Sunday, May 26, 2013

Biglow Canyon Wind Farm Analysis


As part of my work for the Wind Energy and Turbulence Lab I was given a large dataset from the Biglow Canyon Wind Farm in Eastern Oregon which covered ten minute average wind speed and direction for 217 some turbines and meteorological masts for all of 2011.



We visited the Biglow Canyon Wind Farm too! Pretty darn pretty!
 It was a pretty large dataset, and worked out to be 11.6 million data points, and I learned firsthand why programs like MATLAB were developed in the first place. A good chunk of time was taken up by getting the data simply organized, dealing with blank values (not the proper NaN), and teaching myself how to use linear indexing.

Originally to sort the data I wrote some loop heavy code that by my estimate (extrapolated tic toc) would take 6,000 some minutes to run... A few modifications later I got down to 2,000 minutes. Since this was still unacceptable, I set out to redo the code in a more linear index way, and got the run time down to 2.6 minutes! I was pretty proud of that.

Once I had the data sorted, I thought that I could get down to actually analyzing it, but I quickly realized that I had to learn some visualization techniques first to make sure the data actually made sense in its raw form of speed and direction. Some simple graphs, wind bins, and histograms showed a lot, but since the data was time and location based, I needed something more robust. I loaded up a USGS terrain elevation dataset, combined it with the tower coordinates, created a windspeed mesh colormap overlay, and started making movies MATLAB movies of the results. I'm not linking any movies, but here is a still frame.





Windspeed in m/s on 2011-04-01 at 18:10





What you can see immediately from the above image is that you have some very different wind speeds and directions in a relatively small area. The terrain is definitely a factor, with all sorts of pressure gradients being imposed upon the site. There is also a clear diurnal cycle present in the farm's data. The rest of the analysis aside, this was a great learning experience for me, and I never would have thought that I would get so much pleasure from spending days reading MATLAB documentation


No comments: