Using detailed upwind and nacelle-based measurements from a
General Electric (GE) 1.5sle model with a 77 m rotor diameter, we calculate
power curves and annual energy production (AEP) and explore their
sensitivity to different atmospheric parameters to provide guidelines for
the use of stability and turbulence filters in segregating power curves. The
wind measurements upwind of the turbine include anemometers mounted on a
135 m meteorological tower as well as profiles from a lidar. We calculate
power curves for different regimes based on turbulence parameters such as
turbulence intensity (TI) as well as atmospheric stability parameters such
as the bulk Richardson number (
Power performance testing and annual energy production (AEP) assessments rely on accurate calculations of wind turbine power curves. Previous work on power performance highlights the role of turbulence intensity (TI) and wind shear in influencing power production (Elliot and Cadogan, 1990; Hunter et al., 2001; Kaiser et al., 2003; Sumner and Masson, 2006; Gottschall and Peinke, 2008; Antoniou et al., 2009; Rareshide et al., 2009; Wharton and Lundquist, 2012a, b; Clifton et al., 2013a; Dörenkämper et al., 2014). Wharton and Lundquist (2012b) also found that vertical TI and turbulence kinetic energy (TKE) affect power performance and Rareshide et al. (2009) found that veer affects power performance. Atmospheric stability induces deviations of power from the manufacturer power curve (MPC) (Motta et al., 2005; van den Berg, 2008; Vanderwende and Lundquist, 2012; Wharton and Lundquist, 2012b), and atmospheric variations across the rotor disk can influence power performance results (Sumner and Masson, 2006; Wagner et al., 2009; Choukulkar et al., 2016).
Because the power curve so closely impacts AEP, factors that influence power performance typically influence AEP calculations as well. As suggested by the works mentioned above, the two most closely explored atmospheric factors with regard to AEP are TI and wind shear, but the existing studies do not agree on the influence of TI and wind shear on AEP. The simulation-based study of Antoniou et al. (2009) found that low wind shear supported high AEP. For low wind speeds, the highest AEP occurred during conditions of high TI, but at higher wind speeds, the highest AEP occurred when TI was low. In contrast, based on data from a number of wind farms in the continental US, Rareshide et al. (2009) also compared AEP calculated with different TI and shear combinations, and found that AEP typically decreased with increasing TI, but increased with increasing shear.
In this study, we also investigate the influence of different atmospheric stability and turbulence regimes on wind turbine power curves and AEP calculations, incorporating a broad set of atmospheric parameters as well as different approaches to measuring these parameters. In Sect. 2 we describe our data set, which includes an upwind meteorological (met) tower with measurements spanning the rotor disk as well as a wind-profiling lidar. In Sect. 3 we present our data analysis methods, which include filtering the data by atmospheric parameters like shear, TI, and atmospheric stability. The effects of atmospheric parameters on power curves and AEP are presented in Sect. 4, and in Sect. 5 we summarize conclusions about the effects of atmospheric stability and inflow turbulence on power curves and AEP calculations.
The measurements used in this analysis were collected at the US Department of Energy (DOE) National Wind Technology Center (NWTC, Fig. 1) at the National Renewable Energy Laboratory (NREL), located just south of Boulder, Colorado, and about 5 km east of the Colorado Front Range (Clifton et al., 2013b; Aitken et al., 2014). The prevailing wind direction at 80 m (hub height) at this site during this campaign (29 November 2012–14 February 2013) was west-northwesterly.
Left panel: local map of the NWTC with instrument locations and topographic contours in meters above sea level. Right panel: the regional setting of the NWTC between the greater Denver metropolitan area and Boulder, with the Front Range of the Rocky Mountains shown in the higher topography west of the site. (Courtesy of Joshua Bauer and Billy Roberts at NREL.)
135 m met tower instrument information.
This wind direction also dominated a 14-year period from a neighboring met tower at the NWTC (Clifton and Lundquist, 2012). During the winter, the downslope flow from the nearby Rocky Mountains is frequently channeled through Eldorado Canyon, located just west-northwest of the NWTC (Banta et al., 1996; Poulos et al., 2000, 2007; Clifton et al., 2013b; Aitken et al., 2014). The NWTC site slopes upward with about 20 m in elevation change toward the west for about 1.5 km before dropping off 20 m towards the highway on the western edge of the site. The surface is mostly short grass.
Upwind measurements were taken using a Renewable NRG Systems (NRG)/LEOSPHERE WINDCUBE v1 vertically profiling Doppler lidar (Courtney et al., 2008; Rhodes and Lundquist, 2013) and a 135 m met tower. The tower supports several levels of cup anemometers, vanes, sonic anemometers, and temperature sensors, along with precipitation and air-pressure sensors (Fig. 2, Table 1) all on booms pointing in the dominant wind direction (west-northwest). Data were collected during the winter season, typically the season of the strongest winds at the NWTC (from 29 November 2012 through 14 February 2013). The lidar is located about 216 m (2.7 D) west of the General Electric (GE) 1.5sle turbine on the NWTC site. The met tower is located approximately 160 m (2.0 D) west-northwest of the turbine (Fig. 1). Because different instruments employ different averaging methods, Fig. 3 demonstrates that all wind speed data sets were synchronized and illustrates how the power output responds to changes in wind speed.
Configuration of 135 m meteorological tower with some key heights labeled. This tower varies slightly from the M4 tower described in Clifton et al. (2013b), but data are available online (NWTC, 2016).
The NRG/LEOSPHERE WINDCUBE v1 lidar measures volumetric-averaged wind speeds
and directions every 20 m from 40 to 220 m, thereby spanning the entire
vertical extent of the turbine rotor disk. The wind speeds are measured with
an accuracy of 0.2 m s
The M5 met tower (NWTC, 2016, similar to the M4 tower at the site, which was
studied by Rinker et al., 2016) is instrumented with cup anemometers at 3,
10, 30, 38, 55, 80, 87, 105, 122, and 130 m, and vanes at 3, 10, 38, 87, and
122 m (Fig. 2 and Table 1). Barometric pressure and
precipitation sensors are located at 3 m and temperature sensors at 3, 38,
and 87 m (Table 1). Sonic anemometers are mounted at 15, 41, 61, 74, 100,
and 119 m (Fig. 2 and Table 1). The tower booms are
directed at 278
A GE 1.5 MW turbine (GE 1.5/77 sle) with an 80 m hub height was chosen for
this study. The GE 1.5 MW is the most widely deployed utility-scale turbine
in the world with more than 12 000 turbines deployed around the globe as of 2009
(GE Energy, 2009). The supervisory control and data acquisition (SCADA)
system of the turbine provides 10 min averages of nacelle wind speed,
nacelle orientation, turbine power, blade pitch angles, and generator speed
set point. These measurements can be compared with the upwind measurements
to quantify power curves and AEP. The cup anemometer mounted on the nacelle
of the turbine is a NRG IceFree Hybrid XT turbine control anemometer. The
GE 1.5sle reaches its nameplate capacity, 1.5 MW, at a wind speed of 14 m s
Time series from 11 January 2013 from 08:00 to 17:00 MST (Mountain
Standard Time):
Before calculating atmospheric parameters, all meteorological and turbine data are checked for data quality as described in Sect. 3.1.
All lidar-measured wind speed measurements are filtered by CNR: only
measurements with a CNR greater than
Quality control filtering methods performed on the met tower data discard data that are flagged for a number of reasons, including irregular timing (the time between measurements is inconsistent), insufficient percentage of data points within an averaging period (less than 95 %), low standard deviation (less than 0.01 % of the mean) or constant values during the measurement interval (which indicate icing events), empty data channels, bad values as defined by manufacturer limits, or when an instrument records a “NaN” in place of a real measurement. After filtering for quality-control purposes, the met tower provides horizontal wind speeds and directions and temperatures about 90 % of the time at all levels during this study.
Several spikes in wind speed occur in the raw sonic anemometer data. Therefore, a de-spiking filter is applied based on the change in wind speed from each data point to the next. Data points are removed if they are preceded and followed by changes exceeding the lowest 99 % of all changes. After filtering the spikes in the sonic anemometers as well as the previously discussed quality-control procedure, the sonic anemometers provide wind speed and temperature about 90 % of the time at 15 m and about 60 % at 74 m during this study.
Although the dominant wind direction at the site is west-northwesterly,
other wind directions do occur. To ensure the lidar and met tower
measurements are upwind of the turbine, we consider only data from time
periods of hub-height wind from the 235–315
Wind roses for
After filtering for quality control as well as wind speed and direction, a large number of times occur when the turbine is producing significantly less power than expected – underperforming – as seen in Fig. 5a. We test two methods to isolate and discard the cases where the turbine is producing significantly lower power, inconsistent with “normal operation”. The first approach relies on blade pitch angle to segregate data and flag most of these underperforming periods; this approach could be used by wind plant owner-operators with access to limited SCADA parameters. When more SCADA parameters are available, such as generator speed set point, these values may be used in a more rigorous way to filter on curtailment and to define normal turbine operation.
Without access to the turbine control system or data more refined than
10 min averages, typical blade pitch angles can be quantified as a function
of wind speed (Fig. 5b). The median value for blade
pitch angle for each wind speed bin as well as
After filtering for hub-height wind speed and direction, positive power production, and blade pitch angle, 1240 out of 7949 lidar 80 m wind speed data points remain (16 %), and 2235 out of 9918 met tower 80 m wind speed data points remain (23 %). Concurrent lidar, met tower, and turbine data that fulfill the various screening criteria occur during 1107 10 min periods.
Access to a number of turbine control parameters from the SCADA on the DOE
GE 1.5sle allows for a more accurate definition of normal turbine operation,
mostly based on generator speed set point filtered on curtailment. However,
from cut-in wind speed until around 5.5 m s
After filtering for hub-height wind speed and direction, positive power production, and normal turbine operation, 1227 out of 7949 lidar 80 m wind speed data points remain (15 %), and 2249 out of 9918 met tower 80 m wind speed data points remain (23 %). Concurrent lidar, met tower, and turbine data that fulfill the various screening criteria occur during 1127 10 min periods.
The turbine operation filters described in Sect. 3.3.2 not only filter out
all of the times when the turbine is producing significantly less power than
expected but also allow the use of about 2 % more data points deemed “bad”
by the blade pitch angle filtering method described in Sect. 3.3.1. Many of
the data points that would be discarded using the blade pitch angle
filtering method are between cut-in wind speed and 10 m s
Scatter power curve using the tower 80 m wind speed. Blue dots show points filtered out using turbine control parameters described in Sect. 3.3.2. Red dots show data points that passed this filtering process. The grey dashed line marks rated speed.
Power curves after filtering for wind speeds between 3.5 and 25 m s
Power curves based on wind speeds normalized by air density following the
International Electrotechnical Commission (IEC 61400-12-1, 2015) can be
used to evaluate turbine performance. The observed power curves, comparing
power production to 80 m tower anemometer wind speeds (Fig. 7a), 80 m lidar wind speeds
(Fig. 7b), and nacelle anemometer wind speeds (Fig. 7c), generally show good agreement with an
approximation of the MPC (GE Energy, 2009). This approximated MPC is
determined by placing the publicly available MPC for the GE 1.5sle on a grid
(with dimensions of 0.5 m s
The nacelle-mounted anemometer does not observe the ambient wind speed that the rotor disk experiences because the wind that flows through the rotor disk and along the nacelle during operation is modified by the blades and nacelle (Antoniou and Pedersen, 1997; Smith et al., 2002; Frandsen et al., 2009; Zahle and Sørensen, 2011). However, power curves calculated using nacelle wind speeds are shown here along with power curves calculated using upwind measurements in order to compare the different methods. In many cases, operators calculate these nacelle-based power curves due to lack of other data.
Lidar 80 m wind speeds compared to tower 80 m wind speeds filtered
for wind speeds between 3.5 and 25.0 m s
The power curves created from 10 min tower and nacelle-mounted anemometer measurements (Fig. 7a and c, respectively) show less variability than the lidar power curve (Fig. 7b). It is especially apparent from the power curve created from 10 min lidar measurements (Fig. 7b) that the lidar variability at this particular site is vulnerable to inhomogeneity in the flow. Although lidars are widely available and used in the field (Clifton, 2015), the variability between the lidar and tower measurements (Fig. 8) indicates sufficient inhomogeneity in the flow at this particular site (as observed by Aitken et al., 2014) to cause us to discuss and show only the upwind data from the tower from this point forward. Note, however, that not all sites are subject to the inhomogeneity seen at the NWTC, and all instruments available for wind measurement should be considered. Concurrent met tower and turbine data that fulfill the screening criteria occurred during 2240 10 min periods, equivalent to about 373 h of data, which is more than twice the 180 h of data that the IEC 61400-12-1 (2015) standard recommends for power performance testing.
Numerous approaches are available for classifying the atmospheric stability
of a given 10 or 30 min time period. Bulk Richardson number (
Defined stability regimes.
Obukhov length (
When the
TI
TI can also be used to describe atmospheric conditions, as demonstrated by
Rareshide et al. (2009), Wagenaar and Eecen (2011), and Wharton and
Lundquist (2012a). TI is typically defined as
When the atmospheric stability regimes are compared to the TI regimes
defined here (Fig. 14), the
TI distribution using thresholds in Table 3. Includes data filtered
for tower 80 m wind speeds between 3.5 and 25 m s
TI vs.
To further understand the turbulence characteristics demonstrated during
this campaign, we also calculate TKE using the 74 m 3-D sonic anemometer
mounted on the M5 met tower. Although TI is a parameter typically calculated
and analyzed in the wind industry, TKE has the advantage of including the
vertical component of the wind:
Defined turbulence regimes.
TKE distribution using thresholds in Table 3. Includes data filtered
for tower 80 m wind speeds between 3.5 and 25.0 m s
Many cases with relatively high TI or TKE are considered neutral and stable
according to our stability definitions in Table 3. Depending on whether TI,
TKE,
To estimate the effect of the wind speed vertical profile across the rotor
disk, the wind shear exponent or power law exponent parameter,
For this period of time at this site, however, it was rare for the rotor
equivalent wind speed (REWS) to deviate significantly from the hub-height
wind speed (Sect. S2). Therefore, shear exponents are separated into regimes
simply by splitting the shear exponent distribution into thirds (Table 2,
Fig. 16). Other approaches to classify stability
regimes using shear exponents such as combining with other stability
measures such as
To explore the variability in the power curves, we apply filters to the
power curves based on factors such as atmospheric stability and TI. We apply
a new method to calculate AEP using these classifications. We can consider
periods with low TI to be approximately “stable” by
The NWTC site generally exhibits high TI throughout this data collection
period. Even so, some differences in power produced emerge at wind speeds
between 5 and 7 m s
Shear exponent distribution using thresholds in Table 2. Includes
data filtered for tower 80 m wind speeds between 3.5 and 25.0 m s
Nacelle anemometer power curves with
Nacelle anemometer power curves shown as the anomaly from the
neutral or medium power curve of the
On the other hand, power curves separated by
Distinct differences between power curves calculated from nacelle winds and
power curves calculated from upwind tower winds occur in the power curves
of both of these atmospheric parameters. Statistically distinct wind speed
bins in power curves calculated from nacelle winds tend to be similar to
those in power curves calculated from tower winds near rated speed. At lower
wind speeds, however, between about 5 and 9 m s
Agreement between the TI and
The large variability reported in the literature (and herein) regarding power production can be understood by recognizing the interactions between a pitch-controlled turbine and the atmosphere: the operation of control algorithms changes with wind speed, with varying effects depending on the ambient turbulence.
Sensitivity to atmospheric turbulence occurs at low wind speeds, near cut-in wind speed. In these conditions, the turbine generator speed (revolutions per minute, RPM) increases, as does the generator torque. As a result, the blades will often pitch backward, changing the angle of attack to generate more lift, and the power production ramps up. At low wind speeds and higher turbulence, the turbine can react to the higher variation in wind speed and can capitalize on the variation seen in the wind flow because of the additional lift resulting from the blade pitch, and the turbine produces more power. Conversely, at low wind speeds with lower turbulence, the variation in wind speed is lower, and so the turbine experiences more consistent wind than in highly turbulent conditions and therefore produces less power.
At higher wind speeds, closer to or just below rated speed, control mechanisms seek to maintain rated generator speed, rather than continuing to increase generator speed. The blades will pitch forward (or “feather”), allowing the power production to maintain rated power. This process effectively decreases the amount of lift when compared to lift generated by a non-feathered blade. At these wind speeds during periods of high TI, a turbine reacts to the high variation in wind speed with subtle changes in blade pitch. For example, if the turbine detects a drop in wind speed, the blades may pitch back to generate more lift, but then if the wind speed increases quickly after, the blades will pitch forward again. If the blade pitch cannot immediately respond to increases in wind speed, then power losses occur. At these higher wind speeds, lower turbulence enables consistent blade pitch to match atmospheric conditions, and so the turbine can capture more power.
Weibull parameters for the case of no stability or turbulence filter as well as for each turbulence and stability class.
It is also important to mention the strong connection between turbulence and shear: high shear will eventually erode turbulence (Wharton and Lundquist, 2012a). Periods of high shear generally coincide with periods of low turbulence and vice versa. With low shear, the mean wind speed is more consistent over the height of the rotor disk. However, since we did not see significant differences in power curves for different shear regimes here, we cannot speculate further on this in this analysis. Finally, if veer occurs in the wind profile (as in Vanderwende and Lundquist, 2012, and Dörenkamper et al., 2014), which usually occurs only in stable or low turbulence atmospheric conditions, that veer will generally undermine power production as the turbine blades are not oriented perpendicular to the flow at all vertical levels.
