Luke Dosiek­­­­­­­­­­­­­­­­­­1 and Pragasen Pillay2AbstractMethodologyResultsModeling of a Stand Alone Horizontal Axis WindTurbine Luke Dosiek1 and Pragasen Pillay2Department of Electrical and Computer EngineeringAbstractThis paper addresses the design of a horizontal axis wind turbine simulation usingMATLAB/SIMULINK. A novel approach to simulating the wind over a long period of time with aresolution of one Hertz is also discussed in this paper, as the wind turbine simulator will take a simulatedwind profile as the input and convert it to an output of electrical power. The results of the work done thusfar are compared to actual data to verify the validity of the Wind Generator (WG) proposed here.MethodologyIn order to accurately reproduce wind speeds over a time scale longer than about 10 minuteswhile retaining an accurate resolution of about 1 second, both the turbulence and the long term windcomponents need to be modeled. Unfortunately, there is no single model that can accurately cover bothshort term fluctuations and long term changes (Nichita 2002). As stated by Nichita (2002), a method formodeling long term wind called the Van der Hoven model is accurate for long range simulation butprovides an inaccurate representation of turbulence. On the other hand, Nichita (2002) states thatmodeling turbulence can be achieved using the Von Karmen power spectrum, but it fails to provideaccurate results once time horizons on the scale of tens of minutes are reached. Another technique ofmodeling turbulence alone is the Shinozuka method (Jeffries 1991). To get around the problem of modeling long term wind with a high resolution, Nichita used amethod where he combined both the Van der Hoven and Von Karmen models. The Van der Hoven modelis used to generate a new wind speed data point every 10 minutes or so. This Van der Hoven point is thentaken as the average wind speed over the 10 minute interval that is filled with Von Karmen generatedwind data. Since the Von Karmen psd has average velocity as a variable, this moving average generatedby the Van der Hoven equation gives both accurate short- and long-term wind modeling. The only1. Class of 2004, Electrical Engineering, Clarkson University, Honors Program, Oral Presentation2. Professor, Department of Electrical and Computer Engineering, Clarkson Universityproblem with this is the heavy computational time involved as described by Nichita (2002), but this canbe alleviated with faster computers.For the wind model used in this paper, the Shinozuka model was chosen as the method forsimulating turbulence. For the problem of generating long term wind data, it is hypothesized that theShinozuka method could in fact be used for both the turbulence and the long term fluctuations based onthe premise that short term fluctuations is a relative idea. If one was looking at several hours’ worth ofwind data, then a scale of seconds could be considered as short term fluctuations; this of course is thetraditional view of what turbulence is. If, however, one was looking at several days’ worth of wind data,then the smallest time scale of concern may be on the scale of minutes or even tens of minutes. Thislower resolution might be acceptable since the entire set of wind data covers several days, and literallythousands of these ten minute intervals. If this is the case, then small samples of these once-every-ten-minute points would be turbulence in the presence of days of data. If true, then the Shinozuka turbulencemethod will work to describe small numbers of these points. This concept is similar to that of fractals in nature. For example, a pine tree, when examinedclosely, has needles as its smallest significant part, with cells making up those needles. When examinedfrom afar, the smaller branches are the smallest significant part, with the needles making up the branches.When examined from even farther, the trees are the small parts making up the forest. Which parts aresmall and which are large is all relative to the size of the system being examined; this is the very conceptbehind using the Shinozuka turbulence method for both the long and short term models.The task of combining the models into a coherent long term wind profile of high resolution isbased on the variable average ideas outlined in Nichita’s paper (2002). The Shinozuka method is used tocreate a data point once every 10 minutes for any number of hours. This data point is then used to fillevery 10 minute interval with second by second data points, again using the Shinozuka method. Thisresults in hours of data points at a resolution of one second per point with long term average andturbulence information that is specified by the user. The method results in accurate wind modeling thatholds true if the user is looking at hours of data or at a few seconds.ResultsAs hypothesized above, turbulence depends on the size of the system. For a second-by-secondsimulation of wind over a long period of time, there needs to be two separate time horizons at work,similar to the method used by Nichita (2002). Since the wind engineering standard for measuringturbulence intensity is 10 minutes (Manwell 2002), that was chosen as the larger time horizon. Thesmaller, is of course, one second. This means that if, for example, 5 hours of data were to be generatedthe simulator would first generate a data point every 10 minutes using the Shinozuka formula and the Von1. Class of 2004, Electrical Engineering, Clarkson University, Honors Program, Oral Presentation2. Professor, Department of Electrical and Computer Engineering, Clarkson UniversityKarmen psd, creating 30 points that will become the “moving average.” These 30 points will adhere tothe statistical properties as governed by the inputs which are long term mean, average turbulenceintensity, integral length scale of both large- and small-scale turbulence, and the number of hours tosimulate. The next level of the program takes each of the 30 data points and creates 10 minutes worthsecond-by-second data, or 600 points for each of the 30 averages. These points are created using theShinozuka formula with Von Karmen psd and will adhere to the behavior given by the inputted turbulenceintensity and length scale, but each set of 600 will have its own average pulled from its correspondingpoint in