For offshore wind farms, wake effects are among the largest sources of losses in energy production. At the same time, wake modelling is still associated with very high uncertainties. Therefore current research focusses on improving wake model predictions. It is known that atmospheric conditions, especially atmospheric stability, crucially influence the magnitude of those wake effects. The classification of atmospheric stability is usually based on measurements from met masts, buoys or lidar (light detection and ranging). In offshore conditions these measurements are expensive and scarce. However, every wind farm permanently produces SCADA (supervisory control and data acquisition) measurements. The objective of this study is to establish a classification for the magnitude of wake effects based on SCADA data. This delivers a basis to fit engineering wake models better to the ambient conditions in an offshore wind farm. The method is established with data from two offshore wind farms which each have a met mast nearby. A correlation is established between the stability classification from the met mast and signals within the SCADA data from the wind farm. The significance of these new signals on power production is demonstrated with data from two wind farms with met mast and long-range lidar measurements. Additionally, the method is validated with data from another wind farm without a met mast. The proposed signal consists of a good correlation between the standard deviation of active power divided by the average power of wind turbines in free flow with the ambient turbulence intensity (TI) when the wind turbines were operating in partial load. It allows us to distinguish between conditions with different magnitudes of wake effects. The proposed signal is very sensitive to increased turbulence induced by neighbouring turbines and wind farms, even at a distance of more than 38 rotor diameters.
Wake effects are one of the largest sources of losses in offshore energy yield assessment. This makes wake modelling very important, and much research is ongoing to improve wake model predictions. In the latest offshore CREYAP benchmark exercise (Comparative Resource and Energy Yield Assessment Procedure), wake modelling was found to be the prediction with the highest variation among the participants (Mortensen et al., 2015).
In order to use a wake model for validating the performance of an
operating offshore wind farm (Mittelmeier et al., 2017),
prediction uncertainties need to be reduced. Hansen et al. (2012) studied wake effects at the offshore wind farm Horns Rev in
different atmospheric conditions and revealed an influence on the wake
magnitude. They also compared turbulence intensities for different stability
classes as a function of the wind speed. Below 7 m s
Stability classification is based on measurements from met masts, buoys or whether it is assisted by remote sensing devices, such as light detection and ranging (lidar) or sound detection and ranging (sodar). For offshore use, these devices are very expensive and therefore often not permanently available. In several studies, lidars have been used to assess the wind speed recovery behind the turbine, and wake models have been tuned to match the measured wind speed (Beck et al., 2014; More and Gallacher, 2014).
The purpose of this paper is to investigate wind farm operational data and establish methods of identifying correlations between SCADA statistics and wind turbine wake behaviour caused by different atmospheric conditions.
For this investigation, we select three offshore wind farms, alpha ventus, Nordsee Ost and Ormonde. The first two wind farms have a well-equipped met mast and provide high-quality measurements of hub-height wind speed, wind direction, shear intensity, turbulence intensity and water temperature.
The wind farm alpha ventus (AV) is located about 45 km north of the island of Borkum in the North Sea. It consists of 12 turbines of the 5 MW class with a rotor diameter of 126 m and was commissioned in April 2010. The six northern turbines (AV1–AV6) were manufactured by Senvion. The six turbines in the southern part of the wind farm were manufactured by Adwen and are not considered in this analysis. The FINO1 research met mast is located only 3.2 D (rotor diameters) west of turbine AV4.
The layout of the northern part of alpha ventus (See Fig. 1) allows for an investigation of the wake behaviour in single- and double-wake conditions for westerly wind directions. Data were used from March 2011 to January 2015. After January 2015 no data were used because the installation of the Trianel wind farm in the west is suspected to have changed the environmental conditions of alpha ventus by adding extra turbulence to the inflow.
Schematic layout of the northern part of alpha ventus wind farm (circles) and FINO1 met mast (red square) with free-flow sector and distance in rotor diameters.
