Wind Turbine Clutter Mitigation for Weather Radar by Adaptive Spectrum Processing Fanxing Kong1, 2, Yan Zhang1, 2, Robert Palmer1, 3 1 Atmospheric Radar Research Center, University of Oklahoma 2 School of Electronic and Computer Engineering, University of Oklahoma 3 School of Meteorology, University of Oklahoma 120 David L. Boren Blvd, Norman, OK, 73072, USA
[email protected] Abstract— Wind Turbine Clutter (WTC) is the radar clutter caused by strong backscatter from large wind turbines within the radar vicinity. Due to the rotation of the rotor blades, the Doppler spectrum of WTC varies from scan to scan. This timevarying radar signature results in the failure of classic ground clutter filter techniques. The Adaptive Spectrum Processing (ASP) algorithm proposed in this paper explores the spatial and spectral characteristics of weather returns in weather radar, and adaptively generates band pass filters based on the moment estimates of uncontaminated weather in detected wind farm region. The actual NEXRAD weather radar data have been used to verify this algorithm. The experiment shows significant improvement on moment estimation in terms of both bias and variance.
[5]. The major scattering components of a wind turbine such as tower and nacelle are cautiously shaped to avoid specular reflections [6]. Partial treatment of wind turbine blades with Radar Absorbing Material (RAM) has also been studied [7]. Although it will be difficult to implement these techniques practically without compromising the structural integrity, it is inspiring to work around the WTC problem from both ends.
The impact of WTC on aviation radar was detailed in [2]. The essential difference in altitude between wind turbines and aircrafts has been used. Independent concurrent low and high beam channels are applied to mitigate WTC effect above the wind farm region [3]. The concept of gapfiller solution [4] has also been proposed to complement the radar system by installing supplemental radars within the wind farm area.
The negative effects of wind turbines to weather radar have also been reported [8, 9]. Intense WTC caused by operating wind turbines can bias moment estimates and may further lead to misidentification of thunderstorms, false estimates of precipitation accumulation and incorrect storm cell identification, etc. Unlike aviation radar, the targets of interest of weather radar are various forms of precipitation, such as rain, snow, hail, etc., which are often spatially inseparable from wind turbines for radar resolution at low elevation scans. Thus the success of mitigating WTC in aviation radar cannot be reproduced in weather radar. Traditional ground clutter filter techniques based on notch filters have also failed mitigating WTC due to its time varying Doppler spectrum. The difficulty to separate WTC and weather return signal in both spatial and Doppler frequency domain has prompted ideas to combine the information from both domains. The three dimensional spectra interpolation technique [10] was used to correct the moment estimation bias caused by WTC contamination. The range-Doppler signal processing method [11] exploited the difference in the rangeDoppler domain and alleviated Doppler spectral bins identified as WTC. The Adaptive Spectrum Processing (ASP) will further explore the spatial and spectral difference between weather signal and WTC, and adaptively filters the contaminated Doppler spectra while preserving valuable weather information. The aim of all WTC mitigation algorithms for weather radar, though, is to recover moment estimates with less bias and variance.
Another effort worth mentioning in WTC control is the stealth wind turbine technology. RCS reduction techniques are applied in customizing low-reflection wind turbines, which would ideally appear stealthy to radars. The reflectiveantenna-like active coating for rotor blades has been designed
The covariance method widely used in weather radar applications to estimate spectrum moments and the bias introduced by WTC will be studied in Section II. The assumptions and implementation of ASP will be detailed in Section III. In order to evaluate the performance of ASP
I.
INTRODUCTION
The increasing need for energy has urged the seeking for replacement of fossil fuels. Wind energy can be converted to electricity by wind turbine generators. The development of wind energy, not only helps diversify the energy portfolio, but could also become one major component of the power infrastructure in the near future. The Department of Energy has explored a modeled energy scenario in which wind provides 20% of U.S. electricity by 2030 [1]. As the wind industry thrives, many utility scale wind turbines have been installed across the country. However, these increasingly large rotating structures have been discovered to have potential Electromagnetic Interference (EMI) to current radar networks.
This work has been supported by NWC-Radar Operational Center (ROC) through grant # NA17RJ1227
978-1-4673-0658-4/12/$31.00 ©2012 IEEE
0471
quantitatively, the actual NEXRAD weather radar data will be used to demonstrate the improvement of moment estimation in Section IV. Section V will provide a summary of this research. II.
MOMENT ESTIMATES
In weather radar, moment estimates refer to the three lowest moments of the Power Spectrum Density (PSD), namely, reflectivity, mean radial velocity and spectrum width. The covariance calculation [12] has been widely used to estimate these moments for its simplicity, by estimating the Auto Correlation Function (ACF): ∑
(
ℓ) ( )
(1)
where V(m) is the IQ time series and ℓ is the lag. The ACF is known to be a Fourier transform pair with the PSD. Thus, for filters implemented in the frequency domain, the PSD is first estimated by means of the periodogram. And then the ACF estimates are obtained as the Inverse Fourier Transform (IFT) of processed PSD. The ASP will follow this same procedure to minimize potential changes to the signal processor.
Fig 2.
Time varying Doppler spectrum of WTC PSD Pre-filtered
20 0 PSD(dB)
(ℓ)
-20 -40
Weather&GC Weather&WT True Weather
-60 -80 -0.5
Reflectivity is proportional to the power estimate:
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.1
0.2
0.3
0.4
0.5
0.2
0.3
0.4
0.5
PSD Filtered 20
(2)
where N is the noise estimate. Under the assumption of Gaussian PSD, which is the case for most weather signal, the mean radial velocity can be derived from lag 1 estimate of the ACF: (1)
(3)
0 PSD(dB)
( ) ( )
0
Fig 1. shows a typical weather PSD estimate and its reconstructed Gaussian PSD. The PSD of WTC is more complicated and its time varying signature is shown in Fig 2. The zero Doppler frequency line can be attributed to the stationary components of the wind turbine such as tower and nacelle. The periodic flashes occur when one of the rotor blades turns to be perpendicular to the radar Line of Sight (LOS)[13]. 15
-20 -40
-80 -0.5
-0.2
-0.1
0
Weather&GC Weather&WT Weather
-0.4
-0.3
-0.2
-0.1 0 0.1 Normalized Dopper Frequency
(a) PSD Pre-filtered 20 0 -20
Weather&GC Weather&WT True Weather
-40 -60 -0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.1
0.2
0.3
0.4
0.5
0.2
0.3
0.4
0.5
PSD Filtered 20 0 -20
Weather&GC(filtered) Weather&WT(filtered) True Weather
-40 -60 -0.5
-0.4
-0.3
-0.2
-0.1
0
Reconstructed Gaussian PSD 20 0 -20
Weather&GC Weather&WT Weather
-40 -60 -0.5
10
-0.3
-60
PSD(dB)
(4)
| ( )|
-0.4
20
PSD(dB)
√
Weather&GC(filtered) Weather&WT(filtered) True Weather
Reconstructed Gaussian PSD
PSD(dB)
̂
-40
-80 -0.5
where λ is the radar wavelength, Ts is the Pulse Repetition Time (PRT) and (1) is the phase of lag 1 estimate of the ACF. The spectrum width can be estimated by: /
-20 -60
PSD(dB)
∑
(0)
-0.4
-0.3
-0.2
-0.1 0 0.1 Normalized Dopper Frequency
Power spectrum density (dB)
5
(b) 0
Fig 3. Two examples of moment estimate bias caused by WTC: (a) severe estimate bias; (b) Moderate estimate bias. GC is for Ground Clutter, WT for Wind Turbine. The Clutter to signal ratio is the same for both cases
weather signal PSD Gaussian model
-5
-10
-15
-20
Fig 1.
-25
-20
-15
-10
-5 0 5 Radial velocity (m/s)
10
15
20
25
Estimated PSD of weather signal and the reconstructed Gaussian model based PSD
978-1-4673-0658-4/12/$31.00 ©2012 IEEE
The time varying radar signature of WTC depends on many factors such as wind turbine blade position, yaw motion, rotation speed, radar scanning rate, etc. Therefore, the moment estimate bias can be significantly different from scan to scan. In Fig 3. , the two cases of (a) and (b) have been selected from
0472
two radar scans of the same wind turbine.. The weather IQ signal has been coherently added to simulatte the scenario of WTC contaminated weather signal. Groundd clutter IQ time series has also been mixed with the same w weather signal for comparison. A simplified GMAP (Gaussiann Model Adaptive Processing) [14] ground clutter filter has beeen applied to the PSD of both Weather & Wind turbine (WW W) and Weather & Ground clutter (WG). It is obvious that thee filtered PSD of WW is still noisy compared to the PSD of fiiltered WG. Thus, the moment estimates of WW are all biaseed for both scans. However, due to the time varying characterristic, the PSD of WTC changes significantly from scan too scan, resulting different levels of bias shown in Fig 3. III.
the same time, an isolated storm waas observed at the similar range northeast of the radar.
CESSING ADAPTIVE SPECTRUM PROC
Assuming the spatial extent of the weathher signal is larger than the wind farm, which is true for most cases, some radar resolution volumes are severely contaminated; some are moderately contaminated, and others havingg no wind turbines in them are uncontaminated. It is believed thhat the statistics of radial velocity estimates from the latter two cases can provide a prior knowledge about the weather in the ccontaminated area. The ASP mitigation scheme based on this inference can be implemented as follows.
Fig 4.
PPI of the uncalibrated reflectivitty estimate at KDDC radar site. GMAP has been implemented to remo ove nearby ground clutter
Fig 5.
PPI of moment estimates of wind farm, weather, weather & wind farm and the ASP reecovery
The detection algorithm [15] will first bbe implemented to confirm the coexistence of weather and WTC C, and identify the severely contaminated volumes. Starting froom the wind farm area, an omnidirectional growing window is used to iteratively expand the area until: the variation of thhe mean velocity estimate is less than 0.5 m/s and the mean abbsolute error of the normalized radial velocity distribution is lesss than 0.1: Δv Δ
N
∑
N
∑
v
N
∑
N
| ( )
v
(5)
( )|
(6)
where v is the radial velocity estimate from the nth volume detected as moderately contaminated orr uncontaminated (both referred as uncontaminated later on forr convenience), Ni is the number of uncontaminated volumes w within the area of ith iteration, pi(v) is the normalized discrette distribution of radial velocity estimate of ith iteration and M is the number of velocity bins. If Δv and Δ do not conveerge, the iteration stops when the expanded area is twice as llarge as the wind farm area. p(v) from the last iteration proviides an estimated likelihood of radial velocity and will be usedd to filter the PSD of those identified as contaminated: ( )
( )· ( )
(7)
The filtered PSD ( ) is transformed to AC CF by IFT and the moment estimates can be recovered througgh equation (2) to (4). The ASP is essentially a band pass filterr that adapts to the weather. IV.
N OF ASP PERFORMANCE EVALUATION
The data from the NEXRAD site “KDDC C” in Dodge City, Kansas are used to demonstrate the improveement of moment estimation by using ASP. The Gray County w wind farm located about 40 km southwest of the radar site is m marked in Fig 4. At
978-1-4673-0658-4/12/$31.00 ©2012 IEEE
The IQ signal of the wind farm is i coherently added to the weather signal gate by gate to simu ulate the Weather & Wind farm (W&W) mixture. The momen nt estimates of the wind farm only, weather only, W&W an nd the ASP recovery are shown in Fig 5. The Reflectivity esstimates of the wind farm area are high due to the large Radaar Cross Section (RCS) of wind turbines; radial velocity estim mates are random because the blade rotation is asynchronous from f one wind turbine to another; spectrum width are consisteently broad due to the fact that the radial velocity continuously increases from the root to
0473
the tip of each blade, all of which contribute to the PSD. The weather only estimates will be considered as the true values to evaluate the estimation bias and variance caused by WTC contamination. Compared to the true values, the contaminated (W&W) estimates within the wind farm area clearly have large bias and variance. However, after implementing the ASP, the moment estimates are mostly recovered as Fig 5. shows. TABLE I.
MOMENT ESTIMATTION IMPROVEMENTS
Bias/ variance
Moment estimates comparison
Reflectivity (dB)
Radial velocity (m/s)
Spectrum width (m/s)
Contaminated
13.02/16.89
-6.74/9.63
7.25/8.15
ASP recovered
7.38/11.5
0.59/2.38
0.37/1.00
TABLE I. compares the moment estimation bias and variance before and after the ASP process. The estimation performance has been significantly improved, especially for radial velocity and spectrum width. V.
CONCLUSIONS AND FUTURE WORK
WTC can severely affect weather radar moment estimation accuracy and degrade radar data quality. The ASP has been proposed to mitigate this negative effect by adaptive filtering based on the statistics of radial velocity estimates of uncontaminated weather. The experiment of ASP on actual weather radar data has shown promising improvements in reducing the bias and variance of moment estimations. The spatial correlation will be further explored in future works by incorporating a general model to address the different significance of uncontaminated estimates adjacent to each contaminated volume. REFERENCES [1] Dept.-of.-Energy, "20% Wind Energy by 2030, Increasing Wind Energy's Contribution to U.S. Electricity Supply," 2008. [2] G. J. Poupart, "Wind Farms Impact on Radar Aviation Interests - Final Report," QinetiQ 2003.
978-1-4673-0658-4/12/$31.00 ©2012 IEEE
[3] L. Sergey, O. Hubbard, Z. Ding, H. Ghadaki, J. Wang, and T. Ponsford, "Advanced Mitigating Techniques to Remove the Effects of Wind Turbines and Wind Farms on Primary Survellance Radars," presented at the IEEE Radar Conference, Rome, 2008. [4] E. Aarholt and C. A. Jackson, "Wind Farm Gapfiller Concept Solution," presented at the ERAD, Paris, France, 2010. [5] B. Chambers, K. L. Ford, and A. Tennant, "Active coating for wind turbine blades," in Antennas and Propagation Conference, LAPC, Loughborough, 2008, pp. 137-140. [6] J. Pinto and et al., "Stealth technology for wind turbines," IET Radar Sonar & Navigation, vol. 4, p. 126, 2010. [7] L. Rashid and A. Brown, "Partial Treatment of Wind Turbine Blades with Radar Absorbing Material (RAM) for RCS Reduction," presented at the 4th EuCAP, Barcelona, 2010. [8] R. J. Vogt, J. R. Reed, T. D. Crum, J. T. Snow, R. Palmer, B. Isom, and D. W. Burgess, "Impacts of wind farms on WSR-88D operations and policy considerations," presented at the 23rd IIPS, San Antonio, Tx, 2007. [9] R. J. Vogt, T. D. Crum, J. T. Snow, R. Palmer, B. Isom, D. W. Burgess, and M. S. Paese, "An Update on Policy Considerations of Wind Farm Impacts on WSR-88D Operations," presented at the 24 IIPS, AMS Annual Meeting, New Orleans, LA, 2008. [10] B. M. Isom, R. D. Palmer, G. S. Secrest, R. D. Rhoton, D. Saxion, T. L. Allmon, J. Reed, T. Crum, and R. Vogt, "Detailed Observations of Wind Turbine Clutter with Scanning Weather Radars," Journal of Atmospheric and Oceanic Technology, vol. 26, pp. 894-910, 2009. [11] F. Nai, R. D. Palmer, and S. M. Torres, "Range-Doppler Domain Signal Processing to Mitigate Wind Turbine Clutter," presented at the IEEE Radar Conference, Kansas City, MI, 2011. [12] R. J. Doviak and D. S. Zrnić, Doppler Radar and Weather Observations, 2nd ed. Mineola, New York: Dover Publications, INC., 2006. [13] F. Kong, Y. Zhang, R. Palmer, and Y. Bai, "Wind Turbine Radar Signature Characterization by Laboratory Measurements," presented at the IEEE Radar Conference, Kansas City, MI, 2011. [14] A. D. Siggia and R. E. Passarelli, "Gaussian model adaptive processing (GMAP) for improved ground clutter cancellation and moment calculation," presented at the ERAD, Gotland, Sweden, 2004. [15] K. Hood, S. Torres, and R. Palmer, "Automatic Detection of Wind Turbine Clutter for Weather Radars," Journal of Atmospheric and Oceanic Technology, vol. 27, pp. 1868-1880, Nov 2010.
0474
