International Scholarly Research Notices

Volume 2016 (2016), Article ID 5790464, 12 pages

http://dx.doi.org/10.1155/2016/5790464

## The Impact of Variable Wind Shear Coefficients on Risk Reduction of Wind Energy Projects

Engineering Department, Dalhousie University, Faculty of Agriculture, Truro, NS, Canada B2N 5E3

Received 15 June 2016; Accepted 4 October 2016

Academic Editor: Shafiqur Rehman

Copyright © 2016 Kenneth W. Corscadden et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

#### Abstract

Estimation of wind speed at proposed hub heights is typically achieved using a wind shear exponent or wind shear coefficient (WSC), variation in wind speed as a function of height. The WSC is subject to temporal variation at low and high frequencies, ranging from diurnal and seasonal variations to disturbance caused by weather patterns; however, in many cases, it is assumed that the WSC remains constant. This assumption creates significant error in resource assessment, increasing uncertainty in projects and potentially significantly impacting the ability to control gird connected wind generators. This paper contributes to the body of knowledge relating to the evaluation and assessment of wind speed, with particular emphasis on the development of techniques to improve the accuracy of estimated wind speed above measurement height. It presents an evaluation of the use of a variable wind shear coefficient methodology based on a distribution of wind shear coefficients which have been implemented in real time. The results indicate that a VWSC provides a more accurate estimate of wind at hub height, ranging from 41% to 4% reduction in root mean squared error (RMSE) between predicted and actual wind speeds when using a variable wind shear coefficient at heights ranging from 33% to 100% above the highest actual wind measurement.

#### 1. Introduction

Wind has been a major contributor to renewable energy sources, with large wind, both onshore and offshore, dominating the energy mix in many countries [1]. This trend will continue with wind expected to generate 12% of global electricity by 2020 and 20% by 2030. Much of this development will be in large scale wind; however, small wind is becoming a major player for both grid-tied distributed power generation and off-grid generation [2]. Accurate assessment of the wind resource is crucial in order to secure funding for projects, with many funding agencies requiring wind measurements at two-thirds of the proposed hub height; however, with increasing turbine sizes this is becoming more difficult, even with 60 m meteorological towers. Suitable forecasting models are an essential component of the assessment of proposed wind projects and the subsequent control and integration into grid-tied systems. Forecasting models and control technologies have received significant attention, evidenced by recent comprehensive reviews; Foley et al. [3] identify the significant advances in forecasting methods, encompassing statistical and physical models over varying time horizons and Mahela and Shaik [4] and Jain et al. [5] provide an equally detailed and enlightened review of the control strategies and prediction methods used to integrate wind into transmission and distribution networks. The necessity to accurately forecast wind speed is well documented and is generally based upon the Weibull probability density function, which is estimated from time series wind data typically obtained from a meteorological tower over extended testing periods [6]. The Weibull probability density function is a two-parameter function which is used to produce a wind speed profile for a particular site. Two parameters of shape (*k*) which is dimensionless and scale (*c*) in m/s are sufficient to characterize the Weibull function and are estimated from time series wind data typically obtained from a meteorological tower over extended testing periods. Of several recent regional studies testing the Weibull parameters, Weisser [7] analyzed two years of meteorological data in Grenada, West Indies, demonstrating the value of long term wind data to account for seasonal wind speed variations and the need for capturing the variation of wind speeds over the course of a day to account for diurnal changes but further highlight that typically two years of wind speed data is insufficient. Similarly, both Zhou et al. [8] and Lun and Lam [9] analyzed wind speed data, 1-year and 30-year data sets, respectively, both highlighting the value of the Weibull distribution, the need for long term data sets, and the need for wind-related assessments and evaluations. Statistical analyses of estimating the Weibull distribution of wind speed time series data are analyzed using various methods such as maximum likelihood, modified maximum likelihood, least squares, chi square, regression, graphical method, and methods of moments [10–12], with Seguro and Lambert [13] suggesting “maximum likelihood” as the recommended method, while Genc et al. [11] report the least squares method as producing better results for large sample sizes, while chi square is reported as providing the best overall fit [6]. The resulting Weibull distribution is then used to provide an estimate of potential power generation capacity, the basis for economic evaluation of a wind energy project for a particular site. Meteorological towers provide characteristics of wind speed near the earth’s surface and it is therefore necessary to extrapolate wind speed to higher levels in the planetary boundary layer, particularly with higher turbine hub heights. This is achieved using a wind shear exponent or wind shear coefficient (WSC), variation in wind speed as a function of height, with two mathematical models of power law (PL) and Logarithmic Law (LL) used for extrapolation [14]. PL extrapolation is the most commonly utilized method for predicting wind speeds at a higher height than what is measured and historically uses a default exponent of 1/7th (0.142); however, research indicates that this value is neither stable on a diurnal, weekly, or seasonal basis, nor accurate for all sites due to varying surface roughness factors, atmospheric influences, and measurement heights [15, 16]. Firtin et al. [17], in a review of available WRAP (Winds Resource Analysis Program) data, found that 91.9% of wind shear coefficients were above 0.14, a clear indication that a default WSC may in some cases result in under- or overestimation of wind speeds and subsequently turbine Actual Energy Output (AEO). These findings were further supported by Rehman et al. [18] and Schwartz and Elliot [19], who identified a more realistic range of WSC of between 0.15 and 0.25. With the recognized issues of using a default 1/7th PL exponent for extrapolation, researchers have sought to modify the standard power law methodology and other extrapolation methods to better predict wind speeds at higher heights; however, in many cases, a fixed WSC is used based on long term average time series wind data. Farrugia [20] reported that while PL and LL were generally accepted for extrapolation up to heights of 100 m, significant variation in WSC occurred based on the month and time of day, a result that was also reported in a substantial study conducted by Bailey [21]. Ray et al. [22] conclude that there is little difference between the performance of the PL and LL models but noted greater variation in WSC with more complex terrain. Other researchers have investigated fundamental factors that impact wind shear, which includes its impact on turbine structures [23] and composite turbine blades [24], atmospheric stability, upwind terrain, surface roughness, sky condition (which contributes to night time radiative cooling), temperature, air-pressure, and humidity, in daily, seasonal, and directional trends. A common assumption however, that the WSC remains constant, has been identified by a number of authors as a contributing factor to increasing uncertainty in wind speed extrapolation. Lubitz [25] investigated the level of uncertainty associated with the PL model, concluding that the mean absolute error of extrapolated wind speed increased with increasing height above the measured wind speed and Irwin [26] proposed that variations in the power law exponent were impacted mostly by surface roughness and atmospheric stability, a factor that has more impact closer to the earth’s surface. Fox [27] used friction velocity instead of a fixed WSC and applied this to utility scale turbines based on heights up to 150 m, claiming greater accuracy in wind speed extrapolation. Mikhail [28] used an alternative method referred to as a modified power law expression, claiming better accuracy. The degree of uncertainty in wind speed extrapolation has a much greater influence on energy production estimates [29]. Firtin et al. [17] investigated the impact of wind shear coefficients on electrical energy generation suggesting up to 49.6% error in energy production estimates using a PL extrapolation, with a fixed WSC. Several researchers such as Altunkaynak et al. [30] and Gualtieri and Secci [31] have attempted to address the uncertainty of WSC using a distribution, particularly the Weibull probability distribution to incorporate the temporal variation in WSC, using tower data. Şen et al. [32] consider an additional approach and combine the Weibull probability distribution with perturbation theory (which includes the standard deviations and covariance of wind speed at different elevations) to produce an extended PL, again incorporating time variations. Ðurišić and Mikulović [33] utilized a method of least squares (LES) and varied the shear exponent on a time-varying basis as a method of improving upon the traditional PL methodology. This methodology removes the concept of surface roughness and takes in to account the significant variation in WSC found on a diurnal and seasonal basis. Smedman-Hogstrom and Hogstrom, [34] developed a modified PL empirical model that incorporates the surface roughness in to the WSC calculation and Panofsky and Dutton developed a modified PL semiempirical model that estimates WSC as a function of surface roughness and stability [31]. Haque et al. [35] propose a new strategy for using computing models to predict short term wind speed, a method that has potential for both shear prediction and control systems. Significant advances have been made in the prediction of wind speed and there is evidence that such methods are now beginning to be considered for wind shear calculations as shown by Sintra et al. [36]. The evaluation of wind speed and wind shear is also inherently linked to control systems as such information is a prerequisite in the development of predictive control methodologies. The multivariable temporal variations in wind shear could be addressed using control theory and represented as a multivariable disturbance model, which has been demonstrated in other industries to result in improved controller performance [37]. This along and hybrid forecasting are emerging research opportunities [38]. Remote sensing of wind speed data using SoDAR is reported to provide wind speed measurements that correlate with anemometer data [39]. It has been proposed by Jeannotte [40] that SoDAR technology may have some limitations when used in complex terrain; however, there is merit in the contribution that SoDAR can make in addressing uncertainty associated with wind shear estimation, especially when applied to noncomplex terrain. Several researchers including Altunkaynak et al. [30] and Farrugia [20] have attempted to address the uncertainty of WSC using a distribution, particularly the Weibull probability distribution, to incorporate the temporal variation in WSC, using tower data. Şen et al. [32] consider an additional approach and combine the Weibull probability distribution with perturbation theory (which includes the standard deviations and covariance of wind speed at different elevations) to produce and extended power law, again incorporating time variations. A number of authors [35, 38, 41] have used advanced models to improve the characterization of wind speed and wind power estimates including neurofuzzy inference systems [42] and Markov Chain Models [43]. Bilgili et al. [44] utilized ANN to predict mean monthly wind speed at a target site using local reference wind tower data with some success but concluded that there is a need to ensure that reference wind towers must have a reasonable correlation factor (0.59). Fadare [45], Lee et al. [46], and Carolin Mabel and Fernandez [47] have all applied ANN models to specific geographic areas, with the authors highlighting the success of the ANN models in achieving reasonable accuracy in predicting wind speeds and the subsequent power output of wind turbines. However, while the use of ANN as a valuable tool is not disputed, Li and Shi [48] state that due to the number of different ANN models available and developed there are currently a number of factors that will influence forecasting accuracy including model structures, learning rates, and variation in required inputs. These models provide estimation at a single height, typically hub height using time series analysis, linear, nonlinear, and Artificial Neural Network (ANN) models, and subsequent hybrids, but there is little evidence of the same methods being applied to wind shear estimation. The estimation of WSC using a single fixed variable is an oversimplified approach creating challenges and increasing uncertainty in power production estimates for wind power projects. One must question the impact of such oversimplification and is the motivation behind the research presented in this paper. WSC research is possible using remote sensing, in this case, SoDAR technology to evaluate the potential of using a wind shear distribution (WSD), instead of a fixed WSC. This paper contributes to the research of wind shear estimation by presenting the results of an applied regional project conducted in the province of Nova Scotia, Canada. This research demonstrates the improvement in accuracy of wind speed estimation achieved using a WSD, validated using power prediction estimates for a commercial turbine, based on 60 m wind data with proposed hub heights of 80 m, 100 m, and 120 m. The paper examines the reduction in uncertainty and error obtained by using a variable of WSC creating a distribution instead of a fixed WSC to evaluate the accuracy of wind speed predictions at heights above measurement height.

#### 2. Materials and Methods

Wind speed data was collected at eighteen different sites in Nova Scotia using a Vaisala Triton® Sonic Wind Profiler and SoDAR (Sonic Detection And Ranging)**,** which uses the Doppler effect to reliably and accurately determine wind speed, wind direction, quality, and other operational parameters at heights ranging from 40 m to 200 m. The data collected by the SoDAR is sent via satellite to a “skyserve” website every 10 minutes, which is then downloaded to an excel file. The SoDAR, Figure 1, was mounted on a mission trailer for easy deployment and transportation between sites, the specification of which is listed in Table 1.