The wind farm Nordsee Ost (NO) is located about 35 km north-west of the island of Heligoland in the North Sea. The 48 Senvion turbines have a rated power of 6 MW each and a rotor diameter of 126 m. The met mast is located in the south-western corner of the wind farm (See Fig. 2). In the south, the neighbouring wind farm Meerwind Ost/Süd reduces the sector of free flow for the met mast and the possibilities to study multiple wakes higher than triple-wake conditions without disturbing effects from Meerwind.
The wind farm Nordsee Ost was fully commissioned in 2015. Data for this analysis are selected from November 2015 to November 2016. A correlation analysis (described in Sect. 3.2) is performed and the data from a long-range lidar are analysed. This lidar measurement campaign took place within the European research project “ClusterDesign”.
Nordsee Ost wind farm (blue circles) with neighbouring wind farm Meerwind Süd (green triangles), met mast (red square) and distance in rotor diameters (D). The orange area indicates the plan position indicator (PPI) scan from the Windcube 200S mounted on the helicopter platform of NO48 (described in Sect. 2.5).
The Ormonde wind farm consists of 30 Senvion turbines with a rated power of
5 MW each and a rotor diameter of 126 m. The wind farm is located in the
Irish Sea 10 km west of the Isle of Walney. The selected data are from January 2012 to January 2014. During this period, neighbouring wind farms were operational.
Located in the south-west are Walney 1 (51
The farm layout displayed in Fig. 3 is structured in a regular array which allows for a comparison of several multiple-wake situations. The inner farm turbine distance for the investigated wake situation from the south-west is 6.3 D and from the north-west it is 4.3 D.
The SCADA data from all wind farms and the meteorological data consist of
10 min statistics. Each turbine provides wind speed, wind direction, active
power, yaw position and pitch angle. The operational condition of the wind
turbine, which is used for the correlation with the met mast turbulence
intensity, is categorised by the minimum active power > 10 kW, the
maximum pitch angle < 3
Within the ClusterDesign research project funded by the European Union,
a long-range lidar measurement campaign was realised. A Windcube 200S
(WLS200S) lidar with scan head was placed on the helicopter platform of NO48
(See Fig. 2) from November 2015 to May 2016. A differential GPS system composed
of
three antenna GNSS systems of the type Trimble SPS855/SPS555H allows for
additional measurements of turbine yaw, nacelle pitch and roll angle.
One lidar measurement cycle takes about 200 s. It includes five plan position
indicator (PPI) scans followed by one range height indicator (RHI) scan.
Both scans cover a sector of 30
Upper plot: visualisation of the horizontal plan position indicator (PPI) scans downstream of NO48. Wind speed is normalised with the inflow wind speed measured at the met mast. The black crosses are the locations of the wind speed minima derived from a Gaussian fitting for each measurement distance. Bottom plot: normalised wind speed as a function of the distance in rotor diameters extracted from the top plot for the Gaussian-fitted minima (black crosses).
The horizontal 10 min average wind speed is calculated on a well-defined grid under the assumption of a negligible vertical component of the wind. The average of the wind component measured by the lidar during the considered time interval is included in the region of interest of the addressed grid point. The average 10 min wind direction is provided by the met mast. When the latter measurement is not available, the turbine yaw provided by the differential GPS system is used to estimate the wind direction. A detailed description of the lidar data preprocessing can be found in Schneemann et al. (2016).
For the assessment in this paper we are using averages of 10 min periods of
horizontal wind speed data evaluated from PPI scans of the wake behind NO48
(See Fig. 2). A multiple-wake situation can be observed at a wind direction
of 237.5
The lower graph of Fig. 4 shows the resulting normalised wind speeds over the normalised distance from NO48. The black line with the corresponding black crosses refers to the fitted values of the Gaussian fits.
Following the objective of proposing a signal based on SCADA data that enables us to identify the magnitude of wake effects, we first establish stability classification based on Monin–Obukhov surface-layer theory (Monin and Obukhov, 1954) and met mast turbulence intensity. Both approaches can be found in many publications and their influence on the magnitude of wake effects is reported. Secondly, SCADA signals which are affected by turbulence intensity are proposed and the ability of these signals to classify low, medium and high wake effects is analysed.
For the determination of atmospheric stability we follow the approach
suggested by Ott and Nielsen (2014).
Their iterative method is implemented in the software AMOK and derives the
inverse Monin–Obukhov length
Both classifications and their impact on wake effects are compared with FINO1 data in Sect. 4.1 and alpha ventus data in Sect. 4.2.
At wind farms with no met mast we have to rely on other signals to describe
the differences in power production under different atmospheric conditions.
To find the best substitute for a met-mast-measured turbulence intensity,
several SCADA signals that are affected by turbulence are correlated to the
met mast turbulence intensity, which is defined as
Definition of stratification with turbulence intensity and the
dimensionless Monin–Obukhov length
Adjusting PO
In this section, we describe the methodology to establish thresholds for
subsetting the measurements into weak, medium and strong wake effects. The
thresholds are estimated with a three-step approach. First we select the
normalised power (waked turbine normalised by the power of a free-flow
turbine) for a small sector (10
The third step divides the data set into high wake effects (values < 0)
and low wake effects (values >
For the selected period at alpha ventus, Fig. 6 shows the stability
distribution with the proposed thresholds of
Distribution of stability based on
Figure 7 presents the data from FINO1 with bin-averaged met mast turbulence intensity measured at 100 m (TI_100), turbulence intensity from the turbine nacelle anemometer at hub
height (AV4_TI), the power law coefficient from 40 and 90 m
of height (alpha_40_90) and the standard
deviation of the power divided by the average power (AV4_POTI). Each signal is plotted as a function of the wind speed for the three
proposed stability classes based on
Wake effects in alpha ventus (AV) under different atmospheric conditions classified by met mast turbulence intensity. Power of downstream turbine normalised with free-flow turbine. Upper row: single wake, bottom row: double wake. Left column: normalised power as a function of wind direction, right column: normalised power as a function of wind speed.
For all signals in Fig. 7 the differences between the stability classes are visible. Whereas shear (See Fig. 7c) has problems in light winds, it gives the best results for higher wind speeds. Turbulence intensity (See Fig. 7a) is the most constant signal for the selected wind speed range. The turbulence intensity measured at the nacelle (See Fig. 7b) is much higher due to the location of the anemometer behind the rotor. This introduces variation and weakens the ability to distinguish between the stability classes. The relationship between the standard deviation of the power divided by the average power (See Fig. 7d) and the wind speed is very dominant and can be approximated with a third-order polynomial.
Alpha ventus data from almost 4 years of operation are used to evaluate
the influence of atmospheric stability and turbulence intensity on the wake
development. Figure 8 shows the different wake
behaviour under different turbulence conditions. The top row of plots shows
the
single-wake condition of turbine AV5 in the wake of AV4. The second row
displays the same evaluation but for the double-wake condition of AV6 in the
wake of AV4 and AV5. The left side is a normalised power deficit as a function
of the wind direction for a wind speed range from 7 to 9 m s
Wake effects in alpha ventus (AV) under different atmospheric
conditions classified by
Wind speed recovery at wake centre at hub height behind NO48 for different turbulence stability classes. The wind speed is normalised with the inflow wind speed and the distance from the lidar on NO48 downstream is displayed in multiples of rotor diameters.
For the single wake, a clearly distinguishable difference between the stable
and unstable power deficit is visible. The largest deviation is found in the
full wake. The second wake has a less pronounced difference in power, which
can be explained by the fact that the first turbine operating in the wake
supports the mixing with the ambient wind speed. Another interesting effect
is noticeable in the top left plot. The difference in power for the
different stabilities is higher at the right-hand side of the deficit in
the downstream direction. This right drift of the wake in stable conditions has
also been observed in LES simulations by Vollmer et
al. (2016). This effect is even more pronounced
when the data are distinguished with
The turbulence intensity for this classification has been measured at 100 m,
which is the largest height at the FINO1 met mast. The second height of the
FINO1 met mast (90 m) is closer to hub height (92 m), but the strong mast
structure and the boom orientation of 135
For the wind farm Nordsee Ost (NO), only 1 full year of SCADA data and
6 months of lidar data are available for this investigation. Eighteen PPI scans
(as described in Sect. 2.5) with full-wake conditions are available and
categorised according to the classification in Table 1. Figure 10 displays the wind speed measured with the lidar normalised with
the inflow wind speed measured at the met mast. The wind speed recovers
faster for the unstable turbulence class than for the neutral and stable
classes. The second and third turbine in the row for the investigated wind
direction are marked with blue vertical lines. The decreased wind speed in
the induction zone in front of each downstream turbine is clearly visible.
The error bars indicate the standard error of the mean. For the stable class
In the next step, we correlate the SCADA signals described in Sect. 3.2 with the turbulence intensity measured at the mast. In Fig. 11 a panel plot is displayed. The graphs on the diagonal present the histogram and density distribution for the respective variable. The panels above the diagonal provide the Pearson correlation coefficients. The lower panels are scatter plots for the two variables with a fitted linear regression line. The colours of the points indicate the three stability classifications (blue: stable, green: neutral, red: unstable) determined with the met mast turbulence intensity.
The correlation between met mast and turbine TI in the panel at row 1, column 2 equals 0.55. This poor result can be explained by the nacelle wind speed measurement position behind the rotor, which induces additional disturbance in the flow.
The highest correlation with the met mast TI is obtained with the standard
deviation of the turbine power divided by its average active power
(PO
Correlation matrix. Turbulence intensity from met mast
(
To check the validity of these results, we use data from Nordsee Ost (NO). Figure 12 provides the information corresponding to Fig. 11 but for a different turbine type, met mast and location in the North Sea.
The correlation reveals the best result for the PO
Both correlation analyses show that the new artificial SCADA signal, derived
from the standard deviation of the power divided by its average active power
PO
Correlation analysis for Nordsee Ost. Turbulence intensity
(
Normalised SCADA signal
(
The thresholds displayed as vertical lines representing the median of the density distribution for the signal of interest.
In Sect. 4.2 we demonstrated the correlation of the SCADA signal PO
For AV4 the constants of the third-order polynomial are
With the methodology from Sect. 3.3 we obtain different density
distributions for high and low wake effects (See Fig. 14). The thresholds
are derived from the median of the data distribution. For
Comparing the two turbulence intensities in Fig. 14a and c, one can
clearly see the increased turbulence behind the rotor, especially for low TI
values. The distribution for the nacelle measurement is compressed in a way
that the low measurements have up to 4 % difference but the high
turbulence intensities are more or less comparable. This effect reduces the
ability of the nacelle turbulence intensity to distinguish between the wake
magnitudes. A more clear separation between the wake classes can be achieved
with the newly proposed PO
The quality of the established relationship in terms of dependence on turbine type, layout and location of the wind farm is tested by applying the same classification on a different wind farm where no met mast is available.
In the next sections, the ability of this new signal to distinguish between different environmental stratifications is analysed. Table 2 shows the proposed thresholds for the different parameters under investigation.
Summary of thresholds for the different classes of interest at alpha ventus.
The classification of wake effects by the PO
Stability classification with the
A clear difference in power production between stable and unstable cases can be identified in the single wake. The differences in double wake are again less pronounced. Compared to the TI classification, the curves for the neutral case are not as clear as between the stable and unstable curves, and in the normalised power curve plots (right column) the stable conditions can only be highlighted up to the wind speed of rated power for the free-flow turbine. This can be explained by the fact that at rated power the pitch controller rather than the power variation governs the turbine reaction to turbulence intensity. This leads in Eq. (4) to a significant decrease in the numerator and keeps the denominator constant.
The classification of wake effects in Fig. 16 is based on
PO
Wind speed recovery behind NO48 for different
This result states the ability of PO
Finally, the transferability of classification boundaries to other wind farms where no met mast is available is of interest.
First we have a look at the sensitivity of the signal PO
The signal
The grey area represents the geometrical location of the neighbouring wind
farms Walney 1 and 2. The closest distance to OR24 is Walney 1 at
approximately 38.8 D (SWT-3.6-107 Siemens). The two peaks at 208
and 255
Secondly we have a look at the influence on the wake recovery. With the south-westerly wind direction, we focus on single-wake,
double-wake and triple-wake conditions behind turbine OR27 for a sector of 10
For the north-westerly wind direction, Fig. 18 provides a view on different wake
effects for the proposed classification. The normalised power for each
turbine in the row behind OR23 is displayed (wind from left to right). Wind
speed is filtered for 8
The south-westerly wind direction is analysed in Fig. 19, which is a similar illustration as in Figs. 8 and 9. It is still possible to identify different wake behaviour for the different classes, but the effect is less pronounced than in the previous examples. A higher level of inflow turbulence intensity contributes to the mixing of the wake with free wind. Hence at lower inflow turbulence levels the effect of the wake-added turbulence is larger.
Further investigations are necessary to account for controller properties and to fill the normalised wind speed range (0.75–1) beyond the rated wind speed of the turbine in free-flow conditions.
Normalised power for each turbine along the row behind OR23 for a
wind direction of 312
In performance monitoring of offshore wind farms, the newly aggregated SCADA signals can be used as an auxiliary quantity to classify different atmospheric conditions. Advanced engineering wake models, which are able to take turbulence intensity or stability parameters into account, may be parameterised by these artificial turbine signals in order to improve their prediction of wind turbine power production under waked conditions.
Wake effects in Ormonde (OR) under different conditions. Power of downstream turbine normalised with free-flow turbine. First row: single wake, second row: double wake and third row: triple wake. Left column: normalised power as a function of wind direction, right column: normalised power as a function of wind speed.
Measured data from three different offshore wind farms, two met masts, one
buoy and one long-range lidar have been analysed to identify different
influences on power production for turbines operating in the wake. We have
validated the method described in Dörenkämper (2015),
which proposes using the turbulence intensity to describe the power
production in the wake, and compared it with the atmospheric stability
evaluation proposed by Ott and Nielsen (2014). In this case, turbulence intensity could better distinguish between
the magnitudes of wake effects. A correlation analysis was performed, and for
wind speeds in partial load operation, the standard deviation of the power
divided by its average power (PO
Both signals can distinguish between stronger and weaker wake effects. The
magnitude of influence of the PO
Using PO
Data are not available to the public.
The authors declare that they have no conflict of interest.
This article is part of the special issue “The Science of Making Torque from Wind (TORQUE) 2016”. It is a result of the The Science of Making Torque from Wind (TORQUE 2016), Munich, Germany, 5–7 October 2016.
The presented work is partly funded by the Commission of the European Communities Research Directorate-General within the scope of the project “ClusterDesign” (project no. 283145; FP7 Energy). We would like to thank Deutsche Offshore-Testfeld und Infrastruktur GmbH & Co. KG (DOTI), Research at alpha ventus (RAVE), Forschungsplattformen in Nord und Ostsee (FINO1), Innogy SE, Vattenfall Wind Power and Senvion SE for making this investigation possible. Furthermore, special thanks go to the R Core Team for developing the open-source language R (R Core Team, 2015). Edited by: Christian Masson Reviewed by: two anonymous referees