The power curve estimation unit may estimate the power curve using interpolation based on the calculated mean value of power for each bin as an input for the interpolation, and the interpolation may be cubic B-spline interpolation.
The limit search unit may search for the accurate power curve limits while shifting the power curve leftwards and rightwards or upwards and downwards.
The limit search unit may repeatedly shift the power curve leftwards and rightwards by a distance of ΔV until the following inequality is satisfied:
PDLi−PDLi−1<βshift(i>1)
The limit search unit may repeatedly shift the power curve upwards and downwards by a distance of ΔP until the following inequality is satisfied:
PDLi−PDLi−1<γoffset(i>1)
PDL may be defined by equation below:
The method may further include: after setting the input data, receiving the calculated standard deviation of power for each bin from the power calculation unit and calculating a mean value thereof by a data determination unit; comparing, by the data determination unit, the calculated mean value of the standard deviations with that of previous standard deviations calculated in a previous algorithm loop and determining whether a variation in mean value is less than a constant for determining whether to terminate a power curve calculation algorithm; and determining that the accurate power curve limits have been calculated and terminating execution of an entire algorithm loop when the variation in mean value is less than the constant, and repeating the entire algorithm loop using the new input data when the variation in mean value is greater than the constant.
As described above, according to the present invention, even if wind speed-power data measured in a wind turbine includes a large amount of faulty data in addition to normal data, power curve limits can be automatically calculated by calculating a mean value and a standard deviation of power data input for each bin and estimating a power curve using interpolation, followed by searching for optimal limits while shifting the power curve. Therefore, power curve monitoring techniques can be easily applied to actual fields to rapidly recognize and cope with malfunctions in terms of an entire wind turbine system, thereby maximizing efficiency, reliability, and economic feasibility in the operation of a wind farm. In addition, the algorithm for automatically calculating power curve limits according to the embodiments of the invention can be used as a principal monitoring algorithm of a SCADA system or a condition monitoring system (CMS) for operation and maintenance of a wind farm.
Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
It should be noted that the drawings are not to precise scale and may be exaggerated in thickness of lines or size of components for descriptive convenience and clarity. In addition, terms used herein are defined by taking functions of the present invention into account and can be changed according to user or operator custom or intention. Therefore, definition of the terms should be made according to the overall disclosure set forth herein.
Referring to
The input data classification unit 10 classifies input data entered during wind power generation, using variable speed bins. This is aimed at classifying wind speed-power data measured through specific measurement units to calculate a mean value and a standard deviation of power for each bin.
The power calculation unit 20 calculates a mean value and a standard deviation of the power input data for each bin classified through the input data classification unit 10. The mean value is used as an input value for estimating a power curve through interpolation, and the standard deviation is used to detei mine whether an entire algorithm loop is terminated.
The power curve estimation unit 30 receives the mean value of the input data for each bin calculated through the power calculation unit 20 and estimates a power curve using interpolation.
The limit search unit 40 excludes faulty data included in the input data while shifting the power curve, estimated through the power curve estimation unit 30, leftwards and rightwards or upwards and downwards.
The data extraction unit 50 extracts data, as new input data, between power curve limits obtained by excluding the faulty data while shifting the power curve leftwards and rightwards or upwards and downwards, by the limit search unit 40.
The data determination unit 60 calculates a mean value of the standard deviations of the power for the respective bins, calculated in the power calculation unit 20, by targeting the new input data extracted through the data extraction unit 50 and determines whether to terminate or repeat the entire algorithm loop by comparing the mean value with that of standard deviations calculated in the previous algorithm loop.
Next, a method of automatically calculating power curve limits for power curve monitoring of a wind turbine according to one embodiment of the invention will be described in detail with reference to a flowchart shown in
In operation S10, an input data classification unit 10 classifies input data measured through specific measurement units during wind power generation, using variable speed bins.
Referring to
Bin Width=1/Number of repetition of entire algorithm loop (m/sec) [Equation 1]
In determination of the bin width by Equation 1, the bin width has a value of 1 msec at first and a gradually decreasing value, such as ½ msec, ⅓ msec, and so on, as an entire algorithm loop is repeated.
The reasons for using the variable speed bins are as follows. First, much faulty data may be included in the initial input data and therefore, a wide bin is used to minimize an influence of the faulty data. Second, as shown in
In operation S20, a power calculation unit 20 calculates a mean value and a standard deviation of the classified power input data for each bin. This is for the purpose of using the mean value of power for each bin as an input value for estimating a power curve through interpolation by a power curve estimation unit 30 and to use the standard deviation of power for each bin in determining whether to terminate the entire algorithm loop by a data determination unit 60.
Referring to
In operation S30, the power curve estimation unit 30 inputs the mean value of power for each bin which has been calculated in the power calculation unit 20 and estimates a power curve using interpolation.
Referring to
After estimation of the power curve, in operation S40, a limit search unit 40 excludes as much faulty data as possible while shifting the estimated power curve leftwards and rightwards or upwards and downwards, thereby searching for optimal power curve limits including only normal data if possible. That is, the limit search unit 40 excludes as much faulty data as possible from the existing input data to select only normal data as new input data for a next algorithm loop.
As shown in
The limit search unit 40 determines whether optical upper and lower power curve limits have been obtained, based on an inequality such as Equation 2 below. In this case, the limit search unit 40 repeatedly shifts the power curve leftwards and rightwards by a distance of ΔV until the inequality of Equation 2 is satisfied. In
PDLi−PDLi−1≦βshift(i>1) [Equation 2]
where i denotes the number of repetitions of a limit search algorithm, and βshift is a constant for determining whether optimal upper and lower power curve limits have been determined at the present stage. In
As shown in
In the same way, the limit search unit 40 determines whether the optical upper and lower power curve limits have been obtained at the present stage, based on an inequality such as Equation 4 below. In this case, the limit search unit 40 repeatedly shifts the power curve upwards and downwards by a distance of ΔP until the inequality of Equation 4 is satisfied. In
PDLi−PDLi−1<γoffset(i>1) [Equation 4]
where i denotes the number of repetitions of a limit search algorithm, PDL is defined by Equation 3, and γoffset is a constant for determining whether optimal upper and lower power curve limits have been determined at the present stage. In
In the limit search function, the optimal upper and lower power curve limits have been obtained through the leftward and rightward movement of the power curve limits and then through the upward and downward movement thereof. However, the order is not limited thereto, and the optimal upper and lower power curve limits may also be obtained in reverse order.
In operation S50, a data extraction unit 50 extracts only data present between the upper and lower power curve limits as new input data, by targeting the upper and lower power curve limits set to exclude faulty data from the input data.
As shown in
In operation S60, the data determination unit 60 receives the standard deviations of power for the bins which have been calculated in the power calculation unit 20 and calculates a mean value of the standard deviations in order to determine whether to terminate the entire algorithm loop, by targeting the input data newly extracted by the data extraction unit 50.
After calculating the mean value of the standard deviations, in operation S80, the data determination unit 60 compares the currently calculated mean value of the standard deviations with the previous mean value of standard deviations calculated in the previous algorithm loop and determines whether a variation in mean value is greater than a100 which is a constant for determining the termination of the power curve calculation algorithm.
The determination as to whether the variation in mean value is greater than aloop is based upon the inequality of Equation 5 below.
(Mean value of standard deviations of power for bins)k−(Mean value of standard deviations of power for bins)k−1<aloop [Equation 5]
where k (k>1) denotes the number of repetitions of the entire algorithm loop, and aloop is a constant for determining whether to terminate the entire algorithm loop for automatic calculation of power curve limits. In
When it is determined that the variation in mean value of standard deviations is less than the constant aloop as expressed by the inequality of Equation 5, in operation S80, the data determination unit 60 yields the upper and lower power curve limits, obtained by the limit search unit 40, as optimal (i.e., accurate) power curve limits and terminates the entire algorithm loop. In contrast, when it is determined that the inequality of Equation 5 has not been satisfied, the operation flow returns to operation S10 to perform the entire algorithm loop again. In this case, the process of classifying input data through variable speed bins is restarted using the new input data extracted by the data extraction unit 50, in operation S10.
Referring to
Although some embodiments have been described above with reference to the accompanying drawings, it should be understood that these embodiments are given by way of illustration only, and that various modifications, variations, and alterations can be made without departing from the spirit and scope of the present invention. Therefore, the scope of the present invention should be limited only by the accompanying claims and equivalents thereof.
| Number | Date | Country | Kind |
|---|---|---|---|
| 10-2012-0103396 | Sep 2012 | KR | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/KR2013/001874 | 3/8/2013 | WO | 00 |
| Publishing Document | Publishing Date | Country | Kind |
|---|---|---|---|
| WO2014/046355 | 3/27/2014 | WO | A |
| Number | Name | Date | Kind |
|---|---|---|---|
| 4525633 | Wertheim | Jun 1985 | A |
| 6856039 | Mikhail | Feb 2005 | B2 |
| 7228235 | Grzych | Jun 2007 | B2 |
| 7259471 | Basteck | Aug 2007 | B2 |
| 20090033313 | Jurkat et al. | Feb 2009 | A1 |
| 20090299780 | Sarkar et al. | Dec 2009 | A1 |
| 20100283245 | Gjerlov | Nov 2010 | A1 |
| 20120091713 | Egedal | Apr 2012 | A1 |
| 20120139244 | Bonnet | Jun 2012 | A1 |
| 20120261919 | Nakashima | Oct 2012 | A1 |
| 20120292903 | Merkel | Nov 2012 | A1 |
| Number | Date | Country |
|---|---|---|
| 10-2010-0031897 | Mar 2010 | KR |
| 10-2012-0095710 | Aug 2012 | KR |
| Entry |
|---|
| Korean Office Action dated Nov. 11, 2013 issued in corresponding Korean Patent Application No. 10-2012-0103396. |
| A. Kusiak, et al., “Models for monitoring wind farm power,” Renewable Energy 34 (2009), pp. 583-590. |
| A. Verma, “Performance monitoring of wind turbines: a data-mining approach,” University of Iowa, Iowa Rearch Online, dissertation, 2012. |
| E. Hau, “Wind Turbines: Fundamentals, Technologies, Application, Economics,” 2nd Edition, pp. 495-503. |
| M. Sheehy, “Wind Energy: Optimising Operations,” The Engineers Journal 237, pp. 237-239. |
| A. Verma, et al., “Predictive Analysis of Wind Turbine Faults: A Data Mining Approach,” Department of Industrial Engineering, The University of Iowa, Proceedings of the 2011 Industrial Engineering Research Conference, pp. 1-9. |
| Z. Zhang, “Power Curve: Concepts and Application” (Presentation Material),Intelligent Systems Laboratory, The University of Iowa. |
| P. Caselitz, et al., “Rotor Condition Monitoring for Improved Operational Safety of Offshore Wind Energy Converters,” Journal of Solar Energy Engineering, May 2005, vol. 127, pp. 253-261. |
| A. Kusiak, et al., “On-line Monitoring of Power Curves,” Renewable Energy, 34 (2009) pp. 1487-1493. |
| M. Heath, “Scientific Computing: An Introductory Survey,” Second Edition, pp. 330-331. |
| N. Forsman, “An analytical tool for the evaluation of wind power generation,” Dept. of Energy and Environment, Chalmers University of Technology, Goteborg, Sweden, 2011, pp. 1-50. |
| E. Sainz et al., “Robust filtering for the characterization of wind turbines: Improving its operation and maintenance,” Energy Conversion and Management, 2009, pp. 1-12. |
| A. Llombart-Estopinan, “Improving the operation and maintenance of wind farms: determination of wind turbine performance,” RE&PQJ, vol. 1, No. 6, Mar. 2008, pp. 8-14. |
| A. Marvuglia et al., “Learning a wind farm power curve with a data-driven approach,” World Renewable Energy congress 2011—Sweden, May 8-13, 2011, pp. 4217-4224. |
| German Office Action issued in Application No. 11 2013 004 526.5 dated Dec. 21, 2016. |
| Number | Date | Country | |
|---|---|---|---|
| 20150198144 A1 | Jul 2015 | US |