AEP allows developers and operators to quantify the projected energy production of a turbine. To quantify the impact on AEP of these stability- and turbulence-driven differences in power curves, we use a Weibull distribution for wind speed and calculate AEP with no filter, as well as with TI and stability filters. These turbulence and stability filters for the AEP calculations can be further explained as AEP calculated using the power curves calculated from nacelle winds (Fig. 17a and b) as well as the power curves calculated from upwind tower winds (Fig. 17c and d). These power curves are used together with a sample wind distribution using Weibull distribution parameters based on wind speed data separated into each stability class (Table 4) as suggested by IEC 61400 12-1 (2015) for a site-specific AEP. For each of these filters, separate AEP calculations are made for each regime, weighted by the number of data points in that regime, and then added together to compare with the AEP calculated with no atmospheric filter. Note that although data are collected only during 2.5 months in the winter of 2012, AEP is calculated for an entire year to show values closer to a representative AEP value.
Results in Table 5 show a higher AEP when using no filter, followed by an AEP calculated with a TI filter and then a stability filter. The lower AEP calculated when separating by stability and turbulence regimes suggests that the AEP calculated using no filters may be overestimating the production, perhaps because the higher and lower extremes of the parameter ranges bias the averages in each bin.
AEP in megawatt-hours per year calculated for different atmospheric and turbulence regimes using a Weibull distribution with a scale and shape parameter associated with the corresponding wind speed distribution.
AEP results in Table 5 also show that the AEP calculated using nacelle winds underestimates the AEP when compared with an AEP calculated using upwind tower measurements. This underestimation of the nacelle anemometer-calculated AEP is true for both the AEP calculated for the entire data set as well as with each stability or turbulence filter and is likely because the nacelle anemometer underestimates the ambient wind speed due to flow interference of the rotor disk and nacelle.
When the AEP's low and high regimes are compared to the medium regimes of
their respective atmospheric parameters, the AEP for medium-TI periods is
higher than that for low-TI periods and for high-TI periods for both the
nacelle anemometer-calculated AEP and the tower-calculated AEP (Table 6).
Using low- and high-TI power curves results in an AEP smaller than that
calculated using the medium-TI power curve. These results are likely
obtained because the low-TI power curve loses production at lower wind
speeds and the high-TI power curve loses production around rated speed. When
using a stability filter, the AEP calculated with the low-
Using 2.5 months of data from upwind and nacelle-based instruments, we calculate power curves for different regimes of atmospheric stability and turbulence as well as AEP with and without these atmospheric filters. This work not only focuses on the idea of calculating different power curves for different atmospheric conditions for power performance testing but also highlights the differences in AEP that can emerge from the application of stability- or turbulence-dependent power curves. We also summarize extensive data quality-control methods, including two approaches for filtering out turbine underperformance or curtailments.
AEP in percentage calculated for different filter regimes using a Weibull distribution with a scale factor and a shape factor representative of the corresponding wind speed distribution. Medium regime is set at 100 % and low and high regimes are percentages compared to the medium regime. The highest value within each row is italicized.
Statistically significant differences emerge among power curves segregated
by TI and
After calculating an AEP for each regime and comparing the high and low
regimes with the medium regime, differences between AEP calculated using
different atmospheric filters are revealed. An AEP calculated with no
atmospheric or turbulence filter is higher than any AEP calculated with
these filters. In addition, the AEP calculated using a TI filter shows that
the AEP calculated with the medium TI regime is greater than the AEP
calculated with the low or high TI regimes. The AEP calculated with the
As a small percent difference in AEP leads to a large deviation in cost for both operators and manufacturers, calculating different power curves for different atmospheric conditions may not only be a practical approach but may also lower the financial risks for both operators and manufacturers.
As discussed by Rareshide et al. (2009), manufacturers increasingly filter out data that represent what they consider anomalous or extreme atmospheric conditions for power performance testing. The IEC-61400-12-1 (2015) standard calls for at least 180 h of data to be used in a power performance test. Consequently, if manufacturers filter out data based on higher TI values, for instance, this means that more data must be gathered to make up for the discarded data. We see this discarding of data as unnecessary and potentially more costly. We suggest that instead of discarding these data, different power curves be calculated for different conditions. This approach can allow for a more nuanced understanding of how a turbine operates in different atmospheric conditions, and may lead to a more accurate and reliable performance result and AEP calculation.
Data from the M5 tower are available for download at
The authors express appreciation to the Center for Research and Education in Wind for supporting this work, to Thomas Fischetti and Peter Gregg at GE Renewable Energy for their assistance in turbine data collection and interpretation, and to the reviewers of a previous version of this work. This work was supported by the US Department of Energy under contract no. DE-AC36-08GO28308 with the National Renewable Energy Laboratory. Funding for the work was provided by the DOE Office of Energy Efficiency and Renewable Energy, Wind and Water Power Technologies Office.
The US Government retains and the publisher, by accepting the article for publication, acknowledges that the US Government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this work, or allow others to do so, for US Government purposes.